Two Investigations of Adverbs and

Clause Structure in English

Erik Zyman
Senior Thesis (2011-2012)
Princeton University

Advisor: Prof. Edwin Williams

Second Reader: Prof. Robert Freidin
 Abstract

This thesis consists of two investigations. In the first, we test the predictions of the
Cinque hierarchy (Cinque 1999) in English by determining whether pairs of adverbs dis-
play the rigid ordering restrictions predicted by the hierarchy to hold universally. The re-
sult is that the Cinque hierarchy makes many incorrect predictions concerning English,
and therefore faces substantial difficulties in accounting for the facts of English adverb
ordering. It is shown that the counterexamples to the Cinque hierarchy cannot be due to
“direct attachment” of the adverbs that occur unexpectedly low, because all the Cinque
adverbs can directly attach to a wide variety of constituents, but not all of them can occur
low enough to disobey Cinquean adverb orders. (It is argued that, in the few cases where
an adverb seems unable to attach to constituents not lying on the clausal spine, this is not
because it is not a “direct attacher,” but because it must be licensed by an element that
only occurs in the extended projection of V.) The second investigation explores the se-
mantic, syntactic, and prosodic characteristics of English “interpretive escape” sentences,
in which an adverb apparently in an embedded clause is interpreted as modifying a higher
clause. An example is John considers Mary - ↑probably a murderer ‘John probably con-
siders Mary a murderer’. An analysis of “escaped” adverbs as parentheticals is argued to
incorrectly predict that adverbs should only be able to semantically escape to the matrix
clause, not to intermediate clauses. It is shown that interpretive escape can occur out of
small clauses, infinitivals, and finite clauses, and even across multiple clause boundaries.
This suggests that it is in principle unbounded, even though it becomes more effortful the
greater the distance it traverses. It can also affect all the Cinque adverbs, with two excep-
tions—almost and retrospective just—which can be explained by appealing to independ-
ently motivated lexical properties of these adverbs. Some tentative suggestions are made
as to what an analysis of interpretive escape may look like. My hope in this thesis is to
have brought to light: 1) a number of empirical difficulties that the Cinque hierarchy
faces in accounting for English, which it should ideally overcome in order to remain a
viable hypothesis; 2) evidence suggesting that perhaps all adverbs can directly attach to
nonspinal constituents; and 3) some characteristics of the phenomenon here dubbed “in-
terpretive escape.”

Keywords: Cinque hierarchy, adverb orders, adverb classes, clause structure, cartogra-
phy, direct attachment, interpretive escape, semantic displacement, prosody

2

 List of abbreviations used in examples

Adv adverb
ANT anterior tense
DO direct object
EPIST epistemic modality marker
EVID evidential
FUT future
id. idiomatic translation
lit. literal translation
NOM nominative
PASS passive
PERF perfect aspect
PM politeness marker
Q question marker
PROG progressive
PROSP prospective aspect
PST past
semilit. semiliteral translation

Typographical conventions used in example sentences

• Italics indicate the presence of a prominent stress and a High pitch accent.

• SMALL CAPS indicate the most prominent stress and High pitch accent within a par-
ticular prosodic domain.

• Boldfacing has no prosodic meaning, but rather serves to highlight elements of inter-
est in particular examples.

3

 Overview of contents

I. Introduction
A. General background
B. The Cinque hierarchy
1. The traditional hypothesis: adverbs as adjuncts
2. A problem for the traditional view: adverb ordering restrictions
3. Cinque (1999), part 1: From ordering restrictions and the logic of tran-
sitivity to a hierarchy of adverbs
4. Cinque (1999), part 2: Defending the AdvP-in-Spec Hypothesis
against its alternatives
a. The Extended Projection Hypothesis
b. The Multiple Specs Hypothesis
c. The AdvP-in-Spec Hypothesis
5. Cinque (1999), part 3: Matching the adverb hierarchy to a hierarchy of
functional heads
a. The hierarchy of functional heads
b. Matching the two hierarchies
6. Cinque’s classes of apparent counterexamples
7. A few alternatives to the Cinque hierarchy
8. Summary
C. The project
1. Method
2. A note on the data
II. Testing the predictions of the Cinque hierarchy in English
A. The data
1. Moodspeech act° (honestly) > Moodevaluative° (unfortunately)
2. Moodevaluative° (unfortunately) > Moodevidential° (allegedly)
3. Moodevidential° (allegedly) > Modepistemic° (probably)
4. Modepistemic° (probably) > T(Past)° (once)
5. T(Past)° (once) > T(Future)° (then)
6. T(Future)° (then) > Modirrealis° (perhaps)
7. Excursus: Modepistemic° (probably) > Modirrealis° (perhaps, [almost] cer-
tainly)
a. Modepistemic° (probably) > Modirrealis° (perhaps)
b. Modepistemic° (probably) > Modirrealis° ([almost] certainly)
8. Modirrealis° (perhaps) > Modnecessity° ([not] necessarily)
9. Modnecessity° ([not] necessarily) > Modpossibility° (possibly)
10. Modpossibility° (possibly) > Asphabitual° (usually)
11. Asphabitual° (usually) > Asprepetitive(I)° (again)
12. Asprepetitive(I)° (again) > Aspfrequentative(I)° (often)
13. Excursus: Asphabitual° (usually) > Aspfrequentative(I)° (often)
14. Aspfrequentative(I)° (often) > Modvolitional° (intentionally)
15. Modvolitional° (intentionally) > Aspcelerative(I)° (quickly)
16. Aspcelerative(I)° (quickly) > T(Anterior)° (already)
17. T(Anterior)° (already) > Aspterminative° (no longer)

4

 18. Aspterminative° (no longer) > Aspcontinuative° (still)
19. Aspcontinuative° (still) > Aspperfect(?)° (always)
20. Aspperfect(?)° (always) > Aspretrospective° (just)
21. Aspretrospective° (just) > Aspproximative° (soon)
22. Aspproximative° (soon) > Aspdurative° (briefly)
23. Aspdurative° (briefly) > Aspgeneric/progressive° (characteristically)
24. Aspgeneric/progressive° (characteristically) > Aspprospective° (almost)
25. Aspprospective° (almost) > AspSgCompletive(I)° (completely)
26. AspSgCompletive(I)° (completely) > Voice° (well)
27. Voice° (well) > Aspcelerative(II)° (fast/early)
28. Aspcelerative(II)° (fast/early) > Asprepetitive(II)° (again)
29. Asprepetitive(II)° (again) > Aspfrequentative(II)° (often)
30. Aspfrequentative(II)° (often) > AspSgCompletive(II)° (completely)
31. Domain adverbs > Moodspeech act° (honestly)
B. Conclusions
C. Discussion
1. A possible explanation: the Direct Attachment Hypothesis
2. Testing the Direct Attachment Hypothesis
a. What nonspinals to use?
b. Data and discussion
c. Why are some of the Cinque adverbs unable to attach to non-
spinals?
d. Conclusions
e. Direct attachment and (or vs.) the Cinque hierarchy
III. When adverbs are interpreted higher than they appear
A. Introduction
1. Semantics and syntax
2. The prosodic characteristics of the inverse scope reading
3. Do the surface scope and inverse scope readings differ in information
structure?
B. Some possible explanations for interpretive escape
1. Escaped adverbs as parentheticals
a. A first attempt
b. A second attempt
2. The two-fragment analysis
3. Summary
C. Testing the limits of interpretive escape
1. Interpretive escape out of small clauses
2. Interpretive escape out of infinitival clauses
3. Interpretive escape out of finite clauses
4. Discussion
a. Summary of findings
b. Interpretive escape and processing effort
c. Interpretive escape across multiple clause boundaries
d. Explaining the counterexamples just and almost
i. Why just and almost cannot be semantically displaced

5

 ii. Why the just-sentences are worse than the almost-
sentences
D. Toward an account of unusually low adverbs
IV. Conclusion
A. The Cinque hierarchy faces substantial difficulties in accounting for English
adverb orders
B. Findings regarding interpretive escape

6

 Two Investigtions of Adverbs and Clause Structure in English1

I. Introduction

I.A. General background

At least in generative linguistics, there is a view that modification is one of the least
well understood areas of grammar, and hence deserving of its status as a controversial
topic. This view is expressed by Delfitto (2006:85), who cites a claim by Chomsky
(1995:382, n. 22) that “we still have no good phrase structure theory for such simple mat-
ters as attributive adjectives, relative clauses, and adjuncts of many different types.”2 One
modification-related controversy concerns what exactly the syntactic relation is between
adverbs and the rest of the syntactic tree. Are adverbs adjuncts (the traditional view), or
are they specifiers of dedicated functional projections with specific semantic content, as
argued by Cinque (1999)? This issue is taken up by Wilson and Saygın (2001), Ernst
(2002), Nilsen (2003), Haider (2004), Cinque (2004), and others,3 and will be discussed
in more detail in §I.B.
Regardless of whether you agree with the way these researchers frame the issues,
Chomsky and Delfitto are certainly correct in their claims that there is much we do not
yet understand about modification. This is true even of modification structures in English
(presumably the most intensively studied language in modern linguistics)—as should be-
come clear in this thesis, even as we attempt to understand certain modification-related
phenomena better.

I.B. The Cinque hierarchy

The specific subclass4 of modifiers with which we will be concerned is the adverbs.
Multiple hypotheses concerning the syntax of adverbs are being entertained today. One of

























































1I am grateful first and foremost to my advisor, Edwin Williams, for our illuminating discussions and for
his guidance throughout these investigations. I would also like to thank Robert Freidin for kindly agreeing
to be my second reader. For valuable discussion, I am indebted to James Collins, Amy Rose Deal, Michael
Yoshitaka Erlewine, Christiane Fellbaum, Adele Goldberg, Valentine Hacquard, Stephanie Harves, Richard
Kayne, Joshua T. Katz, Jeff Lidz, Salvador Mascarenhas, Mark Norris, Colin Phillips, Norvin Richards,
Anna Tchetchetkine, Anie Thompson, Alexander Williams, Amy Zhou, and Jesse Zymet. None of these
people should be assumed to endorse the material here presented. I would like to thank Michael Yoshitaka
Erlewine for assistance with LaTeXiT. For keeping me company during much of the writing of this thesis,
in an atmosphere of thoroughgoing comradeliness and solidarity, I am deeply grateful to Nick Bellinson,
Derek Gideon, Mitha Nandagopalan, Krithin Sitaram, Alisa Tao, Anna Tchetchetkine, Evan Warner, Lily
Yu, and Amy Zhou. Finally, this thesis would not have been possible if I weren’t fortunate enough to have,
in all things, the unconditional support of my parents, Nancy Carrasco and Samuel Zyman, who have al-
ways been willing to do anything and everything for me. All errors are mine alone. The title of this thesis
was inspired by that of Postal (1998): Three Investigations of Extraction.

2
I am grateful to Robert Freidin for reminding me of this quote.
3
See also Grohmann (2005:1, 3) and Wenger (2009:2-3, §3), and the list of references at the beginning of
the latter (p. 1). It was through this list that I became aware of several of the works cited in the first para-
graph of the text. For responses to Cinque (1999) that do not deal specifically with what Wenger (2009:1)
calls the “specifier vs. adjunct” debate (cf. (8) herein), see Bobaljik (1999) and Manninen (2005).
4
This wording is not meant to imply a commitment to the hypothesis that Adv(erb) is a primitive category,
although it seems plausible that this is in fact the case at least for familiar languages like English (see

7

them, the Cinque hierarchy (Cinque 1999), will play a central role in our discussion; and
so it will be fitting to describe this hypothesis in detail, starting with the intellectual con-
text in which it arose and the empirical facts that motivated it.

I.B.1. The traditional hypothesis: adverbs as adjuncts

The traditional view regarding the syntax of adverbs is that these elements are ad-
juncts, and consequently their placement is relatively free. At least at first glance, this
certainly appears to be borne out by facts from English:

(1)00 a. *‘Carefully, Ashley picked up the wounded bird.

b. *‘Ashley carefully picked up the wounded bird.
c. *‘Ashley picked up the wounded bird carefully.

On the ADJUNCTION HYPOTHESIS, as I will call it, these sentences fundamentally have the
structures in (2a-c), respectively. These diagrams are drawn in accordance with the “old-
school” phrase structure rule S → NP VP, which most generative linguists today would
reject.5 I do this here to avoid certain complications that immediately arise in a strictly
endocentric framework (with IP or TP instead of S, etc.). These concern what exactly the
adverbs are attached to, and are irrelevant to the illustration of adjunction.

(2)00 a. S

AdvP S

carefully

*‘ Ashley picked up the wounded bird

b. S

NP VP

Ashley
AdvP VP

carefully
*‘ picked up the wounded bird


























































Delfitto 2006:86-89, §2; Pullum 2010, §3.2.4). Whether it is or not, though certainly an interesting ques-
tion, will be largely orthogonal to our aims here.
5
The trees also gloss over the issue of the exact label of the constituent Ashley and treat it as an NP for sim-
plicity.

8

 c. S

NP VP

Ashley
VP AdvP

carefully
*‘ picked up the wounded bird

To recapitulate, the picture of adverb syntax we have so far is one in which the fol-
lowing hold:

(3)00 a. *‘There is a class of words (“adverbs”) belonging to the category Adv, which
*‘includes carefully, slowly, well, …
b. *‘Adverbs are incorporated into the tree by adjunction.
c. *‘Adverbs may adjoin to S or VP, at least.

I.B.2. A problem for the traditional view: adverb ordering restrictions

Now, it is clear that (3) is not even remotely the end of the story. One problem with
the view in (3) is that certain adverbs display ordering restrictions, which nothing in (3)
predicts. Consider the following Italian data from Cinque (1999).
In this language, già ‘already’ and più ‘any longer’ can only cooccur in that order:6

(4)00 a. *‘All’ epoca non possedeva già più nulla.

