A Reasonable Way To Proceed
A Reasonable Way To Proceed
A Reasonable Way To Proceed
ss
aysi
nho
nor
ofJ
i
mMc
Cl
os
key
A reasonable way to proceed
Essays in honor of Jim McCloskey
Jason Merchant, Line Mikkelsen, Deniz Rudin, and Kelsey Sasaki (eds.)
Santa Cruz—Berkeley—Chicago
March 2018
Typesetting by Kelsey Kraus
Illustration by Kim Bennett
Cover design by Nicholas Merchant
Cover photo by Catherine Cronin
This book is an open access publication made possible by the University of Cali-
fornia’s eScholarship Repository. Its contents are the intellectual property of the
authors, licensed under a Creative Commons Attribution-ShareAlike 4.0 Inter-
national License.
Table of Contents
Dedication 1
Foreword 3
Cathal Goan
1
Foreword
Jim McCloskey’s contributions to our field—theoretical syntax in particular—
are myriad. He has made major advances in our understanding of resumption,
ellipsis, existentials, clefts, subjecthood, relative clauses, verb-initiality, and the
syntax-prosody interface, to name only a few. Though his work has been pre-
dominantly syntactic, it has consistently displayed an uncommonly sophisticated
engagement with semantics and with phonology.
The above alone would comprise a noteworthy career for any theoretician. This
makes Jim’s career all the more remarkable, because the above is only half of the
picture. Jim’s work has not only grappled deeply and productively with theo-
retical issues of relevance to linguistic theory writ large, it has done so through
an equally deep attention to the fine-grained messiness of the empirical terrain,
and a sensitivity to dialectal variation at an almost microscopic level. The theory
is put to work to help make sense of complicated data, and that data is used to
illuminate the larger theory. Jim’s lifelong engagement with Irish and Irish En-
glish has provided a model for how to combine rigorous theoretical work with
careful, community-involved fieldwork, a pairing that has come to be emblem-
atic of the Santa Cruz style of syntax. His commitment to working on and in the
Irish language reaches beyond theoretical linguistics, into the vitally important
realms of language documentation, revitalization, and advocacy. In this respect
as well, Jim has been an inspiration to several generations of Santa Cruz linguists.
In essentially all of Jim’s work, there is a masterful balance between the global
view and the (hyper)local. This skill is one of the marks of an expert story-
teller. Indeed, Jim’s characteristic habits of thought, as reflected in his writing,
his teaching, and even his comments on advisees’ drafts, are narrative. He is at-
tentive to the story of how the field arrived at its current state of understanding,
and to how his contributions advance that narrative, always trying to push our
understanding forward rather than trying to have the final say. The title of this
volume, a favorite Jim-ism of ours, is a perfect encapsulation of this narrative
temperament. This quality is reflected in Jim’s engaging and adventurous papers
and lectures—always enlivened by Jim’s friendly-yet-incisive brand of humor—
which have an uncanny ability to "catch the mind off guard and blow it open."
3
On a personal level, Jim is unfailingly affable, warm, and actively engaged in mak-
ing our community at Santa Cruz such a pleasant place—a community in the
truest sense, linked by camaraderie and affection, not just professional courtesy.
He is a widely beloved figure in a field that, as Jim famously pointed out, some-
times seems to “thrive on a . . . diet of anger, polemic, and personal abuse.” Jim’s
presence in the field has long been a welcome counterweight to this tendency.
Jason Merchant
Line Mikkelsen
Deniz Rudin
Kelsey Sasaki
4
On the exponence of gender in the Irish DP*
Paolo Acquaviva
University College Dublin
7
is the marking of a configuration of gender agreement inside DP. Its realization
is subject to a number of constraints, particularly complex in the case of the
complement of a lexical noun. It is this relation between mutation and gender
agreement that is subject to a significant weakening; when gender has a different
realization, its systematic morphological realizations are stable. An empirically
successful theory must account for this state of affairs.
8
1.3 Grammatical and semantic gender
Discussion of gender in Irish is typically bound up with the issue of “semantic”
gender assignment in pronominal anaphora: while a pronoun referring to a pre-
ceding DP generally takes the same gender value, this can be overridden. In (1),
the DP cailín deas is grammatically masculine (as shown by the lack of initial le-
nition on the adjective deas),1 but it is resumed by a feminine pronoun í; and in
(2), the noun ainm ‘name’ is preceded by the feminine variant of the article an,
which does not insert a -t- in front of the initial vowel, yet the pronoun resuming
it is the masculine é (both examples from Ó Sé 2000: 87):2
(1) Cailín deas is ea í
girl nice cop pred 3sg.f
She is a nice girl.
(2) Is é an ainm a bhí uirthi ná Móire
cop 3sg.m the name prt be.pst on.3sg.f prt Móire
Her name was Móire.
This phenomenon is undoubtedly significant, but it concerns the choice of a gen-
der value in anaphora, as opposed to its realization in DP. Like the well-known
parallel phenomena in languages like German (das [neuter] Mädchen ... sie [fem.],
‘the girl ... she’), it shows that a value arising from the interpretation of DP can
override the value determined (syntax-internally) by agreement between D and
N, when it comes to determining the gender value imposed by a DP to a referen-
tially dependent pronoun. However, this does not mean that the gender of D is
chosen on the basis of meaning, rather than on the basis of grammatical agree-
ment with N. It is the value of DP as a whole, as agreement controller, which in
such cases is determined semantically; but D itself is regularly realized accord-
ing to the value determined by N: das Mädchen, never *die [fem] Mädchen. This is
not a particularly insightful or novel observation, but it should be remembered
in order to avoid taking facts like (1)-(2) as evidence that grammatical gender
is superseded by semantic gender. As the in-depth discussion of Lindau (2016)
makes clear, it is better to think that the featural makeup on D in such cases has
the full complement of values, those determined by morphological agreement
(“concord”) and those determined by semantics (“index”; Lindau 2016: 978 use-
fully explains it as ‘the grammaticalized content of the semantic denotation of
the noun’). Since cases like (1)-(2) are quite frequently reported for all varieties of
Irish, the conclusion to draw is that pronominal anaphora often disregards the
morphological marking on D but this marking is still there.
It bears stressing that the pattern shown in (1)-(2) is not evidence that a gender
value is attributed on the basis of “natural” gender, as if it was an objective cate-
gorization opposed to a language-internal one. This is because nouns for inani-
1 Cailín is grammatically regularly masculine, despite its meaning, because it is formed with -ín,
which is here a noun-forming suffix and not a noun-modifying diminutive one; see 4.1 below).
2 I use the following abbreviations: cop = copula, f = feminine, gen = genitive, M = masculine,
9
mates too can enforce or favour feminine agreement with pronominal anaphora,
as (2) shows with the noun ainm. The personalization of inanimate nouns like bád
‘boat’, carr ‘car’, geansaí ‘jersey’, and even of recent borrowings like rólar ‘roller’,
all referred to by feminine pronouns, was one of the subcategories identified by
Ó Siadhail (1984: 176) for inanimate nouns, alongside non-personalized cases like
áit ‘place’ and uair ‘hour’ which are resumed by masculine pronouns despite be-
ing grammatical feminine. But there is nothing “natural” in associating feminine
with certain inanimate referents: semantics does play a role in gender assign-
ment, but it is not extra-linguistic properties of the referents that determine a
value.
Having clarified the role of the semantic motivation of gender value in pronom-
inal anaphora, I will now focus on the grammatical, language-internal aspects of
this category; in particular, on the determination and expression of gender value
inside DP.
10
Nominative/accusative case Genitive case
an bata ‘the stick’ an bhata ‘of the stick’
The contextual changes triggered by the articles interact with the inflectional
variability of lexical stems, so that for example seoid becomes seoide in the geni-
tive. Adjectives have their own inflectional behaviour: some are invariable, oth-
ers change in form depending on number, and some depending on gender too.
A more general, and for us more relevant, exponent of gender for adjectives is
the lenition of the initial consonant, which accompanies feminine singular at-
tributive adjectives. Given the order article – noun – adjective, the lenition on
the adjective can be seen both as a marker of DP-internal feminine agreement,
and as an effect of the linearly preceding noun. Some illustrations are provided
in (4):
11
• misalignment of the exponents of feminine gender (lenition and realized
morphology)
The first heading does not refer to cases where an opposition in gender ex-
presses two semantically distinct readings, distinguishing what can be legiti-
mately viewed as two homonymous, inflectionally identical nouns (like mám,
masculine as mountain pass, feminine as handful, or ráth, masculine as earthen
rampart, feminine as shoal of fish; Ó Curnáin 2007: 506 identifies such a semantic
split between the masculine and the feminine use of méid, respectively as amount
and size). The oscillation consists instead in nouns that have the same mean-
ing but can occur with either gender value. This sort of anomaly is prominent
enough to have been recorded even in descriptions of the (artificially codified)
standard. The official standard originally issued in 1958 (an Caighdeán Oifigiúil,
CO), followed by the official 1978 dictionary by Ó Dónaill, mentions distinct gen-
itive forms of the masculine talamh earth, namely the masculine talaimh and the
feminine talún. What is an isolated exception in the standard, however, takes dif-
ferent proportions in the dialects, which alone reflect native states of linguistic
competence. Ó Siadhail (1989: 147) lists the following examples of nouns that are
masculine in some dialects but feminine in others:
More specific descriptions reveal more cases where the gender value in one di-
alect contrasts with the value of others, or the standard (typically in the sense of
all other varieties). In his overview of Munster varieties, Ua Suilleabháin (1994)
discusses among other discrepancies the feminine ainm noun, name, dlí law, guth
voice, srian bridle. For Ulster, Hughes (1994: 629) reports that tír land, country is
usually feminine but fails to lenite the following adjective in the fixed phrase tír
mór mainland.
More interesting are the cases where both gender values are reported as cur-
rent, in the same dialectal area: Ua Súilleabháin mentions iúna wonder, brí strength
(fem. in Corca Dhuibhne, masc. or fem. in Muskerry), and the series of loan-
words tae tea, siúcra sugar, and béile meal; in addition, he mentions loch lake,
which Ó Siadhail (1989: 147) calls a particularly good example, where the genitive
singular fluctuates from dialect to dialect between the feminine na locha, na loiche
12
and the masculine an locha with no discernible pattern. As for the Connacht di-
alect, the classic description of the variety of Cois Fharraige by de Bhaldraithe
(1977) states on its very first page that a few nouns have two genders, listing the
following examples: aistir journey, cleachtadh habit, practice, deatach smoke, éis-
teacht ‘hearing’, eolas knowledge, fad length, leabhar book, méid amount, size,
oiread amount, talamh earth. The overview of Connacht dialects by Ó hUiginn
(1994: 562) also discusses am time, and adds the observation like some of these
words, they take the masculine form of the article, but often the feminine form
of the adjective after them, for example an cleachtadh but cleachtadh mhaith, an
t-eolas but eolas mhaith [translation PA]. Ó Siadhail (1984: 174) noticed that this ir-
regular feminine pattern of lenition triggered by a usually masculine noun takes
place in the absence of an article:
As we will see directly, however, even this subregularity proves less than abso-
lute when the analysis is more detailed. The more recent and vastly more de-
tailed (four volumes) single-dialect description by Ó Curnáin (2007) features far
longer lists, with five pages devoted to nouns that are feminine and three pages
for nouns that are masculine in the Connacht dialect described, but have the op-
posite value elsewhere or in closely related dialects (vol. I, 497-502, 503-505). In
several of the nouns listed, masculine or feminine are in fact both recorded.
The next two headings are more important. Ó Curnáin (2007) explicitly dis-
cusses the pattern of gender dependent on case (p. 505), where a noun has the
morphology of one value in the nominative/accusative case but that of the op-
posite value in the genitive. For instance, leabhar book is regularly masculine
in the genitive, but which can be masculine or also feminine in the nomina-
tive/accusative (the example produced is an leabhar dhubh sin that black book.
Ó Curnáin also mentions talamh earth, trinse trench, and especially in fixed col-
locations am time, scéal story, and Gleann valley, the last in a placename). Even
more striking is the simultaneous presence of masculine and feminine morphol-
ogy side by side. Such is the case of cleachtadh, which takes the masculine article,
i.e. an cleachtadh, but like a feminine noun, lenites a following adjective. This
is very common in the phrase without the article cleachtadh mhaith (Ó Curnáin
2007: 506). The same happens with iomaire ridge, in the phrase an t-iomaire mhór
the big ridge (Ó Curnáin 2007: 504; masculine article, feminine lenition on the
adjective mór; this represents a counterexample to the generalization tentatively
put forward in Ó Siadhail 1984).
What matters, then, is not so much that a few or many words vary in gender
assignment across the dialects (in itself, a natural state of affairs), but that the
exponence of gender value is often irregular and sometimes inconsistent, to the
point of making it problematic whether a noun really has a unique gender value
13
in a single, quite homogeneous dialectal variety, or indeed in individual com-
petences. Ó Curnáin (2007: 500-501) is explicit on this point when noting, for
instance, that in query both sábh maith and sábh mhaith were offered [good saw,
masculine and feminine respectivelyPA], or that a speaker is hesitant between
an iascach and an t-iascach [fishing, fishery] and actually self-corrects. This last
example should be related to the masculine an t- in the phrase an t-iomaire mhór,
mentioned a few lines above. Together, they show that the instability does not
concern just the juncture between noun and adjective, but also that between ar-
ticle and noun. It is interesting that most cases of exceptional, irregular feminine
lenition by a noun are attested when the noun is not preceded by an article, as
noted by Ó Siadhail; but this is only a tendency and does not seem to be man-
dated by some grammatical principle.
14
Some caution is needed before drawing conclusions from Frendas figures, of
course, and not only because the relevant potential agreement targets are a small
number (the results are statistically relevant within the corpus, but we cannot a
priori know how representative the sample is of the spoken Irish of the relevant
generations). Firstly, in the light of the variation in gender assignment for nouns
shown by Ó Curnáin (2007) (occasionally even with the same speaker), some in-
stances of wrong agreement might simply go back to a non-standard gender
assignment. Secondly, the data conflate together initial mutation and choice of
article form, but the two types of exponents might be partly decoupled (I will
develop this suggestion later on). The lack of lenition on adjective or noun is
taken as correct masculine agreement, in the appropriate context; but it could as
well signal the lack of any morphological marking, as Frenda acknowledges in
response to a referee (he also adds, correctly, that the gender opposition would
still be morphologically active, in view of the other cases). This observation gen-
eralizes: in many cases gender may simply fail to be expressed, something which
Ó Curnán (2007: 497) also draws attention to by pointing out that cases of nonle-
nition are often ambiguous as to gender, particularly when there is a possibility
of homorganic nonlenition.
Still, two important results remain: first, lenition of an attributive adjective
after a feminine noun seems decidedly recessive; second, this contrasts with the
realization of agreement between noun and article (in both forms it takes, leni-
tion noun and form na of the article), which also fails sometimes but generally
holds its ground.
The tendency towards nonlenition of attributive adjectives after feminine
nouns contrasts with the situation reported by Ó Curnáin (2007: 1736) for a ho-
mogeneous local variety: Feminine singular nouns, not inflected for genitive
case, regularly lenite attributive adjectives, less regularly nouns. We should also
note that a case like an t-iomaire mhór, with masculine an-t and feminine mhór,
goes against the tendency reported by Frenda (2011): either the noun is mascu-
line, and then the lenited adjective is an overapplication of feminine lenition; or
it is feminine, and then what is deviant here is the agreement with the article.
However, the contrast might not be as sharp as it seems. Ó Curnáin takes
into account the fact that several nouns have a non-standard masculine gender
value; his hinting that nonlenition could say nothing about gender is impor-
tant, because Frendas data might follow from lenition being generally less reg-
ular in general, more than from gender not being grammatically represented.
Ó Curnáin (2007: 506) clearly identifies a reduction of feminine gender mark-
ing in nominals as a feature of younger speakers (born after 1960). Interestingly,
however, he calls this gender depletion: it is not as if masculine and feminine
were grammatically equivalent choices, and one is encroaching on the other, but
rather that gender marking usually means feminine marking, and this surfaces
less and less frequently.
I will develop this line of interpretation, but with an important twist mo-
tivated by Frendas data. The fact that lenition between articles and nouns sur-
vives better than that between nouns and adjectives, in proportion (so, making
15
allowance for the fact that the second configuration is much less common), is not
easily compatible with the idea that what is disappearing is the feminine gender.
Because if it were so, we would expect to see a generalized rise of masculine mor-
phology, with many more cases of wrong masculine article in front of feminine
nouns, like an t-iarnáil the [masc] iron cited by Ó Curnáin (ibid.) for a speaker
who otherwise follows the older practice of treating as feminine loanwords in
-áil, like an ghesáil the [fem] guessing.
3 A puzzle
The preceding section should have made clear that gender is undoubtedly still
a morphological category of Irish nominals, but its exponence has oscillations
and inconsistencies which are definitely not usual for an inflectional category. In
particular, the fact that the morphological reflexes of feminine gender are much
better preserved on articles than between nouns and adjectives, speaks against a
simple interpretation of this situation as a system in flux. Stated in these terms,
the situation may be unclear, but not particularly puzzling. However, two facts
make it harder to make sense of it.
16
as shown by pairs like both hut - bothóg shanty, cabin (beside bothán), but it is
no longer productive in that capacity (Doyle 1992: 122-130). Its function is that
of deriving nouns for individual entities characterized as small, from nouns (as
in bábóg doll from báb baby, or béalóg gap, muzzle, mouthful from béal mouth)
or from adjectives (bog soft - bogóg shell-less egg; ciar dark - ciaróg beetle). It is
consistently feminine. Derivations like meabhróg thoughtful girl from meabhair
mind or plandóg shapely woman from planda plant suggest a female-denoting
function; however, Ó Curnáins overview of Irish word formation (2016: 2796)
explicitly characterizes it as predominantly non-personal, adding that it is par-
ticularly common in Ulster varieties (some examples are tonnóg duck, beachóg
bee, sopóg sheaf, bachóg sprout). In sum, a reasonably common noun-forming
suffix expresses the feminine gender value consistently and yet independently of
semantic motivation (it is compatible with, but not restricted to, female refer-
ence).4 But then it is not nouns in general, as a lexical category, which are losing
gender as a lexically fixed grammatical property.
17
as incompatible with his generalization that, in Italian, nouns entailing maleness
(unlike person) are feminine.5 The conclusion is clear: Irish grammar cannot just
select the gender value of a DP on the basis of the semantic motivation of gender
assignment. No matter how unstable gender morphology might be, the seman-
tic value expressed on D cannot be detached from the value morphologically
expressed on N; there is no le / la journaliste.6
The puzzle, then, stands in these terms. On the one hand, the gender as-
signment of nouns is unstable in significant parts of the nominal lexicon, and
feminine gender marking shows signs of being quite often suspended specifi-
cally on nouns, as opposed to articles. On the other hand, there are no signs that
grammatical gender on nouns is being replaced by a semantically-determined
assignment of value based on the reference of DP; on the contrary, Irish lacks
constructions where gender is determined at DP level overriding the grammat-
ical gender on the noun, something which happens in other languages.
The first assumption states that, despite all signs of instability, gender should still
be seen as a grammatical feature operative inside DP, and has not been reduced
to an English-style DP-level categorization for referents. This was motivated in
the previous sections and will not be argued for in what follows. We can now
focus on the second proposition.
5 Percus (2011: 192) opts to keep the generalization but interpret the entailment as referring to
entities that are either females or like females in certain relevant respects. This plays down the fact
that such terms highlight precisely the contrast between the referent and the gender value of the
noun: the fact that the noun refers to males is not linguistically irrelevant but essential for lexical
meaning.
6 The apparent lack of mated nouns is one of the factors that make it difficult to resort to ellipsis
phenomena to investigate the status of gender on nouns. In the general absence of pairs where
gender minimally distinguishes readings like uncle-aunt, male-female teacher, or actor-actress, we
cannot use patterns like He is a good Nmasc , but she is also [a good Nfem ] to test whether the gender
marking can be teased apart from the noun under ellipsis (see Merchant 2014 for a study of Greek
which concludes that not only does the status of gender different across lexical classes, but ellipsis
itself is in fact two distinct phenomena). Another factor is that the Irish pronominal equivalent of
one is ceann, which to my knowledge is never sensitive to the gender of its antecedent. That said,
it is likely that a study of noun ellipsis in Irish will find alternative ways to bring some facts to bear
on the issue. I thank Jason Merchant for this insightful suggestion, which I cannot pursue here (but
might be pursued by others).
18
4.1 The gender of nominalizing suffixes
The second assumption addresses the theoretical question posed by suffixes like
-óg. Plainly, there are suffixes that impose a gender value; the problem is how to
model the observation that the value in these cases is stable, while in other cases
(nouns built with other suffixes or underived) it is not. If we assume, with Borer
(2005) and Harley (2014) (and by now many others), that lexical categories like
nouns are by definition structurally complex objects built around a category-
free root, we can follow De Belder, Faust, and Lampitelli (2014), and Déchaine et
al. (2014) and analyze diminutives in nominalizing function as realizations of a
morpheme that qualifies the root as a noun. Since such diminutive-nominalizers
often specify a count interpretation (cf. French glace ice - glaçon ice cube/frag-
ment; German Brot bread - Brötchen bread roll; Italian zucchero sugar - zuccherino
sugar cube), we can associate it with the general function of Division of reference
in Borers (2005) structure. In Irish, as in other languages, these morphemes im-
pose a gender value to the noun. When the same -óg morpheme expresses only a
diminutive content, as in cnapóg little lump from cnap lump, it is still marked as
feminine (as a morphological property of the affix) but it realizes a higher eval-
uative node. This is schematically represented in (7) and (8):
(7) [ D ... [ Number ... [Evaluation ... [ Division ... root ]]]]
-óg [Fem] ciar-
(8) [ D ... [ Number ... [Evaluation ... [ Division ... root ]]]]
-óg[Fem] cnap-
Déchaine et al. (2014) and De Belder, Faust and Lampitelli (2014) have pursued
the same insight, centred on an inner and an outer locus for diminutive mor-
phemes (with relevant differences: the latter source, in particular, cogently ar-
gues that the innermost diminutive appear on other categories too, and so are not
really nominalizers). These analyses converge on distinguishing two functions
for diminutives, noun-forming and noun-modifying; and it stands to reason that
the former should determine a gender value.
The nominalizer -óg, then, acts like a light noun which encodes feminine.
The same holds of those suffixes that rigidly correlate with one gender value.
Other nouns lack this type of morpheme: either they contain nominal suffixes
which do not unambiguously specify a gender value, or they have no distinct
morphemes in their structure. In this case, a gender value is associated with
the whole complex that makes up the noun, but without being the content of
any specific morpheme (I will return to lenition directly); in short, gender has
no direct exponent, as these nouns do not formally encode their gender value
through a grammatical element. When the association with a gender value is
a property of the whole word, it can be reinforced by regularities in form, like
the often-cited tendency to associate palatal word endings with feminine, or in
meaning, as in effects of semantic contiguity (so that a noun may take over the
gender of a semantically close noun). But these are generalizations that arise
19
from the usage of words, sharply different in nature from gender-form pairings
that are explicitly part of the grammar.
This, then, is what the articles and gender-specific suffixes have in common:
they both directly spell out a gender value (in addition to the rest of their con-
tent). By contrast, the noun endings compatible with masculine or feminine may
be more or less strongly associated to one value, but they do not realize it as an
exponent realizes a grammatical feature.
20
other genitive constructions which do (as in (12); both taken from de Bhaldraithe
1977: 262-263):
This asymmetry is well documented, and indicates that the lenition on a de-
pendent noun must be sensitive to syntactic and semantic factors which break
down this configuration into a number of distinct subcases; see de Bhaldraithe
(1977: 262-266), Ó Siadhail (1989: 121), Ó Sé (2000: 61-63), and especially Ó Curnáin
(2007: 1737-1747). The distribution is complicated and probably less than system-
atic, since de Bhaldraithe also lists on the same page aimsir gaoith anoir weather
of-wind from East, with no lenition, and aimsir bháistí weather of-rain, with
lenition. Whatever underlying pattern explains the alternating presence and ab-
sence of lenition in this configuration, in the sources just mentioned it reflects a
genuine property of native linguistic competence. This stands out clearly when
it is contrasted with the occurrence of lenition on adjectives, as witnessed in
the same sources. For these reasons, I will set aside the whole issue of lenition
marking on a dependent genitive noun, which varies according to independent
(if, for this writer, unclear) syntactic and semantic factors. What matters more
directly for the question of gender marking is instead the lenition of attributive
adjectives; and in this connection we have seen that lenition is relatively regular
in traditional varieties, but is infrequent and shows strong signs of collapsing
judging by the data in Frenda (2011).
All of these properties suggest that lenition has an indirect relation with the
realization of gender; not only because it also realizes other grammatical con-
tents, but because even as an exponent of gender it prone to being omitted, or
overgeneralized, or realized only on one link of the agreement chain rather than
on all agreeing elements. Gender involves the choice of forms in an agreement
configuration, and gender values identify agreement classes. There is, then, a
purely morphological side to it (the forms selected) and a syntactic side (the
agreement configuration). In syntactic terms, we can represent the structure
containing the noun controller and the two potential targets for gender agree-
ment in the following standard notation, where the controller values the feature
representations of the targets D and Adj (I specify the case on D as genitive, but
it could be nominative/accusative):
21
These featural representations are spelled out by the appropriate morphemes,
and the lexical root in the case of N and Adj (where N is a shorthand for a more
complex structure, as discussed in 4.1 above). A form like na realizes gender,
number, and case on D (in case the feature values are feminine, singular, and
genitive), but otherwise gender does not find a specific morphological realiza-
tion (excepting the specific feminine forms of the adjective, nowadays rare, il-
lustrated by bhige in (4) above). In particular, N lacks a dedicated exponent for
this feature, unless it contains a suffix explicitly marked for a gender value, like
-óg. My suggestion is that lenition should be seen as a morphological formative
which is conditioned by (among other triggers) the feminine gender value on
the feminine singular form of D, N, and Adj (for D, also nominative/accusative),
but which can fail to be expressed even though the feature value is still syntac-
tically present. As a secondary exponent (Noyer 1997), it accompanies the mor-
phological spellout of an abstract lexical and syntactic representation, specifying
word forms in ways that can only take effect when the abstract representation
has been linearized as seems natural for a morpheme that is only visible as a
sandhi phenomenon. Lenition, as a specification of word forms, is then a reflex
of the feminine gender value on a whole agreement chain, and only becomes the
exponent of this feature value when no other morphemes spell it out on the same
chain; for instance, in a structure like an bhean bhocht the poor woman. This, I
think, is the difference between lenition and the more usual exponence of femi-
nine by means of morphemes like na and -óg; it is in principle a side-effect of the
marking [Gender: Feminine], and in this sense it is more loosely related to this
marking than inherently feminine morphemes. It can, then, be subject to oscilla-
tions in use, independently of the truly feminine morphemes and of the syntactic
marking [Gender: Feminine] on the whole DP (as long as this is recoverable on
at least some element). So, for instance, in (12) lenition may fail to appear, while
the feature value is there and conditions the choice of the article form na. In
addition, lenition may surface only on some of the chain links on which it is
licensed: supposing that it may appear on N but not on D (which by hypothe-
sis is still feminine), this would model what happens in inconsistent phrases like
an t-eolas (the knowledge, no lenition) but eolas mhaith (good knowledge, with
lenition on the adjective following eolas), without needing to posit two distinct
gender assignments for the noun.
22
an irregular choice of gender value according to the tables in (12).
If a speaker hesitates between an t-iascach and an iascach (the fishing) in the
same utterance, as Ó Curnáin reports (2007: 500), it is possible that the first form
of the article an t-, simply recruited the prevocalic t- of the masculine as a hiatus-
avoiding strategy, rather than being a genuine gender marker. It certainly seems
more plausible that speakers should hesitate about such sandhi phenomena, than
about the gender of nouns. The same would apply to the string an t-iomaire mhór
(see 2.2 above), where the two apparently contradictory markings appear side
by side.
The hypothesis that lenition may be to some extent decoupled from fem-
inine gender is of course still compatible with the idea that feminine is simply
lost in some (or many) cases. Ó Curnáin (2007: 506) identifies a tendency towards
gender depletion for younger generations of speakers, resulting in feminine be-
ing no longer realized because it is simply no longer there. There is no need to
dispute this interpretation, as long as the data are compatible with a generalized
masculine morphology and do not at the same time suggest that feminine is still
there, but only partially expressed. The (intended) strength of the reinterpreta-
tion I have offered is that it allows us to account for some of these more puzzling
data, not only because the lack of lenition does not automatically imply lack of
feminine gender, but also because it could be D or N that fail to lenite despite
being feminine. If Ulster varieties usually treat tír land as a feminine (genitive na
tíre), but feature an irregularly non-lenited adjective in the phrase tír mór main-
land, literally big land (see again 2.2 above), it seems more plausible that lenition
is simply suspended, especially in similar collocations, rather than thinking of an
unstable gender value for this noun. In particular, a generalized retreat among
younger speakers from the lenition triggered by N, as opposed to that triggered
by D, would model the pattern found by Frenda (2011), with feminine gender
agreement between N and Adj much less frequent than that between D and N
(respectively 42% versus 87%, discounting the few proper names).
Again I must emphasize that this is not intended to explain away all the grey
areas in Irish gender marking as an only apparent puzzle, because many cases
are still better viewed as inconsistent assignments of gender value. Ó hUiginn
(1994: 563), for a final example, cites for Connacht the two forms of the noun
bróigín, formed by the diminutive -ín on the feminine (in -óg!) noun for shoe: in
the nominative, the suffix regularly retains the gender of the base noun, resulting
in the feminine an bhróigín (with lenition on N triggered by the article). But he
adds that the genitive of the same noun behaves like a masculine in the phrase
barr an bhróigín the tip of the shoe where lenition appears on both nominative
and genitive, so it does not surface too little but, if anything, too much; the mor-
phologically feminine noun in question simply has a masculine genitive (in that
phrase).
This sort of cases might suggest that lenition too, like t-insertion, might
sometimes act just like a sandhi phenomenon detached from the exponence of
gender, especially in common collocations like cleachtadh mhaith. Such sequences,
then, would no longer count as evidence for a feminine value of the noun. I sug-
23
gest that this is probably the correct interpretation in at least some cases, but
with the essential qualification that it cannot be accepted as an explanation with-
out independent reasons for thinking that the noun is not feminine. Otherwise,
simply claiming that lenition, effectively, doesnt count in some cases (the prob-
lematic ones) would deprive the proposed interpretation of all predictive power.
As I dont have independent evidence to bring to bear, I will simply mention this
possibility without pursuing it.
Gender marking on Irish nominals is, objectively, messy. I have proposed an
interpretive framework that can help make sense of this messiness: the irregu-
larities concern some nouns more than others because of their morphological
make-up,8 and they have substantially to do with lenition not appearing where
it would be expected to. Alternative interpretations, especially the restatement-
like position that, on the face of the facts, it is more economical to simply accept
inconsistent gender assignments, should explain why this does not happen when
initial mutation is not an exponent. I have not proposed that the irregularity is an
epiphenomenon, but I have suggested that it would not be so common if nominal
gender did not rely so much on initial mutations and other sandhi phenomena
for its realization.
24
dition, the lenition of the attributive second noun in a sequence Noun 1 - Noun 2
[genitive] is sometimes suspended, especially in traditional petrified phrases like
crích Banba the land of Banba, or where the first noun has quasi-prepositional
value, as in lár mí an Meithimh the middle of June. These are fixed collocations,
and therefore they have a special status; still, by itself being a fixed collocation
does not necessarily cancel the internal syntactic structure of a lexicalized phrase
(proper names are certainly not immune from lenition). In addition, many fixed
collocations, even petrified as place-names, preserve genitive forms, like Drom
Domhnaigh (Dromdowney), County Cork, the ridge of the church, from domh-
nach church (https://www.logainm.ie/719.aspx).
These data independently confirm the conclusion reached in the past sec-
tion: lenition realizes gender in a fashion that is less regular and systematic than
affixes. It makes sense, then, to attribute a substantial role in the attested in-
stability of nominal gender morphology to the instability of lenition (and other
mutations) as a morphological marker. At least some of the puzzling cases where
a noun seems to have masculine value (without lenition) after the nominative/ac-
cusative article an, and feminine (with article na, again without lenition) in the
genitive, may well have this explanation. In other cases too, a noun that is syn-
tactically marked feminine, with the corresponding forms of article, might fail
to trigger lenition on a following adjective simply because lenition is suspended.
This is not the explanation for all cases, as noted, but I propose it is a part of it.
(14) a. A peann
poss pen
Her pen
b. A pheann
poss [len]pen
His pen
c. A bpeann
poss [nas]pen
Their pen
Granted that lenition and initial mutations are less than totally deterministic as
morphological exponents, how are we to make sense for the additional observa-
25
tion that this concerns so specifically articles, nouns, and post-nominal adjectives
inside DP, but not possessives?
The answer I suggest is that the Irish gender opposition is perfectly sound
in the pronominal system as a determination of morphemes (including posses-
sives) that express a referential index, but it is receding from the representation
of nouns and, possibly less strongly, of articles, as morphological words. It is not
as a global morphophonological phenomenon that mutations are becoming less
regular (at least not for native competent speakers), but as formants in the rep-
resentation of words as morphological objects; specifically, of nominal words,
nouns and adjectives. This statement encapsulates the claim that lexical items
are represented as structured symbols not only as syntactic objects, but also as
abstract morphological objects which feed phonological realization. While this
is assumed by, or is at least compatible with, most approaches to morphology, it
is not a claim universally accepted. It is implied, for instance, by those analyses
that follow Harris (1991) in positing word-markers as vocalic right edges of lexical
words, required for morphological well-formedness: such would be the final -o,
for instance, which in Italian closes off a noun endowed with the feature values
of masculine and singular, like cavallo horse, but also a non-inflected adverb like
quando when. Such formants are mandated by the morphology of the language,
and not by syntax or phonology, witness the numerous consonant-ending loan-
words like sport or pus (also masculine and singular) which are syntactically in-
flected for pronominal features but do not realize them morphologically through
a vocalic ending (see Acquaviva 2009). In a similar vein, we can view the initial
lenition triggered by feminine singular non-genitive nouns as a constituent of a
linearized morphological representation:
For a word like bliain year, the information that the gender value is feminine
is not encoded by a feminine affix (or through the choice of a declension class
associated with feminine by default). It rather surfaces as a juncture effect, as a
global property of the word in certain phonological and syntactic environments.
My hypothesis is that this property, represented as [LEN] in (16), is obliterated
much more easily in these circumstances than in the case of nouns ending in a
suffix that consistently spells out the feminine value, like -óg. The formant [LEN],
in other words, is stable on words that are morphologically marked as feminine,
where it is a side effect of feminine marking. But where a gender value is not
explicitly marked by a suffix, it ends up having no direct exponence. Ultimately,
this is because Irish nominal morphology does not seem to define declension
classes that are truly consistent in the choice of gender (see Carnie 2008), espe-
cially when one considers the extremely reduced two-case paradigm and the fact
that genitive forms are used less and less.
Articles are different. As functional words, they associate directly a featural
content with a form, which is unique for a given choice of feature values. There
is no lexical, word-depending variability in morphological structure. We then
26
predict that lenition and other initial mutations should prove more stable as an
exponent of gender agreement between article and noun, than between noun
and adjectival modifier. Still, the instances of missing or wrong mutation trig-
gered by articles call for some explanation, over and above the fact that lenition
can be suspended. The missing piece of explanation might lie in the demor-
phologization of such mutations after articles, which increasingly often are used
as phonotactic juncture elements rather than gender exponents. This would be
consistent both with the evidence (occasional) for hesitation and inconsistent
choices by speakers, and with the theoretical claim that articles should retain
more robustly than lexical nouns the property of triggering a mutation on a fol-
lowing word onset.
27
values. But such an asymmetry is not surprising. In realizing the same set of
inflectional features, determiners often define morphological paradigms orga-
nized differently from those of lexical nouns (and adjectives). French is an il-
lustration. Here, determiners oppose an invariable plural (les, ces /le/, /se/) to a
gender-differentiated singular (la - le, ce-cette /la/ - /l@/, /s@/ - /sEt/ ). By con-
trast, adjectives, personal pronouns and those nouns that have a masculine and a
feminine alternant either have no opposition (like joli, jolie, jolis, jolies, all /joli/),
or they primarily oppose masculine and feminine, and only in some cases specify
a number form inside a determinate gender value, like nouveaux /nuvo/, which
can only be masculine and plural. This is shown in (17)-(18):
This familiar example shows that there is nothing unusual in having morphology
organize differently across D and N the syntactic featural information shared be-
tween the two. But the difference is not to be construed as if the non-morphological
information were automatically semantic in nature. The Irish facts, especially
with the constant feminine gender of male-denoting nouns like piteog (see 3.2),
clearly show that what lies behind the weakening of gender morphology on
nouns is not a reduction to semantic-based assignment. This is relevant to cur-
rent debates about gender in natural language, which distinguish between an in-
terpretable and a purely morphological value for this category. A more nuanced
approach would distinguish the mechanisms for gender assignment, which may
be based more on the form or on the content of a linguistic representation, from
the properties of the representation itself. Within these properties, gender has
a syntactic and a morphological facet. The instability of Irish nominal gender, I
have argued, concern the latter.
28
References
Acquaviva, Paolo. 2009. The structure of the Italian declension system. In Fabio.
Montermini, Gilles Boyé, and Jesse Tseng (eds.), 6th Décembrettes: Morphol-
ogy in Bordeaux. 5062. Cascadilla Proceedings Project, Somerville, Mass.
Aronoff, Mark. 1994. Morphology by itself. Cambridge, Mass.: MIT Press.
Borer, Hagit. 2005. In name only. Oxford: Oxford University Press.
CO 2001[1958] = Gramadach na Gaeilge agus litriú naGaeilge: an caighdeán oifigiúil.
Dublin: Oifig an tSoláthair, Rialtas na hÉireann.
Carnie, Andrew. 2008. Irish nouns. A reference guide. Oxford: Oxford Univer-
sity Press.
De Belder, Marijke, Noam Faust, and Nicola Lampitelli. 2014. On a low and
a high diminutive. In Artemis Alexiadou, Hagit Borer, and Martin Schäfer
(eds.), The syntax of roots and the roots of syntax. Oxford: Oxford University
Press, 149163.
de Bhaldraithe, Tomás. 1977 [1953]. Gaeilge Chois Fhairrge:.an Deilbhíocht [The
Irish of Cois Fairrge: Morphology]. Baile Átha Cliath: Institiúid Ard-Léinn
Bhaile Átha Cliath.
Déchaine, Marie-Rose. Raphael Girard, Calisto Mudzingwa, and Martina Wiltschko.
2014. The internal syntax of Shona class prefixes. Language Sciences: 1846.
https://doi.org/10.1016/j.langsci.2013.10.008
Doyle, Aidan. 1992. Noun derivation in Modern Irish. Lublin: Redakcja Wydawn-
nictw Katolickiego Uniwersytetu Lubelskiego.
Frenda, Alessio. 2011. Gender in Irish between continuity and change. Folia
Linguistica 45, 2: 283 - 316.
Harley, Heidi. 2014. On the identity of roots. Theoretical Linguistics 40, 3/4:
225276.
Harris, John. 1991. The exponence of gender in Spanish. Linguistic Inquiry 22,1:
2762.
Hughes, Art. 1994. Gaeilge Uladh [Ulster Irish]. In McCone et al. (eds.), 611660.
Landau, Idan. 2016. DP-internal semantic agreement: A configurational analysis.
Natural Language and Linguistic Theory 34/3: 9751020. Luís and Bermúdez-
Otero (2016). The Morphome debate. Oxford: Oxford University Press.
McCone, Kim, Damian McManus, Cathal Ó Háinle, Nicholas Williams, and Liam
Breatnach (eds.). 1994. Stair na Gaeilge [History of Irish]. Maigh Nuad: Roinn
na Sean-Ghaeilge, Coláiste Phádraig.
Merchant, Jason. 2014. Gender mismatches under nominal ellipsis. Lingua 151:
932.
Noyer, Rolf. 1997. Features, Positions and Affixes in Autonomous Morphological
Structure. Doctoral Dissertation, MIT.
Ó Curnáin, Brian. 2007. The Irish of Iorras Aithneach, County Galway. Dublin:
Dublin Institute for Advanced Studies.
Ó Curnáin, Brian. 2016 Irish. In Peter Müller, Ingeborg Ohneiser, Susan Olsen,
and Franz Rainer (eds.), Word Formation. Volume 4: Word-formation in the
individual European languages. Berlin: de Gruyter Mouton, 27822810.
29
Ó Dónaill, Niall. 1978. Foclóir Gaeilge-Béarla. Dublin: An Roinn Oideachais.
Ó hUiginn, Ruairí. 1994. Gaeilge Chonnacht [Connacht Irish]. In McCone et al.
(eds.), 538609.
Ó Sé, Diarmuid. 2000. Gaeilge Chorca Dhuibhne [The Irish of Corca Dhuibhne].
Dublin: Institiúid Teangeolaíochta Éireann.
Ó Siadhail, Mícheál 1984. A note on gender and pronoun substitution in Modern
Irish dialects. Ériu 35: 173177.
Ó Siadhail, Mícheál. 1989. Modern Irish: Grammatical structure and dialectal
variation. Cambridge: Cambridge University Press. Ua Suilleabháin, Seán.1994.
Gaeilge Mumhan [Munster Irish]. In In McCone et al. (eds.), 479538.
Wigger, Arndt (ed.) 2000. Caint Chonamara: Bailiúchán Hans Hartmann, Vol.
IX: Ros Muc, [The spoken language of Connemara: Hans Hartmanns collec-
tion, Vol. IX: Ros Muc]. Universität Bonn.
30
The derivation of verb initiality in Santiago
Laxopa Zapotec*
Jeff Adler Steven Foley
University of California, Santa Cruz University of California, Santa Cruz
Maziar Toosarvandani
University of California, Santa Cruz
1 Introduction
Languages with verb-subject-object (vso) word order are amongst the most com-
mon in the world, and yet they pose a fundamental mystery. If the subject is
structurally superior to the object, there is no way it can appear in underlying
structure between the verb and object (Anderson & Chung 1977). A prominent
solution in Irish, a language with rigid vso order, invokes head movement of the
verb to a position on the left of the subject (McCloskey 1991, 1996).
For other languages, a different solution has been proposed. Most promi-
nently, for Austronesian languages in which vso alternates with vos, phrasal
movement is commonly thought to be responsible for verb-initial order (Pen-
salfini 1995, Massam 2001, among others). A verbal constituent containing the
verb raises, giving rise to vos when it contains the object, and to vso when the
object has escaped, possibly for reasons involving definiteness or specificity. (For
alternatives, see Chung 1990 on Chamorro, Holmer 2005 on Seediq, or Otsuka
2000 on Tongan.)
Which route a language takes might seem, then, to be a relatively straightfor-
ward choice. If vso alternates with vos, it uses phrasal movement; if it has rigid
vso order, it uses head movement. However, as Clemens & Coon (to appear)
propose for several Mayan languages, head movement can still give rise to al-
ternating word order if the resulting vso order is manipulated postsyntactically,
* We are extremely grateful to Alberto Diaz Robles, Raul Diaz Robles, Fe Silva Robles, and two
other native speakers for teaching us about their language. We are also thankful to audiences at 2017
Annual Meeting of the Society for the Study of the Indigenous Languages of the Americas in Austin
and at UC Santa Cruz. And, of course, we are greatly indebted to Jim McCloskey for his mentoring
and advice in so many different capacities.
31
through operations reordering the subject and object.
We describe the inverse scenario. Santiago Laxopa Zapotec (slz) is, like other
Zapotec languages, rigidly vso (see Sonnenschein 2004: 125 and López Nicolás
2016: 266 on closely related varieties).1
(2) TP
T0
T
DPS
DPO vP
tS v0
v VP
V tO
This rather abstract predicate raising analysis might seem difficult, if not impos-
sible, to distinguish from a more concrete verb raising analysis, in which the verb
alone raises to T. But we take inspiration from Jim McCloskey’s work—say, on
the number and type of subject positions (McCloskey 1997) or the derivational
mechanism undergirding A0 -dependencies (McCloskey 2002)—which shows how
it is often possible to tease apart very similar analytical possibilities by examining
subtle patterns of data in detail and with careful and precise argumentation.
1 The abbreviations we use are: an = animal, caus = causative, cl = classifier, cont = continuous
aspect, comp = completive aspect, def = definite, el = elder, freq = frequentative aspect, hu = non-
elder human, in = inanimate, inch = inchoative, hu = informal, int = intensive, rep = repetitive
aspect, sg = singular, stat = stative aspect, pot = potential aspect.
32
2 Some background on Santiago Laxopa Zapotec
Santiago Laxopa Zapotec (slz) is a Northern Zapotec (Oto-Manguean) language
spoken in the Sierra Norte of Oaxaca, Mexico. We present data from three speak-
ers from the town of Santiago Laxopa itself. We also include data from two
speakers from the nearby towns of San Sebastián Guiloxi and Santa María Yalina.
There are dialectal differences amongst these speakers, but we have observed no
significant variation with respect to the syntactic phenomena under considera-
tion. These southeastern Sierra varieties are most closely related to those of San
Jerónimo Zoochina (López Nicolás 2016) and San Bartolomé Zoogocho (Long &
Cruz 2000, Sonnenschein 2004).
In slz, the subject is structurally superior to the direct object: it is able to
asymmetrically bind an R-expression in that position, inducing a violation of
Condition C (which has been shown to be active in other Zapotec varieties as
well; Lee 2003).
(3) a. Bdi’inn [beku’ tse Pedro1 ] lleba’1/2 .
bite.comp dog of Pedro 3.hu
‘Pedro1 ’s dog bit him1/2 .’ (FA/RM, GZYZ019, 1:30)
b. Betw=ba’*1/2 [beku’ tse Pedro1 ].
hit.comp=3.hu dog of Pedro
‘He*1/2 hit Pedro1 ’s dog.’ (FA/RM, GZYZ019, 13:20)
Assuming that phrases are endocentric, and that the clause is the extended pro-
jection of the verb (Chomsky 1970, Grimshaw 2005), the verb and the direct object
must form a constituent to the exclusion of the subject (pace Broadwell 2005).
(4) vP
DPS v0
v VP
V DPO
Assuming a fairly articulated structure for the verb phrase, the external argument
is introduced in the specifier of a functional head, Spec-vP (Kratzer 1996).
There is evidence that the subject occupies a surface position outside of vP.
The subject of an unaccusative verb (5) occupies the same position relative to
manner adverbs that the subject of a transitive verb does (6).
(5) a. Dz-i-yag Pedro xtidao’.
cont-inch-be.cold Pedro quickly
‘Pedro is getting cold quickly.’ (FA/RM, GZYZ019, 1:19:07)
b. * Dz-i-yag xtidao’ Pedro.
cont-inch-be.cold quickly Pedro (FA/RM, GZYZ019, 1:19:20)
33
(6) a. Udoo Juan=a’ yet=e’n xtido’-yes.
eat.comp Juan=def tortilla=def quickly-int
‘Juana ate tortillas very quickly.’ (FSR, SLZ1009-s, 21)
b. * Udoo xtido’-yes Juan=a’ yet=e’n.
eat.comp quickly-int Juan=def tortilla=def
(FSR, SLZ1009, 31:15)
Regardless of how verb initiality is derived, then, subjects must raise to a position
below the surface position of the verb (see Lee 2006: 49 for additional arguments).
(7) TP
T FP
DPS F0
F vP
Adv vP
v VP
V tS
We call this projection simply FP. The subject raises into its specifier presumably
to satisfy an epp feature, receiving nominative case from T in that position.
The clause in slz is probably even more articulated than this, since verbs
bear rich inflectional morphology, including aspectual, directional, number, and
voice prefixes (Sonnenschein 2004: 118, López Nicolás 2016: 153). If these are
heads, they must attach to the root in some way. We remain agnostic about
whether this happens through head movement or a postsyntactic operation (see
Lee 2006: 53–62).
34
(8) a. Verb raising
TP
T FP
DPS F0
F vP
DPO vP
tS v0
v VP
V tO
b. Predicate raising
TP
T0
T FP
DPS F0
F vP
DPO vP
tS v0
v VP
V tO
35
appear between the verb and subject, such as an object, must evacuate this con-
stituent before it raises to Spec-TP. As we show next, this kind of movement is
independently available in the language.
While the position of the subject is fixed—it must be the first postverbal ar-
gument in the clause—the order of other clausal elements is flexible. Direct and
indirect objects are more or less freely ordered amongst one another (9), as are
any clausal complements (10) or adjunct PPs (11).
This freedom recalls scrambling in the Germanic middlefield, which has been
variously analyzed as base generation with flexible lineariziation and as move-
ment to a functional projection or to adjoin to the verb phrase (see the survey in
Haider 2006).
To maintain a straightforward mapping from syntax to phonology, we as-
sume that the sentence constituents that exhibit flexible word order in (9)–(11) un-
dergo movement to positions that are higher in the clause and are not internally
ordered. For simplicity, we will allow them to simply adjoin to vP, though we are
open to the possibility that they are instead specifiers of one or more functional
projections. Importantly, for the predicate raising account, this movement must
be obligatory, so that vP contains just the verb before it itself moves. While the
36
need for this additional assumption could be interpreted as an argument in fa-
vor of the alternative verb raising account, we reemphasize that the real question
here is why major sentence constituents other than subjects are freely ordered.
For similar facts in other languages, e.g., Germanic scrambling, the answer is not
clear cut and often depends on higher, framework-level assumptions. So, in the
absence of decisive arguments to the contrary, we see no harm in assuming that
this movement in slz is obligatory.
Crucially, wherever these elements move to, it cannot be to a position that
intervenes between the subject in Spec-FP and the position of the verb. A direct
or indirect object cannot intervene between the subject and verb (12a–b); nor can
an adjunct PP (13) or clausal complement (14).
37
Returning to the main issue at hand, it might now seem impossible to dis-
tinguish between the verb raising and predicate raising accounts, if all the ele-
ments inside vP can move out except V, thereby conflating the difference between
moving a head and moving a phrase. But, in what follows, we identify several
elements that can or must remain within vP: aspectual adverbs (Section 4), the
adjectival predicate in a copular clause (Section 5), and the nonverbal element
in a light verb construction (Section 6). These elements are able to move along
with the verb, sometimes appearing between the verb and the subject, just as the
predicate raising account predicts.
Then there are manner adverbs, such as xtido’ (or xtidao’) ‘quickly’, cholazhe’e
‘slowly’, and zishje’ ‘loudly’, which have a superficially similar distribution to tem-
poral adverbs.
38
d. Udoo Juan=a’ yet=e’n xtido’-yes.
eat.comp Juan=def tortilla=def quickly-int
(FSR, SLZ1009-s, 21)
Finally, aspectual adverbs, such as chintje’ ‘just (now)’, ba ‘already’, and ne’e ‘still’,
can only appear immediately before the verb.
(18) a. Chintje’ bta Sonia=’n zah.
just stir.comp Sonia=def bean
‘Sonia just stirred the beans.’ (RD, SLZ2012-s, 17)
b. * Bta chintje’ Sonia=’n zah.
stir.comp just Sonia=def bean (RD, SLZ2012, 43:14)
c. * Bta Sonia=’n chintje’ zah.
stir.comp Sonia=def just bean (RD, SLZ2012, 43:24)
d. * Bta Sonia=’n zah chintje’.
stir.comp Sonia=def bean just (RD, SLZ2012, 43:28)
We call these aspectual adverbs, following Tenny (2000), because they appear
sensitive to the internal structure of the event described by the verb. They can
be contrasted with temporal adverbs, which simply locate the event in time.
We propose to account for the distributions of these three classes by adjoin-
ing them in different, albeit partially overlapping, positions.
(19) The position of adverbs in Santiago Laxopa Zapotec
i) Temporal adverbs can adjoin to vP or to CP.
ii) Manner adverbs can adjoin to vP or fill Spec-CP.
iii) Aspectual adverbs can only adjoin inside vP.
Under our present assumptions, both temporal and manner adverbs must be
able to adjoin to vP, since they can be freely interleaved amongst any nonsubject
arguments or adjunct PPs.
(20) a. Blo’ed Maria cholazhe’e bidao’ ni beku’ lo’ yo’o.
show.comp Maria slowly child this dog in house
‘Maria showed the dog to this child in the house slowly.’
(FA/RM, GZYZ020, 49:53)
b. Blo’ed Maria bidao’ ni cholazhe’e beku’ lo’ yo’o.
show.comp Maria child this slowly dog in house
(FA/RM, GZYZ020, 49:33)
c. Blo’ed Maria bidao’ ni beku’ cholazhe’e lo’ yo’o.
show.comp Maria child this dog slowly in house
(FA/RM, GZYZ020, 49:11)
d. Blo’ed Maria bidao’ ni beku’ lo’ yo’o cholazhe’e.
show.comp Maria child this dog in house slowly
(FA/RM, GZYZ020, 48:52)
39
(21) a. Blo’ed Maria neje bidao’ ni beku’ lo’ yo’o.
show.comp Maria yesterday child this dog in house
‘Yesterday, Maria showed the dog to the child in the house.’
(FA/RM, GZYZ019, 16:23)
b. Blo’ed Maria bidao’ ni neje beku’ lo’ yo’o.
show.comp Maria child this yesterday dog in house
(FA/RM, GZYZ019, 1:45)
c. Blo’ed Maria bidao’ ni beku’ neje lo’ yo’o.
show.comp Maria child this dog yesterday in house
(FA/RM, GZYZ019, 17:03)
d. Blo’ed Maria bidao’ ni beku’ lo’ yo’o neje.
show.comp Maria child this dog in house yesterday
(FA/RM, GZYZ019, 17:19)
The two classes come apart in preverbal position. While temporal adverbs adjoin
to CP, appearing before a wh-phrase (23a–b), manner adverbs occupy Spec-CP,
since they cannot cooccur before the verb either preceding or following a wh-
phrase (22a–b). (In wh-questions, manner adverbs only surface postverbally.)
While we do not fully understand the source of this pattern, one not implausible
explanation is semantic. Since aspectual adverbs are sensitive to the internal
40
structure of the event described by the verb, they must be adjoined closer to it
(Tenny 2000).
Under the predicate raising account, however, the inverted order of aspec-
tual adverbs relative to temporal adverbs is unsurprising. If aspectual adverbs
only adjoin inside vP, they always raise with the verb to Spec-TP, thereby ap-
pearing only to its left.
(26)
= 18a
TP
vP3 T0
t1 v0 T FP
v VP DP1 F0
Advasp VP Sonia’n F vP
chintje’ t2 t3
V DP2
bta zah
By contrast, under the verb raising account, the inverted pattern of aspectual ad-
verbs in slz is unexpected. If they only adjoin inside vP, then the grammatical
sentence in (18a) would incorrectly be predicted to be ungrammatical and the un-
grammatical sentence in (18d) to be grammatical. The only recourse would be to
assume that aspectual adverbs adjoin higher than temporal adverbs–for instance,
to TP—but this would cut against the robust generalization about the hierarchi-
cal position of adverbs across languages. So, the position of aspectual adverbs,
immediately preceding the verb in initial position, supports the predicate raising
account over the verb raising account.
41
5 Copular clauses
There are two copulas: -sua (or - soo) ‘be, live’ and -ak ‘be, happen’.2 We focus on
just the first here. In a copular clause, an adjectival predicate can either follow
(28a) or precede (28b) the subject.
We take the copula to be a verb, as it can host a subject clitic (29a) and aspec-
tual morphology (29b). But it takes a small clause complement, comprised of the
subject and a predicate.
The variable word order in (28a–b) follows straightforwardly under the pred-
icate raising account if AP predicates can optionally undergo the same scram-
bling operation that DPs and PPs undergo obligatorily. When it moves out of vP,
subject-predicate order arises (30a); when it does not, predicate-subject order
arises (30b).
(30) a. TP = 28a
vP1 T0
T FP
v VP
V SC DP2 F0
zua Pedro F vP
t2 t3
AP3 t1
wen
2 Verbs in Zapotec have several aspectual forms; the citation form is a bound morpheme lacking
aspectual morphology.
42
b. TP = 28b
vP1 T0
v VP T FP
V SC DP2 F0
zua t2 AP Pedro t1
F
wen
43
(33) a. Neje [zua wen] Pedro.
yesterday be.cont good Pedro
‘Yesterday, Pedro was well.’ (FSR, SLZ1014, 56:00)
b. * [Zua neje wen] Pedro.
be.cont yesterday good Pedro (FSR, SLZ1014, 56:20)
c. * [Zua wen] neje Pedro.
be.cont good yesterday Pedro (FSR, SLZ1014, 56:27)
d. [Zua wen] Pedro neje.
be.cont good Pedro yesterday (FSR, SLZ1014, 56:34)
Note that, under this account, (33b) is ungrammatical because a temporal adverb
cannot adjoin inside vP, in keeping with the proposal in (19).
While the verb raising account can derive subject-predicate order in (28a)
through head movement of the copula, it cannot derive predicate-subject order
in (28b). To do so, the adjectival predicate would have to move independently to a
position between the copula in T and the subject, thereby violating the adjacency
requirement. This, then, is another argument in favor of the predicate raising
account.
3 This class of light verb constructions contrasts with another, superficially similar class that ex-
hibits a different behavior (see Broadwell 2004 for a parallel contrast in another Zapotec language).
We set these light verb constructions aside because they allow for the nonverbal element to move
independently, e.g., to a preverbal position. As expected, they also allow for the nonverbal element
to intervene between the verb and subject, just as in a copular construction.
44
c. Dzon tsgwa Pablo kar tse=ba’.
do.cont much Pablo car of=3.hu
‘Pablo is showing off his car.’ (lit. ‘Pablo is making much of his
car.’) (FA, GZYZ018, 45:50)
Crucially, the nonverbal element in these light verb constructions must occur
immediately following the light verb, preceding the subject.
Broadwell (2004) proposes that parallel light verb constructions in San Dionicio
Ocotepec Zapotec are compounds. This seems unlikely for slz because some
incorporated adverbs can intervene between the light verb and nonverbal ele-
ment.4
4 Two kinds of elements can encliticize to the verb: “incorporated” adverbs and pronominal cli-
tics. While the former attach to the light verb (36a–c), the latter attach only to the nonverbal element.
At least for the pronominal clitics, these are weak elements which must be licensed syntactically (Fo-
ley, Kalivoda & Toosarvandani, to appear-a, b). They are also selective about what their phonological
host can be, which need not be a verb (Marlett 1993: 95). While they can attach to another clitic,
forming a clitic cluster, they cannot, for instance, attach to an R-expression (Toosarvandani 2017:
131). While the licensing conditions for incorporated adverbs and pronominal clitics are germane
here, we leave off further consideration for reasons of space.
45
c. Dzon-tek tsgwa Pedro kar tse=ba’.
do.cont-a.lot much Pedro car of=3.hu
‘Pedro is showing off his car a lot right now.’
(FA/RM, GZYZ019, 1:25:20)
In Persian, the light verb is typically analyzed as the realization of v, since sim-
ply switching out kardan ‘do’ for another light verb can yield an unaccusative
predicate. Its complement, then, is headed by the nonverbal element (Folli et al.
2005).
With this structure, the position of the nonverbal element immediately after
the light verb follows directly from the predicate raising account. The nonverbal
element can move along with the light verb.
(38) TP = 34a
vP1 T0
t2 v0 T FP
v AP DP2 F0
dzun Pedro
A t3 F vP
yeze’ DP3 t1
kar tse=ba’
The nonverbal element always occurs immediately following the light verb, as it
cannot move on its own.
46
b. * Lazhe dzon Pedro nada’.
lying do.cont Pedro 1sg
Intended: ‘Pedro is lying to me.’ (FA/RM, GZYZ019, 1:21:39)
c. * Tsgwa dzon Pedro kar tse=ba’.
a.lot do.cont Pedro car of=3.hu
Intended: ‘Pedro is showing off his car.’ (FA/RM, GZYZ019, 1:21:19)
By contrast, under the verb raising account, there is no way to understand how
the nonverbal element ends up between the light verb and subject. Unlike with
copular clauses, movement of the nonverbal element on its own appears to be
blocked. And even if it could move, doing so would violate the adjacency re-
quirement in (15), as it would involve movement of the nonverbal element to a
position between T and the subject in Spec-FP. This is a final argument in favor
of the predicate raising account.
7 Future prospects
Verb initiality in slz is derived, then, through predicate raising. This is per-
haps somewhat more difficult to see than in Austronesian, where the object can
sometimes surface inside the constituent that moves, yielding vos. If objects
and other non-subject constituents obligatorily move because they are freely or-
dered amongst one another, then the evidence for predicate raising has to come
from other elements with a more restricted distribution: some adverbs and some
predicates.
So, not all languages with rigid vso word order use verb raising. While this
correlation may not turn out to be crosslinguistically robust, adopting predi-
cate raising for verb initiality in slz allows us to maintain other generaliza-
tions, including the fixed ordering of temporal and aspectual adverbs. And, it
revealed another possible generalization—the adjacency requirement between
the sentence-initial predicate and the subject—which we will hopefully gain a
better understanding of through closer scrutiny of other verb-initial languages.
References
Anderson, Stephen R. & Sandra Chung. 1977. On grammatical relations and
clause structure in verb-initial languages. In Peter Cole & Jerry M. Sadock
(eds.), Grammatical relations, vol. 8 Syntax and Semantics, 1–25. New York: Aca-
demic Press.
Black, Cheryl A. 2000. Quiegolani Zapotec syntax: A principles and parameters
account. Arlington, TX: SIL International.
Broadwell, George Aaron. 2004. The morphology of Zapotec pronominal cl-
itics. In Rosemary Beam de Azcona & Mary Paster (eds.), Conference on
Otomanguean and Oaxacan languages, vol. 14 Survey Reports, 15–36. Berkeley,
CA: Survey of California and Other Indian Languages.
47
Broadwell, George Aaron. 2005. It ain’t necessarily S(V)O: Two kinds of VSO
languages. Proceedings of the LFG Conference.
Chomsky, Noam. 1970. Remarks on nominalization. In Roderick A. Jacobs &
Peter S. Rosenbaum (eds.), Readings in English Transformational Grammar, 184–
221. Waltham, MA: Ginn and Company.
Chung, Sandra. 1990. VP’s and verb movement in Chamorro. Natural Language
and Linguistic Theory 8. 559–619.
Cinque, Guglielmo. 1999. Adverbs and functional heads: A crosslinguistic perspec-
tive. Oxford: Oxford University Press.
Clemens, Lauren Eby & Jessica Coon. To appear. Deriving verb-initial word
order in Mayan. Language .
Foley, Steven, Nick Kalivoda & Maziar Toosarvandani. To appear-a. Forbidden
clitic clusters in Zapotec. Chicago Linguistic Society (CLS) 53.
Foley, Steven, Nick Kalivoda & Maziar Toosarvandani. To appear-b. Gender–
Case Constraints in Zapotec. Workshop on Structure and Constituency in Lan-
guages of the Americas (WSCLA) 22.
Folli, Raffaella, Heidi Harley & Simin Karimi. 2005. Determinants of event type
in Persian complex predicates. Lingua 115. 1365–1401.
Foreman, John Olen. 2006. The morphosyntax of subjects in Macuiltianguis Za-
potec: University of California, Los Angeles dissertation.
Grimshaw, Jane. 2005. Extended projection. In Words and structures, 1–73. Stan-
ford, CA: CSLI.
Haider, Hubert. 2006. Mittelfeld phenomena (Scrambling in Germanic). In Mar-
tin Everaert & Henk van Riemsdijk (eds.), The Blackwell companion to syntax,
vol. 3, 204–274. Oxford: Blackwell.
Holmer, Arthur. 2005. Seediq: Antisymmetry and final particles in a Formsan
VOS language. In Andrew Carnie, Heidi Harley & Sheila Ann Dooley (eds.),
Verb first: On the syntax of verb-initial languages, 175–201. Amsterdam: Ben-
jamins.
Kratzer, Angelika. 1996. Severing the external argument from its verb. In Jo-
han Rooryck & Laurie Zaring (eds.), Phrase structure and the lexicon, 109–137.
Dordrecht: Kluwer.
Lee, Felicia. 2003. Anaphoric R-expressions as bound variables. Syntax 6(1). 84–
114.
Lee, Felicia. 2006. Remnant raising and VSO clausal architecture: A case study of
San Lucas Quiaviní Zapotec. Amsterdam: Springer.
Long, Rebecca & Sofronio Cruz. 2000. Diccionaria Zapoteco de San Bartolomé
Zoogocho Oaxaca. Mexico City: Instituto Lingüístico de Verano.
López Nicolás, Oscar. 2016. Estudios de la fonología y gramatíca del zapoteco de
Zoochina: Centro de Investigaciones y Estudios Superiores en Anthropología
Social dissertation.
Marlett, Stephen A. 1993. Zapotec pronoun classification. International Journal
of American Linguistics 59(1). 82–101.
Massam, Diane. 2001. Psueo noun incorporation in Niuean. Natural Language
and Linguistic Theory 19. 153–197.
48
Massam, Diane & Carolyn Smallwood. 1997. Essential features of predication in
English and Niuean. North East Linguistic Society (NELS) 27. 263–272.
McCloskey, James. 1991. Clause structure, ellipsis and proper government in
Irish. Lingua 85. 259–302.
McCloskey, James. 1996. On the scope of verb movement in Irish. Natural Lan-
guage and Linguistic Theory 14. 47–104.
McCloskey, James. 1997. Subjecthood and subject positions. In Liliane Haegeman
(ed.), Elements of grammar, 197–235. Dordrecht: Kluwer.
McCloskey, James. 2002. Resumption, successive cyclicity, and the locality of
operations. In Samuel David Epstein & T. Daniel Seely (eds.), Derivation and
explanation in the Minimalist Program, 184–226. Malden, MA: Blackwell.
Otsuka, Yuko. 2000. Ergativity in Tongan: University of Oxford dissertation.
Pensalfini, Robert. 1995. Malagasy phrase structure and the LCA. In Robert Pen-
salfini & Hiroyuki Ura (eds.), Papers on Minimalist syntax, vol. 27 MIT Working
Papers in Linguistics, 245–260. Cambridge, MA: MITWPL.
Sonnenschein, Aaron. 2004. A descriptive grammar of San Bartolomé Zoogocho
Zapotec: University of Southern California dissertation.
Stowell, Timothy. 1981. Origins of phrase structure: Massachusetts Institute of
Technology dissertation.
Tenny, Carol L. 2000. Core events and adverbial modification. In Carol L. Tenny
& James Pustejovsky (eds.), Events as grammatical objects, 285–334. Stanford,
CA: CSLI Publications.
Toosarvandani, Maziar. 2017. On reaching agreement early (and late). In Ja-
son Ostrove, Ruth Kramer & Joseph Sabbagh (eds.), Asking the right questions:
Essays in honor of Sandra Chung, 124–138. Santa Cruz, CA: Department of Lin-
guistics, University of California, Santa Cruz.
49
An Old Irish Story of a Woman Poet in
Donegal
Liam Breatnach
Dublin Institute for Advanced Studies
In early medieval Ireland, from at least the eighth century, there is attested a
great variety of texts relating to what a learned poet should know, such as met-
rical tracts, as well as status texts, which distinguish grades of poets according
to their ability and learning.1 These texts include formal studies of grammar
and the nature of language, the best known of which is Auraicept na nÉces ‘The
Primer of the Poets’, the earliest core of which is usually dated to the eighth cen-
tury AD.2 One text in particular, Bretha Nemed Dédenach,3 transcends all these
categories, being a collection of material relating to the status of the poets as
well as a most varied selection of discussions of specific aspects of poetry such
as how to ask for rewards,4 the use of satire, metrics, how sounds and words are
produced,5 and so on, and the title given to it by its first editor, E.J. Gwynn, ‘An
Old-Irish Tract on the Privileges and Responsibilities of Poets’ is a fair enough
description of its contents.6 We know from the second Middle Irish Metrical
Tract that this text was on the curriculum of the poetic schools; the metrical
tract is arranged in the form of twelve-year syllabus of study for the fili ‘learned
poet’,7 and one of the items given in the syllabus for the fourth year is the study
of Bretha Nemed.8 The text was edited by Gwynn (1942) with an introduction
and notes, but without translation, and again by Binchy in CIH pp. 1111-32, with
numerous cross-references. Quite a few passages have since been edited with
translations and annotations, but the main obstacle to a complete edition and
1 These texts consistently distinguish between on the one hand, the fili (pl. filid), usually translated
as ‘poet’ or ‘learned poet’, a person who has undergone a formal training not only in metrics, but also
in grammar, history, genealogy, sagas, law, and specialised in all areas of secular learning (but not
to the exclusion of ecclesiastical learning), and on the other hand, the lower-status bard (pl. baird),
rendered into English as ‘bard’, a term which appears to cover every other kind of versifier, whether
literate or not.
2 The most recent edition is by Ahlqvist (1983).
3 Bretha Nemed can be translated ‘Judgements concerning Dignitaries’. There are two texts, which
mainly concern the learned poets, which have this element in their titles, and they are distinguished/
by the addition of toísech ‘first’ in one, and dédenach ‘last’ in the other; see further Breatnach (2005,
184-91).
4 See Corthals (2010).
5 See Corthals (2007).
6 He provides a more detailed account on pp. 3-6 of his edition.
7 Ed. Thurneysen (1891, 29-66). See also Murphy (1961 v).
8 Thurneysen (1891, 36 §18).
51
translation is the deliberately abstruse poetic language which characterises this
text.9
In this paper I hope to make a further small contribution to the elucidation
of this difficult text, by editing a short anecdote in Bretha Nemed Dédenach which
consists of a prose account followed by a verse piece. Only one complete copy
survives, in the seventeenth-century MS, Trinity College Dublin H 2. 15A, and it
is printed in CIH 1126.7-20. In addition six extracts from this passage are cited in
O’Davoren’s Glossary (see below).
The verse is put into the mouth of a certain Eithne, and concerns her beloved
Echaid, imprisoned on Tory Island as a result of an uprising by the servile peo-
ples of Ulster, over whom he had been appointed as deputy ruler by the king of
Ireland. Two of the names in this anecdote are well-known, viz. Cormac aue
Cuinn, a (probably legendary) pre-Christian king of Tara (and thus of Ireland),
and Fercheirtne, a mythical pre-Christian poet. On the other hand, I have come
across no other mention of Eithne daughter of Amalgaid in the extensive re-
mains of Early Irish literature. Although the same is the case for Echaid Búadach,
his father, Fergus Duibdétach, is not only frequently mentioned as a (probably
legendary) pre-Christian king from the Ulaid ‘Ulstermen’, who became king of
Ireland, but he is closely connected with Cormac in that he is represented as
his immediate predecessor, and as having been killed by Cormac in the battle of
Crinna.10 The only well-known population group mentioned is the Ulaid ‘Ul-
stermen’, and the two placenames, Tara and Tory Island, are also well-known; it
is the latter which makes the connection with County Donegal.
References to Tory, an island seven and a half miles off the coast of County
Donegal, in pre-Norman sources are few. In the Annals of Ulster for the year
617 there is a recorded a sacking of the island,11 which is dated in another set
of annals, the Annals of the Four Masters, to the year 612, and the latter annals
also mention the rebuilding of the church in the year 616.12 There is an interest-
ing tale in Immram Curaig Maíle Dúin ‘The Voyage of Máel Dúin’s Boat’, in my
view one of the best prose works of the Old Irish period, about the coic ‘cook’ of
the church of Tory, who, as the person in charge of the food supply, corruptly
enriched himself by selling the food intended for the members of the church.13
The short anecdote presented here is, then, of all the more interest because of
the geographical setting of the story. The principal archaeological remains are
those of a promontory fort and of an early medieval church foundation, includ-
ing a Round Tower, and given that the temporal setting of the anecdote is in
9 To the list in Breatnach (2005, 185) can be added the edition and translation of the sections of
the text comprising CIH 1111.19-22 in Breatnach (2004, 26-7), of CIH 1114.13-6 in Breatnach (2006,
67-8), of CIH 1125.24-7 in Breatnach (2004, 29-30), of CIH 1126.33-1129.32 in Corthals (2007), and of
CIH 1129.33-1130.37 in Corthals (2010).
10 Cf., for example, in the pre-Norman genealogies, Fergus Dubdētach mac Ḟindchada uno anno rı̄
hērenn co torchair i cath Chrinna la Cormac hūa Cuind ‘Fergus Dubdétach son of Findchad was king of
Ireland for one year, until he fell in the battle of Crinna at the hands of Cormac aue Cuinn’, O’Brien
(1962, 121).
11 Ed. Mac Airt and Mac Niocaill (1983)
12 Ed. O’Donovan (1856).
13 Oskamp (1970, 168-9 §33).
52
the pre-Christian period, the promontory fort is probably the imagined place of
detention of Echaid.
The prose introduction has the look of a set of background notes, while riam
at the end implies that she followed her rhetoric with an explanation of its im-
port. We are not, however, told what ensued. The rhetoric itself brings us to
one of the major outstanding issues in our understanding of early Irish met-
rics. While the greater part of Medieval and Early Modern Irish verse, from the
eighth century to the seventeenth, is stanzaic, with regular syllable count and
end-rhyme, we also have, especially in the Old Irish period, verse compositions
of a different structure, the most notable feature being absence of rhyme. The
question which still awaits an answer is what exactly the intended metrical struc-
ture of compositions such as that edited here was. In our case, we can at least
determine that it is not prose, not only as it is described as a díchetal ‘chant’, but
as it is marked by features of poetic language such as alliteration, hyperbaton,14
asyndeton and prepositionless datives,15 as well as by the three antithetical lines
with which it begins. It goes without saying that only with editing and translat-
ing more of these texts can we expect to come to some deeper understanding of
their metrical structure.
Although the great majority of the documented poets of the pre-Norman pe-
riod in Ireland were male, there is some evidence for female poets as well, per-
haps the most unambiguous being the Uallach whose death is noted in the An-
nals of Inisfallen under the year 934 as Quies Uallaige ingene Muinechain, banfile
Herend ‘Repose of Uallach daughter of Muinechán, poetess of Ireland’.16 The very
fact that she is mentioned in the annals at all indicates that she must have been
particularly highly regarded as a learned poet. It seems to me that the contem-
porary relevance (to eighth- and ninth-century Ireland) of this anecdote, which
is set in the prehistoric past,17 is on the one hand to stress the difficulties facing
any woman who might wish to qualify as a learned poet,18 and on the other, to
acknowledge that women could indeed be up to the task, by representing a com-
petent piece of poetry as having been composed by a woman. It is reasonable to
assume that the compositions put into the mouths of fictional characters in a
work such as Bretha Nemed Dédenach, a text which was on the curriculum of the
fili, are also intended to serve as models of composition.
As is the case with many other texts composed in the pre-Norman period,
the earliest surviving manuscript is much later than the date of composition, and
although the manuscript of this text is particularly late a number of diagnostic
Old Irish linguistic features leave us in no doubt that the text was composed
in the Old Irish period. Some of these are: the nasalising relative clause in do-
14 Such as the tmesis and preposed genitive in ro fine tonn tethraig, where in prose one would expect
ninth.
18 In this regard, the statement that Eithne went do ḟogloim éicsi ‘to study poetic learning’ marks
53
mbert, the deuterotonic verbal form with infixed pronoun Ara-ngabat-sidi, the
infixed pronoun in na mberat, the s-subjunctive (3sg. past) form of as-indet in as-
indised, the suppletive perfective form of ad-fét in ad-cuaid, and the reduplicated
preterites cachain (to canaid) and tethraig (to tráigid). The text presented here is
normalised to an Old Irish standard, with the MS text following the translation.
The only emendations (as opposed to replacing later orthography with that of
OIr) are dpl. cineluibh si to apl. cenéla-sa, gpl. mbliadhan to mblíadnae, asperattar
to as-bertatar, mionn oir to mind n-óir (similarly minn oir in O’Dav. 1563, with
omission of nasalisation after -nn), and (the most serious one) eneclann credidh
to enech ruided.19
I am offering this small piece, set in a county which has always been close to
Jim’s heart, in recognition of Jim’s other academic side, his interest in Old and
Middle Irish, in memory of our years spent together as undergraduates studying
in UCD with Conn Ó Cléirigh and Proinsias Mac Cana, and as a token of our
friendship ever since.
Eithne ingen Amalgaid maic Muiredaig, carad-side Echaid ⁊ ro-fitir a airgabáil ⁊ at-
chobair as-indised i Temair ⁊ ní ticed nech cen dán i Temair. Luid Eithne do ḟogloim
éicsi la Fercheirtne i richt gillai ⁊ ad-cuaid a scél i Temair ⁊ cachain díchetal riam:
case the first word is not written out in full, but as enecl followed by a suspension-stroke (admittedly
the usual abbreviation for eneclann).
54
held by Cormac every third year.
Eithne the daughter of Amalgaid son of Muiredach, she loved Echaid, and she
discovered that he had been taken captive, and she wished to make it known in
Tara, yet no one without an art used to come into Tara. Eithne, disguised as a
boy, went to study poetic learning with Ferchertne, and was able to tell her story
in Tara, and sang a chant before [doing so]:
The face of the pre-eminent one of the island of Ireland has been reddened,22
Tory is securing the golden diadem of the lord of the men of the Irish,23
who has been firmly fixed in fetters and chains,24 who is placed in the raiment of
immobility25
– Echaid Búadach son of Fergus Dubdétach.
between glas, a shackle for the feet, and slabrad, a chain round the neck, see Binchy (1962, 71-2).
55
(CIH 1529.25), 1563 (CIH 1529.27), and 1064 (CIH 1508.16).26 There are only four sig-
nificant variant readings; one of these, tocoisigh Torach in 1563 is superior to do
coisig Toraigh, and the others are machaidh (machad, F) in 1230, against muchadh,
teachrac (tethrach, F) in 1562, against tethraigh, and gemad in 1064, against gennadh.
References
CIH= Binchy, D. A. (1978): Corpus Iuris Hibernici (Dublin).
Ahlqvist, Anders (1983): The Early Irish Linguist. An Edition of the Canonical
Part of the Auraicept na nÉces, with Introduction, Commentary and Indices.
Societas Scientiarum Fennica, Commentationes Humanarum Litterarum 73
(Helsinki).
Binchy, D. A. (1962): ‘The Old-Irish Table of Penitential Commutations’, Ériu 19,
47-72.
Breatnach, Liam (2004): ‘On Satire and the Poet’s Circuit’, in Cathal G. Ó Háinle
and Donald Meek (eds), Unity in Diversity. Studies in Irish and Scottish Gaelic
Language, Literature and History (Dublin) 25-35.
Breatnach, Liam (2005): A Companion to the Corpus Iuris Hibernici. Early Irish
Law Series 5 (Dublin).
Breatnach, Liam (2006): ‘Satire, Praise and the Early Irish Poet’, Ériu 56, 63-84.
Breatnach, Liam (2013), ‘The Lord’s Share in the Profits of Justice and a Passage
in Cath Maige Tuired’, Celtica 27, 1-17.
Clancy, Thomas Owen (1996): ‘Women Poets in Early Medieval Ireland: Stating
the Case’ in Christine Meek and Katherine Simms (eds) ‘The Fragility of her
Sex’?: Medieval Irish Women in their European Context (Dublin) 43-72.
Corthals, Johan (2007): ‘Stimme, Atem und Dichtung: Aus einem altirischen
Lehrbuch für die Dichterschüler (Uraicept na Mac Sésa)’, in Helmut Birkhan
(ed.) Kelten-Einfälle an der Donau. Akten des Vierten Symposiums deutschsprachiger
Keltologinnen und Keltologen (Wien) 127-47.
Corthals, Johan (2010): ‘The Áiliu Poems in Bretha Nemed Dédenach’, Éigse 37,
59-91.
Gwynn, E. J. (1942): ‘An Old-Irish Tract on the Privileges and Responsibilities of
Poets’, Ériu 13, 1-60; 220-36.
Hull, Vernam (1949): ‘Miscellanea Linguistica Hibernica’, Language 25, 130-8.
Mac Airt, Seán (1951): The Annals of Inisfallen (Dublin).
Mac Airt, Seán and Mac Niocaill, Gearóid (1983): The Annals of Ulster (to A.D.
1131) (Dublin).
Murphy, Gerard (1961): Early Irish Metrics (Dublin).
O’Brien, M. A. (1962): Corpus Genealogiarum Hiberniae 1 (Dublin).
O’Donovan, John (1856): Annala Rioghachta Eireann. Annals of the Kingdom of
Ireland by the Four Masters, from the Earliest Period to the Year 1616 (Dublin).
26 References are to the paragraph numbers of the edition by Whitley Stokes (1904), with the rel-
evant page and line number of the edition in CIH 1466.11-1531.24 in parentheses. For an account of
O’Davoren’s Glossary see Breatnach (2005, 100-59).
56
Oskamp, Hans P. A. (1970): The Voyage of Máel Dúin. A Study in Early Irish
Voyage Literature (Groningen).
Stokes, Whitley (1904): ‘O’Davoren’s Glossary’, Archiv für celtische Lexikogra-
phie 2, 197-504.
Thurneysen, Rudolf (1891): ‘Mittelirische Verslehren’, Irische Texte 3, 1. Heft (Leipzig)
1-182.
57
Maximize Presupposition and Types of
Indefinites in Chamorro*
Sandra Chung
University of California, Santa Cruz
1 Introduction
It is by now widely accepted that natural language allows a range of types of in-
definites. For instance, some types of indefinites are scopally unrestricted; others
must have wide scope with respect to semantic operators; still others must have
narrowest scope, or must be within the scope of an operator in order to occur at
all (see e.g. Matthewson 1999 on St’át’imcets). Within semantic theory, particular
attention has been devoted to the types of indefinites that must have narrow-
est scope (henceforth narrow-scope indefinites). Chung and Ladusaw (2004 and
2006; henceforth C&L) account for such indefinites by enriching the inventory of
semantic composition operations that combine predicates with their arguments.
Specifically, they propose that a predicate can be combined with the descriptive
content of an indefinite by Restrict, a composition operation that does not satu-
rate the targeted relatum of the predicate but merely narrows its domain. C&L
make no attempt to connect the limited scope potential of such indefinites to
other aspects of their semantic-pragmatic profile. However, languages such as
Hungarian, Spanish, and Catalan have bare singular NPs that must have narrow-
est scope, are semantically number-neutral, and make no contribution to dis-
course dynamics. This constellation of properties has led Farkas and de Swart
(2003) and Espinal and McNally (2009) to propose accounts in which these defi-
ciencies in scope, number, and discourse contribution are intertwined. The ac-
counts are inspired by the leading idea—traceable to Van Geenhoven’s (1998) sem-
inal work on semantic incorporation—that narrow-scope indefinites are more
* All love to Jim, to whom I dedicate this paper, in recognition of many years of intellectual en-
gagement and companionship. Many thanks to the Chamorro speakers who contributed to this work,
especially Manuel F. Borja, Dr. Elizabeth D. Rechebei, Tita A. Hocog, the late Dr. Rita H. Inos, the late
Maria P. Mafnas, Maria T. Quinata, Francisco Tomokane, the late Anicia Q. Tomokane, and the mem-
bers of the Chamorro Dictionary Revision Project. Thanks also to Michela Ippolito, who pointed out
to me some years ago the relevance of sentences like (30a) for MP. I am indebted to Louise McNally,
Chris Barker, and Deniz Rudin for detailed, insightful commentary, and to Pranav Anand for brief,
perceptive remarks; their comments led me to significantly reconfigure an earlier draft. The research
reported here was supported in part by NSF project BCS-0753240 to the University of California,
Santa Cruz.
59
limited than other types of indefinites along multiple dimensions.
It is worth asking whether narrow-scope indefinites have this deficient char-
acter across languages. Here I offer some novel evidence that suggests that they
do not. The evidence comes from Chamorro, an Austronesian language of the
Mariana Islands.
Alongside the definite article i, Chamorro has two indefinite articles that dif-
fer in scope possibilities. One indefinite article, un, has roughly the same range of
scope options as English a. The other indefinite article, which is unpronounced,
must have narrowest scope. Although I contend that indefinites formed from
the null indefinite article are DPs, they have a realization indistinguishable from
bare NPs. That might lead one to wonder to what extent they fit the profile of
narrow-scope indefinites in Hungarian, Spanish, or Catalan.
I first show that Chamorro indefinites formed from the null indefinite ar-
ticle have semantic number and set up discourse referents that can be referred
to subsequently. In these respects, they appear no more limited than indefinites
formed from un. The rest of the discussion documents a pragmatic dimension
along which the null indefinite article has a less limited distribution than un. This
dimension involves Maximize Presupposition (MP), the pragmatic principle that
urges the speaker to ‘Make your contribution presuppose as much as possible!’
(Heim 1991: 514-515).1 Heim originally postulated MP to account for two effects,
which have been called antipresupposition (following Percus 2006) and presupposi-
tional implicature (following Leahy 2016). The indefinite article un exhibits both
effects. But, surprisingly, the null indefinite article exhibits the second effect, but
not the first.
There is no consensus on the precise formulation of MP, whether it can be
made to follow from Grice’s maxims, or even whether the effects attributed to
it should be explained by the same principle (see e.g. Percus 2006, Singh 2011,
Schlenker 2012, Leahy 2016, and Lauer 2016 for a range of views). However, many
discussions of MP are framed partly in terms of presuppositional scales. Pairs of
lexical items that “differ minimally” (Lauer 2016: 980) in carrying, or not carrying,
a particular presupposition are viewed as arranged in a scale ordered by presup-
positional strength. MP directs speakers to choose the alternative that employs
the strong rather than the weak member of the scale—the alternative that pre-
supposes more—if they can.
In Chamorro, the definite article carries a uniqueness presupposition that the
indefinite articles lack. So we expect these articles to form a presuppositional
scale in which the definite article is strong and the indefinite articles are weak;
this is what happens for presuppositional implicature.2 Why is the null indefinite
article ignored for antipresupposition? I suggest that part of the answer lies in
C&L’s notion of mode of composition. From the perspective of semantic com-
position, the definite article and un are what Horn and Abbott (2014: 334) would
call “natural paradigmatic alternatives”: they are type-shifters that enable the de-
1 In the original:
“Präsupponiere in deinem Beitrag so viel wie möglich!” (Heim 1991: 515).
2 Cf.Hawkins (1991: 426) and Horn and Abbott (2014), who propose that the and a form a scale for
the purposes of scalar implicature.
60
scriptive content of DP to be composed with the predicate via Function Appli-
cation. In contrast, the null indefinite article signals that the descriptive content
of DP should be composed directly with the predicate via Restrict. This, I claim,
is enough to explain why the null indefinite article does not compete with the
definite article for antipresupposition purposes.
Like C&L, Farkas and de Swart (2003) and Espinal and McNally (2009) claim
that the descriptive content of a narrow-scope indefinite is composed directly,
as a property, with the predicate. This means that as far as antipresupposition
is concerned, their accounts could make the same cut among Chamorro’s three
articles as C&L. But to the extent that these other accounts are designed to deliver
narrow scope together with number neutrality and discourse inertia, they are
not well-suited to handle the null indefinite article’s full profile. For this reason,
I maintain, C&L’s approach is more appropriate here.
Section 2 introduces Chamorro and the three articles that are the focus of
investigation. Section 3 sketches some ways of accounting for definites and in-
definites in this language. It also presents evidence that the limited scope po-
tential of the null indefinite article is not correlated with number neutrality or
discourse inertia. Then, section 4 introduces MP and the effects attributed to it.
Section 5 shows that the indefinite articles behave as expected for presupposi-
tional implicature: in contexts in which the definite article’s uniqueness presup-
position is not already known to be satisfied, use of an indefinite article conveys
that the speaker believes that the extra information that the definite article would
have communicated is false.3 Sections 6 and 7 deal with antipresupposition. Sec-
tion 6 zeroes in on the use of articles in possessives. Chamorro possessives differ
from English possessives like Meg’s cat in that the article and the possessor co-
occur and co-vary freely. I first confirm that Chamorro possessives formed from
the null indefinite article have the form and meaning of indefinites, even when
their possessor is definite. I then show that for possessed nouns for which it is
common knowledge that the possessee is unique relative to the possessor (e.g.
gui’ing ‘nose’, nåna ‘mother’), the two indefinite articles pull apart: un displays an-
tipresupposition effects—its use is infelicitous—but the null indefinite article can
be used felicitously. Section 7 uncovers a similar pattern in the use of articles with
nouns whose intended referent is commonly understood to be unique in the real
world (e.g. åtdao ‘sun’). After suggesting an account of these patterns, section 8
concludes with some general remarks about the typology of narrow-scope in-
definites and its connection to the semantics of noun incorporation.
516), the proposition expressed by the corresponding sentence with the definite article is false (or not
known by the speaker to be true).
61
of Guam, and by numerous Chamorros in the continental U.S. The language has
undergone rapid decline in the last half-century, and is now widely believed to
be on the cusp of endangerment.
The language is head-initial and allows a range of null arguments. Clauses
consist of a predicate, which can be a verb, noun, adjective or preposition, fol-
lowed by the predicate’s arguments and adjuncts. When the predicate is a verb
or adjective, the relative order of arguments and adjuncts is flexible, but the un-
marked, most frequent word order is: Verb/Adjective Subject Object Other (see Chung
1998). The inflected verbs are underlined in (1).4
(1) Ha konni’ si Orasima’ i haggan, ya ha po’lu gi buti-ña.
agr take unm Orasima’ the turtle and agr put loc boat-agr
‘Orasima’ took the turtle, and he put it in his boat.’ (from a narrative)
DPs formed from common nouns consist of a determiner (D) followed by an NP
constituent consisting of the noun, its complements, and modifiers. The noun
(underlined in (2)) precedes its complements, but can be preceded or followed by
modifiers.
(2) i ottimu na istoria ni guaha ta’lu sustansiån-ña
the last l story comp agr.exist again substance-agr
‘the last story which again has substance’ (EM 99)
The Ds include quantifiers, demonstratives, and three articles: the definite article
i, the indefinite article un, and the null indefinite article.
At first blush, the three articles have uses broadly similar to the uses of the
definite and indefinite articles in English. The definite article i is used when the
speaker and hearer can uniquely identify the intended referent of DP. In (3a), for
instance, i is used because there is a unique sun in the world. In (3b), from a story,
the old woman is the only individual who has been previously mentioned who is
both female and old. In (3c)—an instruction from a psycholinguistic experiment
involving a computer tablet—there is only one star displayed on the tablet screen.
Finally, in (3d), it is reasonable to infer that the engine is unique relative to the car
under discussion. (The relevant DPs are underlined.)
(3) a. Dokku’ i atdao.
agr.sprout the sun
‘The sun rose.’
4 The Chamorro examples are cited in the official orthography now used in the CNMI. In this
orthography, possessor agreement is separated from the rest of the word by a hyphen. The words
otherwise have not been explicitly decomposed into morphemes. The glosses employ the following
abbreviations: agr = agreement, ap = antipassive, comp = complementizer, emp = emphatic, fut =
future, infin = infinitive, l = linker, loc = local case, obl = oblique case, pass = passive, pl = plural,
prog = progressive, q = question, unm = unmarked case, wh = wh-agreement. Naturally-occurring
examples are taken from: Borja et al. 2006 (EM; a book of stories, essays, and poetry), Cooreman 1982
and 1983 (transcriptions of tape-recorded narratives), Marciano n.d. (a children’s book), the database
for the Revised Chamorro-English Dictionary (CD; illustrative examples created by community mem-
bers for dictionary entries), and the Chamorro New Testament (NT; translated into Chamorro by a
group led by Bishop Tomas A. Camacho). Examples not attributed to a source are from my fieldwork.
62
b. Nina’gogof maguf i biha kumu guaha bisitå-ña.
agr.pass.make.very happy.prog the old.lady if agr.exist visitor-agr
‘The old woman was made very happy when she had visitors.’ (Mar-
ciano n.d.: 1)
c. Chonnik i puti’un guatu gi ...
push the star to.there loc
‘Push the star over to ... [the picture that fits the description].’
d. Ti siña masugun i kareta sa’ mayulang i makina.
not can agr.pass.drive the car because agr.pass.break the engine
‘The car can’t be driven, because the engine is broken.’
DPs formed from either indefinite article display quantificational variability. The
most natural interpretation of (5a) is that each child received a different bunch of
bananas, and of (5b), that each house should have a different exit.
(5) a. Ha dispåtta si nåna i rasimun aga’ ya ha påtti kada
agr separate unm mother the stalk.l banana and agr apportion each
patgon-ña un iting.
child-agr a bunch
‘Mother divided the banana stalk and gave each child of hers a bunch.’
(CD, entry for dispåtta)
b. Gi kada guma’ debi di u guaha sagan fanhuyungan.
loc each house should agr exist place.l exit
‘In every house there has to be an exit.’ (CD, entry for fanhuyungan)
Again following Roberts (2003), I assume that the fundamental difference be-
tween i and the indefinite articles is that i carries a uniqueness presupposition,
but the indefinite articles do not.
63
The rest of the discussion takes it for granted that Chamorro has a three-
way article system that includes two indefinite articles, one of which is not pro-
nounced. Before going further, I should perhaps reiterate the claim that the null
indefinite article exists as a D in its own right, as opposed to being an unpro-
nounced form of one of the other articles or simply not instantiating a syntactic
category at all. (In the latter scenario, what I have been calling ‘DPs formed from
the null indefinite article’ would be bare NPs.) Some evidence supporting this
claim will emerge in section 5. A selection of other evidence is offered below:
(i) Out of context, DPs formed from un or the null indefinite article can serve
as the pivot of an affirmative existential sentence, but DPs formed from i cannot
(see Chung 2006). This familiar pattern reveals the null indefinite article is not a
phonologically reduced form of i.
(ii) DPs formed from un typically have wide scope with respect to negation (C&L
2004: 100-102), but DPs formed from the null indefinite article always have nar-
rowest scope. Moreover, only DPs formed from the null indefinite article can be
the pivot of a negative existential sentence. This pattern, shown in (7a), reveals
that the null indefinite article is not a phonologically reduced form of un. (In ad-
dition, (7b) shows that out of context, the pivot of a negative existential sentence
cannot be a DP formed from i.)
(iii) Chamorro has a syntactic topic position at the left edge of the clause (Chung
1998). DPs formed from i or un can occupy this position, but DPs formed from
the null indefinite article cannot. This is another reason for distinguishing the
null indefinite article from the other two articles.
64
(8) a. I taotao ha oddu’ i balutan magågu gi ilu-ña.
the person agr carry the bundle.l clothes loc head-agr
‘The man carried the bundle of clothes on his head.’ (CD, entry for
oddu’)
b. Parehu yan i simiyan muståsa ni un tåotao ha chuli’ ...
agr.similar with the seed.l mustard comp a person agr take
‘It is like a mustard seed, which a man took....’ (NT 133)
c. * Tåotao gai patgun un låhi.
person agr.have child a boy
(A man had a son.)
(iv) Finally, like other DPs, indefinites formed from the null indefinite article
can have a possessor (see section 6). Possessors in Chamorro occur high in the
structure of DP, outside the NP constituent consisting of the noun, its comple-
ments, and modifiers. This is why noun incorporation, which incorporates NPs
in Chamorro (see (9a)), cannot incorporate an NP accompanied by a possessor
(see (9b-c) and C&L 2004: 85-88).
The fact that indefinites of the type shown in (4b), (5b), (6b), and (7) can host a
possessor reveals that they are constituents larger than NP. More precisely, they
are DPs formed from a D that is not pronounced—the D referred to here as the
null indefinite article.
65
descriptive content of the indefinite to narrow the domain of the predicate’s re-
latum. For Chamorro, this amounts to saying that the null determiner is a se-
mantically vacuous identity element that composes with the meaning of NP to
yield a DP that denotes a property. Restrict then applies to compose this property
directly with the predicate’s relatum. The analysis subtree in (10) illustrates how
Restrict combines the meaning of the property-denoting DP påtgun ‘child’ with
the meaning of the predicate humuyung ‘come out, emerge’ in the semantic com-
position of (4b). (I assume the predicate relation supplied by the verb includes a
Davidsonian event argument.)
(10)
λx λe[emerge’(x)(e)] child’
λx λe[emerge’(x)(e) ∧ child’(x)]
Note that the resulting expression is just as semantically incomplete as it was be-
fore Restrict applied: the value of the targeted relatum has not yet been fixed.
C&L (2004: 11-12) assume that if the targeted relatum is not saturated by further
predicate-argument composition, the incompleteness is remedied by existential
closure, which occurs early enough in the compositional process to ensure nar-
rowest scope.
Farkas and de Swart (2003) develop an approach to semantic incorporation
that links the narrow scope of bare NPs to number neutrality and the failure
to contribute a discourse referent. In their account, which is framed in Dis-
course Representation Theory, a predicate’s thematic arguments normally com-
bine with the meaning of argument nominals by instantiation: the predicate’s
thematic argument is replaced by the discourse referent that the argument nom-
inal introduces. However, only nominals that are specified for semantic num-
ber can introduce discourse referents. When the argument nominal introduces
no discourse referent, its descriptive content is unified with the predicate’s the-
matic argument, forming a complex predicate; this is what gives rise to narrowest
scope.
Finally, Espinal and McNally (2009) construct an account of bare noun ob-
jects in Spanish and Catalan that treats them semantically as verb modifiers. In
their account, a lexical rule suppresses the verb’s internal argument, thereby en-
suring that this argument contributes no discourse referent. The lexical rule
does not, however, block the entailment that two participants are involved in
the event. Verbs whose internal argument has been suppressed are composed
with the descriptive content of a bare noun complement by a special rule of in-
tersective modification. The fact that the descriptive content of the bare noun
combines directly with the verb meaning ensures that it has narrowest scope.
Despite certain recurring themes, there are substantial differences among
these accounts. I want to focus here on the claim that certain types of indefi-
nites—or, perhaps, bare NPs in general—are simultaneously deficient in scope,
66
number, and discourse dynamics. Is this the profile of narrow-scope indefinites
(or bare NPs) crosslinguistically?
Chamorro’s narrow-scope indefinites can be viewed as bare NPs, so the ques-
tion can be raised about them. As a matter of fact, these indefinites do not appear
to be deficient in number or discourse contribution. In these respects, they re-
semble indefinites formed from un.
To begin with, indefinites formed from the null indefinite article are not se-
mantically number-neutral, although it requires a bit of effort to see this. The vast
majority of Chamorro nouns show number inflection only optionally. However,
the language has six nouns that must be inflected (irregularly) for number: låhi
‘man’, palåo’an ‘woman’, påtgun ‘child’, saina ‘parent’, che’lu ‘sibling’, and påli’ ‘priest’.
Indefinites formed by combining these nouns with the null indefinite article are
construed as singular if the noun is in the unmarked form, and as plural if the
noun is in the plural form; they do not have a number-neutral interpretation.5
Further, indefinites formed from the null indefinite article can set up discourse
referents that can be referred to subsequently, just like indefinites formed from
un. This can be seen from the examples in (12), which are taken from narratives. In
(12a), an indefinite formed from un serves as the antecedent of a definite DP later
in the discourse (cf. (4a)); in (12b), an indefinite formed from the null indefinite
article serves as the antecedent of a DP formed from a demonstrative (cf. (4b));
and in (12c), an indefinite formed from the null indefinite article serves as the
antecedent of two (null) pronouns—the object of the transitive verb gu’ut ‘grasp’
and the subject of malingu ‘disappear’.
5 Some details are glossed over because they are irrelevant. E.g. che’lu ‘sibling’ is inflected for sin-
gular, dual, and plural number; the other five nouns are inflected for nonplural versus plural number.
For these five nouns, the nonplural (unmarked) form can be construed as singular or dual; the dual
construal arises if and only if the noun is combined with the numeral dos ‘two’.
67
b. Anai ma baba, humuyung påtgun ... Pues esti na påtgun
when agr open agr.come.out child then this l child secret
sikretu mo’na sigi ha’ di ha poksai.
forward keep.on emp agr raise
‘When they opened it, a child came out....So this child they kept on
raising secretly from then on.’ (Cooreman 1983: 107)
c. Siempri guaha nai manå’i hao dangis hålum ya gigun
surely agr.exist comp agr.pass.give you candle inside and as.soon.as
un gu’ut, malingu.
agr grasp agr.disappear
‘Surely there will be times when you are given a candle inside and as
soon as you grasp it, it disappears.’ (Cooreman 1983: 4)
These patterns suggest that Chamorro’s narrow-scope indefinites are better han-
dled by C&L’s account than by Farkas and de Swart’s or Espinal and McNally’s. I
will assume this for now, returning to the issue at the end.
A more familiar range of options is available for the compositional semantics
of DPs formed from i or un. For instance, i could be a type-shifter from prop-
erties to individuals (e.g. Partee’s (1987) iota) or from properties to generalized
quantifiers (e.g. a version of Partee’s (1987) the in which existence and uniqueness
are presupposed; see Coppock and Beaver 2015: 383); un could be a type-shifter
from properties to individuals (e.g. a choice function; see Reinhart 1997, Winter
1997, Kratzer 1998, C&L 2004) or from properties to generalized quantifiers (e.g.
Partee’s (1987) a). I believe it is unimportant for my purposes which of these op-
tions is adopted. What matters is that i and un are type-shifters that enable the
descriptive content of DP to be composed with the predicate’s relatum by Func-
tion Application. This differentiates them from the null indefinite article, which
signals that the descriptive content of DP is composed with the predicate’s rela-
tum by Restrict.
With this much in place, I now move on to MP, which provides a dimen-
sion along which indefinites formed from the null indefinite article are less con-
strained than indefinites formed from un.
68
(13) #A weight of our tent is under 2 kilos. (Heim 1991: 51)
Second, in contexts in which the uniqueness presupposition carried by the defi-
nite article is not already known to be satisfied, use of the indefinite article impli-
cates that the speaker believes that the extra information that the definite article
would have communicated is false (see note 3). This is what Leahy (2016) calls
presuppositional implicature. In (14), for instance, the use of a pianist rather than the
pianist aggressively invites the inference that the pianist who Richard had a beer
with is not the pianist of the Beaux Arts Trio.
(14) Richard heard the Beaux Arts Trio last night and afterwards had a beer
with a pianist. (Heim 1991: 515)
Although Heim noted that the condition at work in (13-14) is reminiscent of scalar
implicature, she claimed that it could not be accounted for in the same way. Her
reason was that scalar implicatures are usually derived from Grice’s maxim of
quantity (‘Make your contribution as informative as is required’), but given that it
is common knowledge that each concrete object has a unique weight, an example
like (13) is not less informative than its felicitous counterpart (15).
(15) The weight of our tent is under 2 kilos.