*‘at+the time not possessed already any.longer nothing
*‘semilit. ‘At the time (s)he did not possess already any longer anything.’
*‘id.7 ‘At the time, (s)he already didn’t own anything any longer.’
(Cinque 1999:5, (5a))

(4)00 b. ‘*All’ epoca non possedeva più già nulla.

*‘at+the time not possessed any.longer already nothing
(ibid:5, (5b))


























































6
In other words, these two adverbs can only cooccur in the order già…più, not *più…già. I will often de-
scribe such ordering restrictions by saying “A precedes B” or “A obligatorily precedes B.” Such restrictions
will only have visible effects when A and B actually cooccur in the same clause or subclausal structure.
The aforementioned wordings are not meant to imply that A requires the presence of B or vice versa. In
fact, this should be assumed not to be the case unless indicated otherwise.
7
Cinque (1999) often departs from standard practice by providing neither a literal gloss nor a perfectly
idiomatic translation, but merely a single line that falls somewhere in between the two (cf. Manninen
2005:457). This is an English sentence that is sometimes stilted or even ill formed, owing to the desire to at
least approximate the word order and other characteristics of the original. When citing such examples from
Cinque, I have adapted them to the standard format (example, literal gloss, idiomatic translation). Semil-
iteral (“semilit.”) translations are Cinque’s; the fully idiomatic (“id.”) translations are generally mine.

9

Più ‘any longer’, for its part, obligatorily precedes sempre ‘always’:

(5)00 a. *‘Da allora, non ha più sempre vinto.

*’from then not has any.longer always won
*‘semilit. ‘Since then, he has no longer always won.’
*’id. ‘Since then, he hasn’t always won anymore.’ (ibid:6, (11a))

(4)00 b. ‘*Da allora, non ha sempre più vinto.

*‘from then not has always any.longer won (ibid:6, (11b))

Thus far, we have identified two adverb ordering restrictions in Italian: già > più (where
“>” means ‘precedes’) and più > sempre. Because linear precedence is a transitive rela-
tion, these two ordering restrictions can be combined into a larger partial adverb order for
Italian:

(6)00 già > più > sempre (cf. ibid:7, (22a))

‘already’ ‘any longer’ ‘always’

The partial order in (6) predicts that già should precede sempre. This prediction is borne
out:

(7)00 *‘Quando si presenta un problema… (ibid:6, (12))

*‘when SI presents a problem
*‘lit. ‘When a problem presents itself…’
*‘id. ‘When a problem arises…’

(7)00 a. ‘*…lui sa già sempre come fare.

‘*…he knows already always how to.do
*‘…he already always knows how to act.’

(7)00 b. ‘*…lui sa sempre già come fare.

‘*…he knows always already how to.do
*‘…he always already knows how to act.’

The same facts obtain in French (ibid:5, (6); 6, (16-17)).

I.B.3. Cinque (1999), part 1: From ordering restrictions and the logic of transitivity to a
hierarchy of adverbs

Cinque (1999, §1.1-1.6) brings to light many more pairwise adverb orders, on the ba-
sis of data primarily from Italian and French and to a lesser extent from English and
Dutch. I will refer to these adverb orders (which all take the form A > B) as ORDERING
8
STATEMENTS. Cinque then unifies these ordering statements into a macro-ordering
(ibid:16, (70); cf. ibid:11, (44); ibid:13, (58)). Finally, he provides preliminary evidence
for this macro-ordering’s universality in the form of data from English, Norwegian, Bos-

























































8
This term is borrowed from Fox and Pesetsky (2005:8ff.) but used in a very different sense than they do.

10

nian/Serbo-Croatian, Hebrew, Mandarin, Albanian, and Malagasy (ibid., §1.7). (Cinque’s
final proposal concerning the fine structure of the clause—which is based on more types
of evidence than just adverb orders—will be shown later, in (18).)
These ordering statements and macro-orderings are not limited to governing the dis-
tribution of particular adverbs. Although they are often stated in terms of particular ad-
verbs (cf. (6)), each adverb in these orders stands for a semantically defined class of ad-
verbs. For instance, Cinque (1999:106, (92)) uses the evaluative adverb fortunately to
stand in for at least the following AdvPs: (un)fortunately, luckily, regrettably,
strangely/oddly (enough), and (un)expectedly (ibid:85).
One challenge that faces anyone trying to arrive at a total (or even just a large-scale)
adverb ordering for a particular language is that it is often not practical to construct ex-
amples containing as many adverbs as we might like. This is because clauses containing
large numbers of adverbs can be difficult to process,9 and sometimes require elaborate
contexts to be pragmatically felicitous. For these reasons, the compilation of Cinque’s
macro-ordering relies crucially on the logic of transitivity invoked to justify (6).
Thus far, our discussion of the restrictions on adverb placement investigated by
Cinque has been framed entirely in terms of linear precedence. But although adverbs cer-
tainly enter into precedence relations, as we have seen, they also enter into relations de-
fined over hierarchical syntactic structure, such as dominance and c-command. There-
fore, the question naturally arises of whether restrictions on pairs of adverbs are defined
in linear or hierarchical terms. Cinque assumes that these restrictions are hierarchical in
nature, and that the sorts of ordering statements we considered above (e.g., già ‘already’
> più ‘any longer’) are their reflexes at PF.

I.B.4. Cinque (1999), part 2: Defending the AdvP-in-Spec Hypothesis against its alterna-
tives

Cinque (1999:44) argues that the rigid restrictions on adverb orders are “entirely un-
expected” on the traditional Adjunction Hypothesis, rendering the latter empirically in-
adequate.10 Of course, this claim raises the question of what hypothesis should replace it.


























































9
An example is given below. The sentence in (i) is perfectly well formed, but difficult to build up a seman-
tic representation for without conscious effort:

(i). . Mike again often, having gone to the beach often one year, goes to the beach often again the next
year.

This sentence asserts that part of Mike’s life has unfolded as shown in (ii)—where b represents a beach trip,
bienn stands for biennium, and Era 2 contains the utterance time.

(ii) Era 1 [bienn (year bb…) (year bb…)] … [bienn (year bb…) (year bb…)] … [bienn (year bb…) (year bb…)] …
Interlude
Era 2 [bienn (year bb…) (year bb…)] … [bienn (year bb…) (year bb…)] … [bienn (year bb…) (year bb…)] …
10
In addition to providing empirical arguments against the Adjunction Hypothesis, Cinque also argues
against it on conceptual grounds. Briefly, he contends that if even some AdvPs can be profitably analyzed
as occupying Spec positions, this raises the possibility of analyzing them all this way, and perhaps even
eliminating adjunction from the theory (Cinque 1999:44). For further discussion, see ibid. (pp. 44-45).

11

Cinque mentions at least three candidates, which are listed in (8b-d). (The Adjunction
Hypothesis is shown in (8a) for completeness.)

(8)00 Name11 Content

a. Adjunction Adverbs are adjuncts.
Hypothesis
b. Extended Adverbs are heads in the extended projection of V. An adverb
Projection takes a smaller portion of the extended projection of V as its
Hypothesis complement. (Cinque 1999:4)
c. Multiple AdvPs are generated in specifier positions. A head may have
Specs Hy- more than one specifier.
pothesis (ibid:47)
d. AdvP-in- Every AdvP sits in the unique specifier position of a functional
Spec Hy- head in the extended projection of V.
pothesis (ibid:4; ch. 2)

Let us consider these new candidates one at a time.

I.B.4.a. The Extended Projection Hypothesis

The Extended Projection Hypothesis ((8a)) Cinque dispenses with swiftly:

The evidence against this assumption ranges from the fact that adverbs do not block head
movement of various verbal forms, [endnote omitted] to the fact that some of them can un-
dergo Topicalization and Focus Movement, which are open to XPs but not to X°s. (Cinque
1999:4)

An additional argument […] may come form the “grammar” of code switching. As Ma-
hootian and Santorini (1996, 472ff and n. 9) observe, a head always dictates the position of
the complement (irrespective of the “code” of the complement). But with adnominal and ad-
sentential adjuncts[12] (adjectives and adverbs), anything goes, and this is another reason not
to see what follows the adverb as its complement. (ibid:167, n. 3)

I.B.4.b. The Multiple Specs Hypothesis

The Multiple Specs Hypothesis ((8b)), Cinque (1999:47) contends, suffers from the
same empirical inadequacy as the traditional Adjunction Hypothesis: it fails to predict the
ordering restrictions to which adverbs are subject. Cinque does not elaborate on this, but
the logic is clear. If the only syntactic constraint on adverbs is that they must be in Spec
positions (since we haven’t said anything else), and a head can have multiple Specs, then
(9b) satisfies the syntactic constraint on adverbs as well as (9a) does. Hence, flexibility of

























































11
The term AdvP-in-Spec Hypothesis is Cinque’s (1999:45); he also calls it the “LOCATION-IN-SPEC” HY-
POTHESIS (ibid:44). The other terms in this column are mine.
12
It is clear that Cinque is here using the term adjunct merely as a descriptive convenience, since his hy-
pothesis is precisely that AdvPs are not adjuncts but specifiers of dedicated functional projections. The hy-
potheses defended in Cinque (1999) are compatible with Kayne’s (1994) Antisymmetry program (cf. Wen-
ger 2009:9, §8), in which the distinction between specifiers and adjuncts is obliterated (Cinque 1999:44).
—EZ

12

this sort (AdvP1 > AdvP2, AdvP2 > AdvP1) should be extremely widespread—
contrary to fact, it would seem, in light of the rigid adverb orders reported by Cinque.

(9)00 a. HP b. HP

AdvP1 H AdvP2 H

AdvP2 H AdvP1 H

I.B.4.c. The AdvP-in-Spec Hypothesis

As we have hinted at, the option that Cinque (1999, ch. 2) defends is the AdvP-in-
Spec Hypothesis ((8c)). As evidence for this view, Cinque adduces Italian sentences con-
taining multiple adverbs (in the order predicted by his hierarchy, of course). He notes
that, in these sentences, the verb can appear to the left or to the right of any of the ad-
verbs. His argument then proceeds as follows (p. 45). Assume the following:

(10)0 a. Classical X′ theory is correct. In other words, all phrases have the following
structure:

XP

ZP X!
X0 YP

b. AdvPs occupy the specifiers of functional heads (the hypothesis to be de-

fended).
c. Head movement exists (following Pollock 1989, a.o.).

By (10a), any two specifiers that are maximally close to each other will nevertheless have
a head position in between them;13 this is simply forced by the structure of the X′ schema.
As a result, if AdvPs are in specifier positions ((10b)), then two string-adjacent AdvPs
will be separated by at least one head position. Hence, if a language allows overt move-
ment of V° up to a certain height (cf. (10c)), such movement should be able to target the
head position in between any two maximally close AdvPs (assuming that this position is
not higher than V° can raise in the language). Therefore, the assumptions in (10) taken
together predict the aforementioned Italian facts concerning the placement of V relative
to adverbs. So if (10a) and (10c) are well-motivated assumptions, then the hypothesis in
(10b) is supported by the ability of all three together to derive these word-order facts.
By contrast, Cinque (1999:45) argues, the Adjunction Hypothesis fails to derive
these facts. On this hypothesis, it should be possible to adjoin two adverbs to the same

























































13
For instance, in (10a), ZP (which is in [Spec,XP]) and [Spec,YP] (not shown) are separated by the head
X°. Note that it is not clear what it would mean for ZP (= [Spec,XP]) and [Spec,ZP] (not shown) to be
separated by a head position, since the former phrase dominates the latter. This suggests that an explicit
characterization of the relation close-to(A,B) would have to presuppose that neither of A and B dominates
the other.

13

side of one phrase. If this happens, there will be no head position between them to which
V can move. Although Cinque does not mention this explicitly, the Multiple Specs Hy-
pothesis faces the same problem. On this hypothesis, a phrase should be able to host two
or more adverbs in its multiple specifiers. In this scenario, there will once again be no
head positions between these adverbs for V to move to.
Cinque illustrates the relevant V/Adv word-order facts in Italian with past participles
(in active clauses) for the lower adverbs and with finite verbs for the higher adverbs, on
the grounds that finite verbs can move higher than past participles in Italian (ibid:49).
There are two cases where the assumptions in (10) would predict that V should be
able to appear in particular (very low) head positions, and yet in Standard Italian it can-
not. Cinque (ibid:46) argues that the evidence that these head positions exist is provided
by other Romance varieties (Logudorese Sardinian and French), and that their existence
in Italian is obscured by a language-specific requirement that a past participle move
above them both (from its base position at the bottom of the extended projection of V).14
Lastly, Cinque argues that the AdvP-in-Spec Hypothesis is superior to the Adjunc-
tion Hypothesis and the Multiple Specs Hypothesis because it is more restrictive, ruling
out several possible types of languages, and hence more easily falsifiable. I will not re-
hearse this argument here; the reader is referred to Cinque (ibid:48-49) for discussion.

I.B.5. Cinque (1999), part 3: Matching the adverb hierarchy to a hierarchy of functional
heads

So, what does it take for a sentence containing multiple adverbs to be ruled in or out?
The picture that emerges from Cinque (1999) is that, in order to be ruled in, a sentence
must minimally15 conform to the Cinque hierarchy—a universal rich structure full of
functional heads in whose specifiers AdvPs are hosted. These functional heads have spe-
cific semantic content: they typically contribute some flavor of temporal, aspectual, or
modal meaning. Any given functional head stands in a tight semantic relation to the
AdvP it hosts in its specifier. In fact, Cinque posits that each head and its corresponding
AdvP share a semantic feature, which is checked in a Spec–head configuration (cf.
ibid:44).
The hypothesis that the functional heads and AdvPs stand in a relation this tight pre-
dicts that the hierarchy of adverbs should be paralleled, structurally and semantically, by
a hierarchy of heads. Cinque argues that this prediction is correct, and he does so as fol-
lows.


























































14
This is a requirement that a past participle raise at least a certain distance (i.e., that it not appear too low).
But as mentioned above, Italian past participles are also not allowed to move too high. And within the por-
tion of the tree in which an Italian past participle can appear, it can apparently surface anywhere, at least as
regards its linear position with respect to adverbs. The word-order facts on which this description is based
seem to be to a considerable degree arbitrary. Two natural questions here are:

1) By what, if anything, are the apparently optional movements of V motivated?