She suggested that the privileging of the definite over the indefinite article might
instead follow from a new conversational maxim, ‘Make your contribution pre-
suppose as much as possible!’—the maxim now known as MP.
There has been an explosion of research on MP since the turn of the cen-
tury. Its empirical domain has been widened to include other pairs of lexical
items that differ in presuppositional strength (e.g. both and all, know and believe;
see Percus 2006), as well as certain types of inflectional morphology (e.g. tense,
number agreement; see Sauerland 2003 and 2008). Considerable effort has been
devoted to crafting a more precise formulation of MP and integrating it in one
way or another into a Gricean model of reasoning (see e.g. Schlenker 2012, Leahy
2016, Lauer 2016). Other research has probed the question of whether the alterna-
tives compared by MP are lexical items or more complex expressions consisting
of clauses or sentences plus their interpretations (see e.g. Percus 2006, Singh 2011,
Schlenker 2012, Lauer 2016, Collins 2016).
The goal of this discussion is to use the interplay between Chamorro’s article
system and MP to provide evidence for C&L’s modes of composition. Because I
am interested primarily in the effects attributed to MP, I will not need to com-
mit to any formulation of it more precise than Heim’s. In section 8, however, I
will briefly engage with some larger questions that emerge from the Chamorro
material investigated here.
5 Presuppositional Implicature
Given that Chamorro i carries a uniqueness presupposition that the indefinite
articles lack, we expect the indefinite articles to exhibit presuppositional impli-
69
cature effects. In contexts in which i’s uniqueness presupposition is not already
known to be satisfied, use of an indefinite article should implicate that, for all the
speaker knows, the extra information that i would have communicated is false.
This is in fact what happens. Consider a discourse in which the speaker utters
(16), followed by one of the sentences in (17), which differ only in the article used
to form the underlined DP.
An utterance of (17a), with the definite article, can be felicitously used to report
that the speaker saw the bride from the wedding s/he attended earlier that day.
But an utterance of (17b), with un, implies that the speaker saw a bride from a
different wedding, or perhaps the wedding that the speaker attended involved
multiple brides. (As one consultant said, “We are assuming here several weddings
happened or it could be a polygamous wedding!”) That is, (17b) implicates that
there is more than one bride in the domain of discourse. An utterance of (17c),
with the null indefinite article, has the same non-uniqueness implication. (The
consultant commented, “Interesting sentence. Translation: Afterwards, I sighted
(a) bride (could be any bride)”.)
A discourse in which the speaker utters one of the sentences in (18) gives rise
to similar effects. Note that måkina means ‘engine, machine’.
70
The most natural understanding of an utterance of (18a), with the definite article,
is that the speaker drove the car and then the engine of that car broke. How-
ever, an utterance of (18b), with un, implies that a machine broke which might
or might not be the engine of the car the speaker drove. (The consultant com-
mented, “Could mean one part of the car’s engine broke or some other machine
broke, or could be [that] a machine he was transporting in the car broke.”) An
utterance of (18c), with the null indefinite article, licenses this inference as well.
(“Not clear which machine.”)
A comment is in order about the status of (17c) and (18c). In Chamorro, DPs
formed from the null indefinite article routinely serve as arguments in naturally-
occurring discourse and in sentences volunteered by speakers in fieldwork ses-
sions. Consider the sentences below, which are parallel in the relevant respects
to (17c) and the last clause of (18c).
6 The oblique case marker ni merges with the definite article i in (20a) and is unpronounced when
the DP is indefinite.
71
(20) a. Ya gi talu’åni manbisita yu’ ni nobiu.
and loc afternoon agr.ap.visit I obl.the groom
‘And in the afternoon I visited with the groom [= the groom from the
wedding earlier that day].’
b. Ya gi talu’åni manbisita yu’ un nobiu.
and loc afternoon agr.ap.visit I a groom
‘And in the afternoon I visited with a groom.’ [It might or might not
be the groom from the wedding earlier that day.]
c. Ya gi talu’åni manbisita yu’ nobiu.
and loc afternoon agr.ap.visit I groom
‘And in the afternoon I visited with a groom.’ [It might or might not
be the groom from the wedding earlier that day.]
However, not all speakers are comfortable with mini-discourses like (17-20a), in
which the internal argument of the antipassive verb is definite, even though such
constructions are grammatical elsewhere. The loose trade-off between voice,
definiteness, and salience seen above exhibits considerable individual variation;
it clearly deserves further study.)
Abstracting away from the intricacies, the presuppositional implicature ef-
fects in (17-18) reveal that the use of definite and indefinite articles in Chamorro
conforms to MP as expected. I provisionally take this to mean that the three ar-
ticles are arranged in a presuppositional scale in which the definite article i is the
strong member and either un or the null indefinite article is the weak member.
It will be convenient later for me to decompose this into two simpler presuppo-
sitional scales, one consisting of i and un and the other consisting of i and the
null indefinite article. In Horn’s (2001[1989]: 231) formalism, in which the stronger
member of the scale occurs to the left of the weaker member, these scales are <i,
un> and <i, null>.
Over and above this, we have now arrived at another argument that the null
indefinite article exists in the first place (see section 2). In the literature on MP, the
members of presuppositional scales are sometimes assumed to be lexical items
that are members of the same syntactic category (see Percus 2006 for explicit
discussion). If we adopt this assumption, then the fact that the null indefinite
article forms a presuppositional scale with i is evidence that it is a lexical item
and, further, belongs to the same category as i. It is a D that happens not to be
pronounced, in other words.
6 Antipresupposition, Part 1
The claim that Chamorro articles conform to MP leads to the expectation that the
indefinite articles should exhibit antipresupposition effects. Here the empirical
patterns are more complex, so the discussion is divided into two parts. I begin by
exploring antipresupposition effects involving nouns whose meaning, together
with common knowledge, communicates that the possessee is unique relative to
72
the possessor (e.g. gui’ing ‘nose’). But for the investigation to get off the ground,
some background must be installed about the form and meaning of Chamorro
DPs that contain a possessor, which—following Barker 1991—I call possessives.
D and the possessor coexist and covary freely. Consider the examples below,
which make the point that a possessive can be formed from any article—i, un, or
null—and, moreover, the possessor can be any type of DP. (The overt articles in
(22) are in bold-face, and pronoun possessors are represented as pro; see note 7.)
the choice between the linker and possessor is free. However, possessors that are pronouns must be
null, and the possessed noun must agree with them.
73
f. che’chu’ [un tåotao]
work.l a person
‘work of a (i.e. one) person’ (heard at a meeting)
g. patgun [tåotao]
child.l person
‘a child of a person’
Just as for other DPs, the definiteness of a possessive is determined by the defi-
niteness of the D from which it is formed. I take this to be self-evident for posses-
sives formed from the articles un. and un. (For some evidence, read on to section
6.2.) One might wonder about the status of possessives formed from the null
indefinite article, given Barker’s (2011) claim that English possessives like Meg’s
cat, which have no overt article, inherit their familiarity and uniqueness from
their possessor. (Relatedly, it has sometimes been claimed that English prenom-
inal possessives are definite; see e.g. Jensen & Vikner 2002: 200-201 and, for
discussion, Barker 2000 and 2011). However, there is abundant evidence that in
Chamorro, a possessive formed from the null indefinite article has the morphosyntax
and semantics of an indefinite, even when the possessor is definite. Chamorro pos-
sessives do not inherit their definiteness from the possessor, in other words (see
Chung 2006). Here is some of the evidence:
(i) A possessive formed from the null indefinite article can serve as the pivot
of an existential sentence (see (23a-b)), but a possessive formed from the definite
article cannot (see (23c)).
(23) a. Guaha difekton-ña [i adding [i taotao]].
agr.exist defect-agr the leg.l the person
‘The person’s leg has a defect (lit. There is a defect of the leg of the
person).’ (EM 133)
b. Tåya’ patgon-ña [pro].
agr.not.exist child-agr
‘There isn’t a child of hers / She doesn’t have a child.’
c. * Tåya’ i patgon-ña [pro].
agr.not.exist the child-agr
(There isn’t the child of hers.)
(ii) A possessive formed from the null indefinite article cannot occupy the syn-
tactic topic position at the left edge of the clause (see (24a)). However, a possessive
formed from the definite article can (see (24b)).
(24) a. * Ga’-mu [pro] ha na’dånu i gualu’.
pet-agr agr cause.damage the garden
(A dog of yours destroyed the garden.)
b. I ga’-mu [pro] ha na’dånu i gualu’.
the pet-agr agr cause.damage the garden
‘Your dog destroyed the garden.’
74
(iii) Possessives formed from the null indefinite article exhibit quantificational
variability, but possessives formed from the definite article do not, or do so with
greater difficulty. The most immediate interpretation of (25a) is that a different
child of hers gets sick on different occasions, whereas the most immediate inter-
pretation of (25b) is that the same child gets sick on multiple occasions.
(iv) Possessives formed from the null indefinite article are nonreferring in opaque
contexts (see (26a)). However, possessives formed from the definite article are
referring even in these contexts (see (26b)).
(v) Finally, question-answer pairs reveal that a possessive formed from the null
indefinite article does not presuppose uniqueness (maximality). Consider a sce-
nario in which B and C both know that Antonio has siblings, but only C knows
that some of Antonio’s siblings are smart and some are not smart. Suppose that
B asks the question in (27a), with a possessive formed from the null indefinite
article. When asked how C would respond, speakers volunteered the responses
in (27b) as the first or most natural answer. (Other answers are possible.) These
responses reveal that the question was understood to be about one or more of
Antonio’s siblings.8
icates, but only when their possessor is strong; see Chung 2006.
75
Now suppose that B instead asks the question in (28a), with a possessive formed
from the definite article. When asked how C would respond, the same speakers
volunteered (28b). Here, the responses reveal that the question was understood
to be about all of Antonio’s siblings.
The contrast between (27) and (28) is evidence that Chamorro possessives do not
inherit uniqueness from their possessors. Instead, possessives formed the defi-
nite article presuppose uniqueness; possessives formed from the null indefinite
article do not.
The possessor in a Chamorro possessive can bear any semantic relation at all
to the (possessed) noun. Nonetheless, the discussion here focuses on relational
nouns—nouns that denote a relation identical to what Barker (1991 and 2011) calls
the possession relation.
But, surprisingly, the use of the null indefinite article is felicitous. In fact, the null
indefinite article occurs quite often in this context, as the following naturally-
occurring examples are intended to illustrate.
76
(30) a. Kumåtma i bongbung kurason-ña [pro].
agr.calm the beat.l heart-agr
‘The beating of her heart (lit. a heart of hers) calmed down.’ (EM 82)
b. Kada ogga’an, hu mokmuk pachot-tu [pro].
each morning agr rinse mouth-agr
‘Every morning, I rinse my mouth (lit. a mouth of mine).’ (CD, entry
for mokmuk)
c. Chamoru nanå-hu [pro], Chamoru tatå-hu [pro].
Chamorro mother-agr Chamorro father-agr
‘My mother (lit. a mother of mine) is Chamorro, my father (lit. a father
of mine) is Chamorro.’ (from a conference presentation)
d. Manågu’ na u ma’utut aga’gå’-ña [si Juan] gi presu.
agr.ap.order comp agr pass.cut throat-agr unm Juan loc prison
‘He ordered that John’s throat (lit. a throat of John) be cut in the prison.’
(NT 27)
e. Singku bibenda linekkå’-ña [atyu na guma’].
five storeys height-agr that l house
‘The (lit. a) height of that building is five storeys.’ (CD, entry for linekka’)
f. Anåkku’ dadalak-ña [i hafula’].
agr.long tail-agr the manta.ray
‘The (lit. a) tail of the manta ray is long.’ (CD, entry for hafula’)
g. Håfa na mampus amariyu kulot-mu [pro]?
what? comp too.much agr.yellow color-agr
‘Why is your color (lit. a color of yours) so yellow?’ [addressed to a
canary] (EM 82)
I should emphasize that the possessives in (30) can be shown to be indefinite by
the same sorts of evidence I used in section 6.1 to make this point more generally
for possessives formed from the null indefinite article. Even when the relational
noun’s meaning, together with common knowledge, communicates that the pos-
sessee is unique relative to the possessor, the possessive can be the pivot of an ex-
istential sentence, cannot occupy the syntactic topic position, and is nonreferring
in opaque contexts (see the Appendix for some relevant examples). Moreover, the
use of the definite article is also felicitous in this context, just as MP leads us to
expect. Compare (30a), (30c), and (30f) with the examples below.
(31) a. Håfa na ti pasifiku i kurason-ña [pro].
whatever comp not agr.peaceful the heart-agr
‘For whatever reason, his heart is not calm.’ (EM 128)
b. U niñukut ni tilipas apuya’ i nanå-ña [pro].
agr pass.strangle obl umbilical.cord the mother-agr
‘It will be strangled by the umbilical cord of its mother.’ (CD, entry for
chathinenggi)
77
c. Hingao i dadalak-ña [i ga’lågu].
agr.hairless the tail-agr the dog
The tail of the dog is hairless.’ (CD, entry for dádalak2 )
What has changed the playing field? Descriptively, it looks like the scale <i, un>
remains in force for antipresupposition, but for some reason, <i, null> has been
suspended.
The idea that the null indefinite article does not form a scale with i for an-
tipresupposition purposes is supported by the discourse patterning of posses-
sives of types (30) and (31). When it is common knowledge that the possessee is
unique relative to the possessor, a possessive that explicitly introduces the dis-
course referent corresponding to the possessee can be formed from the null in-
definite article or from i.
(32) a. Ti siña masugun i kareta, sa’ mayulang makinå-ña [pro].
not can agr.pass.drive the car bec. agr.pass.break engine-agr
‘The car can’t be driven, because its engine (lit. an engine of it) is broken.’
b. Ti siña masugun i kareta, sa’ mayulang i makinå-ña [pro].
not can agr.pass.drive the car bec. agr.pass.break the engine-agr
‘The car can’t be driven, because its engine (lit. the engine of it) is broken.’
78
These patterns suggest that it would not work to try to claim that Chamorro
has two homophonous definite articles, one that differs from un in presupposing
uniqueness and another that differs from the null indefinite article in presup-
posing familiarity (along the lines suggested for Spanish by Alonso-Ovalle et al.
2011). Rather, in antipresuppositioncontexts, the null indefinite article seems not
to enter into competition with the definite article at all.10
7 Antipresupposition, Part 2
Another antipresupposition context is supplied by nouns whose intended refer-
ent is commonly understood to be unique in the real world, such as åtdao ‘sun’ or
långit ‘sky’. Once again, because i’s uniqueness presupposition is already known
to be satisfied, MP leads us to expect that the use of an indefinite article should be
infelicitous. This expectation is realized for un, as can be seen from the minimal
pair below.
What about the null indefinite article? No clear information about (in)felicity
emerges from sentences like (35), because these sentences are simply rejected as
ungrammatical.
79
(37) # Kumahulu’ un dångkulu na åtdao.
agr.go.up a big l sun
‘A big sun went up (higher in the sky).’ [Infelicitous, according to an-
other consultant, because “we all know that there is only one sun (for
earth, that is).”]
One way of describing the pattern in (34-37) is to say that when it is common
knowledge that the noun’s intended referent is unique in the real world, the scale
<i, un> remains in force, but <i, null> is suspended as long as the null indefinite
article is grammatical to begin with.11 Once the issue of grammaticality is factored
in, the pattern replicates what was seen in section 6.2. This suggests that <i, null>
is suspended for antipresupposition in general.
I do not know why (35) is ungrammatical. However, I should point out that
there are naturally-occurring examples in which the null indefinite article is used
with an unmodified noun of this type. A few sentences of this type are cited
below.
(38) a. Bula na amonestasion put tåsi.
agr.much l warning about ocean
‘There is a lot of warning about the sea.’ (CD, entry for amonestasión)
b. Ha chuda’ i mina’kuåttru na ånghit i tason-ña gi hilu’ åtdao.
agr pour the fourth l angel the bowl-agr loc top.l sun
‘The fourth angel poured out his bowl on the sun.’ (NT 476)
Compare the naturally-occurring examples in (39), which are similar but use the
definite article instead.
(39) a. Astaimånu chinago’-ña i lugåt-mu ginin i tasi?
how.far? distance-agr the place-agr from the ocean
‘How far is your place from the ocean?’ (CD, entry for astaimånu)
b. Hu li’i’ un ånghit tumotohgi gi hilu’ i åtdao.
agr see a angel agr.stand.prog loc top.l the sun
‘I saw an angel standing on the sun.’ (NT 481)
Although sentences like (38) are not particularly common, the fact that they are
attested at all supports the claim that <i, null> is suspended for antipresupposi-
tion. (Note that there are no sentences in the CD database or NT in which un
forms a DP with the nouns åtdao ‘sun’, pulan ‘moon’, tåsi ‘ocean’, or långit ‘sky’.)
80
antipresupposition effects; the null indefinite article does not. In this conclud-
ing section, I first tentatively suggest an account of this pattern, and then step
back and survey the larger consequences of the investigation for the typology of
narrow-scope indefinites.
Perhaps the most straightforward story one could tell about the antipresup-
position effects in sections 6 and 7 would claim that i and un form a presupposi-
tional scale for the purposes of MP, but—contrary to what I suggested earlier in
section 5—i and the null indefinite article do not. This would not be particularly
surprising, given that scales—e.g. the scales relevant for scalar implicature—are
known to be lexically arbitrary (see, e.g., Horn 1972 and Hirschberg 1985). The
claim that i fails to form a presuppositional scale with the null indefinite article
could be motivated by appealing to the different composition operations they
signal. Recall that for C&L, the null indefinite article signals that the descriptive
content of DP is composed with the predicate by Restrict, an operation that nar-
rows the domain of the predicate’s relatum but does not saturate it. On the other
hand, i and un are type-shifters that enable the descriptive content of DP to be
composed with the predicate’s relatum by Function Application. This suggests
that i and un are similar enough to count as paradigmatic alternatives—the sorts
of lexical entries that can be members of a scale (see Horn and Abbott 2014)—but
i and the null indefinite article do not satisfy this criterion.
How would this story handle the fact that both un and the null indefinite arti-
cle exhibit presuppositional implicature effects? Most likely, it would have to say
that these patterns do not result from MP, but rather from a generalized scalar
inference that recognizes both <i, un> and and <i, null> as scales.12 This position,
though stipulative, is credible, since it has never been entirely clear whether an-
tipresupposition and presuppositional implicature should be given a unified ac-
count. Leahy makes this point explicitly:
I am not yet prepared to answer. On a different note, Deniz Rudin suggests that speakers’ responses
to (16-17) and (18) could be taken to indicate that the null indefinite article does not exhibit presup-
positional implicature effects, contrary to what I claimed in section 5. Were that the case, the null
indefinite article would be well-behaved with respect to a unified MP—it would not exhibit either of
the effects attributed to that principle—but it would not enter into competition with the other arti-
cles at all. Serious discussion of this idea must await a more detailed study of the sorts of contrasts
presented in section 5.
81
incorporate presuppositional implicature—but not necessarily antipresupposi-
tion—as a special case.
Notice now that the story I have just told about Chamorro indefinites and
antipresupposition could be reconstructed in any approach that posits that the
descriptive content of a narrow-scope indefinite is composed directly with the
predicate.
In this respect, C&L’s account of Chamorro’s narrow-scope indefinites is not
unique. The reasons for preferring it lie elsewhere—in the patterns of semantic
number and discourse dynamics discussed in section 3.
From the standpoint of the typology of indefinites, the pattern of antipresup-
position documented above is noteworthy, because it reveals a dimension along
which narrow-scope indefinites can have a less limited distribution than scopally
unrestricted indefinites in the same language. The observation raises some ques-
tions. How do narrow-scope indefinites in other languages—bare NPs in partic-
ular—fare with respect to antipresupposition? Is the absence of antipresuppo-
sition effects characteristic of narrow-scope indefinites more generally? How, if
at all, does the absence of these effects connect with semantic number and the
ability to contribute a discourse referent?
The broader point to emerge is that there are types of narrow-scope indefi-
nites that are not severely restricted along multiple dimensions. In a way, this is
not surprising, if one takes seriously the semantic-pragmatic parallels between
narrow-scope indefinites and incorporees in morphosyntactic noun incorpora-
tion. Following Van Geenhoven (1998), research has focused on narrow-scope
indefinites whose semantic-pragmatic deficiencies closely parallel the semantic-
pragmatic limitations on incorporees in Mithun’s (1984) Type I and Type II incor-
poration. But Mithun also recognized a type of incorporation (Type III) in which
the incorporee can be construed as familiar or unique. Perhaps the profile of
Chamorro indefinites formed from the null indefinite article can be understood
in this light. I hope to have helped to open up the exploration of this territory; a
more thoroughgoing investigation must be left to another time.
the two-null-determiners proposal, but I believe all are subject to the objections raised in the text.
82
(Their definite counterparts would be excluded just like possessives formed from
i. ) Similarly, possessives formed from a null determiner would be definite, and
felicitous, when their possessee is commonly understood to be unique, as in (30).
(In accordance with MP, their indefinite counterparts would be infelicitous, just
like possessives formed from un. )
The problem with the two-null-determiner proposal is that it makes other
predictions that are incorrect. For instance:
83
The bottom line is that possessives formed from a null determiner do not behave
as if they were systematically ambiguous between definites and indefinites. In
short, the idea that Chamorro might have two null determiners is not viable.
References
Alonso-Ovalle, Luis, Paula Menéndez-Benito, and Florian Schwarz. 2011. Max-
imize Presupposition and two types of definite competitors. Proceedings of
NELS 39, vol. 1, ed. Suzi Lima, Kevin Mullin, and Brian Smith, 29-40. Amherst,
MA: Graduate Linguistics Student Association.
Barker, Chris. 1991. Possessive descriptions. Ph.D. dissertation, University of
California, Santa Cruz.
Barker, Chris. 2000. Definite possessives and discourse novelty. Theoretical Lin-
guistics 26: 211-227.
Barker, Chris. 2011. Possessives and relational nouns. In: Semantics: An inter-
national handbook of natural language meaning, vol. 2, ed. C. Maienborn, K.
von Heusinger, and P. Portner, 1109-1130. Berlin: Mouton de Gruyter.
Borja, Joaquin F., Manuel F. Borja, and Sandra Chung. 2006. Estreyas Marianas:
Chamorro. Saipan, CNMI: Estreyas Marianas Publications. [EM]
Chung, Sandra. 1998. The design of agreement: Evidence from Chamorro. Uni-
versity of Chicago Press.
Chung, Sandra. 2006. Possessors and definiteness effects in two Austronesian
languages. In: Quantification: A cross-linguistic perspective, ed. Lisa Matthew-
son, 179-224. Bingley, UK: Emerald.
Chung, Sandra and William A. Ladusaw. 2004. Restriction and saturation. Cam-
bridge, MA: MIT Press.
Chung, Sandra and William A. Ladusaw. 2006. Chamorro evidence for compo-
sitional asymmetry. Natural Language Semantics 14: 325-357.
Collins, James N. 2016. Reasoning about definiteness in a language without arti-
cles. In: Proceedings of SALT 26, ed. M. Moroney, C.-R. Little, J. Collard, and
D. Burgdorf, 82-102. Linguistic Society of America.
Cooreman, Ann. 1982. Chamorro texts. Ms., University of Oregon.
Cooreman, Ann. 1983. Chamorro texts. Ms., University of Oregon and Saipan,
CNMI.
Coppock, Elizabeth and David Beaver. 2015. Definiteness and determinacy. Lin-
guistics and Philosophy 38: 377-435.
Espinal, M. Teresa and Louise McNally. 2011. Bare nominals and incorporating
verbs in Spanish and Catalan. Journal of Linguistics 47: 87-128.
Farkas, Donka F. and Henriëtte de Swart. 2003. The semantics of incorporation.
Stanford, CA: CSLI.
Hawkins, John A. 1978. Definiteness and indefiniteness: A study in reference and
grammaticality prediction. London: Croon Helm.
Hawkins, John A. 1991. On (in)definite articles: Implicatures and (un)grammaticality
prediction. Journal of Linguistics 27: 405-442.
84
Heim, Irene. 1991. Artikel und Definitheit. In: Semantik: Ein internationales
Handbuch der zeitgenössischen Forschung, ed. Arnim von Stechow and Di-
eter Wunderlich, 487-535. Berlin: Walter de Gruyter.
Hirschberg, Julia Bell. 1985. A theory of scalar implicature. Ph.D. dissertation,
University of Pennsylvania.
Horn, Laurence R. 1972. On the semantic properties of logical operators in En-
glish. Ph.D. dissertation, UCLA.
Horn, Laurence R. 2001[1989]. A natural history of negation. Stanford, CA: CSLI
Publications.
Horn, Laurence R. and Barbara Abbott. 2014. <the,a>: (In)definiteness and impli-
cature. In: Reference and referring, ed. William P. Kabasenche and Mmichael
O’Rourke, 325-355. Cambridge, MA: MIT Press.
Kratzer, Angelika. 1998. Scope or pseudoscope: Are there wide-scope indefinites?
In: Events and grammar, ed. S. Rothstein, 163-196. Kluwer.
Lauer, Sven. 2016. On the status of ‘Maximize Presupposition’. In: Proceedings of
SALT 26, ed. M. Moroney, C.-R. Little, J. Collard, and D. Burgdorf, 980-1001.
Linguistic Society of America.
Leahy, Brian. 2016. On presuppositional implicatures. Topoi 35: 83-91.
Marciano, Dolores. n.d. Mannge’ na alaguan kalamasa. National Dissemination
and Assessment Center, CSULA.
Matthewson, Lisa. 1999. On the interpretation of wide-scope indefinites. Natural
Language Semantics 7: 79-134.
Mithun, Marianne. 1984. The evolution of noun incorporation. Language 60:
847-894.
Partee, Barbara H. 1987. Noun phrase interpretation and type-shifting princi-
ples. In: Studies in Discourse Representation Theory and the theory of gen-
eralized quantifiers, ed. J. Groenendijk, D. de Jongh, and M. Stokhof, 115-143.
Dordrecht: Foris.
Percus, Orin. 2006. Antipresuppositions. In: Theoretical and empirical studies
of reference and anaphora: Toward the establishment of generative grammar
as an empirical science. Report of the Grant-in-Aid for Scientific Research (B),
Project No. 15320052, Japan Society for the Promotion of Science, 52-73.
Reinhart, Tanya. 1997. Quantifier scope: How labor is divided between QR and
choice functions. Linguistics and Philosophy 20: 335-397.
Roberts, Craige. 2003. Uniqueness in definite noun phrases. Linguistics and Phi-
losophy 26: 287-350.
Sauerland, Uli. 2003. A new semantics for number. In: Proceedings of SALT 13,
ed. R.P. Young and Y. Zhou, 258-275. Linguistic Society of America.
Sauerland, Uli. 2008. Implicated presuppositions. In: The discourse potential
of underspecified structures, ed. Anita Steube, 581-600. Berlin: Mouton de
Gruyter.
Schlenker, Philippe. 2012. Maximize Presupposition and Gricean reasoning. Nat-
ural Language Semantics 20: 391-429.
Singh, Raj. 2011. “Maximize Presupposition!” and local contexts. Natural Lan-
guage Semantics 19: 149-168.
85
Van Geenhoven, Veerle. 1998. Semantic incorporation and indefinite descrip-
tions. Stanford, CA: CSLI.
Vikner, Carl and Per Anker Jensen. 2002. A semantic analysis of the English geni-
tive. Interaction of lexical and formal semantics. Studia Linguistica 56: 191-226.
Winter, Yoad. 1997. Choice functions and the scopal semantics of indefinites.
Linguistics and Philosophy 20: 399-467.
___. Database (unedited) for the Revised Chamorro-English Dictionary. [CD]
___. Nuebo Testamento [the Chamorro New Testament]. Saipan, CNMI: Dio-
cese of Chalan Kanoa. [NT]
86
Irish Questions are Relative*
Cathal Doherty S.J.
University of San Francisco
* I would like to thank the staff of the Jesuit library, Milltown Park, Dublin for their help in ac-
cessing materials from the Irish collection.
87
1 Irish relativization
The traditional description of Irish relative constructions, as for example in O’Nolan
(1919) who admits that “the whole matter is somewhat complicated”, envisages
two basic strategies: one used when the relativization site is subject or direct ob-
ject, the other for oblique positions. These are the so-called ‘direct’ and ‘indirect’
strategies, the former involving a gap, the latter a resumptive pronoun:1
In Jim’s standard analysis of the facts, these unbounded dependencies are distin-
guished by syntactic movement: the direct strategy involves cyclic movement (of
a null operator) while the indirect strategy with the resumptive pronoun is not
movement-derived. Moreover, the particles al (followed by the morphosyntac-
tic phenomenon of ‘lenition’) and an (followed by ‘eclipsis’) are taken to be purely
functional elements with no lexical content, i.e. complementizer heads, rather
than relative pronouns of any kind (McCloskey 2001, Sells 1984).2 Nonetheless,
three commonplace constructions are not consistent with this opposition be-
tween direct and indirect strategies. These are not peripheral constructions ei-
ther, but arguably part of the core grammar of Irish. They include the ‘head-
less’ or ‘amount’ relative, as well as prepositional and adverbial relatives. In all of
these, an appears with a gap rather than a resumptive pronoun:
88
(4) an fear [lena raibh tú ag caint _] Prepositional Relative
the man with.an was you talking.prog
‘the man with whom you were talking’
(5) an áit [a raibh mé _ ] Adverbial Relative
the place an was I
‘the place where I was’
It is not easy to see how these constructions can be understood in a model of Irish
syntax that upholds an absolute opposition between the direct and indirect rel-
ativization strategies, (1) and (2) above. One way out of the impasse, however, is
if the particle an is not in fact a purely functional element here, but has pronom-
inal content, as the native traditional grammatical tradition suggests in various
places (e.g. Ó Cadhlaigh 1940, 359; 383). Let us now examine these constructions
in turn.
This construction is puzzling given that it appears to have purely clausal syntax
and yet has the properties, both semantic and syntactic, of a nominal. Note how
headless relatives appear in all canonical nominal positions: subject (both finite
and nonfinite clauses), direct object, prepositional object and genitive:
Subject:
e.g. (Christian Brothers 1919, 91; 1977, 145; Ó Catháin 1922, 171; O’Nolan 1919, 89).
89
(8) Is leat [NP a bhfuil agam. ] (NG 145)
cop.pres with.you an is at.me
‘Whatever’s mine is yours.’
(9) Níor mhaith liom [ [NP a dtáinig arais ] imeacht arís. ]
neg.cond good with.me an came back leave.INF again
‘I wouldn’t like those who came back to leave again.’
Object:
Prepositional object:
(13) Tabhair sin chuig [NP a bhfuil beo de mo bhunadh. ] (OS 134)
give that to an are alive of my people
‘Give that to all of my people who are living.’
(14) Agus d’éirigh sé i ndiaidh [NP a ndúirt an dochtúir. ] (OS 134)
and rose he after an said the doctor
‘And he rose after all the doctor said.’
(15) Tháinig smúid ar [NP a rabh sa teach. ] (OS 134)
came gloom on aN was in.the house
‘Gloom descended on all who were in the house.’
Genitive:
(16) Níl [NP leath [ar imthigh i mbliana ]] ag obair. (OS 134)
neg.be half an .past went this year working.prog
‘Half of all who went this year aren’t working.’
(17) Is leat [NP iomlán [a bhfeiceann tú. ]] (OS 134)
cop.pres with.you entirety an see you
‘All that you see is yours.’
(18) Bhí súile [NP a raibh de dhaoine sa tsinagóg ] dlúite air (OC 384)
Were eyes an were of people in.the synagogue fixed on.him
‘The eyes of all in the synagogue were fixed on him.’
90
This is all the more curious given that the headless relative particle an is syntacti-
cally and morphologically non-distinct from the an particle which appears in the
indirect relative construction, (2) above. No morphological or syntactic prop-
erty obviously distinguishes them. Note that as part of the verbal complex, an
can never separate from the verb, even in co-ordinate structures:
What is more, an fuses with the copula in an identical fashion in both construc-
tions:
91
(24) a. an fear [CP ar cuimhin leis sin ]
the man an .cop memory with.him that
‘the man who remembers that’
b. Sin [NP ar cuimhin leis ]
that’s an .cop memory with.him
‘That’s all he remembers.’
(25) a. an fear [CP arbh eol dó a luach féin ]
the man an .cop.past knowledge to.him his worth own
‘the man who knew his own worth’
b. Sin [NP arbh eol dó. ] (GB 270)
that’s an .cop.past knowledge to.him
That’s all he knew.’
Moreover, note that the particle an is obligatorily absent from relative clauses in
which negation is encoded by the preverbal negative particle nach:
Strikingly, the particle an of headless relatives is also incompatible with the in-
dependent expression of negation. Instead, the clause must be given an overt
nominal head:
92
b. Sin [NP a bhfaca mé _ ]. Headless relative
that’s an saw I
‘That’s all I saw.’
To further illustrate the problem, note that these clauses often contain modifiers
typical of universal quantification, such as partitive expressions beginning with
de (of) and the postnominal modifier eile (other):
To these, we can add examples with the prenominal determiners gach (all / every),
gach de (all of), iomlán (entirety) and leath (half) in initial position:
93
(36) a n-asbiur Wb. 17d12
an I.say
‘what I say’
(37) Dénid anasberat frib Wb. 25c17
Do an say.they to.you
‘Do whatever they say to you.’
(38) Arbeir biuth a mbís for altóir ind ídil Wb. 10c6
eats food an is on altar the idol
‘He eats what food is on the altar of the idol.’
In the headless or amount relative construction, we find exactly the same mor-
phosyntactic variation:
(41) Sílim óna gcuala mé indiu go mbéidh cogadh ann. (OS 134)
I.think from.an heard I today that will.be war in.it
‘I think from what/all I heard today that there will be a war.’
(42) Labhair sé lena raibh ann. (CG 78)
poke I with.an was in.it
‘He spoke with whoever was there.’
Again, these facts are unexceptional if we take the particle an in the headless rela-
tive construction to be a morphosyntactic amalgam of a nasalizing proclitic rel-
ative pronoun with the functional elements of the preverbal complex.
94
3 Prepositional Relatives
Now consider the second recalcitrant construction, in which a preposition and
the particle an fuse in the preverbal complex, corresponding to a gap in the rela-
tivization site:
This construction bears some obvious similarities to pied-piping. In any case, the
particle an here seems to be pronominal, inasmuch as it appears to be a prepo-
sitional complement. Morphosyntactic evidence also corroborates this descrip-
tion since an intrusive nasal appears with certain prepositions as expected:
(17th century) in which the structure in (49) is attested. These are not found anywhere in the contem-
porary language, however, to my knowledge.
95
4 Adverbial Relatives
Relative clauses formed on bare-NP adverbs of time, reason, manner and place
also display the anomalous pattern in which an appears with a gap rather than a
resumptive pronoun:
Time:
(50) Is cuimhin liom an t-am (lá, oíche, etc.) ar gabhadh é. (GB 307)
cop memory with.me the time (day, night, etc.) an .past arrested him
‘I remember the time (day, night etc.) he was arrested.’
(51) Tiocfaidh an lá a mbéimid uilig aosta _. (OS 127)
will.come the day an will.be.pl all old
‘The day will come when we will all be old.’
Manner / Reason:
Place:
Are we dealing with a relative pronoun in this construction also? Note the con-
trast between the following examples (O’Nolan 1919, 90), which is suggestive of
some lexical content in the an particle:
96
Such examples are at least consistent with the description of an as being or con-
taining a nasalizing relative pronoun. Otherwise, we have to lexically specify in
some way that these nominal adverbs demand an rather than the regular direct
relative construction with al .
Finally, note as with the other two constructions discussed above, the use
of a gap is obligatory in these adverbial relatives and a bare pronominal cannot
resume the bare-NP head:
5 an as Relative Pronoun
It seems plausible, therefore, to draw the conclusion that an is, in at least in some
cases, a relative pronoun, such as in these constructions in which it appears with
a gap rather than a resumptive pronoun. Yet this particle, as we have seen, is
indistinguishable in every way from the an particle of indirect relatives. Let us
entertain the strongest possible hypothesis, therefore—an is in all cases a relative
pronoun in Irish, even in the indirect relative construction. The major analytical
task then becomes the accurate distribution of resumptive pronouns and gaps in
the relative system. Recall from (2) above that resumptives are excluded from the
subject position, a fact often attributed to the ‘Highest Subject Restriction’ (e.g.
McCloskey 1990, 210):
97
Resumptives are, however, acceptable in direct object positions, as well as in
oblique positions such as prepositional complement.5
In setting loose a relative pronoun an in the grammatical system, we would
also need to exclude all structures in which it cannot appear with a gap, whether
in subject, direct object as well as prepositional object positions:
(65) All direct object pronouns in Irish are ‘postposed’ to an oblique position
5 Given the VSO order of Irish, potential syntactic ambiguity arises in subject and direct object
relativization. Traditional grammars point to the possibility of a resumptive pronoun in direct object
position as a strategy to avoid such ambiguities. See Ó Cadhlaigh (1940, 376) also McCloskey (1977).
98
This allows us maintain the generalization (66) below:
(67) ∗ NPx [ an x [ V _x ]]
99
It would seem more likely that a bound pronominal requires an overt antecedent
in order to be interpretable and in the case of headless relatives, the proclitic
pronominal an cannot in itself provide such an antecedent. This is far from an
explicit analysis, but it is clear that syntax alone cannot bear the explanatory bur-
den for the absence of pronominal resumptives from this construction, under the
strong hypothesis. Rather, conditions on pronominal reference may be a source
of explanation for this phenomenon. A corroborating observation is that the
nominal méid (amount) cannot be resumed by a pronominal:
(69) * an méid ar ól sé é
the amount an .past drank he it
‘the amount he drank’ / ‘all he drank’ / ‘what he drank’
Again, this is consistent with conditions on pronominal reference as the root ex-
planation for the exclusion of resumptives from headless relatives. Similar con-
siderations extend to the adverbial relatives: if bare-NP adverbs cannot function
as adequate antecedents to resumptives in an argument position, (57) above.
Finally, in the case of prepositional relatives which, as we saw, arguably re-
duce to a form of ‘Pied-Piping’ or wh-movement of a prepositional element, again
we expect a gap rather than a resumptive pronoun, since the relativization site is
prepositional and not nominal.
That is to say, in this construction an does not directly enter into an unbounded
dependency with a referential nominal, which we have previously claimed to be
impossible. Rather, an inflected preposition (lena, ‘with-wh’) is in an unbounded
dependency with a prepositional phrase gap in the relativization site. In other
words, in this construction, we are dealing with an inflected prepositional head
(lenan ) creating an unsaturated verbal complex, rather than a pronominal (an )
functioning as relative operator.
6 Conclusions
As stated at the outset, the goal of this short paper is enlightened description,
rather than the promotion of a particular analysis supported by the Irish facts.
We have seen that there is some evidence that an is a relative pronoun in Irish,
following the suggestions of the native grammatical tradition. We then consid-
ered the analytical task entailed by the strong hypothesis that the particle an is a
generally a relative operator in Irish, concluding that constraints on pronominal
reference may partly explain the alternations between gaps and resumptives, in
tandem with the observation that all pronominal objects are in an oblique (post-
posed) position in the language.
100
Ineluctably, the question of interrogative syntax arises, however, since here
we witness the same opposition between direct and indirect strategies in ques-
tions as relative clauses:
(71) a. Céx a bhí _x anseo inné Direct
who al was here yesterday
‘Who was here yesterday?’
b. Cén fearx a raibh tú ag caint leisx inné Indirect
which man an were you talking to.him yesterday
‘Which man were you talking to yesterday?’
The standard analysis is that these facts are incompatible with the status of the
preverbal particles al and an as pronominals. Instead, the conclusion is that they
must be purely functional, complementizers. The native grammatical tradition,
which of course does not distinguish between relative pronouns and comple-
mentizers, by contrast, often presents interrogatives and relatives as having par-
allel syntax and even refers to the preverbal particles as relative pronouns.6 Could
it be the case that interrogative syntax in Irish in fact reduces to relative syntax?
If we widen our perspective to the extraction of prepositional elements, we
again observe the anomalous pattern, under the standard analysis at least, in
which an binds a gap instead of a resumptive:
(72) a. [Cé leis] a raibh tú ag caint _ ? (OS 104)
who with.him an were you talking
‘Who were you talking to?’
b. [Cé dó] a dtug tú an litir _ ? (OS 104)
who to.him an gave you the letter
‘Who did you give the letter to?’
c. [Ca hair] a raibh sibh ag caint _ ? (OS 104)
what on.it an were you.pl talking
‘What were you talking about?’
In addition, in some dialects and registers at least, even if not usual in everyday
speech, the following pattern is attested, which closely parallels the ‘prepositional
relative’ examined above:
d. [Cén fear] lena raibh tú ag caint _ ? (OC 389)
which man with.an were you talking
‘Which man were you talking with?’
Furthermore, note that with wh-extraction of adverbs, we find the same pattern
as adverbial relatives and an binds a gap:
(73) a. Cad chuige ar labhair tú _ ? (OS 103)
why an .past spoke you
‘Why did you speak?’
6 Ó Searcaigh (1939, 108 par. 213).
101
b. Cá gceannóchaidh tú na caoirigh _ ? (OS 105)
where.an will.buy you the sheep
‘Where will you buy the sheep?’7
c. Goidé an t-am a dtáinig sé _ ? (OS 127)
what the time an came he
‘What time did he come?’
Now if Irish relatives and interrogatives reduce to the same fundamental syntac-
tic structure, then the parallel alternation between direct and indirect patterns
found in both constructions is entirely expected. Moreover, if an is a pronomi-
nal operator, as under the strongest hypothesis pursued here, then those inter-
rogatives in which an appears with a gap rather than a resumptive also become
regular, or at least less surprising, in parallel with their relative counterparts.
Finally, it is important to point to another pattern in the language in which
relative syntax putatively functions as a clause, a cleft-like focus construction
known as the ‘independent’ relative clause (an clásal coibhneasta leithleach) in tra-
ditional grammars “for want of a better term.”8 In this construction, an embed-
ding copula is often missing, as below.
(74) Ø Pádraig a bhí ann.
Patrick al was there
‘Patrick was there / It was Patrick who was there.’
The same traditional grammar categorizes regular interrogative sentences such
as we have been considering in this essay as examples of the same ‘indepen-
dent relative’ (focus) construction. Regardless of one’s theoretical commitments,
therefore, this construction is perhaps a key for better understanding Irish un-
bounded dependencies.
Abbreviations
CG Ceart na Gaeghilge, Ó Cadhlaigh 1916.
GB Graiméar Gaeilge na mBráithre Críostaí, Christian Brothers 1960.
MM Cora Cainte as Tír Chonaill, Mac Maoláin 1943.
NG New Irish Grammar, Christian Brothers 1977.
OC Gnás na Gaeilge, Ó Cadhlaigh 1940.
OS Coimhréir Ghaedhilge an Tuaiscirt, Ó Searcaigh 1939.
RG Réchúrsa Gramadaí, Mac Giolla Phádraig 1963.
References
Adger, David and Gillian Ramchand. 2005. Merge and Move: Wh-Dependencies
Revisited. Linguistic Inquiry 36. 161-193.
7 The wh-word cá (where) here is plausibly fused with an . See Ó Searcaigh (1939, 105 n).
8 See Christian Brothers (1999, 216ff) for a brief description of this construction.
102
Bennett, Ryan, Emily Elfner and James McCloskey. 2016. Lightest to the Right:
An Apparently Anomalous Displacement in Irish. Linguistic Inquiry 47 .169-234.
Bresnan, Joan and Jane Grimshaw. 1978. The Syntax of Free Relatives in English.
Linguistic Inquiry 9. 331-391.
Christian Brothers. 1919. Graiméar na Gaedhilge leis na Bráithreachaibh Críostamhla.
6th edition. Dublin: M.H. Gill & Son.
Christian Brothers. 1960. Graiméar Gaeilge na mBráithre Críostaí. Dublin: M.H.
Mac an Ghoill agus a Mhac, Tta.
Christian Brothers. 1962. Nuachúrsa Gaeilge na mBráithre Críostaí. Dublin: M.H.
Mac an Ghoill agus a Mhac, Tta.
Christian Brothers. 1977. New Irish Grammar. Dublin: Fallons.
Christian Brothers. 1999. Graiméar Gaeilge na mBráithre Críostaí. Dublin: An
Gúm.
Chung, Sandra and James McCloskey. 1987. Government, Barriers, and Small
Clauses in Modern Irish. Linguistic Inquiry 18. 173-237.
Hirschbühler, Paul and María-Luisa Rivero. 1983. Remarks on Free Relatives and
Matching Phenomena. Linguistic Inquiry 14. 505-520
McCloskey, James. 1977. An Acceptable Ambiguity in Modern Irish. Linguistic
Inquiry 8. 604-609
McCloskey, James. 1979. Transformational Syntax and Model Theoretic Semantics: a
case-study in Modern Irish. Dordrecht: Foris.
McCloskey, James. 1990. Resumptive Pronouns, A-binding and Levels of Rep-
resentation in Irish. In Randall Hendrick (ed.), Syntax and Semantics 23: The
Syntax of the Celtic Languages, 199-248. London: Academic Press.
McCloskey, James. 2001. On the morphosyntax of wh-movement in Irish. Journal
of Linguistics 37. 67-100.
McCloskey, James. 2008. Resumption, Successive Cyclicity, and the Locality of
Operations. In Samuel Epstein & T.D. Seely (eds.), Derivation and Explanation
in the Minimalist Program, 184-219. Hoboken, NJ: Wiley, 2008.
Mac Giolla Phádraig, Brian. 1963. Réchúrsa Gramadaí. 3rd edition. Dublin: Long-
man Brún agus Ó Nualláin Teo.
Mac Maoláin, Seán. 1943. Cora Cainnte as Tír Chonaill. 3rd edition. Dublin: Oifig
an tSoláthair.
Ó Cadhlaigh, Cormac. 1916. Ceart na Gaedhilge: Irish syntax. Dublin: Mellifont.
Ó Cadhlaigh, Cormac. 1940. Gnás na Gaedhilge. Dublin: Oifig an tSoláthair
Ó Catháin, Seaghán. 1922. Ceachta Cainnte Gramadaighe: the practical Irish gram-
mar. Dublin: McGill.
Ó Nolan, Gerald. 1919. Studies in Modern Irish: Part 1. Dublin: The Educational
Company of Ireland Ltd.
Ó Searcaigh, Séamus. 1939. Coimhréir Ghaedhilg an Tuaiscirt. Dublin: Oifig an
tSoláthair.
103
On the interaction of head movement and
ellipsis in Danish*
Vera Gribanova Line Mikkelsen
Stanford University University of California, Berkeley
1 Introduction
A key question in investigations of elliptical processes concerns the nature of
the ellipsis site, and its interaction with other syntactic processes. A priori, one
can imagine two logical extremes. One extreme is the hypothesis of complete
opacity, in which ellipsis blocks any and all syntactic processes from applying to
material inside the ellipsis site. On the other end of the spectrum is the hypothesis
of total transparency, in which ellipsis has no effect on syntactic processes
whatsoever.
Existing work makes clear that the complete opacity hypothesis cannot be
correct, since various kinds of movement operations may apply to an element in-
side the site of the ellipsis site (Merchant 2001: among others). In and of itself, this
kind of evidence has been used to argue that constituent ellipsis involves syntac-
tic structure that remains unpronounced (Ross 1969, Merchant 2001), rather than
being a proform (Hardt 1993) and/or subject to semantic or pragmatic reconstruc-
tion (Ginzburg & Sag 2000, Culicover & Jackendoff 2005). Fairly straightforward
examples involve wh-movement out of the ellipsis site (1) and A-movement — in
this case, for passive — out of the ellipsis site (2).
* For generous comments and feedback on the ideas presented here, we thank Emily Clem, Boris
Harizanov, Peter Jenks, Jim McCloskey, Jason Merchant, and the audience of UC Berkeley’s Syntax
Circle. We would like to thank the editors of this volume for putting it together, for providing us
with the opportunity to collaborate, and share our appreciation of Jim McCloskey’s mentorship over
the course of many years.
105
(1) I know which puppy you should adopt, but I don’t know which one you
shouldn’t. (Schuyler 2001: 1)
CP
DP
C TP
which one
DP
T vP
you
[e] ...
shouldn’t
DP
T vP
I
[e] ...
was
106
Second, the pattern is also not due to VP ellipsis being crosslinguistically opaque
for head movement: verb movement out of the site of VP ellipsis results in per-
fectly grammatical verb-stranding VP ellipsis configurations in a diverse set of
languages, including Irish (McCloskey 1996, 2012, 2016), Russian (Gribanova 2013b,a,
2017, To appear), Hebrew (Goldberg 2005b,a), Greek (Merchant 2017), Hindi-Urdu
(Manetta To appear, 2017), Hungarian (Lipták 2013), Swahili (Ngonyani 1996), Finnish
(Holmberg 2001), European Portuguese (Santos 2009), and others.
Recent work by Aelbrecht (2010) and Sailor (To appear) has brought us closer
to an explanation for the behavior of Danish verb movement under ellipsis. One
part of the explanation is the idea, developed independently in Aelbrecht 2010,
that an ellipsis site is closed off for syntactic processes when the licensor of the
ellipsis is merged. The second part of the explanation is that verb movement
in Danish verb second clauses is triggered by C, whereas the licensor of VP el-
lipsis is T (Sailor To appear: 11–12). Taken together with the idea that syntactic
derivations proceed bottom-up, these commitments should result in a deriva-
tional blocking effect, whereby the trigger for verb movement (C) is merged only
after the VP ellipsis licensor (T) is merged. Since the ellipsis licensing head makes
the ellipsis site opaque to further operations, verb movement out of the ellipsis
site is blocked.
What we demonstrate in this paper is that the aforementioned set of commit-
ments yields only a partial explanation for the lack of verb-standing VP ellipsis
in Danish. Citing evidence from Vikner 1995, Sailor (To appear) takes there to be
no movement of V to T independent of V to C. But T and V nevertheless must
be morphologically unified, in both main and embedded clauses. Although main
verb stranding ellipsis is still prohibited in Danish embedded clauses, nothing
about Aelbrecht’s or Sailor’s proposals prevents this from taking place. Under
the standard view of head movement — qua syntactic movement and head ad-
junction — movement of V to T in the embedded clause, in combination with VP
ellipsis, would yield exactly the prohibited effect. If, as Sailor (To appear) claims,
there is no movement of V to T independent of V to C, then the question of how
the main verb comes to bear tense inflection arises but receives no immediate
answer.
The claim we put forth here is that Sailor’s account is missing a crucial and
independently argued for observation about the nature of head movement that
comes from recent work by Harizanov & Gribanova (2018) (henceforth h&g).
Their proposal is that the effects usually modeled by head movement actually fall
into two distinct classes, with principally distinct properties: there is genuinely
syntactic movement of heads on the one hand, and postsyntactic amalgamation
of heads (for purely morphological purposes) on the other. Danish verb move-
ment to C is, they claim, of the syntactic type, while amalgamation of V and T in
embedded clauses is postsyntactic Lowering (Embick & Noyer 2001) and involves
no syntactic operations. Once this observation is in place, both main clause and
embedded clause behavior in Danish fall into place. A Danish verb never under-
goes syntactic head movement below C, hence it can never escape VP ellipsis.
Genuinely syntactic head movement of the verb to C in V2 clauses is blocked by
107
the interaction of the Aelbrecht (2010) licensing schema and Sailor’s derivational
timing proposal, as described above.
The resulting proposal wields a great deal of explanatory power. Beyond ex-
plaining the Danish-specific interaction between VP ellipsis and head movement,
it can also explain crosslinguistic interactions between ellipsis and head move-
ment: why does Danish lack verb-stranding ellipsis, but e.g. Irish, Russian, and
Hebrew have it? This latter question, we claim, can be answered by consider-
ing the range of possibilities in postsyntactic amalgamation. While Danish verb
unification involves Lowering of T to V, in languages that permit verb-stranding
ellipsis, the amalgamation goes in the other direction — a postsyntactic operation
h&g call Raising.
This follows from the standard analysis of Germanic V2 (den Besten 1983, Vikner
1995): the finite verb moves to C via T and a phrasal constituent moves to [spec,
CP]. In the (a) examples, that phrasal constituent is the subject, in the (b) examples
it is the direct object, and in the (c) examples an adjunct wh-phrase. As seen in (5),
non-finite verbs surface below negation, which indicates that they remain in situ
inside the verb phrase.
108
In embedded clauses, there is no V2. Instead the finite verb surfaces below
negation, as seen in the (a) examples. The (b) examples show the impossibility of
V2 in this embedded environment and the (c) examples show that the finite verb
cannot undergo short head movement across negation to T.
(7) a. Jeg undrede mig over hvorfor Mona ikke havde vasket bilen.
I wondered refl over why Mona not had washed car.def
‘I wondered why Mona had not washed the car.’
b. * Jeg undrede mig over hvorfor havde Mona ikke vasket bilen.
I wondered refl over why had Mona not washed car.def
c. * Jeg undrede mig over hvorfor Mona havde ikke vasket bilen.
I wondered refl over why Mona had not washed car.def
(8) a. Jeg underede mig over hvorfor Mona ikke vaskede bilen.
I wondered refl over why Mona not washed car.def
‘I wondered why Mona didn’t wash the car.’
b. * Jeg undrede mig over hvorfor vaskede Mona ikke bilen.
I wondered refl over why washed Mona not car.def
c. * Jeg undrede mig over hvorfor Mona vaskede ikke bilen.
I wondered refl over why Mona washed not car.def
Thus the generalization is that there no head movement of V unless it moves all
the way to C (as argued by Vikner (1995), among others).
(9) a. Mona og Jasper har vasket bilen, eller rettere Mona har .
Mona and Jasper have washed car.def or rather Mona has
‘Mona and Jasper have washed the car, or rather Mona has.’
b. Jeg har vasket bil, selvom Mona ikke har .
I have washed car even.though Mona not has
‘I have washed my car even though Mona hasn’t (washed her car).’
(10) a. Mona og Jasper vaskede bilen, eller rettere Mona gjorde .
Mona and Jasper washed car.def or rather Mona did
‘Mona and Jasper have washed the car, or rather Mona did.’
b. Jeg vasker bil, selv hvis Mona ikke gør .
I wash car even if Mona not does
‘I wash my car even if Mona doesn’t (wash her car).’
109
(11) a. * Mona og Jasper vaskede bilen eller rettere Mona vaskede .
Mona and Jasper washed car or rather Mona washed
intended: ‘Mona and Jasper washed the car, or rather Mona did.’
b. * Jeg vasker bil, selv hvis Mona ikke vasker .
I wash car even if Mona not washes
(12) a. * Mona og Jasper vaskede bilen eller rettere Mona .
Mona and Jasper washed car or rather Mona
intended: ‘Mona and Jasper washed the car, or rather Mona did.’
b. * Jeg vasker bil, selv hvis Mona ikke .
I wash car even if Mona not
The ungrammaticality of (12) is familiar from English. Tense must be expressed
and if no regular verb is available to host tense, a dummy verb must be employed,
as in (10). A Danish-specific puzzle is the lack of verb-stranding VP ellipsis in (11a).
The VP targeted by ellipsis is in a main clause and thus the finite main verb vaskede
is expected to move to C and thereby escape the ellipsis of VP. For some reason
the combination of verb movement to C and VP ellipsis does not result in verb-
stranding ellipsis. This is especially puzzling in light of the fact that analogous
configurations, in which the main verb is stranded outside of an ellipsis site, are
crosslinguistically common, e.g. in Irish:
In the next section we examine the solution to this puzzle put forth by Sailor (To
appear). Sailor’s key idea is that in V2 clauses, verb movement is triggered after
VP ellipsis has been effectuated, and that this derivational timing is responsible
for the lack of verb-standing VP ellipsis in Danish. We argue that this provides
a partial explanation, but that more needs to be said to explain why there is no
verb-stranding VP ellipsis in the absence of verb movement to C, as in (11b).
110
both the semantic side of ellipsis (the givenness requirement) and the phonolog-
ical side: the sister of the head bearing [e] goes unpronounced. If nothing else
constrains ellipsis, we expect the ellipsis site to be completely transparent for
syntactic movement. Aelbrecht’s main claim is that transparency is constrained
by the derivational timing of ellipsis with movement processes that target ele-
ments inside the ellipsis site. At a certain point in the derivation, namely when
the licensor of ellipsis has been Merged, material inside the ellipsis site becomes
in accessible for movement. Only movement that is triggered prior to or at this
point may proceed out of the site of ellipsis.
Aelbrecht implements this derivational timing by separating the [e] feature
from the licensor. In the case of English VP ellipsis, the licensor is T but the
[e] feature resides on a lower head, namely Voice. For ease of reference we’ll
refer to the head carrying the [e] feature as the host. The target of ellipsis is
the sister of the host, which is vP in the articulated clause structure assumed by
Aelbrecht (2010: 175–176). Auxiliaries reside between T and Voice, and negation is
left adjoined to the complement of T. External arguments are introduced in Spec-
Voice. The sister of the host remains transparent until the licensor is Merged.
This defines the window of opportunity for syntactic extraction out of the ellipsis
site. Applying these assumptions to Danish VP ellipsis, we have the structure in
(14) for the auxiliary stranding case in (9).
C TP
T AuxP
[cat[T]]
Aux VoiceP
Voice vP
[e[infl[uT]]]
v VP
...
On this view, auxiliaries survive ellipsis because they are never inside the ellipsis
site (vP). External arguments survive ellipsis because they are never inside the
ellipsis site either; they are base-generated in [spec, Voice]. Internal arguments
may escape VP ellipsis by A-movement to [spec, T] (passive/unaccusative) or by
A-bar movement to the edge of the VoiceP phase.
Consider now an ellipsis derivation of a V2 clause without an auxiliary. If
Danish verb movement to C proceeds through T and that first movement step is
triggered when T is Merged, this set of assumptions predicts verb stranding VP
111
ellipsis, because T is the ellipsis licensor and any syntactic processes triggered
by the licensor or a lower head are unaffected by ellipsis. The core insight of
Sailor (To appear) is to connect the lack of verb-stranding VP ellipsis to the lack
of independent verb movement to T. Recall from the discussion of embedded
clauses in the previous section that in non-V2 clauses, the finite verb doesn’t move
to T — it remains below negation. This suggests that the only trigger for verb
movement in Danish is C, and that the verb moves directly from its base position
to C. C is Merged after the ellipsis licensor (T) and thus unable to interact with
material inside the ellipsis site, including the verb. This is Sailor’s explanation
for the lack of verb-stranding VP ellipsis in Danish V2 clauses.
We don’t question the logic of this explanation, but we believe it is incom-
plete in the following sense: if the verb moves directly to C in V2 clauses, some-
thing else must be responsible for combining V and T and producing a finite
verb. Ideally that would be the same mechanism that produces a finite verb in a
low position in a non-V2 clause. Moreover, under the articulated clause struc-
ture that comes with Aelbrecht’s derivational timing analysis, there is a second
derivational path to verb-stranding VP ellipsis that must be ruled out: namely, V
moving to v and then to Voice, which is outside of the ellipsis site. These prob-
lems are obscured in Sailor’s presentation, because he operates only with a tri-
partite C-T-V structure, abstracting away from v and Voice. Once the details of
Aelbrecht’s system are included, Sailor’s solution is no longer complete.
In the next section we propose that both problems are solved by the under-
standing of head movement developed by h&g.
... ...
X YP X YP
... X Y ...
Y ... tY ...
This formulation raises non-trivial theoretical problems that have been the focus
of much attention in recent years.1 One result of these discussions is a prolifera-
tion of accounts that attempt to reduce head movement to other mechanism(s) —
1 For thorough elaborations, see Matushansky 2006, Roberts 2010, Harizanov & Gribanova 2018.
112
e.g. remnant movement (Koopman & Szabolcsi 2000), PF movement (Chomsky
2001, Schoorlemmer & Temmerman 2012, Platzack 2013), re-projection (Georgi
& Müller 2010), and others. h&g’s contention is that word formation and word
order permutation phenomena are empirically distinct along several important
dimensions, and that this difference indicates a need for two distinct and inde-
pendently needed theoretical mechanisms: Internal Merge in the syntax, and a
postsyntactic operation called amalgamation.
The main point is that certain properties associated with head movement
cluster together in revealing ways, with only one cluster pointing to syntax-like
behavior. This type of head movement can yield interpretive effects (indicating a
syntactic movement, which feeds semantic interpretation); it has locality condi-
tions akin to phrasal movement, violating the head movement constraint (Travis
1984); and it results in word order permutations, but not affixation or other mor-
phological growth. The other cluster has properties which are not typical of
syntax and therefore should be separated out: it never yields semantic effects,
obeys the head movement constraint, and results in morphological growth (but
not necessarily a higher point of pronunciation). This bifurcation, summarized
in (16), has the beneficial result of resolving some of the theoretical issues usually
associated with the traditional formulation in (15).
In h&g’s system, the syntactic type of head movement involves internal merge,
either into the specifier (Matushansky 2006, Harizanov 2014) or as a form of re-
projection (see Harizanov 2017 for details). Postsyntactic head movement consists
of two operations, Lowering (Embick & Noyer 2001) and Raising, its opposite.
Lowering and Raising can apply independently or interact, depending on the spec-
ifications of particular heads in the clausal spine of a given language. Heads must
be considered for amalgamation cyclically, bottom-up. Each head is associated
113
with a morphological feature [m],2 which may have either a positive or a negative
(or no) specification. An [m:+] feature results in Raising of a head; an [m:-] specifi-
cation results in Lowering of a head; and no specification will yield no application
of either operation.
These syntactic and postsyntactic operations will of course interact; we con-
sider this interaction in more detail when we examine how this proposal can be
applied to Danish in the next section.
• the syntactic movement that could bring it outside the ellipsis (v→C) is
triggered after the ellipsis site is rendered inaccessible for syntactic pro-
cesses.
analogous ideas.
3 We take this lower realization point to be v, not V, for reasons having to do with the realization
of the verb relative to indirect objects in double object constructions. None of our conclusions hinge
on this particular view, however.
114
(19) Postsyntactic (dashed) amalgamation in a Danish embedded clause, aux
present
CP
C TP
T[m:-] AuxP
AdvP AuxP
Voice[m:-] vP
v[m: ] VP
V[m:+]
C TP
T[m:-] VoiceP
AdvP VoiceP
Adv Voice[m:-] vP
v[m: ] VP
V[m:+]
In (19) and (20), Lowering and Raising result from the featural specifications of each
head ([m:+/-]). Each head in the syntactic structure is either Lowered or Raised, pro-
ceeding from bottom up. These being postsyntactic operations, once a head has
amalgamated, it does not leave a trace or copy. The combination of these two
assumptions results in the availability of configurations in which amalgamation
of a head is not to the head of its complement, but the relation remains that of
structural adjacency.4 In (20), going from bottom up, V has an [m:+] feature and
will therefore amalgamate into v. v’s [m] feature is not specified, so nothing hap-
4 While the Embick & Noyer’s (2001) definition of Lowering stipulates that a head Lowers to the head
of its complement, this definition is too restrictive in the present theoretical context. Instead, the lo-
cality relation that needs to be established between the two heads participating in amalgamation is
structural adjacency, as pointed out in (17) and (18). We take it that Lowering and Raising into struc-
turally complex specifiers and adjuncts is prohibited by the island status of specifiers and adjuncts
and refer the reader to h&g for further discussion of this point.
115
pens. The next head up, Voice, is specified to Lower and so amalgamates into v.
Once Voice has amalgamated, it exists only as part of the complex head that has
resulted from amalgamation. This means that T, the next head going bottom-
up, will lower directly into v: the elimnination of Voice from the postsyntactic
clausal spine means that v is the next head down in an immediate c-command
relation with T.
With this much in place, we can return to ellipsis. The goal is to understand
why the main verb never raises out of a VP ellipsis site in Danish. Given the struc-
tures above, the answer is now straightforward: in the presence of an auxiliary,
the auxiliary will be merged outside the ellipsis site and stranded if ellipsis takes
place; the main verb will be elided with the rest of vP. Both instances of postsyn-
tactic Lowering in (19) take place independently of the application of ellipsis.5
C TP
... AuxP
Aux[ ] VoiceP
Aux[ ] T[m:-] ... vP
v[m: ] VP
v[m: ] Voice[m:-]
5 Recall from (14) that Voice bears the e-feature in the syntax and will, as a result of an agree
relation with the licensing head T, trigger ellipsis of its complement. The complement of Voice (vP)
is therefore marked for non-pronunciation long before the Voice head Lowers postsyntactically into
the ellipsis site.
116
(22) Postsyntactic amalgamation & ellipsis in a Danish embedded clause, no
aux present
CP
C TP
... VoiceP
... vP
v[m: ] VP
v[m: ] Voice[m:-]
Voice[m:-] T[m:-]
Taken at face value, the configuration above will yield an unattested result; namely,
it will result in a stranded subject outside the ellipsis site (with T unpronounced).
(23) * Jeg vasker bil, selv hvis Mona ikke .
I wash car even if Mona not
There are two ways to explain what goes wrong in (22), thereby yielding the ob-
servation that (23) is ungrammatical. The first possibility is that, as in English,
T has a requirement that its features be exponed. If T is stuck inside the ellipsis
site, the conflict between the requirements of T and the requirements of ellipsis
yields a crash at PF. The second possibility is that (22) fails for selectional reasons:
the Voice head that bears the e-feature may only be selected by an auxiliary, and
not directly by T.6
nance. This is an expository choice, emphasizing the fact that we are dealing with a single syntactic
object occupying distinct structural positions; other implementations are possible as well.
117
be reflected in all other instances of that object; and c) Chain Reduction marks
some instance(s) for non-pronunciation. This set of assumptions will have the
consequence that the pronounced instance will bear the reflexes of postsyntactic
changes to any of the other instances. This will have the desired effect, namely
the pronunciation of the entire verbal complex in the higher position (C) in verb
second clauses. How does C know what to attract? An assumption we make, spe-
cific to Danish, is that the attracting feature on C will probe for a verbal element
— an Aux or v (a verbalizer). Locality conditions on probes and goals dictate that
Aux will be attracted, if it is in the structure (see (24)). Otherwise, it will be v, as
in (25).
(24) Postsyntactic amalgamation and syntactic head movement in a Danish V2
clause, aux present
CP
C TP
... AuxP
Aux[ ] VoiceP
Aux[ ] T[m:-] ... vP
v[m: ] VP
v[m: ] Voice[m:-]
C TP
... VoiceP
... vP
v[m: ] VP
v[m: ] Voice[m:-]
Voice[m:-] T[m:-]
Now consider the interaction of verb movement and ellipsis in each of these
structures. In (24) it is the auxiliary that is attracted to C and the main verb amal-
118
gamates in v. Thus the auxiliary survives ellipsis of vP, the main verb doesn’t.
What prevents main verb stranding if ellipsis applies in a structure with no aux-
iliary, as in (25)? Here, Sailor’s original explanation can be put to use. In a bottom-
up derivation, the derivation will introduce T, the ellipsis licensing head. At this
stage, in accordance with Aelbrecht’s system, everything inside the ellipsis site
(vP) will be rendered inaccessible. C will be introduced next into the derivation,
and at this stage will not be able to attract the v inside the ellipsis site. Syntactic
movement of the main verb out of the ellipsis site is therefore blocked by the logic
of the derivational timing involved in the triggering of ellipsis and the triggering
of verb movement.
6 Conclusion
In this paper we have grappled with the question of how transparent ellipsis do-
mains are for various kinds of movement, and in particular with understanding
the lack of verb movement out of VP ellipsis in Danish. The absence of such
movement, and thus of verb-stranding VP ellipsis in the language, is significant
given two other observations. First, Danish VP ellipsis is transparent for A and
A-bar movement, which indicates that there is regular syntactic structure in the
ellipsis site, and thus a structural basis for verb movement — confirming some
version of the transparency hypothesis with which this paper began its discus-
sion. Second, since Jim McCloskey’s seminal work on verb-stranding VP ellipsis
in Irish, the construction has been widely documented in a range of typologi-
cally diverse languages. So the absence of verb movement out of VP ellipsis in
Danish is a real puzzle for an otherwise very successful theory of ellipsis and its
interaction with syntactic movement.
In a very important analytic move, Sailor (To appear) connects the lack of
verb stranding VP ellipsis in Danish to the lack of movement of V to T indepen-
dently of movement to C in the language. This is what sets Danish apart from
Irish, Russian, Hebrew8 and other languages that do have verb-stranding VP el-
lipsis. The identity of the trigger for verb movement — C vs. T — matters if the
ellipsis site is closed off for syntactic processes once the licensor of the ellipsis
is Merged. This is exactly the analysis of ellipsis developed in Aelbrecht (2010),
where she also identifies T as the licensor of VP ellipsis. Adopting Aelbrecht’s
theory of ellipsis, Sailor proposes that there is no verb movement out of VP el-
lipsis in Danish, because the trigger for this verb movement is C, which is Merged
after T has rendered the ellipsis site inaccessible. This is an elegant and attrac-
tive solution to the puzzle. It is however, incomplete. One question left open by
Sailor is what happens in embedded clauses where there is not movement to C,
in particular what mechanism ensures tense inflection the verb in the absence
of verb movement through T to C. The other question arises from a mismatch
in the clause structures assumed by Sailor and Alebrecht. Sailor operates with a
8 However, see Landau 2017 for a critical reassessment of the claim that verb-stranding ellipsis is
119
simple V-T-C structure and thus the only movement that needs to be excluded to
account for the lack of verb-stranding VP ellipsis is V moving to T or V moving
to C. However, Aelbrecht’s analysis of ellipsis requires a more articulated clausal
spine containing V-v-Voice-T-C, such that the host of [e] (Voice) can be separated
from the licensor (T) so that elements inside the ellipsis site (vP) can undergo
movement to [Spec, Voice] (A-bar movement) or [Spec, T] (A-movement) before
the ellipsis site is closed off.
The additional functional structure between V and T is relevant to Sailor’s
account because it raises the possibility of head movement between these various
projections below T. If V could move to Voice via v, that would be enough to get
the main verb out of the ellipsis site and yield verb-standing VP ellipsis. Since
this is movement below the ellipsis licensor T, the derivational timing pursued
by Sailor account cannot be extended to rule out movement of V to Voice.
In this paper we have argued that these two gaps in Sailor’s account are both
addressed by the general theory of head movement developed by h&g and a
straightforward extension of their analysis to accommodate the articulated clause
structure required for Aelbrecht’s analysis of VP ellipsis. On this view, there are
two kinds of verb movement in Danish: postsyntactic amalgamation of T, Voice
and V in v which takes place in every clause, and syntactic head movement of v
to C which takes place only in V2 clauses. The complete explanation for the lack
of verb-stranding VP ellipsis in Danish is thus two-fold: the verb cannot escape
the ellipsis site by syntactic head movement because that movement is triggered
after the ellipsis site is closed off by the Merger of T (Sailor To appear). Second,
post-syntactic movement amalgamates T, Voice, v, and V in v, which is inside the
ellipsis site.
Turning to languages that exhibit verb-stranding VP ellipsis, we observe that
these are exactly the languages that h&g argue amalgamate high (above v — i.e.
Raising instead of Lowering), and thus the inflected verb survives ellipsis.
We’ll end by returning to the full profile of which verbal elements can and
must be stranded by Danish VP ellipsis. The profile is the same in main and em-
bedded clauses and can be schematized as below:
(26) Profile of Danish VPE (abstracted from (9)–(12))
a. aux
b. gøre
c. * main v
d. * no verb
The focus of Sailor’s paper is the contrast between (26a) and (26c) in main clauses.
In this paper we have added to his analysis so that this contrast is accounted for
in embedded clauses as well. Sailor notes the contrast between (26b) and (26d) in
passing (Sailor To appear: 5) but does not attempt to account for it. With the expli-
cation of Sailor’s analysis provided here, we believe we are a step closer to being
able to explain this contrast as well. While a full account must await another oc-
casion, we’d like to make the following observations. Houser et al. (2011) argue
120
that gøre is an auxiliary, though it is limited in its distribution to environments
where the verb phrase has a non-canonical realization (ellipsis, topicalization and
pronominalization). This suggests that we should treat (26b) as a special case of
(26a): gøre is an auxiliary that takes VoiceP as its complement. What makes gøre
special is that it requires ellipsis. This, we propose, is because [e] is an inherent
part of gøre’s featural specification. With regular auxiliaries ellipsis is optional:
if [e] is present on Voice, vP is elided; if [e] is not present on Voice, there is no
ellipsis. If gøre is present, so is [e], and VoiceP will be elided once T is Merged.
As for the ungrammaticality of (26d), we considered two possible explana-
tions in connection with the derivation in (22). The first is that (26d) violates a
hard requirement that T be exponed. This raises obvious questions of imple-
mentation and explanation: how would one implement such a requirement in
a realizational theory of morphology and is it any more than a restatement of
the observation? The second solution is that the structural configuration that
would lead to (26d), namely the one in (22), never arises because T cannot com-
bine with an [e]-laden Voice head. This is a very different type of account. It is not
explanatory in any way, but it can be straightforwardly implemented within the
assumptions of the rest of the analysis. Moreover, if such a selectional restriction
is imposed, it would rule out verb-standing VP ellipsis (26c) without any appeal
to derivational timing.
References
Aelbrecht, Lobke. 2010. The syntactic licensing of ellipsis. John Benjamins.
Baker, Mark C. 1988. Incorporation: A theory of grammatical function changing.
Chicago: University of Chicago Press.
Chomsky, Noam. 2001. Derivation by phase. In Michael Kenstowicz (ed.), Ken
hale: A life in language, 1–52. Cambridge, MA: MIT Press.
Culicover, Peter W. & Ray Jackendoff. 2005. Simpler syntax. Oxford: Oxford Uni-
versity Press.
den Besten, Hans. 1983. On the interaction of root transformations and lexical
deletive rules. In Werner Abraham (ed.), On the formal syntax of the westgerma-
nia, 47–131. Amsterdam: John Benjamins.
Embick, David & Rolf Noyer. 2001. Movement operations after syntax. Linguistic
Inquiry 32. 555–595.
Georgi, Doreen & Gereon Müller. 2010. Noun-phrase structure by Re-Projection.
Syntax 13(1). 1–36.
Ginzburg, Jonathan & Ivan Sag. 2000. Interrogative investigations. Stanford, CA:
CSLI Publications.
Goldberg, Lotus. 2005a. On the verbal identity requirement in VP ellipsis. Pre-
sented at the Identity in Ellipsis workshop at UC Berkeley.
Goldberg, Lotus. 2005b. Verb-stranding vp ellipsis: A cross-linguistic study: McGill
University dissertation.
121
Gribanova, Vera. 2013a. A new argument for verb-stranding verb phrase ellipsis.
Linguistic Inquiry 44(1). 145–157.
Gribanova, Vera. 2013b. Verb-stranding verb phrase ellipsis and the structure of
the Russian verbal complex. Natural Language and Linguistic Theory 31(1). 91–136.
Gribanova, Vera. 2017. Head movement, ellipsis, and identity. Stanford Univer-
sity.
Gribanova, Vera. To appear. Head movement and ellipsis in the expression of
Russian polarity focus. Natural Language and Linguistic Theory .
Hardt, Daniel. 1993. Verb phrase ellipsis: Form, meaning, and processing: University
of Pennsylvania, Philadelphia dissertation. Distributed as IRCS report 93-23.
Harizanov, Boris. 2014. Clitic doubling at the syntax-morphophonology inter-
face: A-movement and morphological merger in Bulgarian. Natural Language
and Linguistic Theory 4(32). 1033–1088.
Harizanov, Boris. 2017. On the nature of syntactic head movement. Colloquium,
Department of Linguistics. McGill University, Montréal, Québec.
Harizanov, Boris & Vera Gribanova. 2018. Whither head movement? Stanford
University.
Holmberg, Anders. 2001. The syntax of yes and no in Finnish. Studia Linguistica
55(2). 140–174.
Houser, Michael, Line Mikkelsen & Maziar Toosarvandani. 2011. A defective aux-
iliary in Danish. Journal of Germanic Lingusitics 23(3). 245–298.
Koopman, Hilda & Anna Szabolcsi. 2000. Verbal complexes. Cambridge, MA: MIT
Press.
Landau, Idan. 2017. Missing objects in Hebrew: Argument ellipsis, not VP ellipsis.
Lipták, Anikó. 2013. The syntax of emphatic positive polarity in Hungarian: evi-
dence from ellipsis. Lingua 128. 72–92.
Manetta, Emily. 2017. We say “how high?”: Adverbs, negation, and verb move-
ment in a verb-final language. This volume.
Manetta, Emily. To appear. Verb phrase ellipsis and complex predicates in Hindi-
Urdu. Natural Language and Linguistic Theory .
Matushansky, Ora. 2006. Head movement in linguistic theory. Linguistic Inquiry
37(1). 69–109.
McCloskey, James. 1996. On the scope of verb raising in Irish. Natural Language
and Linguistic Theory 14. 47–104.
McCloskey, James. 2012. Polarity, ellipsis and the limits of
identity in Irish. Workshop on Ellipsis, Nanzan University.
http://ohlone.ucsc.edu/∼jim/PDF/nanzan-handout.pdf.
McCloskey, James. 2016. Ellipsis, polarity and the cartography of verb-initial or-
ders in Irish. UC Santa Cruz.
Merchant, Jason. 2001. The syntax of silence: Sluicing, islands and the theory of ellipsis.
Oxford: Oxford University Press.
Merchant, Jason. 2013. Voice and ellipsis. Linguistic Inquiry 44(1). 77–108.
Merchant, Jason. 2017. Verb-stranding predicate ellipsis in Greek. This volume.
Ngonyani, Deo. 1996. VP ellipsis in Ndendeule and Swahili applicatives. In Ed-
ward Garrett & Felicia Lee (eds.), Syntax at Sunset, UCLA working papers in syn-
122
tax and semantics, Number 1, 109–128. Department of Linguistics, UCLA.
Platzack, Christer. 2013. Head movement as a phonological operation. In Lisa Lai-
Shen Cheng & Norbert Corver (eds.), Diagnosing syntax, 21–43. Oxford: Oxford
University Press.
Rizzi, Luigi & Ian Roberts. 1989. Complex inversion in French. Probus 1(1). 1–30.
Roberts, Ian. 2010. Agreement and head movement: Clitics, incorporation, and defective
goals. Cambridge, MA: MIT Press.
Ross, John Robert. 1969. Guess Who? In Robert Binnick, Alice Davidson, Geor-
gia M. Green & Jerry L. Morgan (eds.), Papers from the 5th regional meeting of the
chicago linguistic society, 252–286. Chicago, IL: Chicago Linguistic Society.
Sailor, Craig. To appear. The typology of head movement and ellipsis: A reply to
Lipták and Saab (2014). Natural Language and Linguistic Theory .
Santos, Ana Lúcia. 2009. Minimal answers. ellipsis, syntax and discourse in the ac-
quisition of European Portuguese. Amsterdam/Philadelphia: John Benjamins.
Schoorlemmer, Erik & Tanja Temmerman. 2012. Head movement as a PF-
phenomenon: evidence from identity under ellipsis. In Jaehoon Choi, E. Alan
Hogue, Jeffrey Punske, Deniz Tat, Jessamyn Schertz & Alex Trueman (eds.),
Proceedings of the 29th West Coast Conference on Formal Linguistics, 232–240.
Somerville, MA: Cascadilla Press.
Schuyler, Tamara. 2001. Wh-movement out of the site of VP ellipsis.
Travis, Lisa. 1984. Parameters and effects of word order variation: MIT dissertation.
Vikner, Sten. 1995. Verb movement and expletive subjects in the Germanic languages.
Oxford: Oxford University Press.
123
A definiteness effect with theme passives in
West Flemish?*
Liliane Haegeman
DiaLing Ghent University
Table 1: Germanic double object patterns and passivisation (Haddican and Holmberg 2014)
English (1) Danish Swedish Norwegian German (2) Dutch (3)
√ √ √ √
goal * * (%)
passive
√ √ √ √ √
theme ( ) * ( )
passive
berg (2012, 2014). The standard view on Dutch is that the goal passive is unacceptable, hence ∗, but I
will show presently that this does not represent all speakers, which the % sign is intended to show.
124
(2) a. * Er wurde die Blumen geschenkt.
he.nom was the flowers.acc given
‘He was given the flowers.’
b. Die Blumen wurden dem Mann geschenkt.
the flowers.nom were the.dat Mann given
‘The flowers were given to the man.’
The claim that, like German, Dutch only has the theme passive and lacks a canon-
ical goal passive is the core of Broekhuis and Cornips (2004, 2012), it is replicated
in Broekhuis, Corver and Vos (2015), and is also adopted elsewhere in the gener-
ative literature (Alexiadou et al 2014:10).
(3) a. * Hij werd het eten bezorgd (door mij).
he was the food delivered (by me)
‘He was delivered the food by me.’
b. Het eten werd hem bezorgd (door mij).
the food was him delivered (by me)
‘The food was delivered to him by me.’
However, this claim is clearly a simplification and a subset of speakers of Dutch
accept and produce goal passives (see Declercq 2016). Since the early 20th cen-
tury, attestations of goal passive have been reported for Dutch (Wellander 1920,
de Vooys 1949: 332-3, Van Haeringen 1956, 2006 :70, Kooiman 1963, Langendonck
1968, Den Besten 1981, Declercq 2016 on the recent literature). In the ANS, the
comprehensive grammar of Dutch (Haeseryn et al 1997), as well as in the more
formal literature (Broekhuis, Corver and Vos 2015), these data are often either ig-
nored or set aside as performance errors. However, recent elicitation of Flemish
and Dutch speakers reveals that throughout the Dutch speaking area, a subset
of speakers accept goal passives. Google attestations of goal passives are also
widely found (Haegeman 2016).
I will not address the status of standard Dutch here, but I concentrate on one
of its dialects, West Flemish (WF), in which the passivisation strategy seems to be
markedly different from that in standard Dutch. It is uncontroversial that that
WF has goal passives (cf. section 1.2), and hence the question arises where while
the passivization of the theme argument at first sight leads to unacceptability,
further scrutiny of the data reveals that the what might be interpreted as an un-
grammatical pattern is in fact unacceptable because of information structural
properties.
125
tance rate for goal passives than the same questionnaire distributed in 50 East
Flemish locations.
Observe that when the indirect object is an indefinite DP, as with the nominal
nen student ‘a student’ in (5a), the goal passive remains available, but der insertion
becomes obligatory (5b). The expletive varies in form depending on the preceding
consonant and shows up as der or ter depending on the preceding consonant. I
will refer to it as der insertion.
The insertion of expletive der is fully in line with the grammatical properties of
WF: in this language, der insertion is systematically obligatory with WF indefinite
subjects, as also shown in the active transitive expletive patterns in (6): regardless
of the position of the indefinite subject drie studenten ‘three students’ to the right
(6a) or to the left (6b) of the adjunct gisteren ‘yesterday’, der insertion is obligatory.
The insertion of der is correlated directly with the indefiniteness of the subject;
it does not arise with indefinite direct objects in active sentences as shown in (7):
(6) a. dan ∗(der) gisteren drie studenten dienen boek gekocht een
that-pl ∗(there) yesterday three students that book bought have
b. dan ∗(der) drie studenten gisteren dienen boek gekocht een
that-pl ∗(there) three students yesterday that book bought have
‘that three students bought that book yesterday’
(7) dan ∗(der) de studenten gisteren drie boeken gekocht een
that-pl ∗(there) the students yesterday three books bought have
‘that the students bought three books yesterday
Given that WF has a productive goal passive, the question arises how it fits into
the typology schematized in table 1, i.e. is the theme passive available in WF? At
first sight, the judgments in (8) are puzzling. When the theme is a definite DP,
126
theme passivisation is judged to be highly degraded (8a,b)2 and the alternative
pattern with a goal PP (8c) is highly preferred. On the other hand, when the
theme is indefinite, passivisation is perfectly acceptable, as is the pattern with
the PP goal (8d).
(8) a. *?? dan die nieuwe pillen Valère voorgeschreven wieren
that-pl those new pills Valère prescribed were
b. * dan Valère die nieuwe pillen voorgeschreven wieren
that-pl Valère those new pills prescribed were
c. dan die nieuwe pillen an Valère voorgeschreven wieren
that-pl those new pills to Valère prescribed were
‘that those new pills were prescribed to Valère’
d. dan der (an) Valère andere pillen voorgeschreven wieren
that-pl there (to) Valère other pills prescribed were
‘that other pills were prescribed to Valère’
This contrast is unexpected. One way of handling it might indeed be to con-
clude that WF in fact lacks a theme passive and to analyze (8d) as some form
of impersonal pseudo passive construction. In this paper, I argue against such a
conclusion. I will show first that the indefinite theme passive with the nominal
goal in (8d) is a genuine passive, and I will then show that though the definite
theme passive in (8a) is indeed degraded, this is an epiphenomenon: a range of
repair strategies can salvage the definite theme passive. The data will reveal that
the unacceptability of (8a) is in fact a function of information structural prop-
erties. Based on the discussion, I will thus conclude that WF has both a goal
passive and a theme passive, and thus patterns most closely with Norwegian in
Haddican and Holmberg’s (2014) survey.
2 Observe that in this respect, WF and standard Dutch are clearly different. In relation to the defi-
nite theme passive, I provide just some citations from the literature. Ackema and Neeleman (2016:160,
their (25)) illustrate the standard Dutch definite theme passive in (1): in (1a) the theme, de boeken ‘the
books’, precedes the goal, Jan; in (1b) the theme follows the goal. Observe that in both cases the
finite passive auxiliary werden ‘became’ agrees with the plural theme. As seen in (8), both orders are
degraded in WF with the WF equivalent of (1b) being worse than the WF analogue of (ia).
Similarly, Broekhuis, Corver and Vos (2016: 1601) give (2) as grammatical in standard Dutch. Again
the analogue of this example would be ungrammatical in WF because the definite subject must be
adjacent to the finite verb in second position, thus neither the agent phrase nor the goal Marie can
intervene.
127
The paper is organized as follows: Section 2 discusses the indefinite theme
passive, section 3 shows that though at first sight the definite theme passive is
unavailable this conclusion is incorrect in view of a wider range of data in which
such passives are fully productive. Section 4 explores the rescuing strategies for
definite theme passives and shows that the unacceptability of the definite theme
passive in (8a) is the outcome of the interplay of information structure and syntax.
An additional illustration of the relevance for information structure for accept-
ability will be provided. Section 5 is a brief summary of the paper.
128
(10) dan der Valère almettekeer andere pillen voorengeschreven wieren
that-pl there Valère suddenly other pills prescribed were
‘that all of a sudden Valère was prescribed other pills’
Because the theme subject is indefinite, insertion of expletive (der) in the subject
position is obligatory. This is a property that characterises all indefinite subjects
in WF, as already discussed in relation to (6). As shown in (12), the absence of der
leads to ungrammaticality:
A full analysis of the indefinite theme passive is not the goal of this paper. I pro-
visionally assume the partial representation in (14). I follow Haddican and Holm-
berg (2012) and assume that in theme passives, the goal merges as the specifier
129
of ApplP and receives case from the head Lk. Expletive der merges as SpecvP,
it agrees with VP internal theme, and it moves to specTP. See also Deal (2009).
Crucial is that the case of the goal is unaffected by passivisation; it may remain
VP internally (14a) or it may undergo Object Shift (14b). Observe also that the
Goal must be taken not to block agreement between T/der and the vP-internal
theme.
(14) a. [TP der T [vP der v . . . [LkP [ApplP goal [VP theme V ]]]]]
b. [TP der T goal [vP der v . . . [LkP [ApplP goal [VP theme V ]]]]]
One may try to account for the unacceptability of the definite theme passive by,
for instance, appealing crucially to the role of der as a mediator in T-agreement
and in the satisfaction of the EPP on T. As a specific implementation, one might,
for instance follow Mikkelsen (2002) and propose that the theme is case-marked
by the passive participle and is hence inactive. The expletive der satisfies the EPP
130
on T. Because of the definiteness effect associated with der insertion, with a def-
inite theme there would be no way of satisfying the EPP on T. However, taking
into account a wider range of data will reveal that in fact, WF definite theme
passives are not always unacceptable. Thus, while the patterns in (15a) and (16a)
remain unacceptable, it will turn out that they are not to be used in support of
a claim that WF lacks a definite theme passive. On the other hand, to the best
of my knowledge, the (b)-variants of (15) and (16) are solidly ungrammatical in all
contexts and, this is in line with the grammar of WF in that there is a strict adja-
cency requirement on the definite subject in relation to the complementizer or
the finite verb in C.
erst ‘probably’, al ‘already’), a nominal verleden joare (‘last year’), a PP (in Leuven ‘in Louvain’).
4 The degradation of the definite theme passive and the repair strategies which consist of supply-
ing adjuncts are reminiscent of the observations on English passives in Grimshaw and Vikner (1993).
These authors show that English passives of monotransitives sometimes necessitate what seems to
be an obligatory adjunct, a phenomenon which I return to briefly in sections 4.3.1 and 4.4.2, and which
they account for in terms of Event structure. I will not pursue their account here because, as shown
in Section 4.4.1, the effect discussed for WF also arises with a subset of active sentences, for which
Grimshaw and Vikner’s (1993) account would not be relevant. In addition, with respect to the passive
patterns discussed here, the specific restrictions on the type of adverbial adjuncts noted in Grimshaw
and Vikner (1993) for English do not obtain in WF: to the best of my knowledge, any middle field con-
stituent can salvage the offending definite theme passive.
131
e. dan die posten Karen alle twee beloofd zyn
that-pl those jobs Karen both promised are
‘that Karen was promised both those jobs’
f. dan die pillen Karen niet/wel vuorengeschreven wieren
that those pills Karen not/wel prescribed were
‘that Karen was not/indeed prescribed those pills’
g. dan die pillen Karen toch vuorengeschreven wieren
that-pl those pills Karen part prescribed were
‘that Karen was nevertheless prescribed those pills’
These rescuing strategies do not salvage (15b) or (16b) in which the definite theme
in the passivized pattern is separated from the complementizer by the goal. This
is shown in (17’) below: these minimally differ from the corresponding examples
in (17) by the position of the additional material. I return to this contrast in section
3.2.
(17’) a. ∗ dan Karen die posten gisteren / twee keers beloofd zyn
that-pl Karen those jobs yesterday / two times promised was
b. ∗ dan Karen die posten al beloofd woaren
that-pl Karen those jobs already promised were
c. ∗ dan Karen die posten verzekerst/alleszins beloofd woaren
that-pl Karen those jobs probably/definitely promised were
d. ∗ dan Karen die pillen van den dokteur voorgeschreven wieren
that Karen those pills of the doctor prescribed were
e. ∗ dan Karen die posten alle twee beloofd zyn
that-pl Karen those jobs both promised are
f. ∗ dan Karen die pillen niet/wel vuorengeschreven wieren
that Karen those pills not/wel prescribed were
g. ∗ dan Karen die pillen toch vuorengeschreven wieren
that-pl Karen those pills part prescribed were
132
(15b), (16b) and the primed examples in (17). Like the subject nominal, the definite
theme argument is adjacent to the complementizer: in (18a) the intervention of
the adverbial verzekerst (‘probably’) renders the example ungrammatical. Like the
subject nominal, the definite theme argument agrees with finite verb and with
the complementizer: this is illustrated for the plural nominal die pillen ‘those pills
in (18b) and for the plural nominal die posten (‘those jobs) in (18a) and (18c). When
pronominal, the definite theme is realised as a nominative pronoun, and like
other nominative DPs, it (marginally5 ) allows for pronoun doubling (18d). When
relativized, the definite theme is associated with relative die and the pattern gives
rise to dat/die alternations (18e,f) (cf. Haegeman 1984). I take these data to be suf-
ficient to conclude that the indefinite theme in the non-primed examples in (17)
is indeed the subject of the passive verb, and hence that contrary to first impres-
sions, in addition to the goal passive, documented in (4) and (5), WF also has a
theme passive.
133
by phrase, a floating quantifier or some discourse particle. (19) shows that the
non-primed licit patterns in (17) systematically become ungrammatical when the
goal Karen fails to undergo OS:
(19) a. ∗ dan die posten gisteren/twee keers Karen beloofd zyn
that-pl those jobs yesterday/two times Karen promised zyn
b. ∗ dan die posten al Karen beloofd woaren
that-pl those jobs already Karen promised were
c. ∗ dan die posten verzekerst/alleszins Karen beloofd woaren
that-pl those jobs probably/definitely Karen promised were
d. ∗ dan die pillen van den dokteur Karen voorgeschreven zyn
that-pl those pills of the doctor Karen prescribed are
e. ∗ dan die posten alle twee Karen beloofd zyn
that-pl those jobs both Karen promised are
f. ∗ dan die pillen niet/wel Karen vuorengeschreven wieren
that-pl those pills not/wel Karen prescribed were
g. ∗ dan die pillen toch Karen vuorengeschreven wieren
that-pl those pills part Karen prescribed were
Given that the definite theme DP is adjacent to the complementizer or to the in-
verted finite verb and that it has subject properties, I conclude that it has moved
to the canonical subject position. Apparently, such theme movement is only pos-
sible provided the goal DP undergoes OS, or, using Broekhuis’s (2009) terminol-
ogy, theme movement to the subject position has a “push up” effect. 6 There is
no push up effect when a definite theme becomes the subject of a ditransitive
passive in which the goal is realized as a PP: as shown by the examples in (20), a
PP goal argument may follow all middle field material:
(20) a. dan die posten gisteren/al an Karen beloofd zyn
that-pl those jobs yesterday/already to Karen promised zyn
‘that those jobs were promised to Karen yesterday/already’
b. dan die pillen van den dokteur an Karen voorgeschreven zyn
that-pl those pills of the doctor to Karen prescribed are
‘that those pills were prescribed to Karen by the doctor’
up effect: the goal argument Karen may follow the adjunct or it may precede it. Recall also that the
adjunct itself is not obligatory.
(1) dan der (verleden joare) Karen (verleden joare) twee posten beloofd wieren
that-pl there (last year) Karen (last year) two jobs promised were
134
rect object in WF (and Dutch) ditransitives. As I have demonstrated at length
in my earlier work on order preservation (Haegeman 1993a,b,c, 1994), leftward
movement of a direct object is only possible provided the indirect object under-
goes OS. The pattern is discussed, among others, in den Dikken (1995: 207-208)
Anagnostopoulou 2003: 215-20, Fox and Pesetsky 2005, Broekhuis 2009 etc. For
instance, Broekhuis (2009: 243) states that ‘OS is blocked when it has to cross
the main verb or some co-argument, the moved object forces movement of these
elements as well’. I first briefly sum up the relevant WF data.
In ditransitive sentences with a middle field adjunct and without movement
of the direct object, the indirect object may follow or precede the middle field
adjunct: in (21a) the adjunct verzekerst ‘probably’ precedes both indirect object
and direct object, in (21b) the indirect object has undergone OS and precedes the
adjunct, in (21c) both indirect object and direct object have undergone OS and
precede the adjunct:
(21) a. dat Valère verzekerst zen broers dienen boek angeroaden eet
that Valère probably his brothers that book recommended has
b. dat Valère zen broers verzekerst dienen boek angeroaden eet
that Valère his brothers probably that book recommended has
c. dat Valère zen broers dienen boek verzekerst angeroaden eet
that Valère his brothers that book probably recommended has
‘that Valère has probably recommended this book to his brothers’
When the direct object is relativized, OS of the indirect object is obligatory: (22a),
in which the indirect object DP zen broers ‘his brothers’ follows the adjunct verzek-
erst ‘probably’ is ungrammatical: in the licit (22b), the indirect object precedes the
adjunct. The push up effect is only observed with DP indirect objects. When the
indirect object is realised as PP, it is immune to the push up effect. In (22c), rela-
tivization of the direct object is compatible with the indirect object PP remaining
to the right of the adverbial adjunct verzekerst ‘probably’.
The push up effect is also observed with clitic objects such as t (‘it’) or ze (‘her’,
‘them’) (Haegeman 1993a,b,c). In (23a) the object clitic het (‘it’) precedes the indirect
object Marie. OS of the indirect object is crucial here: in (23b), the clitic precedes
an indirect object which has not undergone OS and the result is ungrammatical.
(23d) shows that the PP indirect object is not subject to the push up effect.
135
(23) a. dat Valère verzekerst Marie dat geld beloofd heeft
that Valère probably Marie that money promised has
‘that Valère probably promised the money to Mary’
b. dat Valère het Marie waarschijnlijk beloofd heeft
that Valère it Marie probably promised has
‘that Valère probably promised it to Mary’
c. * dat Valère het waarschijnlijk Marie beloofd heeft
that Valère it probably Marie promised has
d. dat Valère het verzekerst an Marie beloofd heeft
that Valère it probably to Marie promised has
‘that Valère probably promised it to Marie’
The push up effect also regulates the distribution of indirect and direct objects
in verb projection raising patterns, the indirect object and the direct object may
remain with the ‘raised’ VP, as shown in (24a). However, when the object is moved,
by relativization (24b) or because it is a weak pronoun (24e), the indirect object
must also be extracted, failure to do so leads to severe ungrammaticality (24c,f).
As before, PP indirect objects are not sensitive to the constraint (24d,g).
Returning to the definite theme passive the addition of middle field material sal-
vages what otherwise would be unacceptable patterns. The obligatory OS of the
goal argument is not specific to the definite theme passive and follows from
whatever accounts for the push up effect observed in other contexts.
136
3.5 Additional rescuing strategies
While adding weight via the insertion of middle field material is one productive
strategy for rescuing the ‘unadorned’ indefinite theme passive, other strategies
are available which, at least at first sight, do not affect the middle field. I discuss
two of these here, both of which consist in adding focused material.
The sentence-final additions in (25a-d) are not presupposed, but like extraposed
constituents they constitute part of the focus (new information) (Broekhuis et al
2016:1549). Indeed, as shown in the next section, the addition of focused material
also salvages the ‘unadorned’ definite theme passive.
137
moet je beetje voor werken: er moet iemand anders zijn waarvan ie-
mand denkt/beweert dat hem/haar die slaapmiddelen waren voorge-
schreven.
Tr: ‘you must do some work here: there must be someone else about
whom it is thought that those sleeping pills had been ascribed to that
person.’:
Clearly then, this informant adds contrastive stress to Ciro in order to salvage the
unadorned definite theme passive. (26b) is also much improved: in this example,
the goal is a negative quantifier niemand, which also receives focal stress.
(26) a. ? K peinzigden dan die pillen Ciro voorgeschreven wieren.
I thought that-pl those pills Ciro prescribed were
‘I thought that CIRO had been prescribed those pills.’
b. ? K peinzigden dan die pillen niemand voorgeschreven wieren.
I thought that- plthose pills no one prescribed were
‘I thought that no one was prescribed those pills.’7
Similarly, stressing the lexical verb itself may increase acceptability: in (27) stress-
ing the particle of ‘off’ creates an effect of contrastive or corrective focus on the
verb: (27) is understood to introduce a contrast with other events in which a
course was recommended or alternatively it corrects a previous proposition ac-
cording to which a course was recommended:
of goal and theme (1); for (19b) with the indefinite subject niemand both orders in (2) are acceptable.
Der insertion is obligatory (cf. section 1.2 for discussion).
138
sight, a purely syntactic account that brings together these strategies is not ob-
viously available. In the next section I will propose that in fact the data suggest
that what excludes the unadorned definite theme passive is a deficit in terms of
information structure: the unadorned definite theme passive is informationally
incomplete in the sense of Erteschik-Shir (2007:187-9).
139
(i) the definite subject cannot be separated from the complementizer or from
the finite verb in C;
(ii) the indirect object always precedes the direct object.
When the direct object moves across the indirect object, for instance in the case
of relativization (22) or when it is a clitic, as shown in (23), repeated here in (29a,b),
this ‘pushes’ up the indirect object, which has to undergo OS (Broekhuis 2009).
(29d-e) schematically represents the effect of Push up on clitic movement: basi-
cally, the data amount to saying that the direct object can only reorder with the
indirect object provided the latter undergoes OS. The pattern as such is of im-
portance and had led to wide ranging discussions in the literature (cf. den Dikken
1995: 207-208, Anagnostopoulou 2003: 215-20, Fox and Pesetsky 2005, Broekhuis
2009 and many others). I will not attempt here to provide an account for the push
up effect and refer to the literature cited, but I will focus on its interaction with
and relevance for passivisation in WF.
(29) a. dat Valère het Marie waarschijnlijk beloofd heeft
that Valère it Marie probably promised has
‘that Valère probably has promised it to Marie’
b. * dat Valère het waarschijnlijk Marie beloofd heeft
that Valère it probably Marie promised has
c. dat DOcl IO adjunct IO DO Vlex aux
d. * dat DOcl adjunct IO DO Vlex aux
In the definite theme passives, the theme moves to the canonical subject posi-
tion. By assumption, this is to satisfy the EPP on T. The movement of the object
entails that the goal-DP has to undergo OS (hence it appears to the left of the
middle field adjunct/particle).8
against the ‘push up’ effect (Broekhuis 2009): in (1) the negative IO/goal enters into a concord re-
lation with a negative adverb to its right, suggesting that it has vacated the vP internal position and
undergone neg-movement to a middlefield position (Haegeman 1995, Haegeman& Zanuttini 1993):
140
4.3 Theme passivisation in West Flemish
4.3.1 Theme passives of ditransitives
My starting point in this paper was the observed degradation in acceptability of
the unadorned definite theme passive in WF ditransitive sentences. (30a) has the
schematized representation in (30b). Because the theme argument die pillen ‘those
pills’ moves to the canonical subject position, the goal argument Valère must also
evacuate the VP.
(30) a. ∗?? dan die pillen Valère vuorengeschreven wieren
that those pills Valère prescribed were
b. ∗?? dan die pillen Valère [VP Valère die pillen vuorengeschreven] wieren
The net outcome of the derivation is thus that with respect to ditransitive pat-
terns, the ‘unadorned’ definite theme passive lacks a VP-internal constituent which
can carry the focus. Lacking focus, the ‘unadorned’ definite DO/theme passive
is ‘informationally incomplete’ (in the sense of Erteschik-Shir 2007:187-9). Sal-
vaging the pattern is achieved by supplying a constituent that can constitute a
focus, be this in a middle field position or in an extraposed position.
The degradation of the unadorned definite theme passive in WF (30a) is like
that of the English passive in (31a), which was originally discussed by Grimshaw
and Vikner (1993) in terms of a deficiency in Event Structure and was subse-
quently reanalyzed in terms of an information structural deficiency by Erteschik-
Shir (2007: 189, her (87a)):
(31) b. The book was written last year/in 2005/by Chomsky. (2007:189, (88))
c. The book was revised. (2007:189: (87b))
d. The book was destroyed. (2007:189: (87c))
141
(32) a. dan die pillen an Valère voorgeschreven wieren
that-pl those pills to Valère prescribed were
‘that those pills were prescribed to Valère’
b. dan die pillen [VP an Valère die pillen vuorengeschreven] wieren
The degradation observed for the unadorned definite theme passive does not
arise with an unadorned indefinite theme passive. Because the indefinite theme
remains VP internally (13), the VP contains material eligible for carrying focus.