2) Are the language-specific bans on V appearing too low or too high purely arbitrary, or are they amena-
ble to some sort of explanation?
15
Of course, it must be well formed in every other respect as well. For instance, its multiple adverbs cannot
induce semantic anomaly.

14

I.B.5.a. The hierarchy of functional heads

First, Cinque provides evidence for particular pairwise head orders from numerous
languages. Cinque takes functional heads to be realizable by the following kinds of ele-
ments:

(11)0 a. ‘*auxiliaries, aka “(functional) particles” (cf. Cinque 1999:52)

b. ‘*“nonclosing” (agglutinating) suffixes on V
c. ‘*“closing” (inflectional) suffixes on V16

Cinque also considers evidence from “mixed cases,” in which more than one kind of
element in (11) appears in the same clause.
The rationale for taking suffixes on V to realize functional heads is “(a generalized
version of) Baker’s Mirror Principle” (ibid:52 and references there; 186, n. 2). This prin-
ciple is stated in (12):

(12)0 ‘*The Mirror Principle (Baker 1985:375, (4))

‘*Morphological derivations must directly reflect syntactic derivations (and vice
‘*versa).

The way that (12) is regularly interpreted in practice, at least in relation to verbs and
clause structure, is as follows. The left-to-right linear order of suffixes on the verb is the
mirror image of the top-to-bottom hierarchical sequence of heads in the clause. If this in-
terpretation of (12) is adopted as a working hypothesis, then a verb of the form V-Asp-T-
Mood constitutes evidence for the following order of heads: Mood° > T° > Asp° > V°.
(Here “>” must be interpreted as ‘is higher than’.17)
The different types of elements that Cinque takes to realize functional heads (see
(11) and subsequent discussion) are illustrated below ((13-16)). Gungbe exemplifies aux-
iliaries, Korean “nonclosing” suffixes, and Spanish “closing” suffixes. The Welsh exam-
ple in (16a) is a mixed case, with past tense realized synthetically on the verb ‘be’, but
perfect and progressive aspects realized as auxiliaries. For each of (13-16), the (a) exam-
ple provides the sentence and the (b) example the order of heads that can be inferred from
it.18


























































16
“Closing” and “nonclosing” suffixes are so called because they do and do not close off the word respec-
tively, “thus disallowing/allowing further suffixation” (Cinque 1999:185, n. 1). Cinque attributes these
terms to Nida (1974:85).
17
On standard assumptions, this can be made more specific: ‘c-commands’.
18
The reader who looks up in Cinque (1999) the head orders in the (b) examples here will instead find more
extensive orders when these were deduced from multiple examples (by the logic of transitivity). The orders
shown here include only those heads that are inferable from the relevant (a) examples.

15

(13)0 a. *‘Àsíbá ná tò nà kèkè . [Gungbe]
*‘Àsíbá FUT PROG PROSP buy bicycle the
*‘Àsíbá will be about to buy the bicycle.’19
(Aboh 1993, 1996, via Cinque 1999:65, (52e); see ibid:192, n. 38)

(13)0 b. *‘T(Future) > Aspprogressive > Aspprospective > V (based on Cinque 1999:65, (54))

(14)0 a. *‘Ku pwuni caphisiessesskeysssuptikka? [Korean]

(14)0 a. *‘Ku pwun -i

*‘the person -NOM

(13)0 b. *‘cap -hi -si -ess -ess -keyss -sup -ti -kka?
*‘catch -PASS -PM -ANT -PST -EPIST -PM -EVID -Q

(13)0 b. *‘Did you feel that he had been caught?’

(Sohn 1994, §2.1.3, via Cinque 1999:53, (1))

(13)0 b. *‘Moodspeech act > Moodevidential > Modepistemic > T(Past) > T(Anterior) > Voice
*‘> V20
(based on Cinque 1999:54, (4))

(15)0 a. *‘Esos libros han estado siendo leídos todo el año. [Spanish]
*‘those books have been being read all the year
(13)0 a. *‘Those books have been being read all year.’
(Zagona 1988:5, via Cinque 1999:57, (20b))

(13)0 b. *‘T > Aspperfect > Aspprogressive > Voice > V (based on Cinque 1999:57, (21))


























































19The sequence “FUT PROG PROSP” in the gloss, combined with the fact that Gungbe heads apparently typi-
cally precede their complements (cf. ibid:64-65, (52)), would normally lead us to expect that (13a) would
be translated as follows:

(iii). . ‘Àsíbá will be being about to buy the bicycle.’

with the underlined parts corresponding to the auxiliary tò. However, the prospective auxiliary nà ‘be about
to’ must always be preceded by tò (cf. Cinque 1999:64, (47)). So two possibilities come to mind:

1) (iii) is the best literal translation of (13a), but the sequence tò nà is regularly translated into English
simply as ‘be about to’, because be about to… in English cannot be embedded under progressive as-
pect (#be being about to…), except under extraordinary circumstances (which I will not describe here,
as it would take us too far astray).
2) tò loses its progressive force when preceding the prospective auxiliary nà.
20
Cinque (1999:54) excludes from this partial head order the politeness markers in (14a) (glossed PM), on
the grounds that they are plausibly agreement markers. In Cinque’s system, this move is justified by his
argument that Agr(eement) and Neg(ation) heads do not occupy fixed positions in the functional structure
of the clause but rather can be generated at many different heights (ibid:121, 127; see also ibid:108ff.,
where he argues that AgrPs aren’t even the best way to treat agreement in the first place).

16

(16)0 a. *‘Oedd y bachgen wedi bod yn ymlad. [Welsh]
*‘be.PST the boy PERF be PROG fight
*‘The boy had been fighting.’
(Hendrick 1991:173, (6), via Cinque 1999:67, (63))

(13)0 b. *‘T(Past) > Aspperfect > Aspprogressive > V (based on Cinque 1999:67, (64))

Using the by now familiar logic of transitivity, Cinque (ibid:76, (96)) combines all
the partial head orders obtained in this way into a macro-ordering of heads.

I.B.5.b. Matching the two hierarchies

Having arrived at preliminary versions of the adverb and head hierarchies, Cinque
(1999, ch. 4) “match[es] and refin[es]” them (ibid:77). He does this in order to bolster his
claim that the following situation obtains:

(17)0 a. *’There is a one-to-one mapping between adverb classes and functional

*’heads.
b. *‘The adverb hierarchy and the functional head hierarchy line up perfectly.
c. *‘(17b) is true because, for any given adverb class CAdv, an adverb belonging
*‘to CAdv checks a semantic feature against the functional head corresponding
*‘to CAdv.

The result of “matching and refining” the adverb and head hierarchies is the functional
structure (F-STRUCTURE) for the clause shown in (18):

(18)0 [frankly Moodspeech act° (Cinque 1999:106, (92), via Wenger 2009:4, (6))
[fortunately Moodevaluative°
[allegedly Moodevidential°
[probably Modepistemic°
[once T(Past)°
[then T(Future)°
[perhaps Modirrealis°
[necessarily Modnecessity°
[possibly Modpossibility°
[usually Asphabitual°
[again Asprepetitive(I)°
[often Aspfrequentative(I)°
[intentionally Modvolitional°
[quickly Aspcelerative(I)°
[already T(Anterior)°
[no longer Aspterminative°
[still Aspcontinuative°
[always Aspperfect(?)°
[just Aspretrospective°
[soon Aspproximative°

17

 [briefly Aspdurative°
[characteristically Aspgeneric/progressive°
[almost Aspprospective°
[completely AspSgCompletive(I)°
[tutto(Italian) AspPlCompletive°
[well Voice°
[fast/early Aspcelerative(II)°
[again Asprepetitive(II)°
[often Aspfrequentative(II)°
[completely AspSgCompletive(II)° …]]]]]]]]]]]]]]]]
]]]]]]]]]]]]

Cinque (1999:105; cf. ibid:141) stresses that the structure in (18) is only provisional.21
For discussion of certain aspectual morphemes and adverbs that have not yet been inte-
grated into this structure, see Cinque (ibid., §4.28).

I.B.6. Cinque’s classes of apparent counterexamples

One virtue of the hypothesis that has come to be known as the Cinque hierarchy is
that it makes strong predictions. It predicts that a host of elements will occur in the same
order in every language—or at least in the same hierarchical configuration, since head–
complement pairs surface in different orders (H-C or C-H) in different languages, and
sometimes even in the same language, depending on the categories involved and other
factors.
Admittedly, the Cinquean worldview does allow the universal hierarchy under dis-
cussion to be deformed by movements. But because on this view movement is subject to
constraints (such as Relativized Minimality, Cinque 1999:19, 24; cf. ibid:4), the Cinque
hierarchy still predicts the impossibility of a host of structures.
Needless to say, apparent counterexamples to the Cinque hierarchy need to be ana-
lyzed carefully to determine whether they are real counterexamples or not. For instance,
imagine that we are investigating language L, and the following situation obtains:

(19)0 a. The Cinque hierarchy predicts that adverb A should occur higher than adverb
B, universally.
b. In L, as in English, (19a) essentially entails that the Cinque hierarchy predicts
that A should obligatorily precede B.
c. L actually permits the linear order B > A.

Cinque (1999:3-4) alerts us to six possible sources for such an apparent counterexample
and argues that these kinds of cases are not actually problematic for his hypothesis. I
paraphrase his discussion here:


























































21
Cinque (1999:4-5) refers to the negative elements mica (Italian) and pas (French) as adverbs, and dis-
cusses their placement with respect to other adverbs, but does not include them in this version of the clausal
hierarchy. This is probably because he “[p]rovisionally […] assume[s]” that NegPs, unlike the functional
projections hosting adverbs, are not present syntactically when they are not overtly filled (ibid:121).

18

(20)0 a. B is directly modifying A. (If this is the case, then no material can be inserted
between them.)
b. A portion of the clause containing B has raised over A for reasons related to
information structure.
c. B has wh-moved over A.
d. A and/or B can actually be generated in two different positions in the func-
tional hierarchy, which are semantically distinct.
e. A is being used as a “focusing” adverb, as in Katie oftenB talks to [probablyA
Jake].22
f. A is being used parenthetically.

I.B.7. A few alternatives to the Cinque hierarchy

As we mentioned at the outset, the Cinque hierarchy is not the only current hypothe-
sis concerning what principles rule the distribution of adverbs. Although it would be im-
possible in a work of this scope to compare the various current hypotheses in detail, as-
sessing the empirical successes and shortcomings of each one, I will briefly mention just
a few that we have not yet discussed in any detail.
Wilson and Saygın (2001) investigate adverb orders in Turkish and conclude that ad-
verb orders in this language are ruled in or out primarily by semantic scope considera-
tions. Let us state this view more explicitly:

(21)0 Adverb-Order-by-Scope Hypothesis (Wilson & Saygın 2001)23

Let A and B be adverbs that are both modifying portions of the extended projec-
tion of a head H. Then, A will occur lower than B iff A is a semantic constituent
of the constituent that B is modifying.

On this view, any adverb order will be well formed, all else being equal, unless the mean-
ings of the adverbs involved make it impossible to compute a well-formed meaning for a
part of the structure.24
On Tenny’s (2000:316ff.) hypothesis,25 the clause is divided into “semantic zones,”
with the characteristics in (22). On this view, adverbs are lexically specified for particular
semantic zones, but within a semantic zone, their relative positions are determined by
their scopes.


























































22Much of this thesis will be devoted to investigating subcases of this phenomenon. Cinque’s (1999:4) term
“FOCUSING” ADVERB is discussed in fn. 102.

23
Both the term Adverb-Order-by-Scope Hypothesis and the formulation of the hypothesis in (21) are mine.
24
Wilson and Saygın (2001) do not explicitly take a stand on what Wenger (2009:1) calls the “specifier vs.
adjunct” debate (cf. §I.A and (8)). Suppose, however, that they are right and the restrictions on adverb or-
ders really do fall out from the semantic scope of the adverbs involved. Then it would seem that the best
hypothesis regarding the syntactic status of adverbs is that they are adjuncts, since this view would allow us
to dispense with many of Cinque’s (1999) functional projections. Of course, Cinque would respond that
those projections are needed anyway, to account for auxiliaries, “particles,” and verbal suffixes (see (11)).
25
It was through Wenger (2009:2-3, §3) that I became aware of the hypotheses discussed in the following
two paragraphs.

19

(22)0 Zone Functional spe- Approximate location Corresponding stretch of the
cialization in the clause Cinque hierarchy
1 speaker deixis CP Moodspeech act° → Modepistemic°
2 temporal deixis TP T(Past)° → T(Future)°
3 truth value ModP Moodirrealis° → Modpossibility°
modality
4 middle aspect AspP (grammatical Asphabitual° → Aspprospective°
aspect)
5 core event AspP (lexical aspect)26 AspSgCompletive(I)° →
AspSgCompletive(II)°
(adapted from Wenger 2009:4-5, §3.2)

Lastly, there are hypotheses concerning adverb syntax that fall in between purely ad-
junction-based views such as Ernst’s (2001) and the specifiers-only approach of Cinque
(1999). On Alexiadou’s (1997) and Laenzlinger’s (1998) hypotheses, some adverbs are
adjoined, and others occupy the specifiers of dedicated functional projections. Wenger
(2009:2-3) remarks, “these theories are much closer to the functional specifier approach
than to the traditional XP-adjunction approach, in that they adhere to some uniqueness
condition (i.e. if an adverb is adjoined, it is still adjoined to a specific functional head).”