As a result, the need to supply an alternative focus does not arise.
On the basis of the passivisation data discussed in this paper, I hope to have
shown that WF does have both a productive goal passive and a productive theme
passive. Restrictions such as that revealed by the judgements produced on the
definite theme passive in section 1.2 do not challenge this conclusion.
142
(35) a. da Valère Marie dat werk beloofd eet
that Valère Marie that job promised has
‘that Valère has promised Marie that job’
b. * da Valère dat werk Marie beloofd eet
that Valère that job Marie promised has
c. ?? da Valère Marie t beloofd eet
that Valère Marie it promised has
d. da Valère t Marie beloofd eet
that Valère it Marie promised has
‘that Valère has promised it to Marie’
However, to my ear, the degradation in (35c) is not as sharp as that in (35b), which
violates order preservation. Indeed, the order in (35c), where the object pronoun
follows the indirect object DP, actually becomes acceptable provided the pro-
noun is followed by middle field constituents such as adverbial, polar particle,
discourse particle (36a) or floating quantifier (36b).
It is not clear, though, that such a fully syntactic analysis is required to account
for the degradation of (35c). Again, along the lines developed in section 3, it would
appear that these data follow from the interaction between information structure
and object shift.
It is uncontroversial that the pronominal object t (‘it’) encodes presupposed/given
information and exits the VP by OS. As discussed, object movement entails that
the indirect object, here Marie, is pushed up and must also undergo OS (cf. (23)
143
and (24g)). This is schematized in (38). Because both the object pronominal and
the indirect object evacuate the VP, the resulting sentence becomes information-
ally “incomplete” (in the sense of Erteschik-Shir 2007) and supplying alternative
focus will salvage the sentence.
lows:while direct objects normally follow nominal indirect objects under a neutral intonation pattern
[1a], weak pronominal direct objects normally precede indirect objects. Example [1b] shows that this
holds regardless of whether the indirect object is non-pronominal or pronominal. It should further
be noted that it also holds if the two object pronouns have the same form: the first object pronoun in
dat Peter ‘m ‘m aanbood ‘that Peter offered it to him’ is construed as the direct object.
We may speculate that the degradation in the position of the weak pronoun <‘m> to the right of the
indirect object Marie in (1b) may well be of the same nature as that discussed for WF in the present
paper and thus may be due to the interaction between syntax and information structure.
144
In the paper, the definiteness effect in theme passives, which is initially sug-
gested by the contrast in judgements between the ‘unadorned’ definite theme
passive and its indefinite counterpart, is reinterpreted in terms of information
structure. Definite Theme passivisation ‘pushes up’ the goal argument, a pattern
well known from the literature (cf. Haegeman 1993a,b,c,1994, Broekhuis 2009). Ar-
guments which are extracted out of VP by OS are interpreted are given or pre-
supposed. This entails that in the case of definite theme passivisation, the shifted
goal, which due to push up must undergo OS, must be presupposed. This leads
to an imbalance in information structure: by movement of both the theme and
the goal, the resulting VP lacks the new information focus which is by default
associated with a VP internal constituent and the resulting utterance is ‘informa-
tionally incomplete’ (Erteschik-Shir 2007:187-9). The ‘definiteness effect’ detected
in the initial acceptability judgements is thus reanalyzed as a by-product of the
push up effect on object movement. The diverse repair strategies are unified in
that they all supply a focus. Indefinite theme passives are fully acceptable because
the indefinite theme remains VP-internally, hence there is no push up effect and
the theme itself can constitute a VP-internal focus.
The paper shows that the distribution of WF object pronouns too can be
partly interpreted in terms of the interaction between syntax and information
structure and thus does not necessitate the highly complex syntactic analysis out-
lined in Haegeman (1993a,bc, 1994).
A question for future work is how best to capture the role of information
structure in determining acceptability and to which degree these effects are them-
selves inscribed into the syntax.
References
Ackema, Peter and Ad Neeleman (2016). Features of Person. Ms. University of
Edinburgh /UCL.
Afarli, Tor. 1992. The syntax of Norwegian passive constructions. Amsterdam: Ben-
jamins.
Alexiadou, Artemis, Elena Anagnostopoulou and Christina Sevdali. 2014. Opaque
and transparent datives, and how they behave in passives. The Journal of Com-
parative Germanic Linguistics 17:1-34.
Anagnostopoulou, Elena. 2003. The Syntax of Ditransitives. Berlin: Mouton de
Gruyter.
ANSBesten, Hans. den. 1981. A case filter for passives. In Adriana Belletti, Lu-
cia Brandi and Luigi Rizzi (eds), Theory of Markedness in Generative Grammar.
Proceedings of the 1979 GLOW Conference. Pisa Scuola Normale di Pisa. 65-122.
Broekhuis, Hans. 2009. Holmberg’s Generalization. Blocking and push up. In
Artemis Alexiadou, et al., Advances in Comparative Germanic Syntax. Amster-
dam/Philadelphia: John Benjamins Publishing Company, 219-245.
Broekhuis, Hans and Leonie Cornips. 2004. Undative constructions. Linguistics
32: 173-189.
145
Broekhuis, Hans and Leonie Cornips. 2012. The verb krijgen ‘to get’ as an unda-
tive verb. Linguistics 50: 1205-1249.
Broekhuis, Hans, Norbert Corver and Riet Vos. 2015. Verbs and verb phrases. Vol I.
The syntax of Dutch. Amsterdam University Press.
De Vooys, Cornelis Gerrit Nicolaas 1949/1960. Nederlandse Spraakkunst. Gronin-
gen.
Deal, Amy Rose. 2009. The origin and content of expletives. Syntax 12: 285-323.
Declercq, Kato. 2016. Naar een verbeterde beschrijving van het persoonlijk passief
in de Algemene Nederlandse Spraakkunst. Een studie van de recente taalkundige
literatuur. Master paper. Dutch Linguistics. Ghent University.
Dikken, Marcel den. 1995. On the syntax of verb-particle, triadic, and causative
Constructions. Oxford: Oxford University Press.
Dhaenens, Gilles. 2014. Perspectivische variatie van de actieve dubbelobject-
constructie: passieve en receptieve varianten. Bachelor paper, Ghent Univer-
sity.
Erteschik-Shir, Nomi. 2007. Information structure: the syntax-discourse interface.
OUP.
Féry, Caroline and Vieri Samek-Lodovici. 2006. Focus projection and prosodic
prominence in nested foci. Language 82: 131-150.
Fox, Dany and Pesetsky, David. 2005. Cyclic Linearization of Syntactic Structure.
Theoretical Linguistics 31: 1-45.
Grimshaw, Jane and Sten Vikner. 1993. Obligatory adjuncts and the structure of
events. Knowledge and language concepts. Dordrecht: Kluwer, 143-55.
Haddican, William and Anders Holmberg. 2010. Theme-goal ditransitives and
theme passivisation in British English dialects. Lingua 120: 2424-2443.
Haddican, William and Anders Holmberg. 2012. Object movement symmetries
in British English dialects: experimental evidence for a mixed case/locality
approach. Journal of Comparative Germanic Linguistics 15: 189–212.
Haddican, William and Anders Holmberg. 2014. Kinds of Object Symmetry. In
L. Veselovská and M. Janebová (eds), Complex visibles out there: Proceedings of
Olinco 2014, Palacký University, Olomouc: 145-162.
Haegeman, L. 1984. Die and dat in West Flemish relative clauses. In: Hans Ben-
nis and W. Van Lessen Kloeke (eds.). Linguistics in the Netherlands. Dordrecht.
Foris. 83-91.
Haegeman, L. 1986a. Passivisation in Flemish dialects. In Frits Beukema and
Aafke Hulk (eds), Linguistics in the Netherlands. Dordrecht: Foris. 71-78.
Haegeman, L. 1986b. The double object construction in West Flemish. The Lin-
guistic Review 5: 281-300.
Haegeman, L. 2016. The canonical goal passive in Dutch and its dialects. Ling-
buzz: lingbuzz/003105
Haegeman, Liliane 1993c. The distribution of object pronouns in West Flemish.
In: Lars Hellan and Luigi Rizzi (eds.). Eurotyp working papers. Programme in
language typology. European Science Foundation.
Haegeman, Liliane. 1993a. Some speculations on argument shift, clitics and cross-
ing in West Flemish. In: Werner Abraham and Josef Bayer (eds.). Linguistische
146
Berichte, Special issue Dialektsyntax: 131-16
Haegeman, Liliane. 1993b. The morphology and distribution of object clitics in
West Flemish. Studia Linguistica 47: 57-94.
Haegeman, Liliane. 1994. The typology of syntactic positions: L-relatedness and
the A/A’ distinction. In Werner Abraham en Jan-Wouter Zwart (eds.). GAGL,
37, Special issue: The minimalist programme: 115-157.
Haeringen, Coenraad Bernardus van 1956. Nederlands tussen Duits en Engels. Den
Haag: Servire. DBNL 2006.
Haeseryn, Walter, Kirsten Romijn, Guido Geerts, Jaap de Rooij and Maarten C.
van den Toorn. 1997. Algemene Nederlandse Spraakkunst (ANS). Groningen:
Martinus Nijhoff/ Deurne: Plantyn.
Holmberg, A. 1999. Remarks on Holmberg’s generalisation. Studia Linguistica: 1-
39.
Kooiman, Klaas. 1963. Als in het Engels, maar niet uit het Engels. De Nieuwe
Taalgids 26: 33-40.
Kratzer, Angelika and Lisa Selkirk. 2007. Phase theory and prosodic spell out.
The Linguistic Review 24: 93-135.
Langendonck, Willy. van. 1968. Het meewerkend voorwerp van de aktieve zin als
onderwerp van de passieve zin. Leuvense Bijdragen. Tijdschift voor Germaanse
Filologie. 57: 101-118.
Mikkelsen, Line. Reanalyzing the definiteness effect: evidence from Danish. Work-
ing Papers in Scandinavian Syntax 69: 1-75.
Neeleman, Ad, Elena Titov, Hans van de Koot and Reiko Vermeulen. 2009. A
syntactic typology of topic, focus and contrast. Alternatives to cartography.
Studies in Generative Linguistics. 100, 15-51
Van den Toorn, Maarten C. 1971. Enkele opmerkingen over het indirect object.
Levende Talen 274: 32-41.
Wellander, Erik. 1920. Over den datief als subject van een passieve constructie.
De Nieuwe Taalgids 14: 291-299.
147
Ellipsis as a Test for Constituency
Jorge Hankamer
University of California, Santa Cruz
1 What is ellipsis?
Some textbooks include "omission" as a constituency test, citing the optionality
of certain modifiers as indicators of constituency:
148
b. The trolls were hiding (under a bridge).
c. The (very large) dog (that I bought) eats too much.
Such examples, however, do not involve ellipsis but merely reflect the fact
that modifiers are optional. Similarly, the fact that some verbs (such as ’eat’) can
be either transitive or intransitive has nothing to do with ellipsis, but reflects
either optionality of the complement or a choice between different lexical items.1
Ellipsis, as the word is commonly used in linguistics, is a subtype of anaphora:2
a systematic relation wherein one linguistic element (the anaphor) depends for
its interpretation on some other element (the antecedent).
Further, the term "ellipsis" will not be used for A-bar phenomena such as the
gaps (primary or parasitic) in such constructions as relative clauses and compar-
ative clauses. To be sure, the generally held assumption is that these gaps re-
sult from A-bar movement, but I have seen "Comparative Deletion" mentioned
among putative "ellipsis" phenomena (en.wikipedia.org/wiki/Ellipsis_(linguistics)).
I will discuss a genuine ellipsis phenomenon occurring in comparative clauses
(Comparative Ellipsis) in section 2.5.
Other phenomena involving silence that I do not count as ellipsis are those
involving (according to most current analyses) silent pro-forms of various kinds:
big PRO and little pro, for example. Whatever analysis is assumed for those
things, their properties are too dissimilar from the ones dealt with in this paper
for it to make sense to call them by the same name. For similar reasons, "argu-
ment ellipsis" (see, for example, Cheng 2013) is not ellipsis either.3
2 A typology of ellipsis
The purpose of this section is to develop a typology of ellipsis processes which
will be useful in subsequent discussion of the suitability of ellipsis as a test for
constituency.
We will start with an inventory of the (more or less) well-studied ellipsis pro-
1 Omission is not a very good constituency test, either:
expressions that are subject to Condition A of the Binding Theory. Thus, to count as ellipsis, an "omis-
sion" must be recoverable from (linguistic) context. I will henceforth follow Hankamer and Sag 1976
in assuming that ellipsis is equivalent to "surface anaphora" and that the relevant context is inevitably
linguistic context (and not, for example, some aspect of the situational context not expressed in lin-
guistic form). I also assume, since no credible evidence contradicts it, that ellipsis is always optional.
3 Unfortunately, some people (e.g. Aoun and Benmamoun 1999, Depiante 2000, Culicover and
Jackendoff 2005) use the term “bare argument ellipsis” to mean Stripping. That, of course, is ellipsis.
149
cesses:4 VP Ellipsis, NP Ellipsis, Sluicing, Stripping, and Gapping. I will argue
that there are (at least) two kinds of ellipsis processes, with very different prop-
erties. We will then discuss several less well-studied ellipsis phenomena in the
light of the resulting typology, with an eye toward developing a fairly complete
picture of the known landscape of ellipsis.
(2) While I have never ridden a camel, Ivan has [ridden a camel].
(3) While I have never ridden a camel, I would like to [ride a camel].
(4) While I have ridden a camel, Ivan has not [ridden a camel].
(5) While I won’t ride camels, Ivan will [ride camels].
(6) I didn’t ride the camel. Ivan did [ride the camel].
The name is not entirely apt, however, because complements of the copula, which
can be DP, AP, or PP, can also be elided, and there is no reason to believe that this
is a different process:
(7) The children weren’t ready to go home, but the parents were [ready to go
home].
(8) If your GPS tells you you are in Nebraska, you probably are [in Nebraska].
(9) Harvey thinks he’s not a loser, but he is [a loser].
Adopting an insight of Zagona (1982) and Lobeck (1995), we will characterize "Verb
Phrase Ellipsis" as the elision of the complement of a T, Neg, or Aux (where, of
course, the copula is an Aux).5 Thus VPE is not characterized as ellipsis of a VP,
but as the ellipsis of the complement of one of the set of licensing heads T (in-
cluding present and past, Modals, and infinitival ‘to’), Neg, and Auxes.6
Support for this characterization comes not only from the ellipsis of non-
VP complements of the copula, but from several constructions in which a VP
is not the complement of one of the licensing heads, and such VPs cannot be
4 I use the term "ellipsis process" to denote the grammatical representation of any of several dis-
tinct phenomena involving the interpretation of a stretch of silence by direct reference to a linguistic
antecedent. My use of this term should not be taken as a commitment to any set of assumptions about
the exact mechanism by which this relation is established.
5 It won’t do to think that Auxes appear to license VPE simply because they have raised to T, be-
cause even an Aux which cannot have raised to T can license VPE:
(i) Harvey is upset, and we don’t want him to be.
(ii) We hope Harvey didn’t steal the car, but he might have.
(iii) We think Harvey stole the car, but he might not have.
This is on the standard assumption that only one Aux raises to T.
6 It is a mystery what unites this class of licensors. Mostly they are heads that select VP comple-
150
elided. First consider VPs modified by VP-initial adverbs like ’never’ and ’seldom’
(chosen because they apparently cannot re-adjoin to the right, escaping the VP):
(10) Sue has never seen a camel.
(11) * Sue has seen a camel never.
(12) * Sue has never seen a camel, and I have never [seen a camel] either.
(13) * Sue has seldom complained about money, and her mother has also sel-
dom [complained about money].
In these cases the inner VP is not the complement of a licensing head, and cannot
be elided (for example, in (13) the complement of have is the VP also seldom com-
plained about money; the inner VP complained about money is not the complement
of a licensing head).
Another case where a VP can occur without a licensing head is in Small Clause
constructions, and here too VPE is impossible:
(14) * I saw a man beating a donkey, but I didn’t see a woman [beating a don-
key].
(15) * I saw a man beating a donkey, and I saw a woman [beating a donkey]
too.
(16) * I wanted my eggs seasoned with paprika, and Harvey wanted his rice
[seasoned with paprika].
(Examples (14)-(16) are of course grammatical where what I saw or didn’t see was
a woman, and what Harvey wanted was his rice.)
Noun Phrase Ellipsis (NPE), first identified by Jackendoff (1971), appears to
elide an NP:
(17) Sue’s snide remark about the Dean was not as clever as Harvey’s [snide
remark about the Dean].
(18) You scratch my back and I’ll scratch yours [back].7
Not just any NP can be elided, however, because unless the NP is the complement
of a Possessive D (or a demonstrative D, or whatever D hosts demonstratives in
its specifier, and similarly for quantifiers), the ellipsis is not licensed:
(19) *Ivan was riding a big ugly camel, and I rode a little pretty [camel].8
7 It’s amusing that pronominal possessors in English have a special form that they must appear in
anaphora:
(i) Ivan was riding the big ugly camel, and I rode the little pretty one.
One(s) anaphora, while providing an excellent constituency test, is not ellipsis according to the stan-
dard I have adopted (it’s a kind of Deep Anaphora), so I will not discuss it further in this article. Note,
though, that there are languages in which the one-anaphor is silent:
(ii) Juan montaba el camello grande, y yo montaba el chico. [Spanish]
Juan rode the camel big and I rode the small
151
I will assume that NPE, like VPE, is licensed by certain heads, and consists in the
elision of the complement to that head (under suitable conditions of recoverabil-
ity, which are not the focus of this paper, though I will continue to assume that
recoverability requires reference to a linguistic antecedent).
An interesting subcase of NPE, investigated thoroughly in LaCara (2010), in-
volves ellipsis of the VP complement of the possessive D in constructions like
(20-21):
(20) I could stand the cat’s wailing all night, but I couldn’t stand the baby’s
[wailing all night].
(21) The cat’s wailing all night was easier to stand than the baby’s [wailing all
night].
Here the elided material is a VP, but the licensing head is clearly the possessive
D, so I will assume, along with LaCara, that what is involved here is not VPE,
but NPE. Whatever it is, note that here as in the VPE cases discussed above, if
the VP is separated from the licensing head by an adverb, the ellipsis becomes
impossible:
(22) * I was used to Marie’s quietly complaining about the food, but I was sur-
prised at Harvey’s loudly [complaining about the food].
Of course, if the adverb is included in the ellipsis, the result is impeccable:
(23) I was used to Marie’s quietly complaining about the food, but I was sur-
prised at Harvey’s [quietly complaining about the food].
Sluicing, first identified by Ross (1969), is characterized by elision of everything
in a WH Question clause except the fronted WH phrase:
(24) Somebody broke my favorite glass, and I want to know who [broke my
favorite glass].
(25) They stole some pictures, but which pictures [they stole] isn’t clear.
Since the C is generally silent in WHQ clauses, it is not immediately clear what
constituent is being elided in examples like (24-25). One possibility (and the one
that I will adopt) is that the elided part is a TP, complement of the interrogative
C.9
As is well known, Sluicing does not apply in Relative Clauses, though their
internal syntax looks just like that of WH Question clauses:
9 The one context in which the C is audible is when the WH clause is a root Question, in which
152
(26) * Somebody broke my favorite glass, and I plan to kill the person who
[broke my favorite glass].
Property 1
First, as noted above, each of these processes can be characterized as the elision
of the complement of a particular kind of functional head: in the case of VPE, the
licensing heads are T (including pres, past, to, and the Modals); Neg (or possibly
more generally Sigma or Pol); and the Auxiliary verbs have [PERF], be [PROG], be
[PASS], and be [COP]; in the case of NPE, they are D[POSS], D[DEM], and possibly
certain quantifiers; in the case of Sluicing, the licensing head is C[+Q,+WH].
The remaining notable properties of these ellipses can be generally character-
ized as a remarkable insensitivity to structure. They cannot happily go backward
(ellipsis site preceding antecedent) in coordinate structures, but no anaphoric
processes can do that:
further manipulation. As Ross (1969) observed, when the fronted WH phrase originates as a comple-
ment to a P, the P can be apparently stranded by the ellipsis (the phenomenon known as “swiping”
(Merchant 2002)):
(i) She went to the movies, but we don’t know who [she went to the movies] with.
An alternative account would be that the P is pied piped in the normal manner along with the fronted
+WH DP, and a separate operation reverses the order of P and DP; but there appears to be no inde-
pendent motivation for that operation, which would have to occur only in PPs fronted by WH move-
ment and then stranded by Sluicing. I won’t have anything very interesting to say about this problem,
though it does bear on the question of ellipsis as a constituency test.
11 One of the several reasons for not classifying but as a coordinator is that it does permit backward
153
Aside from that, structural relations between the ellipsis site and the antecedent
do not seem to matter.
Property 2
These ellipses can go across sentence and speaker boundaries:
(30) You know I’ve never ridden a camel. Ivan has [ridden a camel].
(31) I smelled the camel’s breath. You smelled the merchant’s [breath].
(32) A: Somebody stole the emeralds.
B: Do you know who [stole the emeralds]?
Property 3
They are unbounded, in the sense that the antecedent and ellipsis site can be un-
boundedly far apart in hierarchical structure:
(33) I’ve never ridden a camel, but it’s pretty clear that we can be sure that Ivan
has [ridden a camel].
(34) If you’ve smelled a camel’s breath, you don’t need to be afraid of anyone
trying to scare you by making you smell a donkey’s [breath].
(35) Somebody stole the emeralds, and I am sure the police think nobody be-
lieves we’ll ever find out who [stole the emeralds].
Property 4
In addition to being unbounded, these ellipses are not subject to island con-
straints (the ellipsis site and the antecedent may be separated by island bound-
aries):
(36) I don’t carry a cellphone, and people who expect me to [carry a cellphone]
are bound to be disappointed.
(37) I have my own Porsche, so the assumption that I was driving my mother’s
[Porsche] when I robbed the bank is unjustified.
(38) They hid something in the trunk of the stolen car, and the only man who
knows what [they hid in the trunk of the stolen car] is in a jail in Phoenix.
Property 5
Finally, ellipses of this type can go backward (the ellipsis site can precede the
antecedent), as long as it doesn’t go backward in a coordinate structure:
(39) Unless you want me to [mention this to your mother], I won’t mention
this to your mother.
(40) Anyone who has seen Peter’s [yacht] will not be impressed by Harvey’s
yacht.
(41) Nobody knows why [rabbits are gentle creatures], but rabbits are gentle
creatures.
154
Properties 2, 3, 4, and 5 have to do with the relation between an ellipsis site and
a potential antecedent. I will call such properties range properties. I will hence-
forth call ellipses that have these characteristics Type A (or, to anticipate a bit,
head-licensed ellipses). In the following subsection I will turn to ellipses that do
not share these properties.
(iii) Harvey couldn’t open the window, but he could [open] the door.
Example (iii), however, is not Gapping, but Pseudo-Gapping, to be discussed more thoroughly in
section 2.5. From here on I will continue to assume that Gapping always involves elision of T in the
gapped clause.
13 Examples like (i) indicate that the main Verb is not always part of what is elided:
(i) Sue could have opened the window, and Harvey [could have] closed the door.
The story of this, though, might be more complicated.
155
(46) I got sick because I ate too much, and Harvey [got sick] because he drank
too much.
(47) Most of these kids can run a mile in six minutes, and a few [can run a mile]
in under five.
Despite the accepted name of the process, the "gap" is not always in the middle:14
Further, and as will become important later, the elided material does not always
constitute a contiguous stretch (at least not in any obvious way):
(52) Bill bet ten dollars on the fight, and Harvey [bet] five dollars [on the fight].
(53) Bill put ten dollars on the table, and Harvey [put] five dollars [on the table].
(54) Mary put rocks in the can, and Betty [put] marbles [in the can].
(55) He called me an idiot, and I [called] him [an idiot].
I will assume that there is an ellipsis process with the following basic character-
istics, and call it "Gapping":
• The elided part usually also contains a verb, and if it does, it contains ev-
erything between T and that verb.15
14 Some in the GPSG tradition use the term "Left Peripheral Ellipsis" for examples like (48-51); but
so far as I am aware, there is no particularly good reason to believe that Left Peripheral Ellipsis and
Gapping are not reflections of the same process. For the purposes of the present paper, it will not
matter, because as will be shown, if there is a separate process of Left Peripheral Ellipsis, it is in the
same family as Gapping in the typology of ellipsis processes.
On the other hand, there is a possibility that these kinds of examples do not involve ellipsis at all,
but rather coordination of big VPs as a complement to little v. When the big Vs head-raise across-
the-board to little v, the resulting structure is [DP V [DP DP] and [DP DP]], providing the observed
pattern with no ellipsis.
This analysis is of course not available for examples like (52-55).
15 Jackendoff (1971, p. 23) noticed this: “Gapping cannot take place if there are unlike auxiliaries . . .
If the auxiliaries are alike, Gapping must delete the second auxiliary as well as the verb ...”. “Gapping
cannot tolerate unlike adverbs preceding the verb, either ...”.
156
• The ellipsis leaves two remnants16 which are contrastive with correspond-
ing parts of the antecedent structure.
2.2.2 Stripping
Stripping is a term introduced in Hankamer 1971, 1979 to denote the ellipsis found
in fragment answers to WH-questions:
I assumed then (and still do) that essentially the same process is at work in several
other contexts:
(58) Sally brought Bill to the party, not [Sally brought] John [to the party].
(59) Sally didn’t bring John to the party, but [Sally brought] Bill [to the party].
(60) Sally brought John to the party, and [Sally brought] Bill [to the party].[FN]
Some sources use the term "Fragment Answer Ellipsis" for the cases in (56-57), and
it is conceivable that the differences in context may justify distinct processes for
(56-57) and (58-60). As will be seen, though, if there are such distinct processes,
they all share the same essential characteristics from a typological point of view.
I will proceed to use the term "Stripping" for all of these cases.
Stripping: an ellipsis process (or processes) in which material is elided from a
domain (I will call it the target domain) under identity with corresponding parts
of an antecedent domain, leaving behind one remnant constituent which con-
trasts with a corresponding constituent in the antecedent domain.17
As characterized here, an immediately striking difference between VPE, NPE,
and Sluicing on the one hand and Gapping and Stripping on the other is that in
cases of Gapping and Stripping there is no obvious head to license the ellipsis,
which as we have seen is characteristic of ellipses of type A. As we shall see, Gap-
ping and Stripping (which I will henceforth call type B ellipsis) exhibit starkly
different behavior in terms of structure-sensitivity as well.
16 For many speakers, the number of remnants in Gapping must be exactly two. For these speakers
(i) is ungrammatical:
(i) Harvey lent a book to Paul, and Sue [lent] a magazine to John.
Other speakers find (i) perfectly grammatical. When necessary I will refer to the first group as
observing a two-remnant constraint on Gapping. One might think that (ii) is Gapping with a single
remnant:
(ii) My oldest daughter went to college, and my youngest [went to college].
I see no reason, however, not to regard (ii) as a case of Stripping, to be discussed immediately below,
which always leaves a single remnant.
17 In this paper, I will choose examples in which the antecedent and target domains are clauses,
157
In one respect type A and type B ellipses are similar: they cannot go backward
in coordinate structures (examples repeated from section 2.1):
(63) * Sally might have written a book, and Marie thinks that Harvey [might
have written] a poem.
(64) * I want Harvey to write a novel, and Mary wants Bill [to write] a play.
(65) A: Who did Sally bring to the party?
B: * Well, she denies (that) [Sally brought] John [to the party].
(66) A: Who brought Sally to the party?
B: * Well, I doubt (that) John [brought Sally to the party].
(67) * Sally brought Bill to the party, so we can be pretty sure that not [Sally
brought] John [to the party].
(68) * Sally didn’t bring John to the party, so we can be pretty sure that [Sally
brought] Bill [to the party].
To summarize, there are at least two very different kinds of ellipsis. In type A
ellipsis, there is a licensing head whose complement undergoes elision, and this
kind of ellipsis is structure-insensitive in that it is unbounded, is insensitive to
18 In these examples, I have been careful to avoid the possibility of a parenthetical interpretation
for the material that is intended to force the ellipsis site into an embedded context.
158
island constraints, and can go backward. In type B ellipsis, the elided material
cannot be characterized as the complement of any particular licensing head, and
type B ellipsis is very sensitive to structure: the antecedent and target domains
must be directly in construction with each other, and the antecedent must pre-
cede. The range properties of the two types of ellipsis are radically distinct. Type
B ellipses do not share any of the distinctive properties (properties 1-5) of type A
ellipses.
The evidence is that when the remnant precedes the polarity morpheme, the el-
lipsis is unbounded, is insensitive to islands, and can go backward, as shown in
(73-76)a:
(73) a. Susana leyó el dictamen pero Maria dijo que Juan no.
’Susan read the report but Mary said that John didn’t.’
b. * Susana leyó el dictamen pero Maria dijo que no Juan.
(74) a. Juan juega al tenis todos los fines de semana, y creo que Susana dijo
que Pedro piensa que Maria no.
’John plays tennis every weekend, and I think that Susan said that
Pete thinks that Mary doesn’t.’
b. * Juan juega al tenis todos los fines de semana, y creo que Susana dijo
que Pedro piensa que no Maria.
(75) a. Juan no entregó el trabajo a tiempo, pero existe el rumor que Maria
sí.
’John didn’t turn in the homework on time, but there’s a rumor that
Mary did.’
b. * Juan no entregó el trabajo a tiempo, pero existe el rumor que sí
Maria.
(76) a. Jorge sí, pero Maria Laura no tiene una gata.
’George does, but Mary Laura doesn’t have a cat.’
b. * Sí Jorge, pero Maria Laura no tiene una gata.
159
When the remnant follows the polarity morpheme, the ellipsis is bounded and
cannot go backward, as seen in (76b). Depiante makes sense of this, in terms of
the typology of ellipsis developed here, by analyzing examples like (71) as a type
A ellipsis licensed by the polarity morpheme. The remnant is moved to the front
by an independently necessary topicalization, and the complement of the polar-
ity head, minus the extracted constituent, is elided under identity with the an-
tecedent clause. In examples like (72), a process equivalent to Stripping applies,
and everything except the polarity head and the remnant is elided by a type B
process. The point of this example is that it is not always obvious, on superficial
inspection, what kind of ellipsis you are dealing with. To know that, some non-
trivial analysis might be necessary. At the end of this paper, I will argue that you
have to know what type of ellipsis you are dealing with before you can use it as a
diagnostic for constituency.
(77) I could see some dogs. – Oh, yeah? How many dogs [could you see]?
It is important to notice, however, that this apparent sluicing of a C-bar can only
occur in root clauses, since that is the only environment in which the element in
T can have moved to C. In this context, however, it is perfectly plausible that the
ellipsis observed is due to Stripping, and not to Sluicing.
(80) * The kid will eat cauliflower, but he will never [eat] cabbage.
160
(81) * You might catch him playing video games, but you won’t catch him
[playing] cards.
Levin (1986) observed that one T that does not license Pseudogapping is the in-
finitival ‘to’:
(82) * While I’m willing to kiss Sue, I don’t want to [kiss] Louise.
(83) * You can eat the shrimp, but you had better not [eat] the mussels.
(84) * You can put the gold coins in the box, but I would prefer that you not
[put] the jewels [in the box].
This makes me suspect that in (85) the licensor might be the T, to which Neg has
adjoined.
161
In summary, Pseudogapping shares more properties with VPE than it does with
Gapping: it is head-licensed, involves the elision of (most of) the complement of
the licensing head, and has most of the range properties of a type A ellipsis. In par-
ticular, it is not confined to coordinate (or any other) structures, has unbounded
range, and is not sensitive to islands. As with many things in linguistics, it is badly
named.
2.5.2 N-Pseudogapping
We can find something like Pseudogapping in the nominal domain. Jackendoff
(1971) exhibited examples like (89-90), and called them "N-gapping":
(89) Harry’s wines from France and Bill’s [wines] from Italy
(90) Harry’s stories about fishing and Bill’s [stories] about hunting
(91) I didn’t taste Harry’s wines from France but I did taste Bill’s from Italy.
(92) Harry’s stories about fishing were more interesting than Bill’s [stories]
about hunting.
(93) While I don’t think I’ve heard Harry’s stories about fishing, I’m quite sure
I have heard Bill’s [stories] about hunting.
The examples above show that this ellipsis process is not confined to coordinate
structures and indeed not restricted to any particular kind of syntactic environ-
ment; that it can cross sentence and speaker boundaries and can go into embed-
ded clauses; and that it is not affected by island constraints. It has, in short, most
of the properties of a type A ellipsis process. It is, to be sure, not very good going
backward:
(94) * While I haven’t tasted Harry’s [wines] from France, I have tried Bill’s
wines from Italy.
(95) * While I don’t think I’ve heard Harry’s [stories] about fishing, I’m quite
sure I have heard Bill’s stories about hunting.
behavior, in any case, makes it very clear that this process is not to be assimilated to Gapping.
162
an extra remnant in addition to the possessor DP. I propose that this is not an
instance of Gapping in the nominal domain but rather the nominal equivalent
of Pseudogapping. I will consequently call it N-pseudogapping. Like Pseudo-
gapping, N-pseudogapping elides most of the complement of a licensing head,
except for one constituent which survives as a remnant.
(96) We picked more apples than we could carry [0] in the wheelbarrow.
(97) We picked more apples than [0] would fit in the wheelbarrow.
Bresnan attributed this kind of gap to a process she called "Comparative Dele-
tion". The standard assumption since Chomsky (1976) is that such gaps result from
A-bar movement of a silent operator from the gap site to the specifier of the com-
parative clause. The particular mechanism will not concern us here, since we are
interested in a further possibility of omission of material, a genuine ellipsis this
time, called Comparative Ellipsis, which is exhibited in (99):22
21 I assume, still following Bresnan (1973), that there is a gap in the indicated position in examples
like (i):
(i) The table is longer than the door is [0] wide.
Bresnan called the process producing this gap "Subdeletion". I actually believe this is the same process
as Comparative Deletion, but that is a matter for another paper.
22 CE is not the only ellipsis process that can apply in comparative clauses. VPE (of the antecedent-
163
(98) Martha peeled more apples than Harry peeled [0]. [CD]
(99) Martha peeled more apples than Harry [peeled [0]]. [CE]
Note that CD, where possible, is obligatory; CE, on the other hand, is always
optional, as is expected of an ellipsis process.
What sort of ellipsis process is CE? First let us note that CD, like other A-bar
processes, is unbounded, though it obeys island constraints:
(100) We picked more apples than the bosses thought we could carry [0] in
the wheelbarrow.
(101) * We picked more apples than the boss hired a man to carry [0] in the
wheelbarrow.
While we are checking things, note also that VPE and Pseudogapping within
comparative clauses are unbounded just as expected:
(102) They ate more pancakes than we thought they would [eat [0]].
(103) They order pancakes more often than they do [order] omelettes.
(104) We liked the pancakes more than we did [like] the omelettes.
(105) They liked the sausages just as much as we already knew they did [like] the
bacon.
(106) Martha peeled more apples than we thought Harry peeled [0].
(107) * Martha peeled more apples than we thought Harry [peeled [0]].
T must be elided, so there is no way for ‘appear’ to get tense. This probably needs
more investigation.
Comparative Ellipsis elides parts of a comparative clause, always including
the T, that are identical with corresponding parts of the matrix clause, always
leaving one remnant. It has the range properties of a type B ellipsis.
164
2.7 Conclusion
In this section I have established that there is a typology of ellipsis processes, with
at least two major types. Type A, which could be called head-licensed ellipsis, in-
volves the ellipsis of the complement of a licensing head, with the possibility of
a single remnant surviving the ellipsis. Since the licensing head can sometimes
be silent, its overt presence cannot be used as a secure diagnostic. Type B ellipsis
does not involve the elision of the complement of any particular head, but it does
involve the elision of a head (T in the case of Gapping, Stripping, and Compara-
tive Ellipsis), together with other material, leaving behind one or two remnants.
The two types of ellipsis differ radically in their range properties: type A ellipses
are unbounded, insensitive to island constraints, and can go backward. Type B
ellipses are very sensitive to syntactic environment, are always bounded in the
sense that the ellipsis site must be in a domain that is linearly adjacent to the an-
tecedent domain, with at most a coordinator or a complementizer intervening,
are therefore bounded, and cannot go backward. I have no theory about why
there should be two types, or why the head-licensed type should sometimes per-
mit a remnant. I am sure, though, that these types exist, and that there are few if
any other types.
4 Conclusion
The conclusion is that Ellipsis is not a test for constituency unless you know what
kind of ellipsis it is. If you know that an ellipsis is the result of a type A (i.e. head-
licensed) process, you can safely conclude that the stuff elided by that process is
either a constituent or a constituent minus a remnant (so you can conclude that
the gap + remnant is a constituent). If the stuff is elided by a type B process, it
165
appears that the remnants are always constituents, but the elided portion is gen-
erally not. In fact, I suspect it is always not, because type B processes always elide
the head of the ellipsis domain, and as far as I can tell they always also elide some-
thing else. In other words, there seems to be no ellipsis process that only elides a
simple head. Here is a place where the theory of ellipsis should tell us why things
are the way they are.
References
Aoun, J. and Benmamoun, E. (1999). Gapping, PF merger, and patterns of partial
agreement. In Lappin, S. and Benmamoun, E., editors, Fragments: Studies in
Ellipsis and Gapping, chapter 175–192. Oxford University Press.
Bresnan, Joan. 1973. ‘Syntax of the comparative clause construction in English’.
Linguistic Inquiry 4.3, 275–343.
Bresnan, Joan. 1975. ‘Comparative deletion and constraints on transformations’.
Linguistic Analysis 1(1), 25–74.
Chao, Wynn (1988) On Ellipsis. Garland Press. (Outstanding Dissertations in Lin-
guistics)
Cheng, Hsu-Te (2013) Argument Ellipsis, Classifier Phrases, and the DP parameter
(U Conn Dissertation)
Chomsky, Noam 1976 “On Wh-Movement”. Linguistic Inquiry
Chung, Sandra, William Ladusaw, and James McCloskey (1995) “Sluicing and Log-
ical form”. Natural Language Semantics 3.3: 239-282.
Chung, S., W. Ladusaw, and J. McCloskey. 2011. Sluicing: Between Structure and
Inference. In Representing Language: Essays in Honor of Judith Aissen.
Culicover, Peter 2009. Natural Language Syntax. Oxford University Press.
Culicover, Peter and Ray Jackendoff. 2005. Simpler Syntax. Oxford University
Press.
Depiante, Marcela (2000) The Syntax of Deep and Surface Anaphora: A Study
of Null Complement Anaphora and Stripping/bare argument ellipsis. PhD
Dissertation, University of Connecticut
Depiante, Marcela (2004) “Dos casos de elipsis con partícula de polaridad en Es-
pañol: Evidencia a favor de una vision no uniforme de la elipsis. RASAL Lin-
guistica 1: 53-69.
Hankamer, Jorge (1979) Deletion in Coordinate Structures. Garland.
Hankamer, Jorge and Ivan Sag 1976 Deep and Surface Anaphora. Linguistic In-
quiry
Jackendoff, Ray 1971 “Gapping and Related Rules”, Linguistic Inquiry, Vol. 2, No.
1, pp. 21-35.
Jayaseelan, K. A. 1990 “Incomplete VP Deletion and Gapping”, Linguistic Analysis
20.1-2, pp. 64- 81.
Johnson, Kyle. 1996. In search of the English middle field. Ms., University of
Massachusetts, Amherst.
166
Johnson, Kyle 2000 “Few dogs eat Whiskas or cats Alpo,” in University of Mas-
sachusetts Working Papers in Linguistics, (eds.) Kusumoto, Kiyomi and Elis-
abeth Villalta, 47–60.
Johnson, Kyle. 2001. What VP-ellipsis can do, what it can’t, but not why. In Mark
Baltin and Chris Collins (eds.), The handbook of contemporary syntactic the-
ory, 439- 479. Blackwell:
Johnson, K. 2009. Gapping is not (VP) ellipsis. Linguistic Inquiry 40, 2, 289-328.
LaCara 2010 Verbal Ellipsis in the Nominal Domain. MA Thesis, University of
California at Santa Cruz.
Lappin, Shalom, and Elabbas Benmamoun 1999, Fragments: Studies in Ellipsis
and Gapping. Oxford University Press.
Lasnik, H. 1995. “A note on pseudogapping”. MIT Working Papers in Linguistics
27: 143-163.
Lasnik, H. 1999. “Pseudogapping puzzles”. In S. Lappin and E. Benmamoun (eds.)
Fragments: Studies in Ellipsis and Gapping. Oxford University Press, pp. 141-
174.
Levin, Nancy 1986 Main Verb Ellipsis in Spoken English. Outstanding Disserta-
tions in Linguistics. Garland.
Lobeck, Anne. 1991. The phrase structure of ellipsis. In Susan Rothstein (ed.),
Perspectives on phrase structure, 81-103. Academic Press: San Diego.
Lobeck, Anne. 1995. Ellipsis: Functional heads, licensing, and identification. Ox-
ford University Press: Oxford.
Merchant, Jason. 2001. The syntax of silence: Sluicing, islands, and the theory of
ellipsis. 2001. Oxford University Press: Oxford.
Merchant, Jason. 2002. “Swiping in Germanic”, in C. Jan-Wouter Zwart and
Werner Abraham (eds.), Studies in Comparative Germanic Syntax, 295-321. John
Benjamins, Amsterdam. Miller, Philip. 1990. “Pseudo-gapping and do so sub-
stitution”. Proceedings of the 26th Meeting of the Chicago Linguistic Society,
293-305.
Morris, Amanda (2008) Polarity Ellipsis and Negative Stripping. MA Thesis, Uni-
versity of California at Santa Cruz.
Ross, John R. 1969. Guess who? In Robert Binnick, Alice Davison, Georgia Green,
and Jerry Morgan (eds.), Papers from the 5th regional meeting of the Chicago
Linguistic Society, 252-286. Chicago Linguistic Society: Chicago, Ill.
Ross, John 1970. Gapping and the order of constituents. In Progress in linguis-
tics: A collection of papers, M. Bierwisch and K. Heidolph (eds.), 249-259, The
Hague: Mouton.
Sag, Ivan and Jorge Hankamer (1984) “Toward a theory of anaphoric processing”,
Linguistics and Philosophy 7: 325-345.
Toosarvandani, Maziar (2013) Gapping is low coordination (plus VP-ellipsis): A
reply to Johnson. Massachusetts Institute of Technology June 11, 2013.
Zagona, Karen 1982 Government and Proper Government of Verbal Projections.
PhD Dissertation, University of Washington.
167
Matching Light Elements
Junko Ito and Armin Mester
University of California, Santa Cruz
1 INTRODUCTION
(1) a. *I don't know where Tom's. (I don't know where Tom is.)
b. Where's Tom? (Where is Tom?)
Familiar as it is, this fact is still surprising since ω(Tom's) looks like a bona fide
prosodic word, just like ω(where's). It should therefore be wellformed in any position,
including phrase-final, as is the homophonous possessive in the utterance This book is
Tom's. The explanation we will pursue here builds on the basic fact that, because of
wh-movement, [is/'s_ ] constitutes an entire syntactic phrase, and as such must
correspond to a non-vacuous phonological phrase.
The impossibility of phrase-final enclisis needs to be seen in the context of the
whole system of cliticization of English. The simple non-vacuity explanation turns
out to have important consequences for Match Theory (Selkirk 2011; Elfner 2012; Ito
and Mester 2013) since it motivates a conception of Syntax-Prosody (SP)-Match
constraints that is rather different from the generally accepted one. The new
conception is purely existential and non-gradient, insisting merely on the existence of
some corresponding prosodic constituent, not on exact correspondence. The latter is
enforced by other constraints that are already part of the theory, such as classical
syntax-prosody Alignment and standard faithfulness, evaluated gradiently.
We start out with one class of English function words, including monosyllabic
determiners, auxiliaries, and prepositions (2), that forms proclitic structures. Possible
mappings to prosodic structure are given in (3) (Selkirk 1996; Ito and Mester 2009),
where "σ" stands for "syllable", "ω" for "prosodic word", and "φ" for "phonological
phrase". Peperkamp (1997) shows that all of (3b-d) are instantiated in Italian dialects.
1
We are very pleased to contribute this paper to a festschrift in honor of Jim McCloskey, a
friend and colleague of many years. Most of the work we have done on syntax-prosody matters
took place in close collegial contact with him, it would not have come into existence without
his input. We wish him many more years of productive research. Many thanks to the festschrift
editors, and to Nick Kalivoda and Lisa Selkirk for productive discussion. We are especially
grateful to Line Mikkelsen for her thorough and helpful comments on an earlier version.
168
(2) FncP a. DP b. IP c. PP
Fnc LexP D NP I VP P NP
Lex N V N
… ω ω … … ω ω … σ ω …
…
… σ ω …
Two views regarding the prosodic structure of English proclitics have been proposed.
The majority of researchers (including (McCarthy 1993; Booij 1996; Vigário 1999; Ito
and Mester 2007; 2009) argues that they are affixal clitics (3c). The other view (Selkirk
1996; Hall 1999) identifies them as free clitics (3d). The two different structures are
contrasted in (4ab).
(4) a. φ b. φ
σ ω … σ ω …
the students … the students …
can work … can work …
at home … at home …
169
2 REQUIREMENTS ON LEFT EDGES
Since the beginnings of metrical phonology it has been known that left edges of
prosodic constituents are subject to more stringent requirements than right edges. An
example is the initial dactyl requirement in English (Prince 1983, 49): feet/stresses are
right-aligned, but words beginning with unfooted/unstressed syllables are avoided:
(Tàta)ma(góuchi), not *Ta(tàma)(góuchi). Recently, Selkirk (2011, 470) has proposed
STRONGSTART, a generalized version of this kind of left edge requirement (informally:
"Beginnings of prosodic units are strong."). STRONGSTART is responsible for a wide
variety of prosodically motivated effects. The first is promotion of the initial
constituent of the utterance, as in Xitsonga, where preposed constituents which
would normally be parsed as phonological phrases are boosted into full intonational
phrases (see Kisseberth 1994 for the original empirical generalizations; Selkirk 2011,
442–445). On the other hand, clitics are often banned from first position and appear
in peninitial second position (Wackernagel 1892), or are moved to a position later in
the sentence, as in Bulgarian (Harizanov 2014) and Irish, as illustrated in (6) (from
Bennett, Elfner, and McCloskey 2016:171).
(7) a. English (Weir 2012) Have you got milk? It is a nice day today.
b. German (“pronoun zap”, Ich hab das schon gelesen Das hab ich schon gelesen.
Ross 1982; Haider '(I) have already read it' '(that) have I already read'
1986)
170
There are even modes of resolution deeply embedded in the morphosyntactic system,
such as redundant agreement to create a branching first constituent (cf. Elordieta
2007), or the morphosyntactically unmotivated doubling of agreement clitics, such as
the initial masculine clitic rà in (8), on unary initial constituents discovered by
Ostrove (2016) in a dialect of Mixtec.
171
(9) a. affixal: φ b. free: φ
ω-adjoined fnc φ-attached fnc
ω
σ ω … σ ω …
the students … the students …
(10) a. I can eat more than Ray can. [kæn] *[kən] *[kn̩]
b. If you think you can, go ahead and do it. [kæn] *[kən] *[kn̩]
c. I don't know where Ray is. [ɪz] *[əz] *[z]
d. Wherever Ray is, he's having a good time. [ɪz] *[əz] *[z]
e. What did you look at yesterday? [æt] *[ət]
f. Who did you do it for that time? [fɔr] *[fr̩]
Selkirk (1996, 202) translates the observations in (10) directly into a constraint:
ALIGN-RIGHT(φ,ω), requiring every phonological phrase to end in a prosodic word
(and not in a prosodically deficient function word). The effect of this is shown in (11).
(11) [ What did Mary [VP look [PP at _ ]PP ]VP last time]
(φ What did Mary look (ω àt)ω ) φ (φ last time)φ
172
(12) What did Mary [look at __ ] Align-R(φ,ω) PWdCon(ω,x)
► φ( ωlook ωàt ) *
φ( ωlook ət
σ ) *W L
PWdCon
173
The most natural interpretation of Match constraints then is one that applies them to
all projections, including auxiliary-verb structures such as can eat that are
"intermediate projections" of T in traditional understanding. In order to stay with
familiar terminology, we refer to all projections of X as "XP", making no distinctions
in bar level.
The recursive structure φ( φ( Michelle) φ( can)) wins over the flat structure φ( Michelle)
φ( can) because it matches the higher TP with a φ. One might argue that the nonap-
plication of the rhythm rule (Michèlle cán, not *Mìchelle cán) favors the flat structure,
but this is not probative if the domain is in fact φmin (see Elordieta 2015; Selkirk and
Lee 2015 for recent overviews of recursive category structure in phonology).
ωleaving ))
( ( Tim )
φ φ ω φ( ωìs *
ωleaving))
( ( Tim əs )
φ φ ω φ( ωleaving )) *W
The take-home question here is: Are there cases where Align-R(φ,ω) ("StrongEnd") is
actually needed in English and elsewhere—because the function word does not
constitute a syntactic XP all by itself?
174
4 MORPHOSYNTACTIC ENCLITICS
In an apparent violation of the ban on weak phrase-final fnc, object pronouns in
English can appear here in a weak form (cf. Selkirk 1972; 1984), besides in their
strong form. The phonetic realization of these weak forms, and their rhythmic
adherence to the verb, is identical to that of word-final stressless syllables (Selkirk
1996).
But there is a fundamental difference between enclitic pro and the proclitics seen
earlier: The host of enclitic pro is always V, whereas proclitics have no such syntactic
category restriction: the bookN, the boringA book, the veryAdv boring book, to goV, to
boldlyAdv go, etc. This suggests that the pronouns have a morphosyntactic signature.
According to Selkirk (1996), whose position we follow, there are two possible
syntactic sources for object pro: as a phrasal object, a full DP (19a), or as a
morphosyntactic enclitic object, an impoverished category (19b) coindexed with a full
DP (see Cardinaletti and Starke 1999 for a theory distinguishing clitic, weak, and
strong pronouns along such lines.) The two syntactic sources for object pro are shown
in the tableaux in (20).
(19) a. VP b. VP
V DP V DP
D V PRO __
φ( see ) *
φ( see φ(σya ) ) *W L
175
Headed- Match- PWd
as an enclitic object vp[see-PRO DP [ __ ] ] ness Phrase Con
φ(ωsee ωyòu ) *W
► φ(ωsee σya )
Summarizing so far, English has a large number of prosodic proclitics (fnc lex): to go,
the student, can meet, etc. There is a small number of specific morphosyntactic
enclitics (lex fnc) which can occur in any position, including phrase-finally, but are
restricted as to their host, which has to be verbal: see ya (V-obj pro, enclitic to verb).
What remains is prosodic enclisis, which is not morphosyntactically restricted to
hosts of a specific category, but which cannot occur in phrase-final position (*Tell me
where Tom's).
5 PROSODIC ENCLITICS
English has half a dozen special forms of auxiliaries that show enclisis, as in (21).