I.B.8. Summary

Cinque (1999) argues that the traditional hypothesis that adverbs are adjuncts is in-
adequate. On the basis of a crosslinguistic survey of adverb and functional head orders,
Cinque argues in favor of the following. AdvPs occur in the specifiers, and functional
heads occur in the head positions, of dedicated functional projections that are rigidly or-
dered universally. The hierarchies of AdvPs and functional heads align perfectly, because
each AdvP checks a semantic feature against its corresponding functional head in a Spec–
head configuration. If this view is correct, then the structure of the clause is far richer
than previously thought.

I.C. The project

I.C.1. Method

The “internalist” approach to language—which was pioneered by Chomsky and oth-

ers and is now pursued in diverse theoretical frameworks—holds that grammar, or lin-
guistic knowledge, exists only in the mind. Assume with me that this is in fact the case.


























































26
Some differences in the terminology used to refer to these two types of aspect are discussed by Cinque
(1999:83). The relevant terminology is summarized here:

(iv) Dahl (1985), Cinque (1999) Smith (1991)

grammatical aspect viewpoint aspect
lexical aspect situation aspect

I have adapted (22) to comply with the terminology that Cinque (1999) borrows from Dahl (1985).

20

Then, if in a particular investigation we are seeking to understand speakers’ knowledge of
their language,27 we will take as our primary object of study particular mental grammars.
There are at least two main ways we can do this when investigating a linguistic phe-
nomenon P in a language La:

(23)0 a. The “breadth” Examine P across a large number of speakers of La.

approach
b. The “depth” Zoom in on a small number of speakers of La, and examine
approach P in them.

The “breadth” approach has the advantage of shedding light on which aspects of P
are subject to interspeaker variation in L and which are not, and on what the patterns of
variation are. But this advantage comes at a cost, namely that we are forced to investigate
P in less detail within each individual speaker.
The “depth” approach has the advantage that it allows us to investigate P in greater
detail, which may expose connections between subphenomena that would otherwise have
remained hidden. The cost of this approach is that we learn less about the relevant pat-
terns of variation, and so it is harder to know how widely applicable our conclusions
are—i.e., how big a subset of the La speech community they are valid for.
In my view, these approaches are both valuable, and complement each other. This
thesis takes the second (“depth”) approach, and investigates in substantial detail how par-
ticular phenomena work in a single mental grammar of American English: my own. The
findings here reported could form the basis for subsequent investigation of how these
phenomena work for other individual speakers, or across large numbers of speakers.

I.C.2. A note on the data

Unless otherwise indicated, all judgments on English sentences in this thesis are
mine. My advisor, Edwin Williams, has explicitly agreed with (or at least not strongly
disagreed with) many of them.
As often happens in linguistics, the phenomena we will be exploring will sometimes
require us to consider fairly unusual sentences. At times the judgments will be subtle.28
But my view in a situation like this is that we should not leave the phenomena uninvesti-
gated, but rather venture boldly into the data.

II. Testing the predictions of the Cinque hierarchy in English

II.A. The data

This section is divided into a number of subsections, one for each adverb order pre-
dicted by the Cinque hierarchy that we will be testing. Each subsection is organized as

























































27
In other words, the rules, principles, and facts that allow speakers to make judgments of acceptability,
ambiguity, synonymy, aptness of paraphrase, and so on.
28
If you are a native speaker of English and find that some of your judgments differ from mine, this will
underscore the fact that this investigation would be well complemented by studies of the interspeaker varia-
tion to which the phenomena that we consider are subject.

21

follows. First comes a statement of the adverb order under discussion. Next come the
relevant data. The data are discussed in the cases where I have deemed this appropriate.
The subsection concludes with a “Preliminary result”: an  where predictions of the
Cinque hierarchy have apparently been falsified and a  where they have not.
Even where I have chosen to discuss the facts immediately after presenting them, I
have generally tried to keep the discussion short. This section is primarily descriptive;
issues of explanation will be taken up in subsequent sections.

II.A.1. Moodspeech act° (honestly) > Moodevaluative° (unfortunately)

Cinque (1999:33) claims that, in English, honestly obligatorily precedes unfortu-

nately. He reports the following judgments (which are repeated in Wenger 2009:6,
(10)29):

(24)0 a. *Honestly I am unfortunately unable to help you. (judgments from

b. *Unfortunately I am honestly unable to help you. Cinque 1999:33)

Cinque is right that (24a) is well formed. As for (24b), however, I find it fine, and so do a
number of other native English speakers I have consulted. Because Cinque does not pro-
vide a source for the judgment in (24b), we have no evidence that anyone would actually
agree with it.30 Therefore, in light of the evidence available, the best assessment of the
situation is that Cinque is simply mistaken. Preliminary result: 

II.A.2. Moodevaluative° (unfortunately) > Moodevidential° (allegedly)

(25)0 a. *Unfortunately, Wayne allegedly committed a murder.

b. *Allegedly Wayne unfortunately committed a murder.
Preliminary result: 

Although, as we have seen, honestly does not reliably precede unfortunately, it does pre-
cede allegedly:

(26)0 A: #But WHY don’t you wanna date her?

B: a. #Honestly, she allegedly has stolen a lot of money.
b. #Allegedly she honestly has stolen a lot of money.31

























































29
Wenger adds a comma after the linearly first adverb in each of (24a-b). For me, this makes both sentences
more idiomatic, or less marked, almost certainly due to its effects on the prosodic structure that is superim-
posed onto the sentences when they are read. However, this issue is orthogonal to the main point: both sen-
tences are well formed with or without the comma (and the corresponding prosodic effects).
30
Of course, even if they did, the judgments of those of us who find (24b) fine would still constitute an em-
pirical problem for the Cinque hierarchy.
31
For me, (26Bb) is actually well formed if the honestly is interpreted as a subject-oriented adverb, like the
cleverly in (va). This is possible because (at least for me) honestly can be interpreted this way, as in (vb)
(the relevant reading of which may be easier to bring out in the variant with quite).

(v) a. Nick cleverly had misanalyzed the data.

b. Emma (quite) honestly has returned the wallet to its rightful owner.

22

 Preliminary result: 

II.A.3. Moodevidential° (allegedly) > Modepistemic° (probably)

This ordering might initially seem to make correct predictions for English, in light of
sentences like the following:

(27)0 a. *Kevin allegedly will probably give up.

b. *Kevin probably will allegedly give up.

But the “B” response in the following sentence is vastly more acceptable than (27b)—
indeed, well formed, I would say:

(28)0 A: Why did the police look into Amanda’s case?

B: She probably had allegedly been tortured.

The same point can be illustrated with the evidential adverb clearly. The ordering state-
ment under discussion—Moodevidential° > Modepistemic°—predicts that clearly too will pre-
cede probably. Although this is clearly possible ((29)), the reverse order is as well ((30)).

(29)0 Clearly John probably will quickly learn French perfectly.

(Bowers 1993:607 and n. 13, via Cinque 1999:33)

(30)0 A: Why did the police help Linda?

B: She probably had clearly been drinking.

Even worse for Cinque’s treatment of these adverbs, the following also seem well
formed. (Admittedly, the judgments here are slightly delicate, probably in part because
the relevant sentences’ meanings are somewhat baroque.)

(31)0 A: Why did the police help Linda?

B: Probably she clearly had probably been drinking.

(32)0 The police helped Linda? Clearly she probably had clearly been drinking.

Preliminary result: 

II.A.4. Modepistemic° (probably) > T(Past)° (once)

As pointed out by Cinque (1999:33), these two adverbs can appear in either order.
Cinque gives the following examples:


























































Pragmatically, however, it would be quite odd to interpret (26Bb) this way, as we do not normally consider
stealing money to be an honest thing to do.

23

(33)0 a. Probably he once had a better opinion of us.
b. Once he probably had a better opinion of us.

Cinque (ibid:13; cf. 175, nn. 40-41, 33) argues that we observe this flexibility because
temporal adverbs can be generated not only in their “proper” place in the hierarchy but
also in a left-peripheral SCENE-SETTING or topic position. (Presumably he analyzes these
adverbs as base-generated high because moving them to the left periphery would result in
a R(elativized) M(inimality) violation when they crossed, say, probably.)
This analysis is plausibly correct; let us adopt it for the sake of argument. Then, a
natural question is why it is correct. That temporal adverbs can be base-generated in topic
position is advantageous, because this allows us to create topic–comment structures with
a time adverbial as the topic even when the comment contains a high adverb like proba-
bly (without violating RM). But consider the following. As far as information structure is
concerned, there is nothing wrong with a topic-comment structure where the topic is (ba-
sically) the manner in which something happens, and the comment contains a high adverb
like probably:

(34)0 *Speaking of doing things CAREFULLY, that’s probably how Liam painted the
*HOUSE. But as far as doing things CARELESSLY is concerned, that’s probably
*how Liam painted the GARAGE.

However, the manner adverbs carefully and carelessly cannot be base-generated in topic
position; and as a result, the following example contains two violations of RM:

(35)0 *CAREFULLY, Liam probably painted the HOUSE. CARELESSLY, he probably

*painted the GARAGE.

Cinque’s base-generation analysis of left-peripheral temporal adverbs is quite possi-

bly correct, but it gets the facts right by stipulating them. In other words, even if it is in-
deed the case that temporal adverbs can be base-generated in topic position but manner
adverbs cannot, I think it’s safe to say that we don’t understand why. I will therefore
mark this subsection “Preliminary result: ()” to indicate that there is something more to
be said on this topic.

II.A.5. T(Past)° (once) > T(Future)° (then)

The predictions of this ordering are incorrect. The once and then at hand32 in fact can
cooccur in the order then > once ((36a)). Indeed, it is the predicted order (once > then)
that apparently cannot occur ((36b-c)).

(36)0 *Celia will then have once been happy.

*Celia will once have then been happy.
*Celia once was then happy.


























































32
As opposed to these adverbs’ other avatars—for instance, the once that forms a scale with twice, thrice /
three times, etc. and the INFERENTIAL then of Don’t have a computer? Then build one yourself!

24

 Incidentally, it is worth questioning whether the head–AdvP correspondences posited
here are on the right track. The claimed matchup between the once under discussion and
T(Past)° does seem reasonable, since this once can cooccur with the past tense (with or
without perfect aspect marking), but not with the present, present perfect,33 or future:

(37)0 a. *I was once a king. [past]

b. *I had once been a king. [past perfect]
c. *I am once a king. [present]
d. *I have once been a king. [present perfect]
e. *I will once be a king. [future]
f. *I will have once been a king.34 [future perfect]

However, it is much less obvious that temporal then should be located in

[Spec,T(Future)P], since it can appear not only in clauses with future time reference (He
will then leave) but also in past-tense clauses (He then left). Preliminary result: 

II.A.6. T(Future)° (then) > Modirrealis° (perhaps)

Cinque claims that, in English, then must precede AdvPs in [Spec,ModirrealisP] such
as perhaps or (almost) certainly.35 He reports the following judgments:

(38)0 a. *He was then [almost certainly] at home. (judgments from

b. *He was [almost certainly] then at home. Cinque 1999:33)

(39) a. *He was then perhaps at home.

b. *He was perhaps then at home.

For me, however, the (b) sentences in (38-39) are as good as the (a) sentences—that is to
say, fine. Because Cinque gives no source for the judgments in (38-39), we once again
have no evidence that they are accurate for anyone. At present, the empirical picture sug-
gests that perhaps and almost certainly can either follow or precede then in English, con-

























































33
The use here of the traditional terms past perfect, present perfect, and future perfect is not meant to sug-
gest that these are tenses in their own right; I regard them as combinations of particular tenses with perfect
aspect.
34
Of course, it is unexpected on this view that this once can cooccur with the future perfect. This fact be-
comes less mysterious in light of the following sentences, in which once occurs in nonfinite clauses:

(vi) a. *I consider Charles to once be a king. [infinitival clause]

b. *I consider Charles to have once been a king.
c. *I {read/interpreted} that character as once being a king. [participial clause]
d. *I {read/interpreted} that character as having once been a king.

The deviant sentences in (via) and (vic) become perfect (no pun intended) when they are marked for perfect
aspect ((vib, vid)). This suggests that once can be licensed by either past tense or perfect aspect. If this is
so, then the oddball sentence in the text is not (37f) but (37d), where perfect aspect is apparently insuffi-
cient to license once. My best guess is that (37d) is somehow blocked by (37a), but I do not know how to
derive that result in a principled way.
35
Naturally, almost certainly is nothing but certainly modified by almost.

25

tra Cinque’s prediction. It therefore appears that temporal adverbs such as then are even
more flexible in their placement than Cinque acknowledged (cf. §II.A.4). Preliminary
result: 

II.A.7. Excursus: Modepistemic° (probably) > Modirrealis° (perhaps, [almost] certainly)

Cinque (1999:181, n. 88) notes that his hierarchy predicts that probably will precede
the AdvPs he locates in [Spec,ModirrealisP].

II.A.7.a. Modepistemic° (probably) > Modirrealis° (perhaps)

He continues: “…this appears to be true for probably and perhaps (which do show
this order, if co-occurring at all: …” (ibid.), and supports this claim with the following
data:

(40)0 a. ?*Probably he will perhaps try again. (judgments from

b. ?*Perhaps he will probably try again. Cinque ibid.)

Almost certainly, a common reaction to sentences like these,36 especially on the part
of naïve informants, would be that they are bad, because it doesn’t make sense for proba-
bly and perhaps to cooccur like that: which is it? But of course, the discussion can’t end
there, because we need to consider the possibility of hierarchical interpretations for (40a-
b). On a hierarchical interpretation of one of these sentences, the two adverbs don’t com-
pete to play the same semantic role. Instead, one modifies a constituent Ca such that the
other one is already modifying a subconstituent of Ca. We can tell that hierarchical inter-
pretations of (40a-b) would be well formed semantically, because we can paraphrase
them with well-formed biclausal structures:

(41)0 a. It’s probably the case [that perhaps he will try again] (but perhaps it isn’t,
if he definitely WON’T try again).37
b. Perhaps it is probable [that he will try again].

Given the semantic and syntactic well-formedness of (41a-b), it is not particularly obvi-
ous why (40a-b) should strike us as so odd (even Cinque (ibid.) questions whether
probably and perhaps “co-occur[] at all” in this way).
That aside, is the Cinque hierarchy’s prediction concerning probably and perhaps
borne out? For me, although the judgment is quite difficult, (40a) is passable; taking this
sentence as a baseline, (40b) is marginal by comparison (“?” at worst). In other words, to
the degree that there is a difference, it goes in the direction predicted by Cinque. How-
ever, we should be wary of giving too much weight to these judgments, because they are
made delicate by the (not fully understood) oddness of probably and perhaps occurring
tautoclausally in the first place. Preliminary result: 


























































36
Thanks to Edwin Williams for discussion of this point.
37
The material in parentheses is added to give the paraphrase some supporting context. The if-clause spells
out explicitly the conditions under which it wouldn’t be the case that he would perhaps try again.

26

II.A.7.b. Modepistemic° (probably) > Modirrealis° ([almost] certainly)

As for probably and (almost) certainly, Cinque (1999:181, n. 88) writes that, “ac-
cording to Richard Kayne (personal communication), [they] seem to admit either order,
which, in the present context, may be problematic.” Although Cinque does not provide
evidence for Kayne’s claim, it seems to be correct, as shown by the following attested
examples.

(42)0 [probably > certainly]

a. Once you have that toaster, you probably will certainly not value modern
models, …38
b. And you probably will certainly see nobody else on the road.39
c. In order to be cautious, ensure you do not end up choosing a stylish wed-
ding decorator who has her place incredibly far out-of-the-way -- they
probably will certainly demand you a lot more.40

(43)0 [probably > almost certainly]

a. …but we probably will almost certainly find fossilised Bacteria, …41
b. …because they have no desire to commit war crimes and probably will
almost certainly never be in a position to do so, …42
c. This shopping cart software will need configuring for your desired pay-
ment gateway and shipping and probably will almost certainly need some
code modifications and template modifications.43

(44)0 [certainly > probably]

a. You might be thinking – “Yes, but my website will be much larger than
just 1 page”. That certainly will probably be the case, but…44
b. (although you certainly will probably be a little more aware of you[r]
footing so you don’t trip and fall into one of these babies!)45
c. Don’t forget, there is the Online Pokédex to help you on your way with
that, because you most certainly will probably need some help with that..
-gulps-46


























































38
http://techforwards.com/what-the-next-after-the-ipad-3.html
39
http://www.activethailand.com/destinations/mae-sot
40
http://www.mypartyplanner.com/search/vendor/Wedding_Decorations/Denver_find42.html
41
http://www.prometheusforum.net/discussion/191/what-would-extraterrestrial-life-look-like-or-be-like-if-
we-found-them/p1
42
http://www.uncsd2012.org/rio20/index.php?page=view&nr=664&type=510&menu=20&template=529&s
tr=Marine%20environment
43
http://www.woodburycinema.com/category/business/entrepreneurs
44
http://www.appnet.com/web-design-anderson-sc.php
45
http://www.virtualtourist.com/travel/North_America/United_States_of_America/California/Joshua_Tree_
National_Park-753376/Things_To_Do-Joshua_Tree_National_Park-Cacti-BR-1.html
46
http://pkmn.net/?action=content&page=viewpage&id=8847

27

(45)0 [almost certainly > probably]
a. …the internal FLASH memory almost certainly will probably not work
at these speeds.47
b. Chris Solomon and Jimmy Frazier each scored three touchdowns for the
back-to-back champions, who almost certainly will probably be moved
from the Southeast Division after this.48
c. …as you almost certainly will probably be working out by yourself at
home.49

Of course, these examples may turn out to differ from one another in important ways. For
instance, certainly cooccurring with a negative element ((42a-b)) could turn out to be
relevantly different from certainly alone. A detailed analysis of (42-45) would take us too
far astray. My purpose here has been to show that the problem highlighted by Cinque in
relation to probably and (almost) certainly is a real one. Preliminary result: 

II.A.8. Modirrealis° (perhaps) > Modnecessity° ([not] necessarily)50

This ordering statement predicts that perhaps should be able to precede necessarily,
which it can ((46a)). But of course, the prediction is stronger than that: perhaps should
obligatorily precede necessarily. As shown in (46b), this does not appear to be the case.

(46)0 a. They will perhaps necessarily be readmitted.

b. A: They will perhaps be readmitted.
B: Well, they’ll NECESSARILY PERHAPS be readmitted, because anything
COULD happen.
(based on Cinque 1999:89, (30a))

One might argue that (46b) constitutes a special case, since the constituent containing the
lower adverb (perhaps be readmitted51) has been repeated from prior discourse. After all,


























































47
http://forums.leaflabs.com/topic.php?id=31
48
http://www.insidesocal.com/tribpreps/2011/12/southeast-divis-3.html
49
http://www.supplement-reviews-online.com/supplement-reviews/workout-in-your-own-home-or-fitness-
center.html

50
Cinque (1999:89) seems to endorse the view that, in Italian sentences containing non necessariamente
‘not necessarily’, [Spec,ModnecessityP] hosts not just necessariamente but the entire sequence non neces-
sariamente, which would then have to be a constituent. But in the one well-formed Italian sentence he gives
involving the meaning ‘not necessarily…’ (ibid., (31a)), non and necessariamente are not adjacent, and
hence do not form a constituent on the surface. (They could conceivably form a constituent underlyingly,
but this would have to be argued for.) This raises the possibility that non and necessariamente (and their
English counterparts) never form a constituent, the non/not simply being the clausal negation. It seems to
me that, on either view, we have to stipulate as a lexical fact about the English necessarily that it only (or
predominantly) occurs outside the scope of negation in high registers. (I am grateful to Anna Tchetchetkine
(p.c., spring 2011) for calling my attention to this fact.)
51
On fairly traditional generative assumptions, (46bA) and (46bB), simplifying somewhat, are underlyingly
[perhaps [will [be [readmitted they]]]] and [necessarily [perhaps [will [be [readmitted they]]]]]. On this
view, the constituent containing the lower adverb in (46bB) is still repeated from prior discourse, but is
larger than it appears to be on the surface.

28

it is known that REPEATED MATERIAL (as I will call it52) can differ grammatically from
nonrepeated material. For instance, the auxiliary will is normally barred in if-clauses
((47a-b)), but this restriction is suspended if the complement of if is repeated material
((47c)):53

(47)0 a. *If Kelly arrives late, I’m going to be really angry.

b. *If Kelly will arrive late, I’m going to be really angry.
c. A: Brian will be speaking.
B: Oh, well if Brian will be speaking, then we should all go!

But these observations do not in themselves make the necessarily > perhaps problem dis-
appear. Advocates of Cinque’s hypothesis need to give some explanation of why the or-
dering statement perhaps > necessarily can apparently be disregarded in contexts like
(46b).
As a matter of fact, the Cinque-noncompliant54 order necessarily > perhaps can oc-
cur even when perhaps is not part of a repeated constituent:

(48)0 You know, we have some reason to be optimistic about tomorrow. Because any-
thing COULD happen, Claire necessarily perhaps will WIN, and we can take some
comfort in that.

Preliminary result: 

The ordering statement under discussion also predicts that perhaps should be able to
precede not necessarily. This prediction is correct ((49a)). However, as in the previous
case we considered, the opposite order is also possible if the lower adverb (here perhaps)
is part of a repeated constituent ((49b)).55


























































52
I will also use the term REPEATED CONSTITUENT, and I will refer to the action that gives rise to repeated
material as DISCOURSE REPETITION.
53
Thanks to Salvador Mascarenhas for discussion of this phenomenon.
54
This term is modeled after Dékány’s (2011:306, 312) term CINQUE-COMPLIANT, but used in a different
sense. Dékány uses Cinque-compliant to describe movements within the extended projection of N that
comply with the restrictions on such movements put forth in Cinque (2005:321, (7)).
55
The crucial sentence in (49b) is the third conversational turn. The necessarily and perhaps in the second
are not tautoclausal (as shown by the bracketing). They therefore tell us nothing about the Cinque hierar-
chy, which orders elements within clauses, not across clause boundaries.

29

(49)0 a. The rich perhaps aren’t necessarily happy.56

b. A: What about the rich—are they happy?

B: Well, I don’t know any rich people, so it would seem that, a priori, [I
would necessarily have to say [that they are PERHAPS happy]].
A: They’re not NECESSARILY perhaps happy—maybe we can rule out
the possibility of their being happy before even meeting any, just by
using our reason.
(based on Cinque 1999:89, (31a))

Interestingly, the order not necessarily > perhaps is bad if perhaps is not part of a re-
peated constituent ((50)). Here, not necessarily is not patterning like necessarily (cf.
(48)).

(50)0 You know, we have some reason to be optimistic about tomorrow.

a. *Claire is not necessarily perhaps going to LOSE, and we can take some
*comfort in that.
b. *Claire is not necessarily even PERHAPS going to lose, and we can take
*some comfort in that.

(50a-b) are unacceptable because perhaps cannot be c-commanded by a clausemate nega-

tive element,57,58 as shown in (51-53).

























































56
I treat =n’t necessarily as equivalent to not necessarily, as I am not aware of any differences between
them that would be relevant here.
57
The following attested sentences are apparent counterexamples to this generalization:

(vii) a. It wasn’t perhaps the height of excitement, nor the riveting adventure that gets immortalized in
print for all time.
http://captainstupie.blogspot.com/2012/01/it-wasnt-perhaps-height-of-excitement.html

b. The outdoors wasn’t perhaps as sinister as the original cathedral.

http://www.mmo-champion.com/threads/1119846-Having-played-the-open-beta-do-you-like-
it/page2

One possible analysis here is that the perhaps in each of these sentences is being used parenthetically (de-
spite not being set off by any punctuation). After all, an adverb that is being used parenthetically can often
occur lower than it would be able to otherwise (cf. (20f)). This analysis immediately raises the question of
why comma intonation, which should force a parenthetical structure, does not seem to be able to salvage
(51b) and (52b). The relevant sentences, which impressionistically seem pragmatically strange, are the fol-
lowing:

(viii) a. #Neil isn’t, perhaps, flying right now.

b. #She wasn’t, perhaps, singing at that precise moment.

One possibility here is that the problem with (viiia-b) is the following. Parentheticals must convey “sup-
plementary” or nonessential information. Perhaps therefore makes a poor parenthetical in (viiia-b), because
an adverb that so seriously undermines the credibility of a factual claim cannot possibly be considered non-
essential. In (viia-b), on the other hand, the main assertion is a subjective evaluation rather than a factual
claim, and a perhaps modifying such an assertion can be construed as nonessential (and hence used paren-
thetically) because we do not attach too much importance to whether a speaker is truly committed to his or

30

(51)0 a. *Neil perhaps isn’t flying right now.
b. *Neil isn’t perhaps flying right now.
(52)0 a. *She perhaps wasn’t singing at that precise moment.
b. *She wasn’t perhaps singing at that precise moment.59
(53) a. *Perhaps nobody was there.
b. *Nobody perhaps was there.

To recapitulate briefly, the order not necessarily > perhaps in (50a-b) is ill formed
because perhaps cannot occur under a clausemate negation. (This generalization does not
apply in metalinguistic negation contexts (fn. 58), or when perhaps is being used paren-
thetically—in which case, however, it is subject to certain constraints on parentheticals
(fn. 57).) Something more needs to be said, though, to account for why not necessarily >
perhaps is fine when perhaps is part of a repeated constituent ((49b)). I leave this an open
question.
At any rate, although the unacceptable (50a-b) are not counterevidence to Cinque’s
ordering perhaps > not necessarily, (49b) is, absent a convincing argument for why ad-
verbs in repeated constituents should be exempt from having to obey the Cinque hierar-
chy. Preliminary result: 

II.A.9. Modnecessity° ([not] necessarily) > Modpossibility° (possibly)

Both orders are possible when the lower adverb is part of a repeated constituent:

(54)0 A: I think he possibly will outdo all his opponents.

B: Well, he NECESSARILY will POSSIBLY outdo them—ANYTHING’S possible!

(55)0 A: I think he necessarily will outdo all his opponents.

B: Well, he POSSIBLY will necessarily outdo them.


























































her subjective evaluation. I cannot here investigate these issues further to determine whether this pragmatic
account is on the right track.
58
Perhaps can occur under negation in a metalinguistic negation context:

(ix) A: The city will perhaps have been destroyed by then.

B: It won’t PERHAPS have been destroyed by then—it’ll DEFINITELY have been destroyed by
then!
59
(51b) and (52b) are not obviously ruled out on semantic grounds. Perhaps can perfectly well occur under
not semantically, as revealed by biclausal structures:

(x) a. It’s not the case [that Neil is perhaps flying right now]. (cf. (51b))
b. It’s not the case [that she was perhaps singing at that precise moment]. (cf. (52b))

31

(56)0 A: He’s also half human, half wolf, meaning he possibly eats babies.60
B: Well, he doesn’t NECESSARILY possibly eat babies—maybe we can rule
out the possibility that he eats babies, just by using our reason.

(57)0 A: She didn’t necessarily rob the bank.

B: Well, she POSSIBLY didn’t necessarily rob the bank, but maybe if we think
about it hard enough, we’ll discover that in fact she DID necessarily rob the
bank.

But as it turns out, both orders seem possible even in the absence of discourse repeti-
tion:

(58)0 A: *Okay, so what do we know about James?

B: a. *Well, he necessarily will possibly have gone home by then.
b. *Well, he possibly will necessarily have gone home by then.