Different from the morphosyntactically enclitic pronouns seen in the previous
section (ya, əm, etc.), the enclitic auxiliaries listed in (22) are single consonants and
hence subsyllabic, and they do not have a morphosyntactic subcategorization frame,
like the enclitic pronouns.
5.1 CHARACTERISTICS
The substantial work on the clitic system of English done in the 1970's by Zwicky,
Selkirk, Kaisse, and others already uncovered most of the characteristics of prosodic
enclitics. They are subsyllabic in size (single consonants); there is a proper subset
relation (wherever reduced auxiliaries can occur, corresponding full verbs can occur
as well, but there are contexts where only the full form is possible.). This is
allomorphy, not productive phonology: Enclitic auxiliaries are lexically listed
allomorphs, not the results of general phonological reduction (Kaisse 1983, 94–95).
For example, while would, could, and should all have reduced forms ([wǝd, kǝd, ʃǝd]),
only would has the idiosyncratic monoconsonantal form [d]: I'd rather be home. In
terms of their position, enclitic auxiliaries are adjoined to the final syllable of the
preceding word, just like the exponent of the plural/3sg/possessive morphemes (23).
176
(23) is/has Matt'[s] gone, but Tom'[z] here, and Bruce'[ǝz] on his way.
plural cat[s], home[z], busse[ǝz]
3sg fit[s], come[z], misse[ǝz]
poss Matt'[s], Tom'[z], Bruce'[ǝz] car
Enclitic forms correspond to auxiliaries, never to full verbs. Thus the word has occurs
both as an auxiliary and as a main verb of possession, but the enclitic form 's
homophonous with that of is) functions only as an auxiliary. Thus in Anderson's
(2008) example (24), the (b) version only has the bizarre reading in which Fred's sister
is a cat.
(24) a. Fred has adopted a new cat, and his sister Joanna has a cat, too.
b. Fred's adopted a new cat, and his sister Joanna's a cat, too.
This indifference regarding the preceding context only holds for 's (is, has), not for the
remainder (Zwicky 1970, 331; Kaisse 1983, 97–98), as shown in (26). We will
henceforth restrict ourselves to these two.
The most important feature of enclitic auxiliaries is that they are prosodically
deficient variants of full forms, consisting of a single consonant. A single consonant,
especially an obstruent, cannot constitute a syllable in English, hence also cannot be a
foot, or a prosodic word on its own. Disregarding their syntactic affiliation, they go
with the word on their left, even if they are syntactically more closely related to the
material on their right. The syntax-prosody mapping is given in (27). Note the
mismatch of the syntactic and the prosodic parse of 's in (27).
Our analysis appears in (28). The winning candidate (28a) parses the TP-initial 's with
the subject and therefore fails to match both the subject NP and TP. It beats the more
faithful candidate that preserves 's in φ-initial position by parsing it at the beginning
of a prosodic word, violating standard positional faithfulness (28b). (To save space, we
177
will from now on suppress the outermost phrase corresponding to the whole
sentence in all candidates).
A brief characterization of the constraints together with their ranking is given in (29).
(31) Inputs:
a. [[Ray] [can]] / I can eat more than __
b. [[Ray] [is]] / I don't know where __
c. [[Tim] [is leaving]]
d. [[Tim] ['s leaving]]
e. [look [at __]] / What did Mary __
178
It contains the six languages in (32).
(32)
final fnc unreduced final fnc reduced
enclitic 's proclitic 's enclitic 's proclitic 's enclitic 's proclitic 's
Lg#1 (English) Lg#2 Lg#3 Lg#4 Lg#5 Lg#6
Lg#1
a. φ( φ( ωRay ) φ( ωcàn ) )
b. φ( φ( ωRay ) φ( ωìs ) )
c. φ( φ( ωTim ) φ( σəs ωleaving ) )
d. φ( φ( ωTim's ) ωleaving )
e. φ( ωlook φ( ωàt ) )
Lg#2
a. φ( φ( ωRay ) φ( ωcàn ) )
b. φ( φ( ωRay ) φ( ωìs ) )
Lg#3
a. φ(φ( ωRay ) φ( σcən ) )
b. φ( φ( ωRay ) φ( σəs ) )
c. φ( φ( ωTim ) φ( σəs ωleaving ) )
d. φ( φ( ωTim's ) φ( ωleaving ) )
e. φ( ωlook φ( σət ) )
179
Lg#4
a. φ( φ( ωRay ) φ( σcən ) )
b. φ( φ( ωRay ) φ( σəs ) )
e. φ( ωlook φ( σət ) )
Lg#5
a. φ( φ( ωRay ) σcən )
b. φ( φ( ωRay ) σəs )
d. φ( φ( ωTim's ) φ( ωleaving ) )
e. φ( ωlook σət )
Lg#6
a. φ( φ( ωRay ) σcən )
b. φ( φ( ωRay ) σəs )
d. φ( φ( ωTim ) φ( ω'sleaving ) )
e. φ( ωlook σət )
The first two languages leave phrase-final fnc unreduced. Lg#1 is English, and Lg#2
differs in showing a faithful phrase-initial parse of 's in (31d), violating word-initial
positional faithfulness, which ranks below MATCH-PHRASE. Lg#3-Lg#6 all allow
phrase-final fnc to reduce. This happens in two ways: In Lg#3 and Lg#4, fnc is its own
phrase while being reduced, violating HEADEDNESS (MATCH-PHRASE, PWDCON >>
HEADEDNESS). Monoconsonantal 's is either enclitic (Lg#3) or proclitic (Lg#4),
depending on the relative ranking of INITFAITH and MATCH-PHRASE. Lg#5 and Lg#6
show reduced final fnc by leaving the fnc-only TP unphrased (HEADEDNESS, PWDCON
>> MATCH-PHRASE). Again, monoconsonantal 's is either enclitic (Lg#5) or proclitic
(Lg#6), depending on the ranking of INITFAITH and MATCH-PHRASE. This typology
seems to reasonably reflect the crosslinguistic options. It can easily be expanded by
including additional possibilities, such as allowing 's to delete, or to remain
unsyllabified at the word level, which are of little interest to our current concerns.
180
5.3 NO PHRASE-FINAL ENCLISIS
We now have all necessary pieces in place to address our main question, the
impossibility of phrase-final enclisis for monoconsonantal clitics. As a reminder, we
give some examples of the phenomenon in (33) (after Anderson 2008).
(33) a. Tim's happier than Kim is/*'s __. John is taller than Harry is/*'s __.
b. Freddie's a werewolf this year for Halloween. Do you know what Tommy
is/*'s __(this year for Halloween)? Tommy has been a werewolf more often
than Freddie has/*'s __(on Halloween).
c. John has known Mary longer than Fred has/*'s __Martha.
d. Who do you think you are/*'r __?
e. Fred's an Independent: he'd no more campaign for a Democrat than he
would/*'d __ for a Republican
f. John is happier with their marriage than his wife ìs/*'s __.
Selkirk (1996, 198, footnote 5) observes that "[i]t is an interesting fact that these
contracted forms are only possible if they are not phrase-final […]. The atypical
prosodic encliticization that they display must somehow reflect this fact. For now,
this remains a puzzle." Anderson's (2008) observes that the TP's in (34a-c) are
wellformed, but not the TP consisting just of the monoconsonantal (34d). This is in
itself unremarkable since it holds for basic syllabic reasons.
(34) a. [TP is happier] b. [TP 's happier] c. [TP is __] d. *[TP 's __]
The real question is why the simple phonological adjustment of reassigning the lone 's
to the preceding phrase, as in (35), is also not a way out.
(35) φ φ φ φ
| |
ω à ω
| |
[. . . ] [kɪm] [z][ _ ] [. . . ] [kɪmz] [ _ ]
Taking up an idea first raised by Selkirk (1984, 366), Anderson's (2008, 11) insight is to
interpret the impossibility of the move in (35) not as an idiosyncratic quirk of Modern
English that could easily be changed, but rather as a reflection of a fundamental
principle: The result of the phonological adjustment would be that the φ originally
built over the phonetic material corresponding to the TP would now be left with no
phonetic content at all. This is impossible. We state the ban on prosodic vacuity in a
preliminary form in (36), and will later derive it from Match Theory.
181
The ban on prosodic vacuity has been argued by Kandybowicz (2015) to motivate a
kind of do-support (ye 'do, make') in Asante Twi. Our question now is how to derive
the ban on prosodic vacuity in our analysis. As things stand, the candidate with
enclisis of 's is the winner in (37b) since MATCH-PHRASE is ranked too low to prevent
this.
We present two different ways of deriving the correct outcome here. The first one
uses M-PARSE (Prince and Smolensky 1993) to select the null candidate instead of the
wrong outcome. In the second approach, is and 's are competing allomorphs, and the
first beats the second. As we will see, both require us to sharpen our understanding of
MATCH constraints. A standard use of M-PARSE is as enforcer of the single-foot
subcategorization (38a) on comparative adjectives in English (pretty, prettier; red,
redder; but beautiful, *beautifuller). As shown in (38b), the null candidate violating M-
PARSE is preferable to a candidate violating the prosodic subcategorization constraint.
Ø * ► Ø *
Where is M-PARSE ranked in the grammar? It cannot be too low in the ranking
because then the null candidate would always win. A first attempt would be to rank it
between MATCH-PHRASE and PWDCON, as in (39). This gets the right result when the
desired candidate only violates a constraint ranked below M-PARSE, such as PWDCON
in (39a), or is in fact the null candidate, as in (39b).
182
(39) Headed- Match- M-Parse PWdCon
a. Tim's leaving if NP [Kim] TP [is ] ness Phrase
correct► φ( ωKim ) φ ( ωìs ) *
Ø *
φ( ωKim ) ( ǝs )
φ σ *
φ( ωKim ωìs) *NP *TP *
φ( ωKim σǝs) *NP *TP
But it goes wrong in (40), where the desired winner in fact violates MATCH-PHRASE,
which dominates M-PARSE.
c. φ( ωKim ) φ( ωìs ) ~ Ø W L
Ranking M-PARSE below MATCH-PHRASE gets the right result when we have a
catastrophic MATCH-violation, akin to a MAX-violation: A failure to have any kind of
phrase whatsoever corresponding to the TP TP['s] in (41a). It gets the wrong result
when the desired winner violates MATCH-PHRASE in a minor way, akin to an IDENT-
violation, by reassigning 's to the subject's PPhrase in (41b). This suggests that we are
in fact dealing with two different constraints. As a preliminary move, to be revised
later, we add an "existential" MAX-type constraint MATCH-∃-PHRASE to the constraint
system which is distinct from general Match-Phrase and is only violated when a
syntactic phrase has no prosodic correspondent whatsoever, of whatever category.
183
With M-PARSE sandwiched between the two types of Match-constraint, the correct
distinctions are derived in (42), where the null-candidate wins in (a), but not in (b).
184
MATCH calls for gradient evaluation, but this has hardly ever been made use of in an
essential way, to our knowledge. The intention has always gone beyond alignment,
and has aimed for prosodic replication of the whole constituent, not just preservation
of its edges (see Ishihara 2014). But checking on whole-scale correspondence requires
the whole set of faithfulness constraints, and is in any case not easily, or profitably,
expressed in a single constraint that can be evaluated gradiently. Elfner (2012, 28), in a
move away from gradience, proposes an all-or-nothing categorical version of
MATCH-PHRASE: "Suppose there is a syntactic phrase (XP) in the syntactic
representation that exhaustively dominates a set of one or more terminal nodes α.
Assign one violation mark if there is no phonological phrase (ϕ) in the phonological
representation that exhaustively dominates all and only the phonological exponents
of the terminal nodes in α." As a categorical constraint, this is easy to evaluate, but it is
unlikely to be workable in real life where standard phonology (such as the ONSET
requirement) routinely leads to small deviations from perfect correspondence. We
subsume Match Theory under General Correspondence Theory, which distinguishes
purely existential MAX (requiring nothing but the existence of a correspondent in the
output, which can be utterly different from the input element) from IDENT and other
faithfulness constraints which deal with detailed aspects of correspondence (here,
instantiated by Al-R(XP,φ)).2
In the M-PARSE approach, /is/ and /'s/ are separate inputs that do not compete. Kim
is leaving and Kim's leaving each win their competition—optionality as lexical choice.
(Tim's leaving if) Kim's loses against the null candidate because the TP [is _ ] has
become phonologically vacuous. (Tim is leaving if) Kim is wins its competition. In an
alternative allomorphy approach, /is/ and /'s/ compete with each other as different
allomorphs of the same input morpheme. All allomorphs enter the same competition.
Priority (Mascaró 1996) (or some economy constraint) prefers /'s/ to /is/. Standard
alignment-based MATCH continues to have the familiar problem: Ceteris paribus,
2
Things are different in the two-stage view of prosodic structure formation developed in
Selkirk (2017), which distinguishes Spell-Out-by-MATCH from the phonology, which
incorporates prosodic structure faithfulness constraints, in a division of labor reminiscent of
the proposal made here.
185
unranked MATCH-PHRASE (preferring is) and PRIORITY (preferring 's) admit both
outputs as winning candidates in 45) (optionality as lack of ranking). But there
continues to be a problem with phrase-final 's (46).
Sentences without gaps do not incur MATCH-∃ violations, so the allomorph variation
arises as before.
186
(48) TP [is/'s Match-∃- Align-R Priority: PWd
a. I wonder if NP[Kim] leaving] Phrase (XP,φ) 's > is Con
► φ(ωKim ) φ( σəs *
ωleaving )
( Kim )
φ ω φ( ωìs * *
ωleaving )
( Kim's )
φ ω φ( *NP
ωleaving )
Priority
TP [is/'s Match-∃- : Align-R PWd
b. I wonder if NP[Kim] leaving] Phrase 's > is (XP,φ) Con
φ(ωKim ) φ(σəs *
ωleaving )
( Kim )
φ ω φ( ωìs * *
ωleaving )
► ( Kim's )
φ ω φ( *NP
ωleaving )
Both analyses (M-PARSE and Allomorph PRIORITY) need the existential version of
MATCH-PHRASE instead of the alignment-based version. Which, if any, of these two
approaches—M-PARSE or allomorphy—is the correct one? It is hard to think of
decisive evidence either way. Kaisse (1983, 95) makes the interesting observation that
the contexts where 's is admitted are not literally a proper subset of those where is is
admitted. There are exceptions to this subset relation in examples such as
There'{s/*is/*has} a new book been written. She notes that this is suggestive of
morphemes with separate listings in the lexicon. Since true allomorphs have one
lexical entry, one could interpret this to favor the M-PARSE approach, where the two
have separate lexical entries.
6 CONCLUSION
By insisting that syntactic constituents must in some form be matched in prosody,
Match Theory provides very simple explanations (i) for positions where weak
elements must appear in their strong form—because otherwise a phonological phrase
would have no head—, and (ii) for positions where a functional element cannot
undergo enclisis—because if it did, a whole syntactic constituent would go
unmatched. In order for this explanation to go through, Match constraints must have
a purely existential force, and merely insist on the existence of a prosodic
correspondent to a syntactic phrase. Detailed correspondence falls to standard
alignment/faithfulness constraints.
187
REFERENCES
Anderson, Stephen. 2008. English reduced auxiliaries really are simple clitics. Lingue e
Linguaggio vii: 1–18.
Bennett, Ryan, Emily Elfner, and James McCloskey. 2016. Lightest to the right: An
apparently anomalous displacement in Irish. Linguistic Inquiry 47: 169–234.
Booij, Geert. 1996. Cliticization as prosodic integration: The case of Dutch. The
Linguistic Review 13: 219–242.
Cardinaletti, Anna, and Michal Starke. 1999. The typology of structural deficiency: a
case study of the three classes of pronouns. In Clitics in the Languages of
Europe, ed. Henk van Riemsdijk, 145–233. Berlin: Mouton de Gruyter.
Chomsky, Noam. 2007. Approaching UG from below. In Interfaces + Recursion =
Language? Chomsky’s Minimalism and the View from Syntax-Semantics, ed.
Uli Sauerland and Hans-Martin Gärtner, 1–30. Berlin: Mouton de Gruyter.
Chomsky, Noam. 2008. On phases. In Foundational Issues in Linguistic Theory. Essays
in Honor of Jean-Roger Vergnaud, ed. Robert Freidin, Carlos P. Otero, and
Maria Luisa Zubizarreta, 291–321. Cambridge, MA: MIT Press.
Chomsky, Noam. 2013. Problems of projection. Lingua 130: 33–49.
Elfner, Emily. 2012. Syntax-Prosody Interactions in Irish. Doctoral dissertation,
University of Massachusetts, Amherst.
Elordieta, Gorka. 2007. Minimum size constraints on intermediate phrases. In
Proceedings of the 16th International Congress of Phonetic Sciences, ed. Jürgen
Trouvain and J.Barry William, 1021–1024.
Elordieta, Gorka. 2015. Recursive phonological phrasing in Basque. Phonology 32: 49–
78.
Haider, Hubert. 1986. V-Second in German. In Verb Second Phenomena in Germanic
Languages, ed. Hubert Haider and Martin Prinzhorn, 49–75. Dordrecht:
Foris.
Hall, T. Alan. 1999. Phonotactics and the prosodic structure of German function
words. In Studies on the Phonological Word, ed. T.Alan Hall and Ursula
Kleinhenz, 99–131. Amsterdam, The Netherlands, and Philadelphia, USA:
John Benjamins.
Harizanov, Boris. 2014. The role of prosody in the linearization of clitics: evidence
from Bulgarian and Macedonian. In Formal Approaches to Slavic Linguistics
22: The McMaster Meeting 2013, ed. Cassandra Chapman, Olena Kit, and
Ivona Kucerova, 109–130. Ann Arbor, MI: Michigan Slavic Publications.
Ishihara, Shinichiro. 2014. Match Theory and the recursivity problem. In Proceedings
of FAJL 7: Formal Approaches to Japanese Linguistics, ed. Shigeto Kawahara
and Mika Igarashi, 69–88. Cambridge, MA: MIT Working Papers in
Linguistics. 73.
Ito, Junko, and Armin Mester. 1992. Weak layering and word binarity. Linguistic
Research Center Report LRC-92-09. University of California, Santa Cruz.
188
[Appeared in Honma, Takeru, Masao Okazaki, Toshiyuki Tabata and Shin-
ichi Tanaka, eds. 2003. A New Century of Phonology and Phonological Theory.
Tokyo: Kaitakusha. 26-65.].
Ito, Junko, and Armin Mester. 2007. Prosodic adjunction in Japanese compounds. In
Formal Approaches to Japanese Linguistics: Proceedings of FAJL 4, ed. Yoichi
Miyamoto and Masao Ochi, 97–111. MIT Working Papers in Linguistics 55.
Cambridge, Massachusetts: MIT Department of Linguistics and Philosophy.
Ito, Junko, and Armin Mester. 2009. The extended prosodic word. In Phonological
Domains: Universals and Deviations, ed. Janet Grijzenhout and Bariş Kabak,
135–194. Berlin: Mouton de Gruyter.
Ito, Junko, and Armin Mester. 2013. Prosodic subcategories in Japanese. Lingua 124:
20–40.
Kaisse, Ellen. 1983. The syntax of auxiliary reduction in English. Language 59: 93–
122.
Kandybowicz, Jason. 2015. On prosodic vacuity and verbal resumption in Asante Twi.
Linguistic Inquiry 46: 243–272.
Kisseberth, Charles W. 1994. On domains. In Perspectives in Phonology, ed. Jennifer
Cole and Charles W. Kisseberth, 133–166. Stanford, CA: CSLI.
Mascaró, Joan. 1996. External allomorphy as emergence of the unmarked. In Current
Trends in Phonology: Models and Methods, ed. Jacques Durand and Bernard
Laks, 473–483. University of Salford, Manchester: European Studies
Research Institute.
McCarthy, John J. 1993. A case of surface constraint violation. Canadian Journal of
Linguistics 38: 127–153.
McCarthy, John J., and Alan S. Prince. 1993. Generalized Alignment. In Yearbook of
Morphology, ed. Geert Booij and Jaap van Marle, 79–153.
Myrberg, Sara. 2010. The intonational phonology of Stockholm Swedish. Doctoral
dissertation, Department of Scandinavian Languages, Stockholm University.
Myrberg, Sara. 2013. Sisterhood in prosodic branching. Phonology 30: 73–124.
Ostrove, Jason. 2016. The prosody of agreement in San Martín Peras Mixtec.
Qualifying paper. UC Santa Cruz.
Peperkamp, Sharon. 1997. Prosodic Words. The Hague: Holland Academic Graphics.
Prince, Alan S. 1983. Relating to the grid. Linguistic Inquiry 14: 19–100.
Prince, Alan S., and Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in
Generative Grammar. Brunswick, New Jersey, and Boulder, Colorado:
Rutgers University and University of Colorado, Boulder.
Ross, John Robert. 1982. Pronoun deleting processes in German, presented at the
Annual Meeting of the LSA, San Diego, California.
Selkirk, Elisabeth. 1972. The Phrase Phonology of English and French. Doctoral
dissertation, MIT.
Selkirk, Elisabeth. 1984. Phonology and Syntax: The Relation between Sound and
Structure. Cambridge, MA: MIT Press.
189
Selkirk, Elisabeth. 1996. The prosodic structure of function words. In Signal to
Syntax, ed. James L. Morgan and Katherine Demuth, 187–213. Mahwah, NJ:
Lawrence Erlbaum.
Selkirk, Elisabeth. 2011. The syntax-phonology interface. In The Handbook of
Phonological Theory. 2nd edition, ed. John Goldsmith, Jason Riggle, and Alan
Yu, 435–485. Oxford: Blackwell Publishing.
Selkirk, Elisabeth. 2017. Syntactic Constituency Spell-Out though Match Constraints
Talk presented at SPOT Workshop, UC Santa Cruz, Nov. 18, 2017.
Selkirk, Elisabeth, and Seunghun J. Lee. 2015. Constituency in sentence phonology:
an introduction. Phonology 32: 1–18.
Spaelti, Philip. 1994. Weak edges and final geminates in Swiss German. In Proceedings
of NELS 24, ed. Mercè Gonzàlez, 573–588. Amherst, MA: GLSA.
Vigário, Marina. 1999. On the prosodic status of stressless function words in
European Portuguese. In Studies on the Phonological Word, ed. T.Alan Hall
and Ursula Kleinhenz, 255–294. Amsterdam Studies in the Theory and
History of Linguistic Science. Amsterdam and Philadelphia: John Benjamins.
Wackernagel, Jacob. 1892. Über ein Gesetz der indogermanischen Wortstellung.
Indogermanische Forschungen 1: 333–436.
Weir, Andrew. 2012. Left-edge deletion in English and subject omission in diaries.
English Language and Linguistics: 105–129.
Werle, Adam. 2009. Word, phrase, and clitic prosody in Bosnian, Serbian, and
Croatian. Doctoral dissertation, Amherst, MA: University of Massachusetts,
Amherst.
Zwicky, Arnold M. 1970. Auxiliary reduction in English. Linguistic Inquiry 1: 323–
336.
190
Passivization of German Double-Object
Constructions: Theory and Usage*
Vera Lee-Schoenfeld
University of Georgia
Using the canonical passive auxiliary werden ‘be’, the only passivization possibility
here is for the direct object to become the subject and therefore be realized with
nominative rather than accusative case. The indirect object keeps its dative case.
192
Interestingly, the passivization possibilities involving a minimally different ex-
ample with the double-accusative verb lehren are much less straightforward (see
e.g. Czepluch 1988, Lang 2007, Duden 2006, 2016, and Lee-Schoenfeld and Diewald
2017). The active version of the sentence is given in (4) and the passivization op-
tions are shown in (5) and (6).1
1 The judgments in (5) are based on my intuitions and informal consultations with other native
speakers.
2 Example (8b) is only acceptable for certain speakers if the indirect object pronoun him is phono-
logically reduced.
193
(8) a. He was taught the rope trick. [be-passive]
b. * The rope trick was taught him.
It is well-established that, in English, only the first accusative object, whose case
feature is valued with structural accusative case by agentive v (in line with Burzio
1986), can become the subject under passivization. The second object can be ar-
gued to be lexically case-licensed by V (see Anagnostopoulou 2003 and Twiner
2016 for an overview of the relevant literature).
While it is obviously the single accusative object that becomes the subject
when the verb is mono-transitive in German as well, the 2016 Duden Grammatik
says, “Bei Verben mit doppeltem Akkusativ wird das personale Objekt zum Sub-
jekt" (Duden 2016: 944) (‘In the case of double-accusative verbs, it is the animate
object that becomes the subject’), but this source marks examples of both type
(5a) and type (5b) with a “?”. It is the 2006 Duden Grammatik (Duden 2006: 952)
that indeed marks example type (5b), with the subject derived from the inanimate
object, with a “*”, i.e. as unacceptable. Likewise, Czepluch (1988: 281) judges an ex-
ample of type (5a), with the animate object as subject, better than one of type (5b),
with the inanimate object as subject:
(9) a. ? dann ist der Junge das Lied gelehrt worden. [werden-pass]
then has the.nom boy the.acc song taught been.pass
b. * dann ist den Jungen das Lied gelehrt worden.[werden-pass]
then has the.acc boy the.nom song taught been.pass
‘Then the boy was taught the song.’
Importantly, both (5c) and (6), the passivization options that are the most readily
acceptable, rely on the animate object of the double-accusative verb being da-
tive instead of accusative-marked. In other words, these options appear to be re-
pair strategies that rely on fitting the exceptional double-accusative construction
into the normal ditransitive pattern of a dative indirect object followed by the ac-
cusative direct object, as shown in (2).3 Given the semantics of this pattern, the
normally accusative-marked animate object seems to be interpreted not as a Pa-
tient but as the Recipient in a scene of transfer (see Lang 2007 and Lee-Schoenfeld
and Diewald 2017). It is the goal of this paper to both capture these observations
formally and investigate them further empirically.
The remainder of the paper is structured as follows. Section 2 provides a
formal account of German double-object as well as bene/malefactive (applica-
tive) constructions and their passivization. Section 3 presents empirical sup-
port for the proposed formal account by examining corpus data. More specif-
ically, section 3.1 takes a brief look at previous corpus-based work on German
double-accusative verbs, and section 3.2 presents the results of a very recent cor-
pus search by Lee-Schoenfeld and Diewald (2017) on the diachronic development
3 In fact, the meaning ‘teach’ is also expressed by the very commonly used verb beibringen, which
follows the normal ditransitive pattern of dat > acc. Thus, speakers may use dat > acc instead of
acc > acc with the exceptional verb lehren in analogy with the more regular verb beibringen.
194
of double-accusative verbs from the double-accusative pattern to the dative-
accusative pattern. Section 4 concludes the paper.
Position (iii), Spec VP, is the structural acc(usative) case position, licensed by
the external-argument-introducing agentive v (Burzio 1986), and it hosts the first
object of a double-accusative verb. This was illustrated in example (4), which is
repeated here as (11).
(11) Jemand hat ihn den Seiltrick gelehrt.
someone has him.acc the rope.trick taught
‘Someone taught him the rope trick.’
This first object becomes the subject under passivization. Without the agentive
v, no structural acc case is available and the object in position (iii) can only be
realized with nom(inative) case. This was shown in (5a) as at least marginally
acceptable and is repeated here as (12).
(12) ? Er wurde den Seiltrick gelehrt.
he.nom was.pass the rope.trick taught
‘He was taught the rope trick.’
4 A sketch of this account is also presented in Lee-Schoenfeld and Diewald 2017.
195
Position (iii) is sister-to-V when position (iv) is not needed. It hosts the direct
object of simple transitive and prototypical ditransitive verbs like lieben ‘love’ and
geben ‘give’.
The direct object, of course, becomes the subject under passivization. This is
shown in (14).
Position (iv), sister-to-V, is the lexical (idiosyncratic) acc case or clausal comple-
ment position, licensed by V, and it hosts the second object of a double-accusative
verb. Example (4) is given here again as (15) with an added clausal complement
option (Seil zu springen ‘to jump rope’) shown as extraposed, following the verb.
The analysis in (10) predicts that the second object does not passivize, which is
obviously correct when it comes to clausal complements and also in line with the
judgments in the Duden Grammatik (2006, 2016) and by Czepluch (1988). This is
illustrated in (16).
Position (iv) also hosts the inanimate dat(ive) object of exceptionally patterning
verbs like aussetzen ‘expose’, an example of which is given in (17). The dat case of
this kind of object is lexical or idiosyncratic, licensed by V itself. As will become
clear in a moment, the proposed analysis makes a crucial distinction between
lexical idiosyncratic case on the one hand and inherent predictable case on the
other.
196
(17) Man hat ihn der Kälte ausgesetzt.
one has him.acc the.dat cold exposed
‘People exposed him to the cold.’
(18) Er wurde der Kälte ausgesetzt / * Die Kälte kriegte ihn ausgesetzt.
he was.pass the.dat cold exposed / the.nom cold got him exposed
‘He was exposed to the cold.’
Position (ii), Spec affectee vP, is the inherent or predictable dat case position,
licensed by affectee v, and it hosts the first object of a double-accusative verb
when this is reinterpreted as the indirect object, the Recipient, of a prototypical
ditransitive construction. Example (19a) shows the double-accusative verb lehren
‘teach’ following the dat > acc pattern, and (19b) shows the prototypical ditran-
sitive verb erklären ‘explain’, which always follows this pattern.
As we know from (6), which is repeated here as (20a), the inherent dat-object
becomes the subject under kriegen (‘get’)-passivization.
Position (ii) also hosts the argument of mono-transitive verbs selecting a dat-
object, like helfen ‘help’, gratulieren ‘congratulate’, and widersprechen ‘contradict’.
An example of helfen is given in (21). The dat case here is again licensed by affectee
little v, which, in this scenario, assigns not a Recipient but a Beneficiary role.
As shown in (22), for some speakers, the inherent dat-object of a verb like helfen
can become the subject under kriegen-passivization (see also Beermann 2011).
197
(22) Er kriegte geholfen.
he.nom got helped
‘He was getting helped.’
I take this to indicate that the dat case here is not idiosyncratic lexical but inher-
ent case. It is predictably assigned to animate arguments and regularly alternates
with nom case for those speakers who allow (22). In this sense, it is similar to
structural acccase. However, in line with Grewendorf 2002 and Haider 2010,
I do not take the dat-nom alternation in examples like this to be evidence for
dat being structural case because, unlike the acc-nom alternation in canonical
passivization constructions, dat-nom alternation depends on the case-changing
nominal playing a certain type of semantic role. It is always some kind of Affectee,
that is, a necessarily animate Goal, Recipient, or Bene/Maleficiary (see also Bader
& Häussler 2013).
Position (ii) can also host a necessarily animate external possessor, a so-called
“free dative” that is compatible with verbs selecting a (potentially possessed) in-
ternal argument and an optional Affectee argument, like ruinieren ‘ruin’ in (23).
Here, dat case is licensed by the Bene/Male-ficiary role assigning version of af-
fectee little v (see Lee-Schoenfeld 2006, 2016).
As predicted and shown in (24), this inherent dat case alternates with nom case
under kriegen-passivization.
Finally, position (i), Spec agentive vP, is the external argument (proto-agent) posi-
tion which hosts the subject of any (di)transitive or unergative verb. Nom case is
licensed at a distance by finite T (tense/agreement). An example with an unerga-
tive verb is given in (25).
198
To sum up, the analysis just presented, as diagrammed in (10), correctly pre-
dicts the (un)grammaticality of the double-accusative facts (active and passive)
known from the literature (see (4)-(6) and (9) of section 1) and it works for regular
(di)transitive and “free dative” (applicative) constructions. Additionally, the pro-
posed account captures well-known co-occurrence restrictions on dat-objects
(see e.g. Maling 2001 and Bosse 2015). Generally, there can only be one dat-object
at a time. The incompatibility of the two dat-objects in (27), for example, falls
out from the analysis because the affected animate arguments, Beneficiary and
Recipient, compete for the same dat-case-licensing position, namely (ii) in (10).
The same holds for the combination of an optional possessor dat and the oblig-
atory dat argument of verbs like helfen ‘help’ in (28).5
Overall, the analysis proposed in (10) is in line with much of the existing litera-
ture on double-object constructions, but there are also some important depar-
tures from previous work that I briefly address here. In line with Woolford 2006
and partly in line with Haider 2010 but contra Anagnostopoulou 2003, the pro-
posed analysis makes a crucial distinction between two types of non-structural
cases, namely idiosyncratic lexical case on the one hand and (regular) predictable
inherent case on the other. Also, as mentioned above, I agree with Grewendorf
2002 and Haider 2010 that the kriegen (‘get’)-passive is not evidence for dat being
structural case, but, contra Haider, I take it to be evidence for this kind of dat
being inherent rather than lexical case and for analyzing dat-object-selecting
verbs like helfen ‘help’ as inherent case licensors (via affectee v) rather than lexical
case-licensors. Note that, following Chomsky 1986, I assume that inherent case
is assigned to a DP by the same head that introduces this DP as its argument and
thus assigns it a thetat-role (in this case, by the affectee v to the DP in its spec-
ifier). On the other hand, given Burzio’s (1986) Generalization, I take structural
accusative case to be assigned by the external-argument-introducing, agentive v
down to any DP in the verbal domain that still needs case (in this case, to the DP
in Spec VP or in the sister-to-V slot, depending on whether the V itself has lexical
case to assign or not).
5 Notice that, according to the proposal in (10), an animate position-(ii) dative should be grammat-
ical when co-occurring with an inanimate position-(iv) dative since these datives do not compete for
the same slot. This prediction seems to be borne out given that examples like Man hat dem.dat armen
Mann das Kind der.dat Kälte ausgesetzt ‘One exposed the poor man’s child to the cold’ are considerably
less degraded than examples (27)-(28). My thanks to Jason Merchant for asking about this.
6 Ethical datives, like mir ‘me’ in Reich mir.dat der.dat Oma doch bitte das Salz! (‘Please pass grandma
the salt, will you?’) are the exception here. They are, however, best analyzed not as verbal arguments
but as discourse markers (see Diewald 2016 for a recent overview of dative usages in German).
199
In line with Bruening 2010 and Bosse 2015, I argue, contra Pylkkänen 2008,
that the base configuration of arguments is dat > acc, with each object being
generated in its own verbal projection, namely in affectee vP (or ApplP) and VP,
respectively. But contra Bruening and Bosse, the account in (10) requires no rais-
ing of arguments into a higher verbal projection in order to establish Pylkkänen
(2008)’s “low applicative” transfer of possession relation. I assume that the lexical
VP and affectee vP can be considered one extended domain after V-to-v raising.
Furthermore, in line with Grewendorf 2002 and contra both Müller 1995 and
Anagnostopoulou 2003, the dat-case licensing position in the configuration in
(10) is above the acc-case licensing position, so that we have the commonly as-
sumed I(indirect) O(bject) > D(irect) O(bject) base configuration. This means that,
unlike in Müller 1995, there is no need to posit A’-movement of the IO to a dat-
case licensing position. Spec affectee vP is a normal A-position. I tentatively pro-
pose that the complication of the IO not being able to bind a DO anaphor, as
shown in (29), can be captured by Grewendorf’s (2002: 63) account, where the
DO anaphor is generated in a projection above the IO-containing verbal layer.
For now, a detailed extension of the analysis to these binding facts needs to be
relegated to future research. The focus of the remainder of this paper is on the
actual usage of double-accusative verbs like lehren ‘teach’ and thus on providing
empirical support for the proposed constellation in (10), abstracting away from
anaphoric objects.
200
object sounds completely grammatical, but the first, animate object becoming
the subject seems to be a bit better than the second, inanimate object doing so, at
least according to the Duden Grammatik (2006, 2016) and Czepluch (1988). There
also seems to be a preference for realizing the first object with dative rather than
accusative case. This section aims to establish that these intuitions and claims
indeed correspond to the actual usage of double-accusative verbs. Subsection 3.1
summarizes corpus-based work by Lang (2007) on the active use of lehren ‘teach’,
and subsection 3.2 presents the results of Lee-Schoenfeld & Diewald’s (2017) di-
achronic corpus study focusing on the use of lehren in both active and passive
constructions.
The passivization facts are reported to be unclear. When dat replaces acc, dat
tends to be used in sentence-initial position, and when there is no dat-marking,
nom tends to be used in sentence-initial position. The prescriptive recommen-
dation is to simply avoid the passive with double-accusative verbs.
The results of Lang’s (2007) corpus search, for which he used newspaper cor-
pora from the Institute of German Language (IDS) Mannheim and focused on
the active use of lehren, yielded a total of 3678 tokens. Only about a quarter of all
occurrences had two realized objects: 12% with the pattern of acc > acc, 6% with
the pattern of dat > acc, and another 6% with the pattern of undifferentiated
acc/dat (as in uns ‘us’) > acc. About half of all occurrences had only one object,
which was either a whole clause (propositional), an undifferentiated acc/dat-
marked one, or an inanimate acc-marked one. An example of the latter is given
in (32).
(32) . . . dass das Spiel einen leichtfertigen Umgang mit Geld lehrt.
that the game a.acc careless handling with money teaches
‘. . . that the game teaches a careless handling of money.’
201
Approximately another quarter had no object at all, as in (33).
Lang concludes that, if lehren occurs with two objects, it is relatively frequently
used with a dat-marked object (dat > acc), not only colloquially.
3.2 The development from acc > acc to dat > acc and what
really happens in the passive
From the literature cited by Lang (2007) as well as his own investigation, Lee-
Schoenfeld and Diewald (henceforth “we”, 2017) conclude that the first object of
double-accusative verbs corresponds to the necessarily animate indirect object
of the prototypical ditransitive pattern. This means that the original animate Pa-
tient/Theme is interpreted as Recipient, a variant of the indirectus (a sympathy-
invoking co-participant, see Lehmann et al. 2004), and that, depending on its se-
mantic features, the second object of double-accusative verbs can be interpreted
as a typical inanimate Theme/Patient. More specifically, we hypothesize that the
second acc-marked object was originally not a typical inanimate Patient but an
adverbial acc (“accusative of measure” in the case of kosten ‘cost’) or an “inner
object” and can often be interpreted as a more Patient-like object. This causes
the prototypical ditransitive schema (a scene of transfer) with a Recipient dat to
become available.
We searched the DWDS core corpus (http://www.dwds.de/ressourcen/
kernkorpus/), which consists of different types of text (fiction and poetry, news-
paper articles, science writing, functional writing) and contains about 100,000,000
words. Unlike Lang, we targeted not only the active use of lehren ‘teach’ but also
its passive use and chose two different time spans for our searches in order to be
able to comment on possible diachronic developments. The following two sub-
sections, reporting on the passive and the active search results, respectively, each
begin with a table summarizing our findings and continue with discussion and
representative examples.
202
3.2.1 Search 1: Passive use of lehren
Time Span 1900-1909 1990-1999
Σ Passive uses (accessible tokens) 39 (138) 29 (44)
dat animate: wird den Kindern gelehrt 5 1
dat/acc animate: wird uns gelehrt 2 0
acc animate: wird die Kinder gelehrt 3 0
acc inanimate/propositional: wird den Seiltrick gelehrt / wird 1 1
gelehrt, dass . . . / wird gelehrt zu . . .
Subject, referential or propositional (inanimate) 36 (34) 28 (27)
Zero-Subject, expletive es 3 1
Passive AUX werden 39 28
Passive AUX kriegen 0 1
Table 1: Passive in first (1900-1909) and second (1990-1999) time span
What this table shows is that passive lehren is rarely used with both a passivized
and an unpassivized object. If it does have an unpassivized object in addition to
the passivized one (28% in first time span and only 7% in second time span), then
we found that, most commonly, the subject is inanimate, while the unpassivized
object is animate and marked with undifferentiated acc/dat-case (as in 34a) or
with dat (as in 34b); acc-marked objects do occur, but only three times in the
first time-span.7
(34) a. Wir leiden unter der eigenen Winzigkeit, unter den engen Grenzen unseres Wissens und Lebens,
seitdem uns die Endlosigkeit von Raum und Zeit gelehrt wird.
since us.acc/dat the endlessness of space and time taught is.pass
animate obj. inanimate subj.
‘We suffer from our own tininess, form the narrow limits of our knowledge and life, ever
since the endlessness of space and time was taught to us.’
(31.12.1903/ Belletristik/ Heyking, Elisabeth von: Briefe, die ihn nicht erreichten. In: Deutsche Liter-
atur von Frauen, Berlin: Directmedia Publ. 2001 [1903], S.32339)
b. Diesen drei jungen Männern ist nie christliche Nächstenliebe gelehrt worden,. . .
these.dat three young men has never Christian brotherly.love taught been.pass
animate obj. inanimate subj.
‘These three young men have never been taught Christian brotherly love.’
(31.12.1994/Belletristik/Jentzsch, Kerstin: Seit die Götter ratlos sind, München: Heyne 1999 [1994], S.153)
There were only two examples with an animate subject and an inanimate object:
one in the first time span where the inanimate object was a propositional infini-
tive complement (35a), and one in the second time span formed with kriegen (35b).
The latter serves as clear evidence of acc-marking of animate objects being re-
placed by dat-marking because, as noted in section 1, the kriegen-passive targets
dat-objects.
7 In all our corpus examples, we only give glosses for the clause containing the relevant object(s)
and/or subject, but we often include more of the sentence in order to provide speakers of German
with as much context as possible.
203
(35) a. Von Haus aus waren die germanischen Pferde klein und unansehnlich;
“sie werden auch nicht gelehrt,” sagt Tacitus (Germ. 6),
they are.pass also not taught says Tacitus
animate subj.
“verschiedenartige Wendungen nach unserer Art zu machen; . . .
various turns according.to our.gen way to make
inanimate obj. (clausal)
‘By nature, the Germanic horses were small and unattractive; “they were also not
taught,” says Tacitus, “to perform different kinds of turns following our technique.”
(31.12.1908/ Wissenschaft/ Fischer, Hermann: Grundzüge der Deutschen Altertumskunde, Leipzig:
Quelle & Meyer 1917 [1908], S. 101)
b. Nee, aber wie stehst du dazu, zu dem,
was du gelehrt kriegst . . .
what you taught get.pass
inanimate obj. animate subj.
‘No, but what’s your opinion of what you’re getting taught?’
(31.12.1991/ Belletristik/ Brussig, Thomas: Wasserfarben, Berlin: Aufbau-Taschenbuch-Verl. 2001 [1991],
S. 179)
This table shows a strong increase of the intransitive use of lehren.8 We identified
8 In the first time span (1900-1909), there were 611 active verbs among 674 accessible tokens (706
total), and in the second time span (1990-1999), there were 180 active verbs among 223 accessible tokens
(283 total). In order to facilitate direct comparison, we reduced the number of tokens from the first
time span to 180, so that it matched the number of tokens from the second time span.
204
two different intransitive meanings: (i) ‘show/illustrate/exemplify’ with an inan-
imate subject (as in 36a), which occurred more frequently in the first time span,
and (ii) ‘be instructor (at a certain school/institution)’ with an animate subject (as
in 36b), which occurred more frequently in the second time span. We take this
to be an indication of a lexical split. A new intransitive verb with meaning (ii)
has been added to the existing (di)transitive verb lehren (‘teach somebody some-
thing’). It seems that meaning (i) is on its way out.
(36) a. Aus diesen beiden Mineralen setzt sich auch,
wie eingehende Untersuchungen gelehrt haben, die Grundmasse selbst zusammen.
as thorough investigations taught have
inanimate subj.
‘As thorough investigations have shown, it is out of these two minerals that the matrix
itself is composed.’
(31.12.1900/ Gebrauchsliteratur/ Jahrbuch des Vereins für wissenschaftliche Pädagogik, 1900, Bd. 32)
b. Er lehrt am Institut d’études européennes an der Universität Paris VIII.
he teaches at.the Institut d’études européennes at the University Paris VIII
animate subj.
‘He teaches at the Institute of European Studies at the University Paris VIII.’
(12.09.1997/ Zeitung/ Die Zeit, 12.09.1997, Nr. 38)
As noted in subsection 3.1, the mono-transitive use of lehren is only possible with
an acc (or undifferentiated acc/dat)-object. We identified two variants of mean-
ing (ii), ‘be instructor of something (inanimate object)’ (as in 37a) and ‘be instruc-
tor of somebody (animate object)’ (as in 37b). The transitive variant of meaning (i)
‘show/illustrate/exemplify’ (with a propositional object, as in 37c) still occurred
frequently in the first time span, but much less so in the second one.
(37) a. Was aber hat man dann eigentlich gelehrt?
what but has one.nom then actually taught
inanimate obj.
‘But what did people actually teach then?’
(31.12.1900/ Gebrauchsliteratur/ Jahrbuch des Vereins für wissenschaftliche Pädagogik, 1900, Bd. 32)
b. Wer andere lehren beziehungsweise bilden will, muß zuvor studieren . . .
who.nom others teach or.else educate wants must beforehand study
animate obj.
‘Someone who wants to teach or educate others, has to study first.’
(28.02.1997/ Zeitung/ Die Zeit, 28.02.1997, Nr. 10)
c. Ein Blick in die Vergangenheit lehrt,
a look into the past shows
daß . . . in der Verwaltung das Gelehrtentum vorherrschte.
that in the administration the learned.class predominated
clausal obj.
‘A look at the past shows that, in the administration, the educated class was in power.’
(31.12.1901/ Gebrauchsliteratur/ Baudissin, Wolf von u. Baudissin, Eva von: Spemanns goldenes Buch
der Sitte. In: Zillig, Werner (Hg.), Gutes Benehmen, Berlin: Directmedia Publ. 2004 [1901], S. 3310)
As for the ditransitive use of lehren, in the first time span, we found consistent
acc-marking of the animate object, but the inanimate object was often proposi-
tional (‘someone/something.nom makes someone.acc learn that something holds’).
We call this the causative construction of lehren (see also Luraghi & Zanchi To
205
appear), which is indeed ditransitive but often shows up without actual double-
accusative-marking because a clause, of course, cannot be case-marked. An ex-
ample is given in (38a). In the second time span, we found slightly more dat-
marked animate objects, but also more inanimate non-propositional acc-objects
(‘someone.nom gives someone.acc/dat something.acc to learn’). We call this the
scene of transfer construction of lehren, which employs the dat case to avoid
double-accusative-marking, as shown in (38b). Also noteworthy is the frequent
use of undifferentiated acc/dat-marking on animate objects (e.g. uns ‘us’) in both
time spans, which, again, results in no actual double-accusative-marking, as ex-
emplified in (38c-d).
(38) a. Man hatte sie nie gelehrt, daß es noch etwas Höheres als sie gab.
one had her.acc never taught that it still something higher than her existed
animate acc-obj. clausal obj.
‘Nobody had ever taught her that there was something superior to her.’
(31.12.1902/ Belletristik/ Janitschek, Maria: Die neue Eva. In: Deutsche Literatur von Frauen, Berlin:
Directmedia Publ. 2001 [1902], S. 36059)
b. Um 1645 lehrte ein Chinese
around 1645 taught a Chinese-man
den Töpfern . . . die Bereitung besserer Schmelzfarben . . .
the.dat ceramists the.acc preparation of.better enamel.colors
animate dat-obj. inanimate acc-obj.
‘Around 1645, a Chinese man taught the ceramists the preparation of better enamel
colors.’
(31.12.1993/ Wissenschaft/ o. A.: Lexikon der Kunst - P. In: Olbrich, Harald (Hg.), Lexikon der Kunst,
Berlin: Directmedia Publ. 2001 [1993], S. 26460)
c. Das hat uns nicht erst
that.acc has us.acc/dat not just
inanimate acc-obj. animate acc/dat-obj.
der Prozess gegen Stellbogen gelehrt. . .
the lawsuit against Stellbogen taught
‘Not just the lawsuit against Stellbogen taught us this.’
(05.04.1900/ Zeitung/ Die Fackel [Elektronische Ressource], 2002 [1900])
d. Was lehrt uns das?
what.acc teaches us.acc/dat that
inanimate acc-obj. animate acc/dat-obj.
‘What did that teach us?’
(30.09.1999/ Zeitung/ Die Zeit, 30.09.1999, Nr. 40)
So, why and how is the double-accusative pattern (acc > acc) becoming the pro-
totypical ditransitive pattern (dat > acc)? With the older ditransitive use of lehren
frequently involving a propositional inanimate object, acc-marking of the ani-
mate object did not typically lead to acc > acc and did not resemble a typical
scene of transfer. Use of dat instead of acc for the animate object only became
necessary or intuitive (in analogy with the prototypical ditransitive dat > acc
verbs in German) when it became more common to use non-propositional (prop)
inanimate objects. Furthermore, the re-interpretation of the formerly causative
(nom > acc > prop) construction as a scene of transfer (nom > dat > acc) con-
struction seems to be taking place via the critical context of the animate object
having a case-undifferentiated acc/dat form, most frequently uns (‘us’).
206
3.2.3 Conclusions regarding the passivization of lehren
To summarize our search results, if an unpassivized object occurs at all, it is
more commonly dat than acc-marked. Due to three unexpected instances of
passivization from the first time span of our search, where the second object
became the subject and the first object kept its acc case, we found the subject
of passivized ditransitive lehren to be slightly more commonly derived from the
underlying inanimate object (as in ‘something was taught him’) than from the an-
imate object (as in ‘he was taught something’). However, what speaks for the pas-
sivization pattern of ‘he was taught something’ over ‘something was taught him’
and thus for the Duden judgment in (5a) and (5b), as well as Czepluch’s (1988) judg-
ments in (9a-b) is that the inanimate object frequently used to be propositional
and would therefore have resisted promotion to subject status. The best solution
to passivizing ditransitive lehren is the kriegen (‘get’)-passive, which targets dat-
objects and goes with the increasing use of dat-marking on the animate object
of double-accusative verbs. We found only one such passivization in the second
time span (35b). A search of a spoken language corpus, though, would likely yield
many more instances of the kriegen-passive.
207
handles the results of these studies are planned as a part of this future research
as well.
References
Alexiadou, Artemis, Anagnostopoulou, Elena, and Christina Sevdali. 2013. Dative-
nominative alternations in Germanic languages: A generative perspective. Lingue
e Linguaggi 9: 7-20.
Anagnostopoulou, Elena. 2003. The syntax of ditransitives: Evidence from clitics.
Berlin: Mouton de Gruyter.
Bader, Markus and Jana Häussler. 2013. How much bekommen is there in the Ger-
man bekommen passive? In Artemis Alexiadou and Florian Schäfer (eds.), Non-
canonical passives. 115-139. Amsterdam: John Benjamins.
Beermann, Dorothee. 2001. Verb semantics and ditransitivity. In Marc van Oos-
tendorp and Elena Anagnostopoulou (eds.), Progress in Grammar: Articles at
the 20th Anniversary of the Comparison of Grammatical Models Group in Tilburg.
Dutch Royal Academy Publishing.
Bosse, Solveig Jana. 2015. Applicative Arguments: A Syntactic and Semantic Investiga-
tion of German and English. New York: Peter Lang Inc., International Academic
Publishers.
Braun, Peter. 1993. Tendenzen in der deutschen Gegenwartssprache. Sprachvarietäten.
3rd Edition. Urban-Taschenbücher, Volume 297. Stuttgart/Berlin/Köln: Kohlham-
mer.
Bruening, Benjamin. 2010. Ditransitive asymmetries and a theory of idiom for-
mation. Linguistic Inquiry 41: 519-562.
Burzio, Luigi. 1986. Italian Syntax: A Government-Binding Approach. Dordrecht:
Reidel.
Chomsky, Noam. 1986. Knowledge of Language. New York: Praeger.
Czepluch, Hartmut. 1988. Kasusmorphologie und Kasusrelationen. Linguistische
Berichte 116: 275-310.