(59)0 A: *Okay, so what do we know about James?

B: a. *Well, he won’t necessarily have possibly gone home by then.
b. *Well, he possibly won’t necessarily have gone home by then.

The outlier here is (59Ba). Above, we argued that the order not necessarily > per-
haps is bad (except, for some reason, when perhaps is part of a repeated constituent) be-
cause perhaps cannot be c-commanded by a negative element in the same clause. The ill-
formedness of not necessarily > possibly in (59Ba) can be explained analogously, since
possibly also cannot be c-commanded by a negative element in the same clause:61,62


























































60
This sentence is adapted from <http://www.cracked.com/article_15201_mock-draft-top-11-movie-basket
ball-players-all-time.html>.
61
Once again (cf. fn. 59), the problem with (60b) is not obviously semantic, since biclausal paraphrases of
this sentence are fine:

(xi) a. It’s not possible [that Emily eats hamburgers].

b. It’s not the case [that Emily possibly eats hamburgers].
62
Unsurprisingly, the ban on possibly appearing under negation is suspended in metalinguistic negation
contexts (cf. fn. 58):

(xii) A: Chase is possibly drinking coffee right now.

B: He’s not POSSIBLY drinking coffee right now—he’s DEFINITELY drinking coffee right now.

For some reason, the ban is also suspended in the presence of certain modals, as shown below. (“»” means
‘outscopes’.)

(xiii) a. I can’t possibly go. (NEG » possibly » can)

b. I couldn’t possibly go there. (NEG » possibly » could)

32

(60) a. *Emily possibly doesn’t eat hamburgers.
b. *Emily doesn’t possibly eat hamburgers.
(61) a. *Possibly nobody was there.
b. *Nobody possibly was there.

What hasn’t been explained is the fact that the order not necessarily > possibly is permis-
sible when the lower adverb (possibly) is part of a repeated constituent ((56)). Prelimi-
nary result: 

II.A.10. Modpossibility° (possibly) > Asphabitual° (usually)

The predictions of this ordering statement seem to be correct:

(62)0 a. *Olivia possibly usually surfs on Sundays.

b. *Olivia usually possibly surfs on Sundays.

Preliminary result: 

II.A.11. Asphabitual° (usually) > Asprepetitive(I)° (again)

A complication arises when testing Cinque’s predictions concerning again. The

complication is that this adverb, on Cinque’s hypothesis, occupies two distinct slots in the
F-structure of the clause: [Spec,Asprepetitive(I)P] and [Spec,Asprepetitive(II)P]. Now, if any ad-
verb could be assigned multiple positions, then the Cinque hierarchy would be vacuous.
However, Cinque holds that an adverb can occupy multiple positions in the hierarchy, but
these positions must be associated with different interpretations (cf. (20d)). In this case,
Cinque argues:
Both types of repetition[63] may quantify over events in which a certain process is involved,
or over the process itself, quantification over events being located higher than quantification
over the process. […] This is not to say that there are two [oftens]. Rather, it seems plausible
to take the same lexical item to be generable in two distinct (scope) positions. [endnote omit-
ted] (Cinque 1999:92)

As we can see, Cinque posits that there is only a single lexical item again, but it acquires
different scope properties depending on which position it’s occupying.64

























































63
I.e., those described by the adverbs again and often, each of which has two slots in the Cinque hierarchy.
—EZ
64
A related issue is how to account for the cases where the “same” adverb seems to take on different mean-
ings depending on its position, beyond just quantifying over different constituents. An example of this is
honestly (Vendler 1984:306, via Cinque 1999:205, n. 41), which has speech act, subject-oriented, and man-
ner interpretations ((xiva-c), respectively). These three honestlys can cooccur ((xv)).

(xiv) a. Honestly, I just don’t see the point.

b. Emma (quite) honestly has returned the wallet to its rightful owner. (= (vb), fn. 31)
c. Max recounted the events honestly.

(xv) Honestly, Lara, honestly enough, confessed to her misdeed honestly.

33

 The hypothesis that again occupies two positions in the Cinque hierarchy predicts
that it should be possible to fill both of these positions at once. This prediction is appar-
ently borne out:

(63)0 Then, Robin again [proofread the letter and then proofread it again].

When testing an ordering statement involving an adverb that Cinque assigns to two
different positions, it is helpful to simply realize the adverb in both of them, to make sure
we know which position we’re dealing with at all times. Applying this strategy to the pre-
sent case, we expect usually to obligatorily precede both agains (cf. the full hierarchy in
(18)). This prediction is not borne out:

(64)0 a. When I run into Alyssa, she usually has again [gotten into trouble and
then gotten into trouble again].
b. A: Last year, Jeff usually SURFED on Sundays, but now he usually FISHES on
Sundays.
B: Nope—NOW he again usually SURFS on Sundays.

However, Cinque (1999:204, n. 36) argues that, like again and often, habitual adverbs
such as usually can be generated in two distinct positions, one quantifying over the proc-
ess and the other quantifying over the event involving that process. His evidence for this
is the fact that usually can either precede or follow already, no longer, and still, coupled
with the fact that, in Italian, the two positions are filled by different, though “apparently
synonymous,” adverbials: di solito or solitamente (higher) and abitualmente (lower).
The predictions of this analysis are slightly difficult to test, because the lower of the
two positions is for some reason not integrated into the total hierarchy that I have repro-
duced in (18). However, given that (18) entails the partial ordering usually > again > of-
ten > again > often (with each again higher than its structurally nearby often), the natural
hypothesis is that integrating the lower usually into this partial ordering would yield usu-
ally > again > often > usually > again > often—in other words, that both “bundles” con-
sisting of a usually, an again, and an often have the same hierarchical structure.
With this hypothesis in mind, let us test the ordering usually1 > again165—using sen-
tences that also contain usually2 and again2, so that we know which adverbs we’re deal-
ing with. The Cinque-compliant order, usually1 > again1 > usually2 > again2, is well
formed ((65a)). However, it appears that we can also construct well-formed sentences
featuring the Cinque-noncompliant order, namely again1 > usually1 > usually2 > again2
(see (65b)).


























































Cinque argues that there is only a single honestly, which has a core meaning common to all three of its
uses, and this meaning combines with the semantic features of the three different positions under discussion
to yield the specialized interpretations we observe in (xiva-c) (the speech act, subject-oriented, and manner
readings of honestly). Cinque admits, however, that we do not yet have a full account of how this would
work (Cinque 1999:205, n. 41).
65
I use numerical subscripts to distinguish the different “versions” of adverbs that putatively occupy more
than one slot in the Cinque hierarchy (following Cinque 1999:27). The lower the number, the higher the
adverb (relative to other versions of the same adverb).

34

(65)0 a. When Nicole emerges from a bout of depression, she usually1 again1 [en-
joys life and usually2 [has pizza for breakfast and then has pizza again2 for
lunch]].66
b. After Nicole has emerged from a bout of depression, she again1 usually1,
in any given month, will usually2, by the end of any given night, have
[[jumped for joy and then jumped for joy again2] many times].

One could argue that (65b) is a special case because each of the two usuallys is immedi-
ately followed by a PP that serves as its restrictor, specifying what kinds of cases it’s
quantifying over (on adverbs of quantification, see Lewis 1975 and Partee 1991). But it is
not obvious how this reasoning would eliminate the counterexample: why should these
PPs salvage a putatively impermissible adverb order? Furthermore, it is not as though the
usuallys in (65a) did not have restrictors. For the usually1, it is the when-clause; and the
usually2 is clearly quantifying over days, as evidenced by the mentions of breakfast and
lunch. Preliminary result: 

II.A.12. Asprepetitive(I)° (again) > Aspfrequentative(I)° (often)

Because the Cinque hierarchy entails the ordering again1 > often1 > again2 > often2,
we will test the ordering again1 > often1 using sentences that also contain again2, to make
sure that we are in fact looking at again1 and often1.
The ordering statement again1 > often1 initially looks promising:

(66)0 a. *Kyle again1 often1 [eats meat on Monday and then eats meat again2 on
*Tuesday].
b. *Kyle often1 again1 [eats meat on Monday and then eats meat again2 on
*Tuesday].

Cinque describes again1 as quantifying over an event (as opposed to the process that that
event involves). In (66a), it appears to be quantifying not over an event but over a state of
affairs, as suggested by the following biclausal paraphrase:

(67)0 There is again a state of affairs involving [Kyle often eating meat on Monday
and then eating meat again on Tuesday].

But surely this is not a problem: we can simply note that again1 quantifies over events or
states (of affairs), whereas again2 quantifies over the processes they involve.67 In this
light, the unacceptability of (66b) is quite interesting. Based on the meaning of again1,
we might expect (66b) to mean the following (any of the three paraphrases will do,
though (68c) may be the clearest):


























































66
The hierarchical structure here, indicated by bracketing, is crucial for getting the intended interpretation.
For instance, the phrase beginning with usually2 is not coordinated with the phrase beginning with usually1,
but with [enjoys life]. The resulting coordinate structure, which contains usually2, is modified by again1.
67
But cf. Cinque (1999:205, n. 39).

35

(68)0 a. Often, [there exists a state of affairs such that [Kyle (habitually) eats meat
on Monday and then eats meat again on Tuesday], that state of affairs hav-
ing existed previously, temporarily ceased to exist, and then come into ex-
istence again].

b. Often, [it is true of Kyle [that he (habitually) eats meat on Monday and
then eats meat again on Tuesday], that having been true of him previously,
temporarily not been true of him, and then been true of him again].

c. Let S be a state of affairs such that Kyle (habitually) [eats meat on Mon-
day and then eats meat again on Tuesday].

Let S′ be the state of affairs of [[S existing] under the circumstance of hav-
ing existed previously, temporarily not existed, and then come into exis-
tence again].

The state of affairs S′ exists often.

But (66b) cannot have such a meaning, which supports the ordering again1 > often1
(again1, again, being the again that quantifies over events or states rather than proc-
esses).68 Preliminary result: 
There is an empirical problem, though. The Cinque hierarchy entails the partial or-
dering again1 > often1 > again2 > often2. This predicts that we should never get an often
higher than an again higher than another again. This prediction seemed to be confirmed
by (66b) (on the interpretation under discussion), but it seems to be disconfirmed by the
following:

(69) Jack often [[drops his pen and then drops his pen again] and then again [drops
his pen and then drops his pen again]].

Suppose that the partial Cinque ordering again1 > often1 > again2 > often2 is correct.
Then, because the often in (69) has agains below it, it must be often1. Therefore, all three
of the agains in (68) must be instances of again2, which quantifies over the process rather
than the event. But this is at least prima facie a problem. Cinque’s hypothesis is that the
various slots specialized for different classes of adverbs are arranged in a rigid universal


























































68
Often1 cannot appear above again1 even if we construct a context that should pragmatically favor the in-
terpretation that such a structure would have:
(xvi)0 A: *Two years ago, Kyle often1 [ate meat on Monday and then ate meat again2 on Tuesday].
*Last year, this was no longer the case. But this year, he AGAIN1 often1 [eats meat on Mon-
*day and then eats meat again2 on Tuesday].
B: *Well, that’s no surprise—these sorts of cycles have repeated themselves so often with Kyle
*that I think it’s fair to say that he OFTEN1 again1 often1 [does things and then does them
*again2 shortly thereafter].



36

order. It seems, then, that Cinque must allow at least some of his functional projections,
including Aspfrequentative(II)P, to iterate—as Rizzi (1997:295-297) does his two TopPs.69

II.A.13. Excursus: Asphabitual° (usually) > Aspfrequentative(I)° (often)

This ordering prediction arises by transitivity from the partial ordering usually1 >
again1 > often1.
Assuming that the Cinque hierarchy entails the partial ordering usually1 > often1 >
usually2 > often2, we should test the ordering usually1 > often1 using sentences that also
contain usually2, to make sure we’re dealing with the right adverbs. The aforementioned
partial ordering predicts that it should be impossible for an often to be followed by two
usuallys, but this does not seem to be true:

(70)0 When Kristen moves to a new city,

a. she usually will often, for any given book club that she joins, usually not
read the books.
b. she often will usually, for any given book club that she joins, usually not
read the books.

In (70a-b), the highest adverb quantifies over instances of Kristen moving to a new city,
and the lowest adverb (usually) quantifies over book club meetings. Preliminary result: 

II.A.14. Aspfrequentative(I)° (often) > Modvolitional° (intentionally)

The Cinque hierarchy entails the partial ordering often1 > intentionally > often2. We
will test the ordering often1 > intentionally in sentences also containing often2, so that we
always know which often we’re dealing with.
The interpretations that the following sentences receive suggest that Cinque is cor-
rect to claim that the lower often quantifies over the process and the higher often quanti-
fies over the event (or, as we have noted, the state) involving that process (cf. the discus-
sion in §II.A.11 preceding (63)).

(71)0 a. *When visiting Cassie, Bill often intentionally [insults her often].
b. *When visiting Cassie, Bill intentionally often [insults her often].

(72)0 a. *Bill often intentionally [insults Cassie often].

b. *Bill intentionally often [insults Cassie often].

The judgments in (71-72) are relativized to the readings indicated by the bracketing.
On these readings, the linearly second often is interpreted under the linearly first often. In
Cinque’s terms, the later often is often2, quantifying over the process, and the earlier one
is often1, quantifying over events involving that process.
However, the linear strings in (72), at least, can be assigned different hierarchical
representations:

























































69
If this is so, then it is no longer obvious whether the higher again in (63) is again1 or again2, or ambigu-
ous between them.

37

(73)0 a. *Bill [often intentionally insults Cassie] often.
b. *Bill [intentionally often insults Cassie] often.

These structures have interpretations on which the core VP70 combines with the linearly
first often (by hypothesis, often2) before it combines with the linearly later one (by hy-
pothesis, often1). In Cinque’s worldview (which is essentially antisymmetric; cf. fn. 12),
the linearly later often cannot be right-adjoined to the bracketed constituents in (73a-b).
Instead, these sentences must respectively be derived from the following underlying
structures (I ignore irrelevant complications):

(74)0 a. *Bill often1 [often2 intentionally insults Cassie].

b. *Bill often1 [intentionally often2 insults Cassie].

by in each case moving the bracketed constituent to the left of often1.

Now, the putative underlying structure (74a) is not compatible with the Cinque hier-
archy as laid out in (18), because it does not conform to the partial ordering often1 > in-
tentionally > often2. Therefore, the Cinque hierarchy predicts that (73a), which by hy-
pothesis is derived from (74a), should be unacceptable. For me, however, there is no
strong contrast in acceptability between (73a) and (73b); both are fine.71 Preliminary re-
sult: 
It is not obvious how Cinque would deal with this problem. One possibility would be
to add another Modvolitional° head beneath Aspfrequentative(II)°, and argue that the two Modvoli-
tional° heads have different interpretations, having to do with (non)volitionally bringing
about a process (the lower head) or a state/event (the higher head). This solution has the
advantage that the revised partial order often1 > intentionally1 > often2 > intentionally2
would not rule in the unacceptable (71b) and (72b), which feature the order intentionally
> often > often. I will not here explore whether the further predictions of this solution are
correct.

II.A.15. Modvolitional° (intentionally) > Aspcelerative(I)° (quickly)

Here again we need to be careful, because quickly can also appear in two different
positions—for Cinque, [Spec,Aspcelerative(I)P] and [Spec,Aspcelerative(II)P]. Unsurprisingly by
now, the lower quickly describes the process, the higher one the event:

(75)0 a. He has been slowly testing some bulbs. (Cinque 1999:20, adapted from
b. He has been testing some bulbs slowly. Thomason & Stalnaker 1973:200)

Regarding these sentences, Cinque (ibid.) writes: “In [(75a)], slowly qualifies (has
scope over) the entire event (each test could well have been rapid), whereas in [(75b)], it
qualifies each test individually [endnote omitted].”
As predicted, both positions can be filled at once:

























































70
Here and throughout, I use the term VP in a fairly traditional sense. I distinguish it from IP/TP, but ab-
stract away from the question of whether what I am calling VP is really, say, vP.
71
My judgment for (77a) is made particularly reliable by the fact that it was arrived at at a time when I had
completely forgotten that the Cinque hierarchy predicted the sentence to be bad.

38

 (76)0 a. Sarah quickly raised her hand fast.72
b. Sarah quickly raised her hand slowly.73

As usual, the ordering at hand will be tested using sentences that fill both positions.

(77)0 a. Sarah intentionally quickly raised her hand fast.

b. Sarah quickly intentionally raised her hand fast.

Cinque’s prediction here is that (77a) should be acceptable and (77b) unacceptable—or at
least that (77a) should be significantly better. The sentences are quite unusual and the
judgments are subtle, but I think it is fair to say that this prediction is not really borne out.
(77b) does not seem worse, provided it is given a hierarchical interpretation (with fast
modifying the core VP, the resulting constituent being modified by intentionally, and the
constituent resulting from that being modified by quickly). Preliminary result: 
Here too (cf. the previous section) Cinque might find it useful to posit a second Mod-
volitional° head, yielding the partial ordering intentionally1 > quickly1 > intentionally2 >
quickly2.

II.A.16. Aspcelerative(I)° (quickly) > T(Anterior)° (already)

At first this prediction seems to be roundly disconfirmed:

(78)0 a. *Matt had quickly already eaten the cake.

b. *Matt had already quickly eaten the cake.

Now, (78b) is not a problem if the quickly is in [Spec,Aspcelerative(II)P] (the lower of the
two positions that quickly can occupy), describing the process rather than the event. Be-
cause quickly in this position gets a manner interpretation, this analysis predicts that (78b)
should be paraphrasable as follows:

(79)0 Matt had already eaten the cake, and he had done it quickly.

This prediction is correct. Sentences containing subject-oriented quickly, by contrast,

cannot be paraphrased analogously. Thus (80a) cannot be paraphrased by (80b):74


























































72
Quickly and rapidly can occupy either position, but fast is restricted to the lower one. For examples show-
ing this, see Cinque (1999:103, §4.25); for qualification of this claim, see ibid:212, n. 73.
73
A possible concern here is that the quickly in (76b) does not obviously describe the event the way the
slowly in (75a) does, as it is difficult to see how the event of Sarah raising her hand could have occurred
quickly if she raised it slowly. The quickly in (76b) instead seems to indicate that Sarah was quick in raising
her hand slowly. In other words, it appears that this quickly is a subject-oriented adverb, not one describing
the event. In the cartographic worldview, this suggests that perhaps the quickly in (76b) and the slowly in
(75a) do not in fact occupy the same position.
74
It is not surprising that paraphrasability by …and DP did it Adv is a diagnostic for manner adverbs, since
manner adverbs can appear after the core VP, but their subject-oriented counterparts cannot, unless they are
right-dislocated (Nick had misanalyzed the data, cleverly).

39

(80)0 a. #Sarah quickly raised her hand slowly.
b. #Sarah raised her hand slowly, and she did it quickly.

(80b) would only be appropriate when discussing circumstances under which the Law of
Noncontradiction for some reason didn’t apply (in a very strange alternate universe, per-
haps).
We have just determined that (78b) is not a problem for the ordering at hand. There-
fore, the paradigm in (78) can be made compatible with the Cinque hierarchy if we can
explain why (78a) is unexpectedly unacceptable. Intuitively, it would seem that (78a)
could be ruled out on semantic grounds, but matters are not so simple.
The semantic argument would go as follows. If the higher quickly is subject-oriented
(as seems to be the case; cf. fn. 73), then (78a) is sharply semantically anomalous. This is
because it is impossible to be quick in already doing something (or rather, in already hav-
ing done something) in the same way that one can be quick in raising one’s hand. There-
fore, it is entirely possible that the ordering statement at hand is right, and (78a) is unac-
ceptable because it crashes at the level of semantic interpretation.
The problem with this argument is that the semantic anomaly it invokes is not real.
Suppose we are discussing a (presumably fictional) alternate universe where people have
the ability to change their own past at will. In other words, if I am an inhabitant of this
universe, and I get to class only to discover that I have forgotten my backpack, I can sim-
ply make myself have brought it. That is, I can, in the present, alter my past, with the re-
sult that I have brought my backpack after all. In discussing this universe, we can use ei-
ther of the following two biclausal structures:

(81)0 a. ?And then Matt was quick [to have already eaten the cake].
b. ?And then Matt was quick in [having already eaten the cake].

where the matrix predicate has the same interpretation as in

(82)0 a. ?And then Matt was quick [to raise his hand].
b. ?And then Matt was quick in [raising his hand].

Admittedly, (81b) is for me more marginal than (81a). But both of them are clearly vastly
better than (78a), repeated here in slightly modified form:

(83)0 (?)*And then Matt had quickly already eaten the cake.

Interestingly, (83) improves if had stays below quickly:

(84)0 (?)?And then Matt quickly had already eaten the cake.

In the context under discussion, (81) and (83-84) are all about someone being quick to
have already done something (roughly, “subject-orientedness above anteriority”). The
contrast between (83) and (84) suggests that, if this is the type of meaning we want to
convey, it is a bad idea to place an overt marker of anteriority (here had) above the sub-

40

ject-oriented adverb (quickly). This is apparently an example of grammar being iconic, or
mirroring, since it is here demanding a matchup between meaning and form.75
Let us take stock. Cinque predicts that the subject-oriented quickly (i.e., quickly1)
should precede already. This configuration led to deviance in (78a). Therefore, the
quickly/already ordering data can only be made consistent with the Cinque hierarchy if
the order quickly1 > already is ruled out independently, for instance by semantics. The
biclausal structures in (81) showed that this order is not semantically anomalous, just in-
applicable to ordinary human experience. In an alternate-universe context, it’s fine.
Therefore, the same context should license the order quickly1 > already within a single
clause. But although (84) is better than (83), it is still deficient76 compared to its quite un-
objectionable biclausal paraphrase (81a). Therefore, something nonsemantic must be
wrong with the order quickly1 > already intraclausally, contrary to Cinque’s prediction.
Preliminary result: 

II.A.17. T(Anterior)° (already) > Aspterminative° (no longer)

This ordering predicts that already should be able to precede no longer, which it can
((85)). It also predicts, however, that the reverse order should be impossible. This is not
the case. No longer can precede already, both when the lower adverb (already) is part of
a repeated constituent ((86)) and when it is not ((87)).

(85)0 John already no longer wins all his games.

(based on Cinque 1999:181, n. 89, (ia))

(86)0 At first it was the case that, in 2011, Carol had already published a book. But after
the time travelers altered the past, Carol had NO LONGER already published a
book in 2011.

(87)0 Thanks to the wonders of time travel and Brad’s diabolical scheme, Carol has no
longer already finished her homework.

Preliminary result: 

II.A.18. Aspterminative° (no longer) > Aspcontinuative° (still)

As Cinque (1999:95, §4.17) points out, one might expect the antonymous AdvPs no
longer and still to occur in exactly the same position—in Cinque’s system, the specifier
of a single functional head. Cinque argues, however, that each one corresponds to a dis-

























































75
Of course, grammar does not always display this property. In the sentence Julia was evidently here, the
adverb evidently appears beneath past tense but is interpreted above it, as revealed by the paraphrase It is
evident [that Julia was here]. I owe the gist of this example to Edwin Williams (class, fall 2010).
76
(84) improves if a pause is placed after quickly:

(xvii) (?)And then Matt quickly - had already eaten the cake.

But this could well be an artifact of processing, and does not let (79) off the hook: normal two-adverb sen-
tences do not require pauses in order to be acceptable.

41

tinct functional head, because their Italian counterparts can marginally cooccur in one
order, but not in the other:

(88)0 a. ‘?Arrabbiato con me, non lo sarai più ancora, spero!

?‘angry with me not it you.will.be more still I.hope
?‘Angry with me, you will no longer still be, I hope!’
(Cinque 1999:207, n. 52, (ia))

(83)0 b. ‘*Arrabbiato con me, non lo sarai ancora più, spero!

?‘angry with me not it you.will.be still more I.hope
(ibid., (ib)).

This decision on Cinque’s part is criticized by Manninen (2005), who argues that because
the supposedly good order ((88a)) is only marginal, it would be better to associate both
AdvPs with a single functional head after all, thereby simplifying the hierarchy.
In English, both orders are quite acceptable when the lower AdvP is part of a re-
peated constituent:

(89)0 A: Were you mad at me on Monday?

B: Yup.
A: How about Tuesday?
B: I was still mad at you then.
A: Well, I hope you’re NO LONGER still mad at me!

(90)0 A: Were you mad at me on Monday?

B: Yup.
A: How about Tuesday?
B: No, I was no longer mad at you then.
A: Well, I hope you’re STILL no longer mad at me!

In each of (89-90), the repeated constituent under discussion is (by definition) old in-
formation. As a result, it is anaphorically destressed (Ladd 2008:231ff., §6.2.2.1), and
hence its sister, the higher adverb, is stressed. Under normal prosodic conditions, on the
other hand—for instance, in an out-of-the-blue context—no longer and still seem to be
unable to cooccur in either order. This is shown in (91a-b), where primary and secondary
stresses are explicitly indicated:

(91)0 a. ?*You ˈknow, I ˈbet ˈFred ˈno ˌlonger ˈstill ˈsleeps with a ˈteddy ˌbear.

b. ??You ˈknow, I ˈbet ˈFred ˈstill ˈno (ˌ)longer ˈsleeps with a ˈteddy ˌbear.

Although (91a-b) could in theory have perfectly sensible hierarchical interpretations, with
the lower adverb interpreted in the scope of the higher one (cf. (89-90)), these sentences
are in fact quite degraded. If there is a contrast between them, (91a) is worse: it seems not
only structurally degraded but also contradictory, as though the speaker were trying to
simultaneously make two versions of the same assertion, one involving no longer, the

42

other involving still. This particular fact makes it seem as if no longer and still were
competing for the same position after all.
To summarize: in Cinquean terms, no longer and still correspond either to a single
head or to two. To the degree that there is evidence in favor of two separate heads (Asp-
terminative° and Aspcontinuative°), Italian suggests that they occur in that order, whereas Eng-
lish suggests the opposite.
Meanwhile, the two AdvPs can cooccur in either order in English when the lower
one is part of a repeated constituent. This could possibly be explained by positing that
both of these adverbials can be “directly attached” to a constituent (in this case, [mad at
me]77), as in (20e). But if this is so, then why isn’t direct attachment available in (91a-b),
rendering them perfect? Preliminary result: 

II.A.19. Aspcontinuative° (still) > Aspperfect(?)° (always)

The predictions of this ordering statement appear at first glance to be incorrect:

(92)0 a. Eliza still always roots for the Yankees.

b. After breaking up with a girl, Chad always still loves her.

Now, the data in (92) may not in fact be problematic for Cinque. This is because he ex-
plicitly claims that the Italian sempre ‘always’, being a quantificational adverb, may be
like usually and often in being generable in two different positions, the lower one quanti-
fying over the process, the higher one quantifying over the event (Cinque 1999:169, n.
12). However, he does not, as far as I can tell, explicitly indicate where the two positions
are, making this prediction very difficult to test. (There is only one always in (18).) I will
therefore mark this subsection “Preliminary result: (),” to indicate that, while the
Cinque hierarchy may well be able to deal with the facts in (92), this must be shown ex-
plicitly.

II.A.20. Aspperfect(?)° (always) > Aspretrospective° (just)

This prediction is correct: English allows the order always > just ((93)) but not just >
always ((94a)), and the latter order cannot be improved by focusing just ((94b)). (The
judgments given in (93-94) apply only to the interpretations of these strings involving
retrospective just, as in I don’t WANNA go there—I just went.)