Diewald, Gabriele. 2016. Non-central usages of datives. In Gabriele Diewald (ed.),
Special Issue: Non-central Usages of Datives. Language Typology and Universals
69. 1-14. Berlin/Boston: De Gruyter.
Duden. 1995. Drosdowski, Günther (ed.) Grammatik der deutschen Gegenwarts-
sprache, Volume 4. 5th edition. Mannheim: Dudenverlag.
Duden. 2006. Die Grammatik. Der Duden in zwölf Bänden, Volume 4. 7th edition.
Mannheim: Bibliographisches Institut.
Duden. 2016. Die Grammatik. Der Duden in zwölf Bänden, Volume 4. 9th edition.
Mannheim: Bibliographisches Institut.
Grewendorf, Günther. 2002. Minimalistische Syntax. Tübingen & Basel: Francke/UTB.
Haider, Hubert. 1984. Mona Lisa lächelt stumm – Über das sogenannte deutsche
‘Rezipientenpassiv’. Linguistische Berichte 89: 32-42.
Haider, Hubert. 2010. The Syntax of German. Cambridge University Press.
208
Lang, Peter. 2007. Grammatik und Norm: Direktes Objekt, indirektes Objekt
und der doppelte Akkusativ. Ms. Universität Zürich.
Lee-Schoenfeld, Vera. 2006. German Possessor Datives: Raised and Affected.
Journal of Comparative Germanic Linguistics 9: 101–142.
Lee-Schoenfeld, Vera. 2016. The syntax of external and internal possessor vari-
ation in German inalienable possession constructions. In Gabriele Diewald
(ed.), Special Issue: Non-central Usages of Datives. Language Typology and Uni-
versals 69. 113-129. Berlin/Boston: De Gruyter.
Lee-Schoenfeld, Vera and Gabriele Diewald. 2017. Passivization possibilities in
double-accusative constructions. Proceedings of the Linguistic Society of America
2, 9:1-14. http://dx.doi.org/10.3765/plsa.v2i0.4050.
Lehmann, Christian, Yong-Min Shin and Elisabeth Verhoeven. 2004. Direkte und
indirekte Partizipation. Zur Typologie der sprachlichen Repräsentation konzeptueller
Relationen (2nd edition). ASSIDUE, Arbeitspapiere des Seminars für Sprach-
wissenschaft der Universität Erfurt Nr. 13.
Luraghi, Silvia and Chiara Zanchi. To appear. Double accusative constructions
and ditransitives in Ancient Greek. Folia Linguistica – Special issue on ditran-
sitives.
Maling, Joan. 2001. Dative: The heterogeneity of the mapping among morpho-
logical case, grammatical functions, and thematic roles. Lingua 111(4): 419-464.
Müller, Gereon. 1995. A-bar Syntax. A Study of Movement Types. Studies in Gener-
ative Grammar 42 .Berlin: Mouton de Gruyter
Plank, Franz. 1987. Direkte indirekte Objekte, oder: Was uns lehren lehrt. Leu-
vense Bijdragen 76: 37-61.
Pylkkänen, Liina. 2008. Introducing arguments. Cambridge, MA: MIT Press.
Reis, Marga. 1985. Mona Lisa kriegt zuviel. Linguistische Berichte 96: 140-155.
Twiner, Nicholas. 2016. Southern coreferential applicatives. Ms. University of
Georgia. (Presented at BLS 42.)
Wahrig. 2003. Dittmann, Jürgen, Thieroff, Rolf, and Adophs, Ulrich (ed.) Fehler-
freies und gutes Deutsch. Gütersloh/München: Wissen Media Verlag.
Wegener, Haide. 1985a. Der Dativ im heutigen Deutsch. Tübingen: Narr.
Wegener, Haide. 1985b. Er bekommt widersprochen. Argumente für die Existenz
eines Dativpassivs im Deutschen. Linguistische Berichte 96: 127-139.
Woolford, Ellen. 1993. Symmetric and asymmetric passives. Natural Language &
Linguistic Theory 11(4): 679-728.
Woolford, Ellen. 2006. Lexical case, inherent case, and argument structure. Lin-
guistic Inquiry, 37(1): 111-130.
209
We say “How high?”: Adverbs, negation, and
verb movement in a verb-final language*
Emily Manetta
University of Vermont
1 The Puzzle
Verb-stranding verb phrase ellipsis (VVPE) refers to a phenomenon in which an
entire verb phrase is elided under identity with a verb phrase in an antecedent
clause, but the verb itself is stranded outside the ellipsis site. VVPE has been dis-
covered and analyzed in detail in a wide variety of unrelated languages, including
Irish (McCloskey 1991), Hebrew (Doron 1991, Goldberg 2005), Portuguese (Martins
1994), Russian (Gribanova 2013a, b), and Greek (Merchant, this volume). I provide
a naturally occurring example from Hindi-Urdu (Manetta, to appear):
For those languages in which it has been investigated, identifying clear instances
of VVPE is challenging when there are other syntactic processes at work that
* It is likely that nearly all of the insights of consequence that underlie the small bit of work done
in this paper have their origin in a talk, class discussion, or paper authored by Jim McCloskey. For
the opportunity to honor Jim, and for their thoughtful comments, I thank the editors of this volume.
I am also grateful to Rajesh Bhatt, Alice Davison, Ayesha Kidwai, and Ghanshyam Sharma for their
contributions, at various points, to the data appearing here. Thanks to Line Mikkelsen, Vera Grib-
anova, and audiences at INALCO, Paris 2016 for useful conversations and exchanges. Above all, my
deepest gratitude and warmest wishes to Jim McCloskey, to whom I am indebted for a number of
fudgies at the Stevenson Coffee Shop, and for my career.
210
can cause internal arguments or other VP-internal material to go missing. For
instance, Hindi-Urdu permits null arguments (e.g. Butt and King 1997). Further,
Simpson, Choudhury, and Menon (2013) claim that Hindi-Urdu has a narrower
ellipsis operation targeting arguments, argument ellipsis.
The task then becomes to establish diagnostics which distinguish VVPE from
these other processes producing similar strings.
One such diagnostic is the so-called “adverb test” : a two-clause sequence in
which the antecedent clause contains a VP-adjoined adverb and that adverb may
be interpreted as present in the VP-ellipsis site in the elliptical clause (this test has
a relatively long history – e.g Matos 1992; Oku 1998; Doron 1991; Goldberg 2005;
Simpson, Chowdhury, and Menon 2013). The reasoning is as follows: if internal
arguments can only go missing due to the presence of a null pronominal or argu-
ment ellipsis, the adverbial reading should be completely absent for the second
clause. If, on the other hand, the adverbial reading is available, the material must
have gone missing as a result of VVPE. For Hindi-Urdu, Simpson, Chowdhury,
and Menon show that temporal adverbials and VP-adverbs modifying manner
can be elided and are optionally interpretable in the site of ellipsis, as in (4b) be-
low. They also show that if the adverb is elided and interpreted in the ellipsis
site, any VP-internal arguments must go missing as well. That is, (4c) indicates
that there is no process permitting adjuncts within the VP to go missing indepen-
dently without arguments doing the same (even though the reverse is certainly
possible). (4b) must then represent a case of true VVPE.
211
However, a puzzle emerges in the implementation of the adverb test in a number
of languages in antecedent-correlate pairs in which the correlate clause includes
negation. Consider (5b), in which the downward entailing environment means
that the situations described by the reading which includes the adverb are not a
subset of the situations described when the adverbial is excluded. If the adverb
reading were available in (5b), it would represent a very strong argument in favor
of VVPE. However, many speakers find that the reading of (5b) that includes the
adverb, what we will call the null adjunct reading, is inaccessible.
The sentence in (5a) asserts that Ram read the paper with care, but it seems that
(5b) has a dominant reading in which Raj did not read the paper at all (carefully
or otherwise). If the null adjunct reading were indeed completely unavailable in
these environments, it would cast doubt on whether VVPE exists in the language
at all, since the operation should hypothetically be possible for any verb phrase,
regardless of its content or context.
The mystery deepens when we consider two additional factors. First, this
apparent failure of the adverb test has been reported for a number of languages
such as Persian and Russian, which have been argued to have VVPE and for which
other diagnostics suggest VVPE is at work. Second, there is a remarkable amount
of variability in the judgements reported for the equivalent of (5b) in these lan-
guages, even among native speaker linguists. For instance, for Persian, Rasekhi
(2014) claims that the null adjunct reading is not available in downward entailing
environments, though a footnote (ftnt 7) admits that some speakers can obtain
these readings with very strong contrastive stress on the equivalent of the ad-
verb “carefully” . On the other hand, Toosarvandani (to appear) states the null
adverb interpretation is indeed available in these environments in Persian with-
out any further discussion (Toosarvandani to appear, p. 18). Turning to Russian,
Vera Gribanova (p.c.) observes that the null adjunct reading is relatively diffi-
cult to obtain in the Russian equivalent of (5b), though Russian has been argued
quite convincingly to feature VVPE (Gribanova 2013a, b; 2017). In Japanese, Oku
1998 claims that the null adjunct reading is not present at all (though this claim is
hedged in a footnote), while Funakoshi (2016) disagrees, claiming that it is indeed
the preferred reading in certain scenarios.
Here, I investigate the apparent failure of the adverb test in Hindi-Urdu as
in (5b) and provide evidence that the null adjunct reading can be facilitated by
additional context and by adjusting the structure of the antecedent (following
methods developed in Funakoshi (2016) for Japanese). The fact that the null ad-
junct reading can be made more palatable supports the potential for VVPE in a
212
given language, but we are still left with a serious question: if VVPE is readily
available, why would it be so difficult to get the null adjunct reading to begin
with?
In this paper, I advance a preliminary analysis of the interaction of contrastive
clauses and ellipsis that explains the inaccessibility of the null adjunct reading for
many speakers in the equivalent of (5b). The analysis draws on insights found in
a rich program of research on the nature of head movement and in particular
recent work investigating the interaction of syntactic processes of verb move-
ment and ellipsis (Hartman 2011; LaCara 2016; McCloskey 2016; Gribanova and
Harizanov 2016; Gribanova 2017; Sailor, forthcoming; i Gribanova and Mikkelsen,
this volume). Locating the position of the verb in the syntax is made all the more
challenging in head-final languages like Hindi-Urdu, since verb movement to
higher functional heads would typically be string-vacuous. The account I pro-
pose for the apparent failure of the adverb test in (5b) hinges on the height of the
verbal complex in the syntactic structure. If, in this kind of negated response, the
verbal complex has moved quite high, outside of TP, then the ellipsis of vP will
be blocked by a constraint like MaxElide (Merchant 2001, 2008; Takahashi and
Fox 2005), which in effect forces ellipsis of the largest possible constituent. Of
course, TP ellipsis would be possible, but that would presumably create a string
distinct from (5b), with the subject missing. Importantly, the analysis I propose
below also provides a formal way to understand the variability in judgments for
the equivalent of (5b) that is evidenced crosslinguistically, and the fact that the
null adjunct reading can emerge more clearly under certain circumstances.
Overall, the work done in this paper is part of a larger effort to better under-
stand head movement in head-final languages, in which the evidence for syntac-
tic verb movement can be relatively subtle (Manetta, to appear). Ellipsis provides
an important window into how high the inflected verb must move to escape verb
phrase ellipsis. The specific contribution in the present paper is a clear account
for a crosslinguistic puzzle dogging analyses of VVPE grounded in current work
on head movement, polarity, and wider conditions on ellipsis.
213
ing in Hindi-Urdu, as in (6)-(8).1
Indeed, Ayesha Kidwai (p.c.) reports that for her, simply additional knowledge
about Raj’s habitual carelessness is sufficient to facilitate the null adjunct reading
in sentences like (5b) above.
1 Thanks to Ayesha Kidwai for her judgments and discussion of these examples.
2 Thereis a related observation concerning the degraded status of following clauses containing
pronouns which refer to the missing internal argument in the alleged VVPE environment.
(1) Ram-ne aapne cake dhyaan-se banay-ee, magar Raj-ne nahiiN banay-ee, aur is-liye wo
Ram-erg self’s cake carefully made-prf.pl but Raj-erg neg made-prf.pl and this-for 3sg
mazedar nahiiN thee!
delicious neg be.pst.pl
‘Ram baked his cake carefully, but Raj did not bake (his cakei ?carefully), and for this reason
it?i was not delicious!
Unsurprisingly these judgments are subject to the same variability and facilitation as the null ad-
junct reading in the text above, so in the interests of space I will omit their discussion here.
214
Crucially, if the internal argument is not missing, the null adjunct reading
cannot be drawn out by any means and remains unavailable:
“low” (restitutive) reading (Johnson 2004) is available for a missing adverb like again (in Hindi-Urdu,
dubara) in an alleged VVPE site, then that reading must be the one obtained from inclusion in the VP-
ellipsis. As (i) illustrates, the restitutive reading does seem to be available. Thanks to Ayesha Kidwai
and Rajesh Bhatt for their judgements.
(1) Ram-ne apnaa darwazaa dubara khol-aa, magar Raj-ne nahiiN khol-aa.
Ram-erg self’s door again open-prf.m but Raj-erg neg open-prf.m
‘Ram opened his door again, but Raj did not (open his door again)’ = Raj did not return his
door to the open state.
215
a. Raj nahiiN paRh-eega
Raj neg read-fut.msg
‘Raj will not read.’ = ‘Raj will not read the new article (?∗ carefully).’
b. magar Raj nahiiN
but Raj neg
‘But not Raj’ = ‘But Raj will not read the new article ?(carefully).’
For (10a), as for (5b) above, in which the subject, negation, and verb remain, the
null adjunct reading is often difficult to obtain without additional context. By
contrast, in (10b), which contains only the subject and a negative particle, the
reading including the adverb carefully is not only available, but in fact the dom-
inant reading. Let us first investigate the ellipsis processes at work in (10b), for
which the null adjunct reading is easy to obtain, in order to better approach (10a).
The analysis I propose below for (10a-b) makes use of well-established ap-
proaches to Hindi-Urdu clause structure. Following a range of previous work
(Bhatt 2003, 2005; Kumar 2006; Butt and Ramchand 2005; Manetta 2011; among
many others), I adopt the basic structure below for a simple Hindi-Urdu clause
as in (11):
(11) TP
AspP T
NegP Asp0
vP Neg
NP
VP v
NP V
I also assume regular syntactic movement of the verb and all inflectional material
via head adjunction to the Aspect head, as argued for in Manetta (to appear) (see
similar proposals adopted in Kumar 2006 and Bhatt and Dayal 2007), and move-
ment of the subject to Spec, TP (along with Bhatt 2003, 2005; Manetta 2011). I fol-
low Dwivedi (1991) and Bhatt and Dayal (2007) in the claim that negation in Hindi-
Urdu heads a right-headed maximal projection NegP which is located between
vP and AspP (contra Kumar 2006). The verbal complex moves through negation
into Asp0 when present, creating either the word order neg+verb+auxiliaries or
verb+neg+auxiliaries in negated sentences (also as in Baker 2014).
It is important to note that sentential negation and the discourse particle
serving as a negative response to a question are the same lexical item in Hindi-
Urdu: nahiiN.
216
(12) a. kyaa Ram Chomsky-ka naya lekh dhyaan-se paR-eega
Q Ram Chomsky-gen new writing carefully read-fut.msg
‘Will Ram read the new article by Chomsky carefully?’
b. HaaN / nahiiN
yes / no
‘Yes’/’no’ = ‘Ram will/will not read the new article by Chomsky care-
fully’
c. Ram Chomsky-ka naya lekh dhyaan-se nahiiN paR-eega.
Ram Chomsky-gen new writing carefully neg read-fut.msg
‘Ram will not read the new article by Chomsky carefully.’
With this background in mind, let us first consider the response in (10b), repeated
here:
I propose that the constituent elided in (10b) is TP - to use the term in Grib-
anova (2017), this is contrastive polarity ellipsis (see also Vicente 2006; Kazenin
2006; Morris 2008; Barros 2014). Stranded outside the ellipsis site are the nega-
tive discourse particle nahiiN in the Polarity (Pol) head, and a contrastive phrasal
remnant (in this case the subject) in the specifier of PolP.
(13) PolP
Raji
TP Pol
nahiiN
ti
AspP T
NegP Asp0
nahiiN+paReega
vP Neg
...
The adverbial reading will naturally emerge here, as the adverb is well-contained
within the elided TP.4 In the interests of space, I gloss over interesting questions
4 This account assumes agreement in polarity between the Pol and Neg head (as discussed in Grib-
anova 2017 for Russian). The Neg head is spelled out morphophonologically as sentential negation
(nahiiN), while the Pol head may be spelled out as a discourse particle of the relevant polarity when
present (haaN/nahiiN). Though space constraints prevent a thorough investigation of negation and
negative polarity items here, this view of agreement between Pol and Neg could help to explain the
facts of NPI licensing in Hindi-Urdu: in particular, the fact that NPIs are licensed in subject position
in the language (of long interest: e.g. Mahajan 1990b; Lahiri 1998; Kumar 2006). I leave this issue to
further research.
217
prompted by the analysis in (13) to make progress toward an account of the failure
of the adverb test which is the focus of this paper.
Let us now turn to the mysterious (10a), which features the subject, negation,
and the fully inflected verb.
Under the account in Manetta, to appear, (10a) has two possible analyses: one
featuring a null/missing internal argument and one featuring VPE. Of course,
if the internal argument has gone missing on its own, the unavailability of the
null adjunct reading is easily explained since no adverb is present in the clause
at all. If, on the other hand, the internal argument is missing as the result of
being contained within a larger ellipsis site, then the reading should in theory be
available. In what follows I show that under unmarked circumstances there are
actually no elliptical parses for (10a).
If the string in (10a) were derived via ellipsis, the entire verbal complex, in-
cluding the verb root, inflectional material, and sentential negation must remain
outside the ellipsis site. I propose here that in an environment of contrastive
polarity which lacks the negative discourse particle, the entire verbal complex
(including negation when present) may be attracted to the Polarity head (Holm-
berg 2001; Gribanova 2017). In this case, there would actually be two constituents
of different sizes which could conceivably undergo ellipsis: vP and TP (bolded in
the tree below).
(14) PolP
TP Pol
nahiiN+paReega
Raj
AspP T
NegP Asp0
nahiiN+paReega
vP Neg
...
This configuration would certainly permit TP ellipsis, but in that case the subject
should be missing, generating not (10a), but instead a string in which the subject
is not present at all, as in (15):
218
While this string is certainly grammatical, the missing agent of the reading action
must be identical to that in the antecedent (Ram). This is not a scenario in which
Ram and Raj are being contrasted, and thus not a scenario in which (10a) can be
produced.
But what about vP ellipsis? Why is it not possible in (14)? I argue here, along-
side Gribanova 2017, that vP ellipsis is impossible here due to the restriction on
ellipsis size termed MaxElide (Merchant 2008). Classically this constraint is in-
voked to block verb phrase ellipsis in environments in which sluicing is possible.
Under Takahashi and Fox’s (2005) formulation of the constraint, in cases where
there is a variable in the elided constituent that is bound from outside (rebinding),
there must be a parallelism domain (PD) containing the variable’s binder which
is semantically identical with another constituent (modulo focus-marked con-
stituents) (Takahashi and Fox 2005:229). MaxElide then requires the constituent
targeted for ellipsis to be the largest possible constituent dominated by the PD.
Hartman (2011) extends the constraint to apply to head movement in addition to
phrasal movement. To the extent that we understand head movement to be a
syntactic phenomenon, it should leave traces interpreted as variables subject to
MaxElide. Gribanova 2017 argues that syntactic head movement to Pol in Russian
leaves behind a variable, forcing the PD to be large enough to include the binder.
For this reason, TP ellipsis but not vP ellipsis is possible in these environments,
since we must choose the largest elliptical constituent in the PD. I propose here
that an analogous situation pertains in (14): given that the parallelism domain
must include the head which has moved to Pol, only TP ellipsis is possible since
vP ellipsis would violate MaxElide.
Given this analysis, it is impossible for the string in (10a) to be generated by
an instance of vP or TP ellipsis. The only way to generate this particular string is
via a null/missing internal argument, and this is why for most speakers, the null
adjunct reading is not readily available in an unmarked context. In this view, (10a)
is quite different from (10b) (or indeed (12b) or (15)) in that it does not represent an
instance ellipsis at all.
We must now ask why the null adjunct reading in a string equivalent to (10a)
can be facilitated with additional context or a change in the nature of the an-
tecedent clause. Given the analysis above, when the null adjunct reading does
emerge, it must be because there exists a Pol head which hosts the V+v+Asp+Neg
complex and may also host the contrastive topic in Spec, PolP. The TP must then
be elided, stranding the contrastive topic and verbal complex. The large ellipsis
site necessarily contains the adverb.
internal arguments are available in the language, (15) could be derived without ellipsis having roughly
the structure: [TP proRAM proARTICLE NEG+VERB.] In this case, the adverb carefully is not present, so
the reading is not available. I believe the interference of this possible parse for (15) is what renders the
adverb less accessible here.
219
(16) PolP
Raji
TP Pol
nahiiN+paReega
ti
AspP T
NegP Asp0
nahiiN+paReega
vP Neg
...
Given this view, what then is the source of the variation we see in the acces-
sibility of the null adjunct reading, both within and across individual speakers
(and indeed, across a range of unrelated languages)? We can speculate that the
Pol head in (16), which attracts a contrastive topic and the verbal complex, might
be in the lexicon of most speakers for scenarios of very strong contrast between
topics, but not otherwise. If the contrastive topic and verbal complex are found
this high in the structure, TP ellipsis is possible. This would explain why rich
context would render the null adjunct reading far more accessible. Alternatively,
it might be that there is inter-speaker variation with respect to this head, and it
is not present in the lexicon of every speaker (or in every variety of the relevant
language). While an extensive empirical investigation would need to be done to
establish the precise facts, the analysis presented here provides a way of couching
this variation in the syntax of ellipsis.
The analysis proposed in (14) and (16) suggests that other strongly contrastive
constituents (non-subjects) should be able to occupy Spec, PolP with similar re-
sults.
(17) Context: We think that Ram will read all of the assigned works by Chom-
sky, but with varying levels of attention.
220
Since the null adverbial reading can be facilitated in (17) by rich context just as
it can in examples in which the subject is contrasted, we have further support
that these are the scenarios in which we have discourse-driven movement of a
contrastive topic and the verbal complex to a high position followed by ellipsis
of TP (easily containing the adverbial).
To summarize the new proposal, under conditions of contrastive polarity,
Hindi-Urdu has several Polarity heads. The first may introduce the negative dis-
course particle, in addition to a contrastive phrase in its specifier. Ellipsis of the
TP dominated by this Pol head will lead to responses like that in (10b) above. In
addition, Hindi-Urdu has a separate Polarity head which can attract the verbal
complex, made concrete here via the feature [v], accompanied by the epp subfea-
ture which serves to trigger head movement. This Polarity head also possess the
[e] feature, prompting ellipsis of TP:
pol: [e]
(18)
[v, epp]
In environments of very strong contrast (and the threshold may be different for
different speakers or for different languages/varieties), a second Pol head may be
available:
pol: [e]
(19) [v, epp]
[topic, epp]
Through this approach to understanding the interaction between adverbs, po-
larity, and ellipsis, we can solve both the initial puzzle of the inaccessibility of
the null adjunct reading in a specific context, and the variability we find in its
facilitation in a wide range of languages with VVPE.
221
b. * magar Raj nahiiN nahiiN paRh-eega
but Raj negpart neg read-fut.msg
Intended: ‘but Raj will not read (the new article by Chomsky (care-
fully)).’
I have claimed here (alongside Gribanova 2017) that in contexts in which the verb
moves as high as Pol, only TP ellipsis is available due to the restriction on ellipsis
termed MaxElide, forcing ellipsis of the largest possible constituent. This means
that the problematic string which would at first glance appear to indicate failure
of the adverb test, is not a string generated by ellipsis at all, but instead by a miss-
ing internal argument. No adverb is present, so the reading including the adverb
is not expected. For many speakers, context is needed to provide a scenario of
sufficient contrast to warrant a version of Pol that can also attract a contrastive
topic to its specifier. In this case, TP ellipsis will strand the contrastive topic and
verbal complex, and the null adjunct reading will emerge. I have posited here
that it is availability of this Pol head in the lexicon of a speaker (and the discourse
pressures which condition it) that is the source of variation in the availability of
the null adjunct reading in Hindi-Urdu.
This apparent failure of the adverb test thus provides an opportunity to fur-
ther our investigation of head movement in a head-final language. In head-final
languages like Hindi-Urdu, the nature of the composition of the morphologically
and lexically complex verb is difficult to probe; any head movement would typi-
cally be string-vacuous as all the heads of the verbal complex appear on the right.
A number of researchers have assumed some degree of verb movement for var-
ious reasons (e.g. Kumar 2006, Bhatt and Dayal 2007, Bhatt 2008). However, it is
challenging to find direct evidence that verb movement has taken place, and tests
for positioning of adverbs, post-verbal material, and subjects relative to the verb
are unrevealing when the verbal complex is clause-final (Pollock 1989; McCloskey
1991; Depiante and Vincente 2012). The position of negation has the potential to be
more useful, but as sentential negation can appear either immediately preceding
or immediately following the inflected verb in the verbal string in Hindi-Urdu,
these tests have not provided unambiguous information (Kumar 2006).
As other researchers working on head-final languages have suggested (Otani
and Whitman 1991; Koisumi 2000; Simpson and Syed 2013), VVPE has the po-
tential to provide just such evidence. The availability of VVPE in Hindi-Urdu
demonstrates that the verb must move at least outside of the vP (Manetta, to ap-
pear). If the solution to the failure of the adverb test is on the right track, the verb
can under certain discourse conditions move even higher, feeding contrastive
polarity ellipsis (ellipsis of TP). Though more research is certainly required, the
analysis I propose here has the potential to be extended to a number of languages
in which the variable availability of adverb interpretations in negated elliptical
structures remains a puzzle.
This small project contributes to the wider program of recent work investi-
gating the nature of head movement and its role in the syntax (Chomsky 2001;
Roberts 2010; Hartman 2011; LaCara 2016; McCloskey 2016; Keine and Bhatt 2016;
222
Manetta, to appear, Gribanova and Mikkelsen, this volume). Discourse-driven
head movement to a relatively high point in the clause (Harizanov and Grib-
anova 2017; Gribanova 2017) may well be at the heart of the apparent failure of
the adverb test.
References
Baker, Mark. 2014. Pseudo-Noun Incorporation as covert noun incorporation:
linearization and crosslinguistic variation. Language and Linguistics 15: 5-46.
Bhatt, Rajesh. 2003a. Lecture Handouts: Topics in the Syntax of Modern Indo
Aryan Languages. MIT, Spring 2003.
Bhatt, Rajesh. 2005. Long Distance Agreement in Hindi-Urdu. Natural Language
and Linguistic Theory 23:757-807.
Bhatt, Rajesh and Veneeta Dayal. 2007. Rightward Scrambling as Rightward
Remnant Movement. Linguistic Inquiry 38:287-30.
Butt, M., Jabeen, F., & Bögel, T. Verb cluster internal wh-phrases in Urdu: Prosody,
syntax and semantics/pragmatics. Linguistic Analysis 40 (3-4). 445-487.
Butt, Miriam, and Tracy Holloway King. 1997. Null Elements in Discourse Struc-
ture. In Papers from the NULLS Seminar, ed. K.V. Subbarao. Delhi: Moti Lal
Banarsi Das.
Butt, Miriam and Gillian Ramchand. 2005. Complex Aspectual Structure in Hindi/Urdu.
In The Syntax of Aspect, Nomi Ertishik-Shir and Tova Rappaport (eds). 117-153.
Oxford: OUP.
Chomsky, Noam. 2001. Derivation by phase. In Ken Hale. A life in language, ed.
M. Kenstowicz. 1-52. Cambridge, Mass.: MIT Press.
Davison, Alice. 1999. Empty categories: anaphoric vs. pronominal coindexing,
ms. University of Illinois.
Davison, Alice. 2005. Phrasal predicates: How N combines with V in Hindi/Urdu.
In Tanmoy Bhattacharya (ed.) Yearbook of South Asian Languages and Linguistics,
83-116. Berlin: Mouton DeGruyter.
Davison, Alice. 2013. Emtpy Categories in the Hindi-Urdu binaa participial clause.
Lingua Posnaniensis, vol. LV (2)/2013. The Poznań Society for the Advancement
of the Arts and Sciences. 25-39.
Dayal, Veneeta. 2003. Bare Nominals: Non-specific and Contrastive Readings
under Scrambling, in Simin Karimi (ed.) Word Order and Scrambling, Malden,
MA: Blackwell Publishers.
Dayal, Veneeta. 2011. Bare Noun Phrases. Maeinborn, von Heusinger and Portner
(eds), Semantics: An International Handbook of Natural Language Meaning, 33.2
1087-1108. Berlin: de Gruyter.
Depiante, Marcela, and Luis Vicente. 2012. El movimiento y la morfología del
verbo. In Brucart and Gallego (eds.) El movimiento de constituyentes, 95-106.
Madrid: Visor Libros.
Doron, Edit. 1991. V-movement and VP ellipsis. Jerusalem: The Hebrew University
of Jerusalem.
223
Dwivedi, Veena. 1994. Syntactic Dependencies in Relative Phrases in Hindi. PhD
dissertation, University of Massachusetts, Amherst.
Funakoshi, Kenshi. 2016. Verb-Stranding Verb Phrase Ellipsis in Japanese. Journal
of East Asian Linguistics. 25(2): 113-142.
Goldberg, Lotus. 2005. Verb-stranding VP ellipsis: a cross-linguistic study. PhD dis-
sertation, McGill University, Montréal.
Gribanova, Vera. 2013a. Verb-stranding verb phrase ellipsis and the structure of
the Russian verbal complex. Natural Language and Linguistic Theory 31(1): 91-136.
Gribanova, Vera. 2013b A new argument for verb-stranding verb phrase ellipsis.
Linguistic Inquiry 44(1): 145-157.
Gribanova, Vera and Boris Harizanov. 2016. Whither head movement? Presented
at the workshop on the Status of Head Movement in Linguistic Theory. Stan-
ford University. September, 2016.
Gribanova, Vera. 2017. Head movement and ellipsis in the expression of Russian
polarity focus. Natural Language and Linguistic Theory 35.1, 1-43.
Gribanova, Vera And Line Mikkelsen. 2018. On the interaction of head move-
ment and ellipsis in Danish. This volume.
Hardt, Daniel. 1993. Verb Phrase Ellipsis: form, meaning and processing. Ph.D. dis-
sertation, University of Pennsylvania.
Harizanov, Boris and Vera Gribanova. 2017. Post-syntactic head movement in
Russian predicate fronting. Annual Meeting of the LSA, Austin, TX; January
5-8, 2017.
Hartman, Jeremy. 2011. The semantic uniformity of traces: Evidence from ellipsis
parallelism. Linguistic Inquiry 42: 367-388.
Hankamer, Jorge, and Ivan A. Sag. 1976. Deep and surface anaphora. Linguistic
Inquiry 7: 391-426.
Hoji, Hajime. 1998. Null objects and sloppy identity in Japanese. Linguistic Inquiry
28: 127-152.
Hoji, Hajime. 2003. Surface and Deep Anaphora, Sloppy Identity, and Experi-
ments in Syntax. In Barss, A. and T. Langendoen (eds), Anaphora: A Reference
Guide, eds. Malden, MA: Blackwell.
Holmberg, Anders. 2001. The syntax of yes and no in Finnish. Studia Linguistica
55(2): 140-174.
Johnson, Kyle. 2001. What VP-ellipsis can do, and what it can’t, but not why. In
Baltin, M. and C. Collins (eds.) The Handbook of Contemporary Syntactic Theory,
439-479. Oxford: Blackwell.
Johnson, Kyle. 2004. How to be Quiet. In Adams, Nikki, Adam Cooper, Fey
Parrill and Thomas Wier (eds.), Proceedings from the 40th Annual Meeting of the
Chicago Linguistics Society, 1-20.
Jouitteau, Mélanie. In prep. Tense Second effects in Breton linear V2: evaluating
the arguments for a post-syntactic analysis. In Biberauer, Teresa, Sam Wolfe
and Rebecca Woods (eds). Rethinking Verb Second. Oxford: Oxford University
Press.
Karimi, Simin. 1999a. A Note on Parasitic Gaps and Specificity. Linguistic Inquiry
30: 704-713.
224
Karimi, Simin.1999b. Specificity Effect: Evidence from Persian. Linguistic Review
16: 125-141.
Kazenin, Konstantin. 2006. Polarity in Russian and Typology of Predicate Ellip-
sis. Ms. Moscow State University.
Keine, Stefan & Rajesh Bhatt. 2016. Interpreting verb clusters. Natural Language
and Linguistic Theory 34: 1445-1492.
Koisumi, Masatoshi. 2000. String Vacuous Overt Verb Raising. Journal of East
Asian Linguistics 9, 227-285.
Kumar, Rajesh. 2006. Negation and Licensing of Negative Polarity Items in Hindi
Syntax. New York: Routledge.
Lacara, Nicholas. 2014. Why there is no verb-stranding in (Mainland) Scandina-
vian. 29th Comparative Germanic Syntax Workshop University of York, UK,
26 September, 2014.
Lacara, Nicolas. 2016. A Germanic verb movement paradox. Workshop on the
Status of Head Movement in Linguistic Theory. Stanford, CA, September 15,
2016.
Lahiri, Utpal. 1998. Focus and negative polarity in Hindi. Natural Language Se-
mantics 6: 57-123.
Mahajan, Anoop. 1990a. The A/A-bar distinction and movement theory. PhD
dissertation, MIT, Cambridge, Mass.
Mahajan. Anoop. 1990b. LF Conditions on Negative Polarity Licensing. Lingua
80: 333-348.
Manetta, Emily. 2011. Peripheries in Kashmiri and Hindi Urdu: The syntax of discourse-
driven movement. series: Language Faculty and Beyond, Amsterdam: John Ben-
jamins.
Manetta, Emily. 2012. Reconsidering rightward scrambling: Postverbal con-
stituents in Hindi-Urdu. Linguistic Inquiry 43:1, 43-74.
Manetta, Emily. 2014. The shape of the causative verbal domain: evidence from
Kashmiri. Syntax 17.3: 235-168.
Manetta, Emily. 2016. Verb phrase ellipsis in Hindi complex predicates. Pre-
sented at the International Conference on Hindi Studies (INALCO), Paris, France.
Sept 14-16.
Manetta, Emily. To appear. Verb phrase ellipsis in complex predicates in Hindi-
Urdu. Natural Language and Linguistic Theory.
Martins, Ana-Maria. 1994. Enclisis, VP-deletion and the nature of sigma. Probus
6: 173-205.
Matos, G. 1992. Construções de Elipse do predicado em Português - SV Nulo e Despo-
jamento. PhD Dissertation. Lisboa: Universidade de Lisboa.
McCloskey, James. 1991. Clause structure, ellipsis and proper government in
Irish. Lingua 85: 259-302.
McCloskey, James. 2016. Interpretation and the Typology of Head Movement:
A Re-Assessment. Workshop on the Status of Head Movement in Linguistic
Theory. Stanford, CA, September 15, 2016.
Merchant, Jason. 2001. The syntax of silence: Sluicing, islands and the theory of ellipsis.
Oxford: Oxford University Press.
225
Merchant, Jason. 2008.‘Variable island repair under ellipsis’. Topics in ellipsis, ed.
by Kyle Johnson, 132-153. Cambridge: Cambridge University Press.
Merchant, Jason. 2013. Voice and ellipsis. Linguistic Inquiry 44(1): 77-108.
Merchant, Jason. 2018. Verb-stranding predicate ellipsis in Greek. This volume.
Morris, Amanda. 2008. Polarity Ellipsis and Negative Stripping. Unpublished
manuscript. http://babel.ucsc.edu/∼ hank/PolarityEllipsisandNegStripping.pdf.
Oku, Satoshi. 1998. A theory of selection and reconstruction in the minimalist perspec-
tive. PhD dissertation, University of Connecticut.
Otani, Kazuyo, and John Whitman. 1991. V-Raising and VP-ellipsis. Linguistic
Inquiry 22: 345-358.
Pollock, Jean-Yves. 1989. Verb movement, universal grammar, and the structure
of IP. Linguistic Inquiry 20: 365-424.
Roberts, Ian. 2010. Agreement and head movement. Cambridge, MA: MIT Press.
Ross, John Robert. 1969. Auxiliaries as main verbs. Studies in Philosophical Linguis-
tics 1: 77-102.
Sailor, Craig. 2014. The variables of VP-ellipsis. PhD Dissertation, University of
California, Los Angeles.
Sailor, Craig. Forthcoming. The Typology of Head Movement and Ellipsis: a
reply to Lipták & Saab. Natural Language and Linguistic Theory.
Saito, Mamoru. 1985. Some asymmetries in Japanese and their theoretical implications.
PhD diss., MIT Press, Cambridge.
Simpson, Andrew, Arunima Choudhury and Mythili Menon. 2013. Argument
Ellipsis and the licensing of covert nominals in Bangla, Hindi and Malayalam.
Lingua 134:123-138.
Simpson, Andrew and Saurov Syed. 2014. Finiteness, negation, and the direction-
ality of headedness in Bangla. Natural Language and Linguistic Theory 32:231-161.
Takahashi, Daiko. 2008. Quantificational null objects and argument ellipsis. Lin-
guistic Inquiry 39:307-326.
Takahashi, Shoichi, and Danny Fox. 2005. MaxElide and the re-binding prob-
lem. In Georgala, Effi and Jonathan Howell (eds) Semantics and Linguistic The-
ory (SALT) 15, 223-240. Ithaca: CLC Publications.
Toosarvandani, Maziar. 2009. Ellipsis in Farsi complex predicates. Syntax 12(1):60-
92.
Toosarvandani, Maziar. To Appear. Persian. In van Craenenbroeck, Jeroen and
Tanja Temmerman (eds) The Oxford handbook of ellipsis. Oxford: Oxford
University Press.
Vicente, Luis. 2006. Short negative replies in Spanish. In Jeroen van de Weijer
and Bettelou Los (eds.), Linguistics in the Netherlands 23: 199-211. Amsterdam:
John Benjamins.
226
Verb-stranding predicate ellipsis in Greek,
implicit arguments, and ellipsis-internal focus*
Jason Merchant
University of Chicago
228
their effects on surrounding material (Thomas et al. 2016). When something is
gone from the linguistic form, but nevertheless speakers produce and hearers in-
terpret the resulting structure with a determinate (within reason) meaning, given
a particular linguistic context, it is up to the analyst to discover the mechanisms
that give rise to these meanings. One of the ways we do this is to determine
whether the syntactic properties of the missing element are the same as those
of its putative overt counterpart. When these align, Ockam’s razor impels us to
conclude that the element is present, but unpronounced. When these properties
do not align, our task is harder, but application of Ockam’s razor in such cases
suggests that the element is not there.
But Greek does not have ellipsis of the verb phrase complement to the perfect
auxiliary exo ‘have’, nor does it allow for sentential negation to appear by itself
(dhen is a clitic on the finite verb complex).
(2) a. *I Maria exi teliosi tin ergasia tis, ke i Anna exi Δ, episis.
the Maria has finished the homework her and the Anna has too
(‘Maria has finished her homework, and Anna has, too.’)
b. *O Petros ine ikanos, ala o Alexandros dhen Δ.
the Petros is capable.m.sg but the Alexander not
(‘Petros is capable, but Alexander isn’t.’)
1 For her many years of judgments on these and countless other, similar examples, my undying
thanks to the infinitely patient Anastasia Giannakidou. Thanks also to the dozens of other Greek
speakers who have heard various parts of this material over the past twenty years and supplied valu-
able feedback and judgments, especially the audience at the 5th annual Midwest Workshop on Greek
Linguistics in 2016, including Natalia Pavlou, Marika Lekakou, and Marina Terkourafi. Thanks also
to Line Mikkelsen, Idan Landau, and Anikó Lipták for timely comments on an earlier draft.
229
These facts are most readily understandable if the target of ellipsis in Greek is a
predicate phrase, which we can conveniently identify with Bowers’ (1993, 2002)
PredP. For an example such as (1a), then, we have the following structure, where
strikethrough indicates the node targeted for non-pronunciation (the position of
the clitic negation dhen= and any potential internal structure of the verb+tense
ine are not relevant here):
(3) TP
DP1
Neg
o Alexandros T VP
the Alexander dhen=
not V tV PredP
ine t1
is Pred AP
ikanos
capable
Clear evidence that PredP is elided, and not merely suppressed with its content
somehow understood, comes from extraction of internal arguments of the elided
predicate head. In (4a), the contrastively focused PP argument of ikanos ‘capable’,
ja dholofonia literally ‘for murder’, is fronted to the clause-initial focus position
in the second clause. This PP is headed by a preposition, ja, that is lexically se-
lected by the adjective ikanos—its appearance here is not predictable from its own
meaning or from the meaning of ikanos, nor is it a default preposition. Standard
assumptions about such idiosyncractic lexical selection, therefore, require that
there be in the syntactic representation of (4a) a head that selects it. This full
structure is precisely what an ellipsis analysis makes available, as shown in (4b).
230
b.
PP
Foc TP
ja dholofonia
for murder
pro1
he Neg
T VP
dhen=
not V tV PredP
ine
is t1
Pred AP
ikanos PP
capable
Additional evidence that the complement of ine is elided (that is, that it is ‘sur-
face anaphoric’ in the sense of Hankamer & Sag 1976 or is true ellipsis, in the
sense of Sag & Hankamer 1984), and not merely suppressed by a general mecha-
nism of argument suppression (that is, that it is not a kind of ‘deep anaphor’ in
Hankamer and Sag’s sense, or model-theoretic anaphora in Sag and Hankamer’s)
comes from the differing interpretations available to the two structures. When
the AP is elided, the predicate of the second conjunct is interpreted exactly as the
first is. This can give rise to covarying or coreferential readings with pronouns
in (5b), for example:
(5) a. O Alexis ine ikanos ja dholofonia, ala o Petros dhen ine.
the Alexis is capable for murder but the Petros not is
‘Alexis is capable of murder, but Petros is not capable of murder.’
b. O Alexis ine perifanos ja ton jo tu, ala o Petros dhen ine.
the Alexis is proud of the son his but the Petros not is
‘Alexis1 is proud of his1/2/3 son, but Petros2 is not proud of his1/2/3
son.’
When just the PP internal argument to an adjective is missing, the adjective re-
ceives a general interpretation, with its internal argument specified only prag-
matically. This means that the internal argument could take as its value murder
or his son, as in (5) above, but it need not; by Gricean principles, the availability of
the expressions in (5) in fact make this reading highly marked. The most natural
interpretation of the following examples is precisely that of their English coun-
terparts, with an unspecified internal argument, and the predicate attributing a
generic individual-level property to the subject.
(6) a. O Alexis ine ikanos ja dholofonia, ala o Petros dhen ine ikanos.
the Alexis is capable for murder but the Petros not is capable
‘Alexis is capable of murder, but Petros is not capable.’
231
b. O Alexis ine perifanos ja ton jo tu, ala o Petros dhen ine
the Alexis is proud of the son his but the Petros not is
perifanos.
proud
‘Alexis is proud of his son, but Petros is not proud.’
Bowers’ Pred has been variously redubbed Voice or v when it is used in the ver-
bal extended projection: it is the head that introduces the external argument.
As Bowers points out, Pred is cross-categorical, given that nouns, adjectives, and
prepositions can take subjects as well. For reasons of perspicuity and consistency
with much recent literature (including Merchant 2013c and Alexiadou et al. 2015),
I will use v as the label for Pred when Pred takes a VP as its sister, but I stress
that this is merely a notational convenience. (In any case, the reader should bear
in mind that this v is the argument-introducing one; some work takes v to be a
categorizing node, a function I would attribute to a V node combining with an
uncategorized root if this were salient.) This means that the structure of the em-
bedded clause in (7a) will be that given in (8); again, because the details of head
movement are not relevant to our concerns, I will follow Bennett et al. (2017) in
representing the result of head movement simply as a vertical stack of labels, and
I will omit additional functional material that associates with verbs in particu-
lar, such as Voice and Aspect (see Merchant 2015 and Spyropoulos et al. 2015 for
exploration of these details in Greek).
232
(8) TP
DP1
to pedhi T vP
the child
v
epitidhes vP
V intentionally
t1
tv VP
ekapse
burned tV DP
tin supa
the soup
The finite verb can invert with the subject in questions and relative clauses (and
even in simple declaratives under certain discourse conditions); if such inver-
sions must be fed by movement of the verb to T, then their presence in Greek is
a further argument for V-to-T movement.
3 Verb-stranding ellipsis
We have concluded that Greek is a language with predicate (PredP/vP) ellipsis
and with V-(to-v-)T movement. The question, then, is whether these two things
can be combined. The combination of the movement of a head H with ellipsis of
HP (or of an XP contained HP, if H moves out of XP) has been the focus of a large
literature (see Funakoshi 2012, Lipták & Saab 2014, Gribanova & Mikkelsen 2018,
Manetta 2018, and Sailor To appear for recent approaches), in particular with re-
spect to the movement of verbs out of elided verb phrases. The primary analytical
issue revolves around examples like the response in (9) and the second clauses in
(10)-(11).
(9) Question: Agorases psomi? Answer: Ne, agorasa.
bought.2s bread yes bought.1s
‘Did you buy bread?’ ‘Yes, I did. (buy bread)’
(10) Epidhi i Anna ithele na agorasi psomi, agorase.
because the Anna wanted subj buy.3s bread bought.3s
‘Because Anna wanted to buy bread, she did. (buy bread)’
(11) Prota irthe ena agori pu agorase psomi. Meta irthe ena koritsi pu
First came a boy who bought bread then came a girl who
episis ithele na agorasi.
also wanted subj buy.3s
‘First a boy came who bought bread. Then a girl came who also wanted
to. (buy)’
233
Examples like these have been the object of sustained and insightful investigation
in Irish in a series of works by Jim McCloskey (McCloskey 1991, 1996, 2017, Bennett
et al. 2017), who has shown beyond a doubt that the finite verb in an Irish example
such as (12) raises to a position outside the target of ellipsis (in his recent work,
the verb moves to Pol, above a lower TP, which can elide).
(13) a. TP b. TP
T vP TE vP
v pro2s v
pro1s
tv VP tv VP
V V
tV DP tV DP
agorases agorasa
bought.2s psomi bought.1s
psomi
bread bread
in situ pro; as Alexiadou & Anagnostopoulou (1998) persuasively argue, there is no reason to believe
that Greek has an EPP requiring a filled specifier of TP.
234
4 Argument omission?
Greek is not at all unusual in having verbal and predicate alternations that appear
to be due to the optionality of certain arguments. This alternation, when it affects
definite pronominal subjects, is generally known as pro-drop, and, generalizing,
when it affects arguments of any kind, as argument drop.
The differences between dropped pronouns, with definite anaphoric refer-
ence, and dropped indefinite arguments, can be seen in the following example,
from Giannakidou & Merchant (1997).
As indicated in the translations, the Greek answer is compatible with two in-
tended readings: either the answerer intends to refer to the individuals in the set
of students introduced in the question (the definite anaphoric reading, compat-
ible with a specific reading of the indefinite), or the answerer intends merely to
affirm that a certain set with the given cardinality came, without being willing
or perhaps able to specify who the members of that set in the actual world might
be. Notice that the English pronoun they in this context lacks the second read-
ing. This is evidence that Greek argument drop, including of subjects, does not
always involve traditional pro-drop. (As Giannakidou & Merchant (1997) claimed,
there are several ways to get to a null DP, including combining a null indefinite
determiner with NP-ellipsis.)
Bare singulars, both mass and count, can also go missing, as in B’s responses
to A’s questions in (15) and (16):
Count singulars with the indefinite article and bare plurals also license omission:
235
(17) Agorasa ena sfungari jati mu ipes na fero (ena).
bought.1s a/one sponge because me told.2s subj bring.1s one
‘I bought a sponge because you told me to bring one.’
(18) Agorasa sfungaria jati mu ipes na fero.
bought.1s sponges because me told.2s subj bring.1s
‘I bought sponges because you told me to bring some.’
But such indefinites do not license omission if they are the objects of preposi-
tions:
When the antecedent is definite, however, whether on a type or token use, this
kind of omission is not possible; this is true both inside and outside of islands:
In this respect, this kind of definite pronominal argument omission is very dif-
ferent from the situation in Hebrew or Russian, both of which permit definite
objects to be dropped when the verb whose argument they are is not inside an
island (Gribanova 2013b). Hebrew also allows this kind of argument omission
inside islands, as Landau (2017) documents.
236
The empirical pattern, then, is somewhat complex. In all the above cases, the
verb whose argument is omitted is not that same as the verb that introduces the
antecedent argument. These examples are chosen in order to help minimize the
possibility that these examples involve a kind of VSE. But as we will see below,
it is not an absolute requirement that the verbs match: instead, contrastive fo-
cus can allow for mismatched verbs. Absent such focus, however, it seems that
Greek has limited ability to license null arguments when the antecedent is a full
DP (whether definite or indefinite). This is particularly clear in the distributive
prepositional case in (19): if arguments could be freely omitted (assuming they
have the appropriate kind of antecedent, as would be the case in (19)), the fact
that (19) is unacceptable without an overt DP complement to the preposition apo
would be unexplained.
This is not to imply that Greek lacks predicates that allow for implicit argu-
ments. It does possess such predicates, and it is important to examine such cases
carefully to distinguish them from VSE.
(22) Implicit indefinite arguments (Fodor & Fodor 1980, Dowty 1981, Mittwoch
1982)
a. John {baked / ate / hunted / read / served the guests}.
b. John {baked a cake / ate a carrot / hunted a rabbit / read a book /
served the guests the salad}.
(23) Implicit definite arguments (Fillmore 1986)
a. Susan {noticed / understood / saw}.
b. Susan {noticed / understood / saw} the error / that something was
wrong.
(24) Implicit reflexive arguments
a. Maxwell {shaved / bathed / scratched}.
b. Maxwell {shaved / bathed / scratched} himself.
(25) Implicit reciprocal arguments
a. Adam and Beth {kissed / screwed / divorced}.
b. Adam and Beth {kissed / screwed / divorced} each other.
237
It is striking to note that all of these kinds of implicit arguments can be found with
predicates which, when they take these arguments overtly, mark them obliga-
torily with lexically selected prepositions.3 This observation, to my knowledge,
has not been made previously in the literature, and indicates that any system that
merely suppresses DPs (or NPs) in these positions, or which posits null DPs, does
not generalize to the full range of facts.
(26) John {flirted (with someone) / was shooting (at something) / argued (with
someone)}.
(27) Susan {agreed (to it / with it / us) / looked (at it)}.
(28) Maxwell is proud (of himself).4
(29) Adam and Beth {are married (to each other) / broke up / argued (with each
other)}.
Not all predicates allow for implicit arguments. Even predicates that have very
similar meanings to those that license implicit arguments do not themselves al-
ways license such arguments. It is not predictable from the meaning of the pred-
icate whether it will allow for an implicit argument, as the following sets of near
minimal pairs with the above show.
238
h i
(35) a. eat cat V, -aux
h i
sel < (D) >
h i
b. ingest cat V, -aux
h i
sel < D >
The crucial device is the parenthesis, following Chomsky (1965), to collapse two
otherwise equivalent lexical entires. This device is employed in more familiar
representations of lexical entries in Levin & Rappaport (1988), Sadock (1991), Pol-
lard & Sag (1994), Bresnan (2001), and Culicover & Jackendoff (2005).
Work on existential implicit arguments of predicates such as eat and read han-
dle the restriction to narrowest scope in a variety of ways. Fodor & Fodor (1980)
use meaning postulate like that in (36), while Dowty (1981) posits the lexical rule
in (37).
However, as the most comprehensive account of the full range of implicit argu-
ments, Gillon (2012), shows, such accounts are inferior to one that uses specific
VP-interpretation rules that depend on the presence of diacritics on the particu-
lar predicates (for existential, definite, reflexive, and reciprocal interpretations),
as in (38):
(38) Let D be the domain of the model and let G be the set of ordered pairs, or
graph, of the binary relation assigned to a lexical entry with the argument
frame of < N P ; N P, q >. Then, the function assigned to q assigns { x :
∃ y ∈ D and < x, y >∈ G } to the VP node of the V node dominating the
lexical entry.
239
(40) I Ana iksere / idhe / katalave.
the Anna knew saw understood
‘Anna knew / saw / understood.’
Note that such uses of these predicates is not restricted to non-island or non-
embedded contexts; they are perfectly acceptable inside islands:
240
This pattern is replicated with verbs that take obligatory PP arguments. In (45a-
c), we see the verb stirizome ‘depend’ without its obligatory PP complement. In
all cases, the verb has an antecedent that heads a VP with the requisite PP. (46)
shows the same for the verb epimeno ‘insist’.
(45) a. A: Stirizete katholu to ikonomiko model stin prostasia tis
depends at.all the economic model on.the protection of.the
ergasias?
labor
‘Does the economic model depend at all on the protection of labor?’
B: Ne, stirizete.
yes depends
‘Yes, it does.’
b. To politiko modelo stirizete stin prostasia tis ergasias —
the political model depends on.the protection of.the labor
dhistixos, to ikonomiko modelo dhen stirizete.
unfortunately the economic model not depends
‘The political model depends on the protection of labor — unfortu-
nately, the economic model does not.’
c. To oti dhio stus tris neus stirizunte stus gonis tus simeni
the that two on.the three young depend on.the parents their means
oti enas stus tris dhen stirizete.
that one on.the three not depends
‘The fact that one out of three young people depend on their parents
means that one out of three does not.’
(46) Parolo pu o dhaskalos epemine stin erotisi dio, o voithos dhen
despite that the teacher insisted on.the question two the aide not
epemine.
insisted
‘Although the teacher insisted on question two, the aide didn’t.’
Another kind of obligatory argument is the definite pronoun in a context that
supports one. These pronouns are proclitic on the finite verb in Greek, making
examples with verbs and clitic objects useless for testing for VSE, since both the
verb and its object will have raised out of the putatively elided vP:
(47) a. A: Idhes tin tenia?
saw.2s the movie
‘Did you see the movie?’
b. B: Ne, *(tin) idha.
yes it saw.1s
‘Yes, I did.’ or ‘Yes, I saw it.’
The only exception to this pattern is found with certain verb+noun idioms, where
the object of the combined verb+noun can be a full DP or a pronoun, but when
241
the verb is used in VSE, both the noun part of the idiom and any potentially
pronominal object must be elided.
(48) A: Pires prefa tin katastasi?
took.2s prefa the situation
‘Did you get wind of the situation?’5
B: a. Pira.
b. *Tin pira.
c. ( Tin) pira prefa.
it took.1s prefa
‘I did/I got wind of it.’
(49) A: Pires xabari tin kopela?
took.2s notice the girl
‘Were you aware of the girl?’
B: a. Dhen pira.
b. *Dhen tin pira.
c. Dhen (tin) pira xabari.
not her took.1s notice
‘I was/I was aware of her.’
What is unusual about these idioms, and sets them apart from regular transitive
verbs like (47) above, is that they also allow for a dropped object without any ellip-
sis, as seen in the (c) examples. It is the possibility of this pronoun-less alternant
that gives rise to pronoun-less VSE in the (a) examples. (Why precisely the other-
wise expected, and indeed possible, definite pronoun is omissible just with these
idioms, I leave for future work, but the solution seems orthogonal to questions
about ellipsis.)
neither the game nor the word outside of this idiom. Cf. cahoots in the idiom be in cahoots with in
English.
242
Very briefly, I give the results of attempting VSE with each of these kinds of
objects, and contrast the licit VSE with illicit pronouns.
The first case comes from NPIs headed by the n-word determiner kanenas
(see Giannakidou (2000). As can be seen below, both negative and positive re-
sponses are possible (see Merchant (2013b) for discussion of this alternation, which
is found in English as well). Responses with a definite pronoun are anomalous:
the NPI does not introduce a referent into the discourse context that the pronoun
could pick up on.
(50) Dhen vrikes kanena meros ja na parkaris to amaksi?
not found.you any spot for subj park.2s the car
‘Didn’t you find any spot to park the car?’
(51) a. Oxi, dhen vrika.
no not found.I
‘No, I didn’t (find any spot to park the car).’
b. Ne, vrika.
yes found.I
‘Yes, I did (find a spot to park the car).’
(52) a. #Ne, to vrika.
yes it found.I
#‘Yes, I found it.
b. #Oxi, dhen to vrika.
no not found.I
#‘No, I didn’t find it.’
Disjunctions deliver a parallel set of facts. A definite pronoun would give rise to
an unwanted existential presupposition in the following example, but the VSE
variant is well-formed.
(53) a. Paratirises i kena i lathi sto xirografo?
observed.2s either gaps or errors in.the manuscript
‘Did you observe either lacunae or errors in the manuscript?’
b. Oxi, dhen (#ta) paratirisa.
no not them observed.1s
‘No, I did not.’
Bare singular noun phrases can have generic meanings (or, in some circumstances,
singular indefinite nonspecific existential readings). These generic readings do
not license following pronouns, but they do participate in VSE:
(54) a. Foras kaskol?
wear.2s scarf
‘Are you wearing a scarf?’ or ‘Do you wear scarves?’
b. Ne, (*to) forao.
yes it wear.1s
‘Yes, I am.’ or ‘Yes, I do.’
243
Quantificational noun phrases can license pronominal anaphora, of course. But
VSE, like VPE in English, gives rise to a second quantificational set, as seen in the
following example:
As can be seen from the two translations in (55b), there are two possible read-
ings to the Greek. In the first, the anaphoric reading, the neuter plural definite
anaphoric pronoun ta refers to the set of books introduced in the question; this
reading is possible only if the indefinites can be read with specific reference, that
is, with the first six of the collapsed examples, and not with the last two (no spe-
cific readings are possible with tipota ‘any’ or the bare plural): the ‘%’ diacritic
means that ta is licit with these first six antecedents, and not with the last two. In
the second possible reading, when the ta is omitted, we have a quantificational
reading: the indefinite inside the ellipsis site is understood with its own quan-
tificational force, and there is no commitment on the part of the answerer to the
set they answer about to be extensionally identical to any set the questioner may
have had in mind—only the cardinality is at stake.
Finally, there are many VP idioms that consist of a verb with its object and
which do not allow an anaphoric pronoun (since there is nothing to be anaphoric
to, on the idiomatic reading). Nevertheless, such idioms allow their object to
omitted;6 the inclusion of the pronoun makes the literal reading (eating wood in
6 The judgments here are somewhat variable across speakers, with some speakers finding all of
244
(56), for example) the only one available, to some amusement of my Greek con-
sultants).
(56) a. To pedhi tha fai ksilo, ke o Kostas episis tha (#to) fai ki aftos!
the kid fut eats wood and the Kostas also fut it eats and he
‘The kid will get hit, and Kostas will, too!’
b. To pedhi tha fai ksilo, ala o Kostas dhen tha (#to) fai.
the kid fut eats wood but the Kostas not fut it eats
‘The kid will get hit, but Kostas won’t.’
The following examples further illustrate the same point, using a wider variety
of Greek VP idioms.
(57) O Dimitris kani tin papia; mono i Ariadne dhen (#tin) kani.
the Dimitris makes the duck only the Ariadne not it makes
‘Dimitris is playing dumb; only Ariadne isn’t.’
(58) a. O Petros efige ke erikse mavri petra piso tu.
the Petros left and threw black stones behind him
‘Petros left and will never return.’
b. Ke i Maria erikse.
and the Maria threw
‘And Maria also will never go back.’
(59) I Elines politiki tazun lagus me petraxilia, ala i Amerikani
the Greek politicians vow rabbits with priests’ habits but the American
politiki pote dhen tazun.
politicians never not vow
‘Greek politicians promise the moon, but American politicians never do.’
(60) O nearos ekane kamaki se mia jineka. Afti tu ipe na figi. Otan
the young.man made advance to a woman she him told to leave when
ksanaekane, ton evrise.
again.made him cursed
‘The young man hit on a woman. She told him to leave. When he hit on
her again, she yelled at him.’
(61) O Janis ke i Maria ithelan na dhosun logo, ala i gonis tus
the Giannis and the Maria wanted to give.3pl word but the parents theirs
dhen ithelan na dhosun.
not wanted to give.3p
‘Giannis and Maria wanted to get engaged, but their parents didn’t want
them to.’
Finally, Greek has particle-verb-like combinations that involve a light verb and
an adverbial particle. These may not be entirely like idioms, since their mean-
ings may be computable from the regular contributions of the pieces, but their
these permit the idiomatic readings, and some more conservative, who accept only a literal reading.
245
behavior is not easily explicable if Greek lacks VSE. One such is perno piso, lit-
erally ‘take back’, meaning take back or get back. As indicated in (62), this particle
can occur anywhere in the clause, even preverbally, though there its placement is
presumably due to focus movement, and it cannot be used out-of-the-blue (and
which I omit for that reason). This verb+particle combination appears with a
direct object and a source PP.
(62) Pire {piso} i Ana {piso} xrimata {piso} apo tin trapeza {piso}?
took.3s back the Ana back money back from the bank back
‘Did Anna get money back from the bank?’
Note that VSE is licit with just the verb remaining. The particle does not, and
cannot, survive VSE.7 This is entirely expected if the verb has moved to T, and if
the arguments and particle must remain inside the boxed elided VP:
(64) TP
T vP
v
V
piso i Ana xrimata apo tin trapeza
pire tv
back the Ana money from the bank
took.3s
All of these data indicate that Greek is not merely dropping pronominal or in-
definite arguments; the data are only consistent with a derivation by VSE.
246
Landau (2017) argues that in Hebrew, there are object gap examples where
the verb stays in situ, and which therefore cannot be due to VSE (but rather are
due to argument ellipsis). His evidence that the verb remains in its base position
comes from the fact that the verb can occur to the right of the lowest adverbial on
Cinque’s hierarchy, namely the frequentative often, and co-occur with completive
completely. Landau takes this to mean that the verb has not raised to T, but rather
has stayed in situ; such a position would rule out a VSE analysis, and yet sloppy
identity in the missing object can still be understood.
Such examples can be produced in Greek as well (modeled on Landau 2017
(34b)):
(65) O Nikos mia fora ksirise to kefali tu en meri afu akuse oti o
the Nikos one time shaved the head his in part because heard.3s that the
Petros sixna ksirizi endelos.
Petros often shaves completely
‘Nikos once shaved his head partially because he had heard that Petros
often shaves his head completely.’
A similar point can be made on the basis of low participles. Such participles do
not move to T (the finite auxiliary verb does), and if such movement were re-
quired to license VSE, then these examples would show that VSE is not available.
(66) To agori exi fai ksilo; to koritsi dhen exi fai.
the boy has eaten wood the girl not has eaten
‘The boy got smacked; the girl didn’t.’
(67) a. Tin exo grameni sta palia mu ta paputsia.
her I.have written on.the old my the shoes
‘I won’t have anything to do with her.’
b. Ki ego tin exo!
and I her have
‘I won’t either!’
Fortunately for the argument in favor of the existence of VSE in Greek, there is
reason to believe that even participles move out of their vP. As seen above in (7ab),
repeated here, participles can appear to the left of relatively ‘high’ adverbs, such
as epitidhes ‘intentionally’ (see Alexiadou 1997).
(68) Itan safes oti to pedhi ixe idhi kapsi epitidhes ti
it.was clear that the child had.3s already burned.participle intentionally the
supa.
soup.acc
‘It was clear that the child had already intentionally burned the soup.’
The evidence from the placement of adverbs in (65) rests on a supposition that
adverbs like often cannot be adjoined higher in the extended projection of the
VP, which I know of no reason to believe is true in Greek.
We can conclude that if the verb raises to at least the lowest Aspect head, then
ellipsis could target VoiceP or vP beneath Aspect.
247
4.2 Extraction
As discussed in Merchant (2013a, 2016), one of the most important and persuasive
diagnostics for ellipsis of syntactic material comes from movement dependen-
cies (see especially the seminal discussion in Hankamer & Sag 1976 and Sag &
Hankamer 1984). Selection is local to particular heads. Thus, when we observe
a selectional relationship that appears to hold between a displaced phrase and
something inside an ellipsis site, we conclude that the ellipsis site contains a head
with the relevant selectional ability or feature. In this respect, ellipses such as VP-
ellipsis in English differ from otherwise interpretationally similar constructions
such as Null Complement Anaphora.
VP-ellipsis after to allows for the extraction of the object of the missing verb:
(69) VP-ellipsis:
a. We need to know which films Anna refused to review, and which ones
she agreed to.
b. We need to know which films Anna agreed to review, and which ones
she refused to.
(70)
which films
she
refused
to VP
review t
The same verbs, when used in their Null Complement Anaphora guises, fail to
license the extraction of the object of an understood complement predicate:
248
We therefore conclude that there is active syntactic structure (licensing ex-
traction, agreement, and other syntactically mediated dependencies) inside ellip-
sis sites. There is no evidence for such structure inside the understood argument
in Null Complement Anaphora. By Ockham’s razor, we suppose that the simplest
explanation of this fact is the absence of such structure.
The Greek case is similar, with the difference that verb-raising occurs before
the ellipsis of the VP: movement of a verb out of an ellipsis site can be combined
with movement of a phrase from within the elided phrase as well.
As in (4b) above, it is crucial that what is extracted in (72) is an l-selected PP, here
ja to deftero ‘about the second one’. The preposition ja ‘for, about’ is selected by
the predicate milao ‘speak’: it is not in any conceivable way an argument of the
matrix verb thelo ‘want’. Yet it appears in the second clause, fronted. This is only
consistent, given any restrictive theory of l-selection, with the PP having been, at
some stage of the derivation, a complement to the head of milao. We can therefore
securely conclude that the second clause contains a missing VP.
The same point is made by the following questions:
249
(74)
me pjon
Pol TP
T Pol i Ana
tT VP
V T
ithele tV
na
milisi tPP
Similar remarks hold for the following examples, which demonstrate PP ques-
tions, PP relatives, case-marked left-dislocated topicalization, and selected PP
left-dislocations:
250
It is important to remember that care must be taken when designing such stim-
uli; for many years, extraction from VP-ellipsis sites in English was thought to
be almost uniformly ungrammatical, except for antecedent-contained deletions.
In fact, such extraction is very sensitive to additional parallelism requirements
(see Schuyler 2001, Merchant 2008) and other factors that are poorly understood;
this is illustrated by (79). Some of the factors that lead to degradation in English
also give rise to similar effects in Greek. For example, though Greek allows ex-
traction of genitive DP possessors from definite DPs (Horrocks & Stavrou 1987),
such extraction is highly degraded in a VSE context:
Compare the ill-formedness of the English as well (as noted for similar examples
in Sag (1976)):
251
l-selection except through the highly implausible suggestion that the purely se-
mantic incorporation of the PP is mediated in some way by the matching prepo-
sition. To my knowledge, no working mechanism with these properties has been
proposed, and in my estimation, doing so would mean making purely idiosyn-
cratic, lexical selectional information available to the semantics. This move has
the same prospects for success as making the height of the vowels in a verb stem
available to the syntactic computation for consideration in triggering verb move-
ment. The syntactic ontology consists of features that determine part of speech
and selectional information, among other things; the semantic ontology contains
things like entities, properties, eventualities, times, etc., but not nouns and verbs.
Needing to conflate the two is the sign of a theory in distress.
252
(85) a. Arnold lost his life in the war, but before he lost it, he had written a
letter to his mother.
b. Arnold lost his life in the war, and #Bernard lost it, too.
Precisely this pattern holds in Greek as well. An acceptable example such as (87)
must in fact be produced with a heavy contrastive focus (realized as a rise-fall
pitch contour; see Arvaniti et al. 2006) on the verb in the question. By adding this
focus, the speaker is explicitly raising the possibility of other verbs being part
of the true answer to the implicit polar question (as well as the sentence with the
given verb being false). This, of course, is just the very nature of contrastive focus
on any element in a question.
If this strong focus is absent, either as contrastive focus or as verum focus, as is the
case in a neutral polar question such as (88), the response with VSE is ill-formed;
in that case, an overt object is required:
(88) Vrike o Petros ena vivlio sti vivliothiki? Oxi. Exase #(ena).
found.3s the Petros a book in.the library no lost.3s one
‘Did Petros find a book in the library? No, he lost one/it there.’
253
Data showing this result were already given in Giannakidou & Merchant (1997),
though the conclusion drawn there was different. Mismatched verbs are possible
as in the following example, where the questioner puts an implicit contrastive fo-
cus on the verb, seeing the addressee with bread and thinking that the addressee
may have stolen it (as opposed to buying it or baking it):
The examples are judged perfect if the same speaker is responsible for both verbs,
because in this case, the speaker can decide beforehand that the verbs will con-
trast, and mark them both accordingly:8
(91) a. Abby tends [ t Abby to work too hard], and Ben does tend [ t Ben to work
too hard], too.
8 It is worth noting that the acceptable examples of verbs differing in VSE in Russian from their an-
tecedent presented in Gribanova (2013b):119 (65)-(66) involve a single speaker, while the unacceptable
Hebrew examples from Goldberg (2005) involve different speakers. Perhaps the requisite contrast
focus is difficult to project back onto a previous utterance from which it was absent.
254
b. We need to know how many people Abby thinks we should
invite thow many people , and how many Ben does think we should invite
thow many .
This follows on any theory of ellipsis resolution that allows for traces to be in-
terpreted as variables, and under which the index of a bound variable does not
matter for the purposes of this computation, such as the LF-identity theory of
Sag (1976), the semantic identity theory of Merchant (2001), or many others.
Note that Gribanova’s claim is not the same as claiming that all head move-
ment is ‘at’ PF, or that head movement leaves no trace, as Messick & Thoms (2016)
do, expanding on Lasnik’s (2003) claim that A-movement leaves no trace. The
idea that head movement leaves no trace was appealing as part of an account of
the Warner facts (Warner 1985), along with the putative constraint in (92) proposed
in Thoms (2015):
Unfortunately, (92) cannot be sustained in the face of examples like the follow-
ing, involving head movement (V2 in Dutch), A′-movement, and A-movement,
respectively (and see the works cited for many more such examples).
(93) [CP Nu gaat [TP zij tnu tgaat ]], maar ik weet niet waarom. Dutch
now goes she but I know not why
‘She’s going now, but I don’t know why.’ (Merchant 2001:21)
a. , *... waarom zij.
b. = ... waarom zij nu gaat.
(94) a. The FBI knows which truck4 they rented t4 , but figuring out from
where they rented it4 has proven difficult. (Merchant 2001:206)
b. This is Washington, where everyone keeps track of who1 t1 crossed
whom2 and when they1 crossed them2 . (Merchant 2001:202)
(95) These facts should be carefully studied, but it’s clear you haven’t carefully
studied these facts. (Merchant 2013c)
This state of affairs is fortunate, given that any claim that A-movement fails to
leave a trace or a copy would leave us in the lurch for understanding passive
of intensional transitives, and reconstructed scope under modals, negation, and
quantificational adverbs, all of which indicate that for semantic reasons, the DP
behaves as though it were in its base position (see Erlewine (2014) for extensive
discussion of the mechanisms of reconstruction):
255
These kinds of predicates can license VP-ellipsis as well, in two relevant varieties.
In the first, in (97), the A-moved antecedent DP of the passive is understood as
taking narrow scope, inside the VP that hosts its origin site (see Bruening (2013)
for a recent defense of the movement approach to the passive), and the VP-ellipsis
involves an active verb. The VP-ellipsis is interpreted as though the indefinite
were inside the elided VP, taking narrow scope with respect to the intensional
verb, modal, negation, or adverb of quantification.
(97) a. A miracle would be needed, and if you do need a miracle then God
help you.
b. Usually, raspberries were easily found on those hikes, but we didn’t
manage to easily find raspberries that particular day.
In the second variety of example showing that A-movement can reconstruct in-
side ellipsis sites, both the antecedent VP and the elided VP involve A-movement
(here, passives, though similar examples can be generated with raising predicates
and intensional adjectives):
(98) a. A unicorn was hoped for, and a dragon was hoped for, too.
b. Raspberries were often/easily found, and strawberries were often/easily
found as well.
c. Raspberries will be easily found, and strawberries will be easily found
as well.
d. A helmet will usually be found in such a grave site, as will a shield
usually be found in such a grave site.
e. A kore wasn’t often stationed in such a temple; a kouros wasn’t often
stationed in such a temple, either.
f. A shield was never made from gold, nor was a sword ever made from
gold.
These examples are important for another reason as well. They clearly demon-
strate that the theory of ellipsis proposed in Heim (1997) is wrong.
Heim assumes a theory of ellipsis resolution that has three ingredients:
Heim shows that these conditions, properly applied, can account for a range of
data from Kennedy (1994) and additional data that she adduces. But, as she ad-
mits, “There would be a problem if the subjects were maximally reconstructed”
(Heim 1997:12). In her discussion of (100a) (her (31)), Heim points out that a fully
256
reconstructed subject, as in (100b), would violate the lexical identity condition.
Instead, she proposes (100c) as the LF, with focus-marked second subject, Mary,
having moved out of the VP and interpreted outside of the VP.
Such a focus-marked subject satisfies the focus-condition, which states the that
the focus-marked element must be contained in a phrase that contrasts appropri-
ately with another phrase. This condition is dubbed the ‘containment’ condition
in Merchant (2001), where the details of Rooth’s proposal are spelled out. Here, I
repeat Heim’s slight restatement of Rooth, given in (101):
The regular semantic value of the antecedent clause in (100) in Heim’s system is
just John called (from Johnx PAST [VP x call]). The focus value of the clause con-
taining the ellipsis and the focus-marked binder of the variable inside the elided
VP is computed from the LF [[MaryF ] y did [VP y call]] and is {that x called: x ∈ D}.
Since John called contains no variables, it is not sensitive to g, and since John called
∈ {that x called: x ∈ D}, and doesn’t overlap with it, the containment condition is
satisfied, and ellipsis is licit.
Heim was right that her system only works if DPs A-moved out of an elided
VP do not have to reconstruct. But unfortunately for her system, and for recent
attempts to revive it, the examples in (98) are precisely the kind of data that are
impossible to accommodate. In (98), the subjects must be maximally (that is, both
the restrictor and the quantificational determiner) reconstructed (or at any rate,
reconstructed to a position inside the VP which is the target for ellipsis, which
comes to the same thing for the purposes of the problem for Heim’s account).
And so the examples show that Heim’s theory fails.
To see in detail why this is, consider first the LF of the passive of the inten-
sional transitive in (98a):
Employing the proposal for the semantics of the passive in Bruening (2013), and
ignoring tense, we have:
257
The crucial point is that the existential force of the indefinite article a can (and in
fact preferentially does) take narrow scope with respect to the intensional quan-
tification. (The descriptive content of unicorn and dragon can in fact be anchored
to the actual world, but this is orthogonal to the question at hand.)
The attested interaction of indefinites with modals, negation, and adverbs
also are fatal for Heim’s proposal.9 Consider the LF for (98e):
Here, negation scopes over the adverb of quantification often, adjoined to VP and
internal to the ellipsis site. Often, in turn, can outscope the contrasting indefi-
nite derived subjects a kore and a kouros. On the most plausible readings of these
sentences, which involve many different statues of young women and men, the
indefinites must totally reconstruct to a position inside the VP, under often:
Note that the problems here are not resolvable by mere reformulation of the con-
ditions, or by retreating from Heim’s conclusion that the VPs denote formulas.
The solution is that we need to allow focus alternatives to be computed for focus-
marked material internal to the ellipsis site. Heim, by stipulation, rules out any
F-marking inside the ellipsis site. This move is wrong. What is true is that there
can be no pitch-accent inside an ellipsis site (since there is no phonological ma-
terial to bear it), so constructions that conspire to require such a pitch accent
(such as the fact that a focus-sensitive operator like only requires a pitch accent
on its associate, as Tancredi 1992 discovered, and Erlewine 2014 discusses) will be
ill-formed. But F-marking per se inside an ellipsis site at LF is fine, as long as the
pitch accent associated with the F-marked material is outside the ellipsis site at
PF. This is the kind of system that I proposed in 2001 (Merchant 2001), building on
Schwarzschild’s givenness system. In that work, I proposed that ellipsis was li-
censed just in case the elided XP and its antecedent were semantically equivalent
to one another modulo F-marking.10
9 See also Jacobson (1998) and Kennedy (2014) for discussion of additional examples that cannot be
Schwarzschild’s type-raising everything to propositional type and comparing entailments. The prob-
lem comes from reversible predicates such as defeat∼lose to or be an older sibling of∼be a younger sibling
of:
(i) a. Abby defeated Ben ↔ Ben lost to Abby.
b. Ben is an older sibling of Abby ↔ Abby is a younger sibling of Ben.
Nevertheless, these predicates don’t license ellipsis of their reversed counterparts:
(ii) a. *Abby defeated Ben, so we know that Ben did lose to Abby.
b. *Ben is an older sibling of Abby, so we can conclude that Abby is a younger sibling
of Ben.
Hartman suggested retreating to an LF-identity condition, but we can simply use the type-flexible
system of Rooth and get the desired result, replacing mutual entailment by semantic equivalence
modulo focus (see (107) below); free variables can be bound by λ-operators for the purposes of the
computation. So (ii.a) will be ruled out because λxλ y [defeat( y )( x )] , λ yλx [lose.to( x )( y )].
258
This point was made most clearly with examples I dubbed ‘contrast sluicing’
in Merchant (2001):150, such as (106), where the quantificational force has to be
calculated within the scope of the modal, and where the restrictors on the quan-
tifiers contrast:
(106) a. There may be nine womenF in the play, but I don’t know how many
menF .
b. ⋄∃x[women( x) ∧ | x | ≥ 9] . . . ?n ⋄ ∃ y [men( y ) ∧ | y | ≥ n]
It is precisely by virtue of this focus that the ellipsis can go through: because both
women and men are focussed, we look at their alternatives when calculating ellip-
tical identity—the ordinary semantic value of the antecedent clause is an element
of the focus-semantic value of the elliptical clause, and vice versa. The problem
is that in these cases, the focused material must be inside the ellipsis site, which
violates Heim’s third clause of (99) (and any theory of purely LF identity, such as
Fiengo & May 1994).11
The technical changes needed to account for the full range of data are trivial12 :
replace the existential closure of free variables of Schwarzschild (1999) with λ-
closure, and replace the entailment condition with an inclusion condition:13
Once we have such a theory that allows us to abstract over focused elements, even
when these reconstruct, or are interpreted inside the ellipsis site, we no longer
need to say anything special about Greek verb movement (or Russian, etc.): the
11 But see especially Rudin (to appear) for a new take on sluicing licensing that differentiates it from
with reversible predicates, but also of handling the example I worried about in (i) (from Merchant
(2001):37 fn 17), where focus-closure and ∃-closure conspired to make VPA and VPE equivalent even
when the focus-marking was anaphoric to some other sentence, not to the one containing the ellipsis:
259
verb can move as usual (successive-cyclically, obeying the Head Movement Con-
straint or Relativized Minimality) and indeed can reconstruct totally, as long as
the verb (or its stem) is focused. This is precisely what seems to be the state of
affairs in Greek, as we’ve seen above.
The calculation of focus alternatives is as Rooth proposed: for a 2-place pred-
icate, the set of alternatives are those in De,et . The parallelism condition on ellipsis
is satisfied in case the ordinary value of the antecedent vP is an element of the fo-
cus value of the elided vP, and vice versa. For the Greek example in (90), repeated
here, this will hold if both statements in (110) are true.
(109) O Petros dhen vrikeF ena vivlio sti vivliothiki — exaseF .
the Petros not found.3s a book in.the library lost.3s
‘Petros didn’t find a book in the library — he lost one there.’
(110) a. Petros found a book in the library ∈ {that Petros P ’ed a book in the library:
P ∈ De,et }
b. Petros lost a book in the library ∈ {that Petros P ’ed a book in the library:
P ∈ De,et }
So Greek VSE is simply subject to the usual condition on VP-ellipsis (semantic
equivalence modulo focus), and verbs in Greek are just like any other moving
element: if focused, they can reconstruct (as predicates typically must, following
Heycock (1995)), but the focus marking on their stems will allow that part of their
meanings to vary (while other material is interpreted outside the vP in any case:
Voice, Aspect, Tense). There is no particular Verbal Identity Requirement at all.
Its effects fall out from focus-marking (or its lack, in certain cases).
260
(112) A: An ngéillfidh siad? B: Caithfidh siad.
q yield.fut they must they
‘Will they yield (on this)?’ ‘They have to.’
What is unusual about this and related examples displaying verum focus is that
the pitch accent falls not on the verb itself, but on the following subject pronoun.
Bennett, Elfner, and McCloskey analyze this unusual pitch placement as es-
sentially an ‘epiphenomenon of phrasing’ (p. 24), the result of the interaction of
constraints favoring rightward accent placement and a special subject pronoun
incorporation process. In any case, it is surely no accident that the most promi-
nent language in which the Verbal Identity Requirement seems to hold is also
the language that seems to have an allergy between focus prominence and ver-
bal stems.14 When no pronoun is available, as is the case in synthetic verb forms
(forms that inflect for person and number), the inflectional ending, not the stem,
takes the accent:
(113) An rabhadar ann? Bhíodar.
Q be.past.3pl in.it be.past.3pl
‘Were they present? They certainly were.’
Most spectacularly, as Bennett et al. note in their footnote 16: “In the absence of
a simple pronoun subject or an appropriate inflectional ending . . . , other means
have to be found to express Verum Focus. . . . the discourse particle muis(e), whose
meaning is, to say the least, unclear, may serve exactly this function in cases like
([114]):”
(114) A: An raibh Colm ann? B: Bhí muis.
Q be.past Colm there be.past particle
‘Was Colm there? He was indeed.’
The crucial empirical question is whether the addition of such a particle would
ameliorate even cases of mismatched verb stems, such as (111) above. If so, then
the problem with (111) may not be the lack of identity of the verb stems per se,
but rather the lack of an appropriate position for the accent to fall, given the
unusual requirements of Irish focal accent placement. The usual cases of VSE in
Irish simply don’t involve such accents, and so can surface as mere verbs, with
no following particle or subject, pronominal or otherwise. It is only in the cases
where the Verbal Identity Requirement is tested that such accent is obligatory,
and imposes these unusual additional requirements.
All of this, I hope, points to a possible solution that ties the appearance of the
Verbal Identity Requirement to something special about how focus is handled in
the grammar of Irish, as opposed to Greek and other languages.
14 Lipták (2012) shows that Hungarian VSE, which can strand either a verb or a phrasal verbal
marker, is also subject to a Verbal Identity Requirement except when the stranded verb and its an-
tecedent are both contrastively focused (see especially Lipták (2013):84 fn 13); a full investigation of the
interaction of focus and prosody and VSE in Hungarian will have to await a future occasion. Likewise
for Brazilian Portuguese, as investigated in Santos (2009), Cyrino & Lopes (2013), and Lopes & Santos
(2014).
261
5.2 Certain adjuncts
Landau (2017), building on Oku (1999), points out an important contrast between
VSE in Hebrew and English VP ellipsis: English speakers very much prefer (in
the absence of contrasting material) to interpret VP adjuncts as being part of an
elided VP, while Hebrew speakers do not normally take such adverbs to mod-
ify a putative stranded verb. (115a), for example, is preferentially interpreted as
‘Beth didn’t clean her flute carefully’, not as merely ‘Beth didn’t clean her flute’.
Likewise for the adverbs in the other cases as well.15
Greek allows such adverbs to be interpreted inside the ellipsis site as well:
When the adjunct can be taken as the sole scope of negation, the two readings can
be readily distinguished. Landau (2017) provides the following Hebrew example
(his (40a)), using the missing antecedent phenomenon (Grinder & Postal 1971) as
the crucial test to diagnose ellipsis. The infelicity of the following anaphora (hi)
shows not just that there was no ellipsis of a VP containing a DP antecedent for
hi to be anaphoric to, but that such an elided VP cannot be posited at all. (So
Hebrew has null definite objects, but not null adjuncts and no VSE at all.)
15 One difficulty with the argument from adverbs is that even adverbs that cannot possibly be inside
the antecedent VP seem to be able to be interpreted as though they were inside an elided VP, as in (i);
perhaps such adverbs are fronted from some position inside the VP.
(i) Abby must consistently have worn her retainer; her sister certainly did consistently wear her
retainer.
262
(117) Yosi afa et ha-uga lefi ha-matkon. hi hayta me’ula.
Yosi baked acc the-cake according the-recipe it was fabulous
‘Yosi baked the cake according to the recipe. It was fabulous.’
a. Gil lo afa __. #hi hayta mag’ila.
Gil not baked it was gross
‘Gil didn’t bake the cake. It was gross.’
b. Gil, lo. hi hayta mag’ila.
Gil not it was gross
‘Not Gil./Gil didn’t. It was gross.’
A similar example in Greek (given with the non-elliptical control in (118) as well)
shows the same pattern:
How can we reconcile these results with the evidence above that Greek does
have VSE? We must seek another reason why the continuation in (118a) is judged
deviant, while (118b,c) are not. That reason has already been hinted at above,
however: some focus-sensitive operators, most famously English only, trigger an
obligatory pitch accent on their associate. If this associate is elided while the op-
erator is not, the result is judged infelicitous.16 Compare the following examples
with and without VP ellipsis:
16 See Beaver & Clark (2008):ch. 7 for some discussion. The requirement is one that applied when
the dependency between the operator and the accent spans the boundary of an ellipsis site. If the
operator itself is also elided, no deviance results:
(i) Abby said she only plays [the flúte]F , and Ben did, too. (say she only plays [the flúte]F )
This is presumably because the requirement is one of actual pitch accent, which secondary occur-
rence focus does not have: secondary occurrence focus shows prominence only through length and
intensity, not pitch movement; see Baumann (2016).
263
(119) Abby will only play [the flúte]F at the recital, not the piano.
a. Ben also will only play [the flúte]F at the recital.
b. *Ben also will only play [the flúte]F at the recital.
As discussed above, Heim mistakenly took such data to mean that F-marking
could not be present inside an ellipsis site. As I have shown, that is incorrect. It
is the requirement that pronounced only be associated with a pitch-accent on its
associate that makes (119b) ill-formed, not the F-marking per se.
And precisely such a requirement holds of Greek dhen as well (but not of
constituent negation oxi, used in the negative stripping example in (118b)). A more
accurate representation of the focus marking of (118c) (similar to the facts studied
in Johnson (1994)) would be as in (120), which makes clear why eliding a phrase
that properly contains the adjunct would be impossible: the pitch accent required
by dhen (falling on the final syllable of the adjunct, jí) could not be realized. There
is no way to reduce or elide any phrase containing the adjunct akoluthondas tin
sintají.
We find the same results when we ensure that ellipsis is present by extracting
from the ellipsis site. Since the PP ja ton baba tu ‘for his father’ is licensed by the
elided embedded predicate ftiaksi ‘make’, not by the matrix predicate borese ‘was
able’, we know that VSE has occurred. Nevertheless, the attempted anaphora is
illicit.
(121) Jan tin mama tu, o Markos borese na ftiaski turta akoluthondas tin
for the mother his the Markos was.able subj make cake following the
sintaji. (Itan nostimi.) Ja ton baba tu, dhen borese. (#Itan aidhiastiki.)
recipe it.was delicious for the father his not was.able it.was disgusting
‘For his mother, Markos was able to make a cake following the recipe. For
his father, he wasn’t. (able to make a cake following the recipe)’
In this case, the pitch accent falls on the negator dhen in the last, contrasting sen-
tence. This stress has the effect of placing the emphasis on the truth of the ut-
terance; it is a kind of a verum focus (or falsum focus, in this case). There is a
remaining, larger question why this negation, and the constituent negator oxi
used in the negative stripping in (118b) above, cannot give rise to a reading that
would make these sentences in effect equivalent to the narrow focus on the ad-
junct (since, of course, one way of ensuring falsity of the whole is to deny the
applicability of the adjunct), but that is a question whose resolution raises ques-
tions beyond the scope of this paper. For our purposes, it is enough to note that
such readings are unavailable with non-elliptical falsum focus sentences.
264
6 Conclusion
Greek has verb-stranding ellipsis, like Irish. Narrow focus on the verb stem can
be used to vary the verb between the antecedent and the elided vP, in line with
other elements that can move out of ellipsis sites (but still be wholly or in part
interpreted inside them), because the ellipsis resolution condition is sensitive to
focus alternatives, not to LF structure per se.
References
Alexiadou, Artemis. 1997. Adverb placement: A case study in antisymmetric syntax.
Amsterdam: John Benjamins.
Alexiadou, Artemis & Elena Anagnostopoulou. 1998. Parametrizing Agr: word
order, V-movement and EPP-checking. Natural Language and Linguistic Theory
16(3). 491–539.
Alexiadou, Artemis, Elena Anagnostopoulou & Florian Schäfer. 2015. External
arguments in transitivity alternations: A layering approach. Oxford: Oxford Uni-
versity Press.
Arvaniti, Amalia, D. Robert Ladd & Ineke Mennen. 2006. Tonal association and
tonal alignment: Evidence from Greek polar questions and contrastive state-
ments. Language and Speech 49(4). 421–450.
Baumann, Stefan. 2016. Second occurrence focus. In Caroline Féry & Shinichiro
Ishihara (eds.), The Oxford handbook of information structure, Oxford: Oxford
University Press.
Beaver, David & Brady Clark. 2008. Sense and sensitivity: How focus determines
meaning. Chichester: Wiley-Blackwell.
Bennett, Ryan, Emily Elfner & James McCloskey. 2017. Prosody, focus, and ellipsis
in Irish. Ms., University of California, Santa Cruz.
Bowers, John. 1993. The syntax of predication. Linguistic Inquiry 24(4). 591–656.
Bowers, John. 2002. Transitivity. Linguistic Inquiry 33(2). 183–234.
Bresnan, Joan. 2001. Lexical-functional syntax. London: Blackwell.
Bruening, Benjamin. 2013. By phrases in passives and nominals. Syntax 16(1). 1–41.
Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, Massachusetts:
MIT Press.
Chung, Sandra. 2000. Close encounters with pronouns in VP ellipsis. In Sandra
Chung, James McCloskey & Nathan Sanders (eds.), Jorge Hankamer WebFest,
http://babel.ucsc.edu/Jorge/chung.html.
Chung, Sandra. 2006. Sluicing and the lexicon: The point of no return. In Re-
becca T. Cover & Yuni Kim (eds.), Proceedings of the annual meeting of the Berkeley
Linguistics Society 31, 73–91. Berkeley, Calif.: Berkeley Linguistics Society.
Chung, Sandra. 2013. Syntactic identity in sluicing: How much, and why. Lin-
guistic Inquiry 44(1). 1–39.
Chung, Sandra, William A. Ladusaw & James McCloskey. 2010. Sluicing(:) Be-
tween structure and inference. In Line Mikkelsen, Eric Potsdam & Rodrigo
265
Gutiérrez-Bravo (eds.), Representing language: Essays in honor of Judith Aissen,
Santa Cruz, Calif.: Linguistics Research Center Publications.
Culicover, Peter W. & Ray Jackendoff. 2005. Simpler Syntax. Oxford: Oxford
University Press.
Cyrino, Sonia & Ruth E. Vasconcellos Lopes. 2013. VP ellipsis, null objects, and
aspect as a licensing category in Brazilian Portuguese. In 43rd linguistic sympo-
sium on romance languages, 21–22.
Dowty, David. 1981. Quantification and the lexicon: A reply to Fodor and Fodor.
In M. Moortgat, M. Hulst & H.v.d. Hoekstra (eds.), The scope of lexical rules,
Dordrecht: Foris.
Erlewine, Michael Yoshitaka. 2014. Movement out of focus. Cambridge, Mass.:
Massachusetts Institute of Technology dissertation.
Fiengo, Robert & Robert May. 1994. Indices and identity. Cambridge, Mas-
sachusetts: MIT Press.
Fillmore, Charles. 1986. Pragmatically controlled zero anaphora. In Proceedings
of Berkeley Linguistics Society 12, 95–107. Berkeley, Calif.: Berkeley Linguistics
Society.
Fodor, Jerry A. & Janet Dean Fodor. 1980. Functional structure, quantifiers, and
meaning postulates. Linguistic Inquiry 11(4). 759–770.
Funakoshi, Kenshi. 2012. On headless XP-movement/ellipsis. Linguistic Inquiry
43. 519–562.
Giannakidou, Anastasia. 2000. Negative...concord? Natural Language and Lin-
guistic Theory 18(3). 457–523.
Giannakidou, Anastasia & Jason Merchant. 1997. On the interpretation of null
indefinite objects in Greek. Studies in Greek Linguistics 18. 141–154.
Gillon, Brendan. 2012. Implicit complements: A dilemma for model theoretic
semantics. Linguistics and Philosophy 35(4). 313–359.
Goldberg, Lotus Madelyn. 2005. Verb-stranding VP ellipsis: A cross-linguistic study.
Montreal: McGill University dissertation.
Gribanova, Vera. 2013a. A new argument for verb-stranding verb phrase ellipsis.
Linguistic Inquiry 44(1). 145–157.
Gribanova, Vera. 2013b. Verb-stranding verb phrase ellipsis and the structure of
the Russian verbal complex. Natural Language and Linguistic Theory 31. 91–136.
Gribanova, Vera. 2017a. Head movement and ellipsis in the expression of Russian
polarity focus. Natural Language and Linguistic Theory 00. 1–43.
Gribanova, Vera. 2017b. Head movement, ellipsis, and identity. Ms., Stanford
University.
Gribanova, Vera. 2017c. Roots in ellipsis and multidominance. In Jason Ostrove,
Ruth Kramer & Joseph Sabbagh (eds.), Asking the right questions: Essays in honor
of Sandra Chung, 1–16. Santa Cruz, Calif.: Linguistics Research Center Publica-
tions.
Gribanova, Vera & Line Mikkelsen. 2018. On the interaction of head movement
and ellipsis in Danish. In Deniz Rudin, Kelsey Sasaki, Line Mikkelsen & Ja-
son Merchant (eds.), A reasonable way to proceed: A Festschrift in honor of James
McCloskey, Santa Cruz, Calif.: University of California, Santa Cruz.
266
Grinder, John & Paul M. Postal. 1971. Missing antecedents. Linguistic Inquiry 2(3).
269–312.
Hankamer, Jorge & Ivan A. Sag. 1976. Deep and surface anaphora. Linguistic
Inquiry 7(3). 391–428.
Harizanov, Boris & Vera Gribanova. 2017. Whither head movement? Ms., Stan-
ford University.
Heim, Irene. 1997. Predicates or formulas? evidence from ellipsis. In Aaron Law-
son & Enn Cho (eds.), Proceedings of salt vii, Cornell University: CLC Publica-
tions.
Heycock, Caroline. 1995. Asymmetries in reconstruction. Linguistic Inquiry 26(4).
547–570.
Horrocks, Geoffrey & Melita Stavrou. 1987. Bounding theory and Greek syntax:
Evidence for wh-movement in NP. Journal of Linguistics 23. 79–108.
Jacobson, Pauline. 1998. ACE and pied-piping: Evidence for a variable-free se-
mantics. In Devon Strolovitch & Aaron Lawson (eds.), Proceedings of SALT 8.
Jacobson, Pauline. 2016. The short answer: Implications for direct composition-
ality (and vice versa). Language 92(2). 331–375.
Johnson, Michael. 1994. The semantics of adverbial adjuncts. Santa Cruz, Calif.:
University of California, Santa Cruz dissertation.
Kennedy, Christopher. 1994. Argument contained ellipsis. Linguistics Research
Center Report LRC-94-03 University of California, Santa Cruz.
Kennedy, Christopher. 2014. Predicates and formulas: Evidence from ellipsis. In
Luka Crnic & Uli Sauerland (eds.), The art and craft of semantics: A festschrift for
Irene Heim, vol. 70, 253–277. Cambridge, Mass.: MITWPL.
Landau, Idan. 2017. Missing objects in hebrew: Argument ellipsis, not vp ellipsis.
Ms., Ben Gurion University.
Lasnik, Howard. 2003. On a scope reconstruction paradox. Minimalist Investiga-
tions in Linguistic Theory 118–124.
Levin, Beth & Malka Rappaport. 1988. What to do with θ -roles. In Wendy K.
Wilkins (ed.), Syntax and semantics 21: Thematic relations, San Diego, California:
Academic Press.
Lipták, Anikó. 2012. Verb-stranding ellipsis and verbal identity: The role of po-
larity focus. Linguistics in the Netherlands 29(1). 82–96.
Lipták, Anikó. 2013. The syntax of emphatic positive polarity in Hungarian: Evi-
dence from ellipsis. Lingua 128. 72–92.
Lipták, Anikó & Andrés L. Saab. 2014. No N-raising out of NPs in Spanish: ellipsis
as a diagnostic of head movement. Natural Language and Linguistic Theory 32(4).
1247–1271.
Lopes, Ruth E. Vasconcellos & Ana Lúcia Santos. 2014. VP-ellipsis comprehension
in European and Brazilian Portuguese. In João Costa, Alexandra Fiéis, M. João
Freitas, Maria Lobo & Ana Lúcia Santos (eds.), New directions in the acquisition
of romance languages, 181–201. Newcastle upon Tyne: Cambridge Scholars Pub-
lishing.
Manetta, Emily. 2018. We say “how high?”: Adverbs, negation and verb movement
in a verb-final language. In Deniz Rudin, Kelsey Sasaki, Line Mikkelsen &
267
Jason Merchant (eds.), A reasonable way to proceed: A Festschrift in honor of James
McCloskey, Santa Cruz, Calif.: University of California, Santa Cruz.
McCloskey, James. 1991. Clause structure, ellipsis and proper government in Irish.
Lingua 85. 259–302.
McCloskey, James. 1996. On the scope of verb movement in Irish. Natural Lan-
guage and Linguistic Theory 14(1). 47–104.
McCloskey, James. 2017. Ellipsis, polarity, and the cartography of verb-initial
orders in Irish. Ms., University of California, Santa Cruz.
Merchant, Jason. 2001. The syntax of silence: Sluicing, islands, and the theory of
ellipsis. Oxford: Oxford University Press.
Merchant, Jason. 2008. Variable island repair under ellipsis. In Kyle Johnson (ed.),
Topics in ellipsis, 132–153. Cambridge: Cambridge University Press.
Merchant, Jason. 2013a. Diagnosing ellipsis. In Lisa Lai-Shen Cheng & Norbert
Corver (eds.), Diagnosing syntax, 537–542. Oxford: Oxford University Press.
Merchant, Jason. 2013b. Polarity items under ellipsis. In Lisa Lai-Shen Cheng &
Norbert Corver (eds.), Diagnosing syntax, 441–462. Oxford: Oxford University
Press.
Merchant, Jason. 2013c. Voice and ellipsis. Linguistic Inquiry 44(1). 77–108.
Merchant, Jason. 2014. Gender mismatches under nominal ellipsis. Lingua 151.
9–32.
Merchant, Jason. 2015. How much context is enough? Two cases of span-
conditioned stem allomorphy. Linguistic Inquiry 46(2). 273–304.
Merchant, Jason. 2016. Ellipsis: A survey of analytical approaches. To appear in
Jeroen van Craenenbroeck and Tanja Temmerman (eds.), The Oxford Hand-
book of Ellipsis.
Messick, Troy & Gary Thoms. 2016. Ellipsis, economy, and the (non)uniformity
of traces. Linguistic Inquiry 47(2). 306–332.
Mittwoch, Anita. 1982. On the difference between eating and eating something:
activites versus accomplishments. Linguistic Inquiry 13(1). 113–122.
Oku, Satoshi. 1999. LF copy analysis of Japanese null arguments. In M. Catherine
Gruber, Derrick Higgins, Kenneth S. Olson & Tamra Wysocki (eds.), Proceed-
ings of cls 34, 299–314. Chicago, Illinois: Chicago Linguistics Society.
Pollard, Carl Jesse & Ivan A. Sag. 1994. Head-driven phrase structure grammar. Stan-
ford: Center for the Study of Language and Information ;.
Rooth, Mats. 1992. Ellipsis redundancy and reduction redundancy. In Steve
Berman & Arild Hestvik (eds.), Proceedings of the stuttgart ellipsis workshop,
Stuttgart.
Rudin, Deniz. to appear. Head-based syntactic identity in sluicing. Linguistic
Inquiry .
Saab, Andrés. 2016. Ineffable narratives in Spanish: Another case of overgenera-
tion by e-GIVENness. Probus 28(2). 367–389.
Saab, Andrés L. 2009. Hacía una teoría de la identidad parcial en la elipsis: University
of Buenos Aires dissertation.
Sadock, Jerrold M. 1991. Autolexical syntax: A theory of parallel grammatical repre-
sentations. Chicago, Ill.: University of Chicago Press.
268
Sag, Ivan A. 1976. Deletion and logical form. Cambridge, Massachusetts: Mas-
sachusetts Institute of Technology dissertation.
Sag, Ivan A. & Jorge Hankamer. 1984. Toward a theory of anaphoric processing.
Linguistics and Philosophy 7. 325–345.
Sailor, Craig. To appear. The typology of head movement and ellipsis: A reply
to Liptak and Saab. Natural Language and Linguistic Theory Ms., University of
Groningen.
Santos, Ana Lúcia. 2009. Minimal answers: Ellipsis, syntax, and discourse in the
acquisition of European Portuguese. Amsterdam: John Benjamins.
Schuyler, Tamara. 2001. Wh-movement out of the site of VP ellipsis. In Séa-
mas Mac Bhloscaidh (ed.), Syntax at Santa Cruz, 1–20. UC, Santa Cruz: Lin-
guistics Department.
Schwarzschild, Roger. 1999. Givenness, avoid F and other constraints on the
placement of accent. Natural Language Semantics 7(2). 141–177.
Spyropoulos, Vassilios, Anthi Revithiadou & Phoevos Panagiotidis. 2015. Verbal-
izers leave marks: evidence from Greek. Morphology 25(299–325).
Tancredi, Christopher. 1992. Deletion, deaccenting and presupposition. Cambridge:
Massachusetts Institute of Technology dissertation.
Thomas, Jens, Chung-Pei Ma, Nicholas J. McConnell, Jenny E. Greene,
John P. Blakeslee & Ryan Janish. 2016. A 17-billion-solar-mass black
hole in a group galaxy with a diffuse core. Nature 532(7599). 340–342.
http://dx.doi.org/10.1038/nature17197.
Thoms, Gary. 2015. Syntactic identity, parallelism and accommodated an-
tecedents. Lingua 166. 172–198.
Warner, Anthony. 1985. The structure of English auxiliaries: A phrase structure gram-
mar. Bloomington, Indiana: Indiana University Linguistics Club.
269