(93)0 a. *Whenever I go over to Jessica’s, she’s always just finished some construc-
*tion project.

(94)0 a. *Jeff just always surfed on Sundays.

b. *Jeff JUST always surfed on Sundays.

Preliminary result: 


























































77
Or [you mad at me] on the Predicate-Internal Subject Hypothesis; cf. fn. 51.

43

II.A.21. Aspretrospective° (just) > Aspproximative° (soon)

It is difficult to test the positive prediction of this ordering statement (namely, that
just can precede soon) without producing semantic anomaly: doing so involves embed-
ding a future-oriented adverb (soon) under a retrospective one (just). This can, however,
be done, as in (95), which I find acceptable:

(95)0 I can’t believe it! I just fulfilled a promise soon relative to when I made it!

The acceptability of (95) is probably easiest to bring out by prosodically phrasing soon
with the material following it,78 as shown ad-hocly in (96a). If soon is instead prosodi-
cally phrased with the material preceding it ((96b)), the structure is still acceptable, but
the relative (to)–phrase is interpreted as focused.

(96) a. I can’t believe it! I just fulfilled a promise · soon relative to when I made it!
b. I can’t believe it! I just fulfilled a promise soon · [relative to when I MADE
it](, rather than [relative to some OTHER point])!

As we have seen, the positive prediction of the ordering statement (just > soon) is
correct. Its negative prediction, however—*soon > just—is not:

(97)0 I will soon have just gone to bed.

Preliminary result: 

II.A.22. Aspproximative° (soon) > Aspdurative° (briefly)

(98)0 a. ?*I will soon briefly tell you my life story.

b. ?*I will briefly soon tell you my life story.

(98b) remains quite degraded even when the lower adverb (soon) is part of a repeated
constituent:

(99)0 A: I will soon tell you my life story.

B: Aw, no!
A: No, don’t worry! ?*I will BRIEFLY soon tell you my life story.79

Preliminary result: 


























































78
I find that this is easiest to do if soon is preceded by a short pause.
79
If some speakers happen to find the briefly soon–sentence in (99) better than the one in (98b), this could
be for the following reason. Because the stress pattern in (99) marks the surface constituent [soon tell you
my life story] as old information, these speakers are not processing it with as much attention as they would
otherwise. As a result, they are less aware that they have heard the sequence briefly soon, which is the
source of the sentence’s unacceptability.

44

II.A.23. Aspdurative° (briefly) > Aspgeneric/progressive° (characteristically)

According to Comrie (1976:41) (via Cinque 1999:98, §4.20), DURATIVE ASPECT

marks a situation as “last[ing] for a certain period of time (or at least, […] conceived as
lasting for a certain period of time).” Cinque (ibid.) argues that the class of durative ad-
verbials encompasses both 1) syntactically simplex elements such as long and briefly and
2) syntactically complex expressions such as for an hour, for a day, for a month, and for
a while.80 With this in mind, consider the following sentence:

(100) In 2011, Lola briefly habitually introduced herself briefly before speaking, but
she soon abandoned the practice.

In (100), the lower briefly could be analyzed as a manner adverb (indicating that Lola in-
troduced herself in a brief manner). It passes the manner adverb test used in (79):

(101) In 2011, Lola briefly habitually introduced herself before speaking—and she
did it briefly—but she soon abandoned the practice.

On the Cinquean view, this might be taken to suggest that this briefly occupies
[Spec,VoiceP], where Cinque (1999:101-103, §4.24) locates manner adverbs in general
(such as well).
However, this briefly (the lower briefly in (100)) seems entirely parallel in interpreta-
tion to the brevemente in the following sentence, which Cinque classifies as a durative
adverb:

(102) ‘Gianni ha appena brevemente parlato con il suo capo.

‘John has just briefly spoken with the his boss
‘John just briefly spoke with his boss.’
(adapted from Cinque 1999:98, (66a))

In light of this, let us suppose that the lower briefly in (100), like the brevemente in (102),
is in [Spec,AspdurativeP]. This can be made consistent with the result of the do it–test in
(101) if we posit that this test diagnoses “manner adverbs” in a broad sense—not just
those that occur in [Spec,VoiceP] but also other adverbs with mannerlike meanings, in-
cluding (at least certain) durative adverbs.


























































80
Cinque (ibid.) claims that these for-PPs cannot occur in functional specifiers in the inflectional portion of
the clause “due to their phrasal nature.” Although it is clear what he means, this is not the best way to ex-
press it, for two reasons. First, even “simplex” adverbials he considers AdvPs—i.e., phrases. Secondly,
AdvPs that are even more obviously phrasal than long and briefly can appear in the inflectional portion of
the clause:

(xviii) Allison will [AdvP almost certainly] have left by then.

Cinque’s (ibid.) example *John has for an hour walked in the park is ill formed because it violates Wil-
liams’s (1982:160) Head-Final Filter. (The well-formed (xviii), by contrast, does not.)

45

 We have decided that the lower briefly in (100) is in [Spec,AspdurativeP]. This means
that the higher briefly must be somewhere else (indeed, higher than AsphabitualP, since it
precedes habitually). Now, this higher briefly could be being used to describe a period of
a month, and Cinque (1999:98, §4.20) counts for a month as a durative adverbial. If an
instance of briefly has a meaning that qualifies as durative, then by the logic of the carto-
graphic enterprise—which assumes a one-to-one mapping between types of meanings
and syntactic positions—this higher briefly must be in [Spec,AspdurativeP]. But this can’t
be—not only because we decided that the lower briefly is located there, but also because
the higher briefly in (100) precedes habitually, whereas AspdurativeP is located below As-
phabitualP according to (18).
All of this is to say that Cinque must either refine his characterization of what counts
as a durative meaning or posit the existence of two AspdurativePs (either of which can host
briefly in its specifier). We will here adopt the second approach.81 The briefly that Cinque
discussed must be the lower one (briefly2), since he places AspdurativeP beneath Asphabitu-
alP, whereas the higher briefly (briefly1) occurs above AsphabitualP ((100)). Therefore, the
briefly that we have added is the higher one. So at this point our ordering prediction is
briefly1 > briefly2 > characteristically.
Before we test this prediction, however, we should note that characteristically can
have not only the “generic” meaning that Cinque (1999:99) attributes to it but also a
manner reading, as in the following attested example:

(103) …a decidedly anti-Israeli leadership that would hold Israel to account every time
it behaves characteristically.82

As one would predict, the two subtypes of characteristically can cooccur:

(104) Spike characteristically behaves characteristically, but occasionally behaves in

a manner you would never expect from him.

We can infer from (104) that the generic characteristically (the one Cinque dealt with)
occurs higher than manner characteristically.83 Combining this inference with our current
ordering statement (briefly1 > briefly2 > characteristically), the ordering statement for
this subsection becomes briefly1 > briefly2 > characteristically1 (generic) > characteristi-
cally2 (manner).
This predicts that a characteristically should never be able to occur between two
brieflys (or, indeed, before any briefly), which is false:

(105) Lola briefly characteristically introduced herself briefly before speaking, but
now this behavior is NO LONGER characteristic of her.

























































81
Note that, in order for this move to be legitimate in the Cinquean framework, the two Aspdurative° heads
(or, equivalently, their specifiers) must be shown to differ in meaning ((20d)).
82
From <http://www.thecornerreport.com/index.php?title=amayreh_israel_caves_to_new_middle_east_&
more=1&c=1&tb=1&pb=1>. That the characteristically in (98) is a manner adverb is confirmed by the fact
that it can be replaced salva veritate with in a characteristic manner.
83
Native-speaker readers will note that it is impossible to reverse the interpretations of the characteristi-
callys in (104)—i.e., to interpret the linearly earlier one as the manner adverb and the linearly later one as
the generic adverb.

46

Of course, it could be that the lower briefly in (105) really is a manner adverb in
[Spec,VoiceP] after all—but then the assignment of the apparently synonymous breve-
mente in (102) to [Spec,AspdurativeP] must be reevaluated.
Let us take stock. Even after we took pains to distinguish between two brieflys and
two characteristicallys, our ordering statement was still not empirically adequate. Advo-
cates of the Cinque hierarchy will therefore have to refine this part of it in order to make
it compatible with the facts. Preliminary result: 

II.A.24. Aspgeneric/progressive° (characteristically) > Aspprospective° (almost)

The generic and manner characteristicallys are distinct enough in meaning that we
should be able to test this ordering even without cluttering up the relevant sentences with
both. The following judgments are strictly relativized to the generic reading of character-
istically:

(106) a. *When his parents get home, Scott characteristically has almost finished his
*homework.
b. *When his parents get home, Scott almost has characteristically finished his
*homework.
Preliminary result: 

II.A.25. Aspprospective° (almost) > AspSgCompletive(I)° (completely)

The predictions of this ordering appear to be correct:

(107) a. #The army had almost been completely defeated.

b. #The army had completely been almost defeated.84
Preliminary result: 

























































84
The written string in (107b) can, at least for me, be felicitous in a different context—for instance, as the
third conversational turn in the following dialogue ((xixA2a)), where it receives an interpretation very simi-
lar to that of (xixA2b) or (xixA2c):

(xix) A1: The army had been - ALMOST defeated, but not QUITE defeated.
B1: The army had NOT been almost defeated but not quite defeated.
A2: a. The army had COMPLETELY been almost defeated but not quite defeated!
b. The army had TOTALLY been almost defeated but not quite defeated!
c. The army had SO been almost defeated but not quite defeated!

The string in (107b) is licensed in (xixA2a) due to a conspiracy of two factors. First, almost appears lower
than it normally would (cf. its position in (107a)) (because it did this in (xixA1), which contains the dis-
course antecedent for the repeated material in (xixA2a)). Secondly, the completely in (xixA2a) is inter-
preted as a marker of emphatic affirmation, like the totally and so in (xixA2b-c). (All three of these are col-
loquial in register.) The completamente that Cinque (1999:100, §4.23, (70-73)) locates in
[Spec,AspSgCompletive(I)P] is parallel in interpretation to the completely in (107a), not to that in (xixA2a).
Therefore, the availability of the order completely > almost in (xixA2a) does not falsify the ordering hy-
pothesis being discussed in the text. Thanks to Edwin Williams (class, fall 2009) for bringing the existence
of emphatic affirmation completely to my attention.

47

II.A.26. AspSgCompletive(I)° (completely) > Voice° (well)

At this point we have to say something about the different COMPLETIVE ASPECT pro-
jections that Cinque posits. According to Cinque, AspPlCompletiveP hosts tutto in its speci-
fier in Italian (cf. (18)). English does not have an element comparable to tutto—i.e., a
word meaning ‘everything’ that can appear higher than the argument position it would be
expected to occupy. Therefore, we obviously cannot test the ordering statements involv-
ing AspPlCompletiveP in English. Now, Cinque does posit two completive aspect projections
whose specifiers can be overtly filled in English: AspSgCompletive(I)P and AspSgCompletive(II)P.
In English, according to Cinque (cf. (18)), both of these projections host in their specifi-
ers the adverb completely. Consider the following two sentences, which are shown with
their possible interpretations:

(108) a. John completely [VP forgot her instructions]. (sentences from Cinque
1) ‘John didn’t think of her instructions at all.’ 1999:178, n. 57)

b. John [VP forgot her instructions] completely.

1) ‘John forgot every part of her instructions.’
2) ‘John didn’t think of her instructions at all.’

On Cinque’s view, to flesh out his brief remarks (ibid.), these facts can be made sense of
if we posit the following. The underlying order at hand is completelyat all > completelyevery
part > VP (where the subscripts refer to the interpretations given in (108)). In English, VP
obligatorily85 raises above completelyevery part (as well as a couple of other very low ad-
verbs). This derives the surface order86 completelyat all > VP > completelyevery part, which
accounts for the interpretation of (108a) and interpretation 1) of (108b). Interpretation 2)
of (108b) is derived by optional further movement of VP past completelyat all, yielding the
surface order VP > completelyat all > completelyevery part.
As it turns out, these two completelys cannot cooccur. The following sentence is only
acceptable if the higher completely is interpreted on its emphatic affirmation reading (see
fn. 84):

(109) #John completely forgot her instructions completely.

Perhaps it is not too surprising that the two completelys exemplified in (108) cannot
cooccur, since one cannot both forget every part of a set of instructions (but remember
that the instructions exist) and forget that they exist at all. Whatever the exact reason,
these two completelys cannot cooccur, and so if we want to test their relative positioning
vis-à-vis well, we will have to be confident that we can distinguish them on the basis of
their meanings—in a sentence also containing well. Unfortunately, it is not clear to me


























































85
One might ask, if this movement is obligatory, why Cinque does not simply base-generate a structure that
would be linearized in the desired way (completelyat all > VP > completelyevery part). The answer is theory-
internal assumptions: in Cinque’s system (which is essentially antisymmetric; cf. fn. 12), there are no
rightward specifiers.
86
I use the term surface order loosely. As we will see shortly, these two completelys cannot cooccur.

48

how this can be done. Therefore, it is difficult to know how to interpret the availability of
both orders of completely and well:

(110) a. Megan [solved the problem completely] well.

b. A: Megan solved the problem well.
B: Did she solve it well COMPLETELY, or only PARTIALLY?

I will therefore have to leave the predictions of this particular ordering statement un-
tested. Preliminary result: —

II.A.27. Voice° (well) > Aspcelerative(II)° (fast/early)

The first thing to note about all three of these adverbs is that they obligatorily follow
the core VP on the surface (though, on fast, cf. fn. 72):

(111) a. *Zach [VP draws canyons] well.

b. *Zach well [VP draws canyons].

(112) a. *Sophie [VP eats chicken wings] fast.

b. *Sophie fast [VP eats chicken wings].

(113) a. *Ben [VP finishes all his homework assignments] early.

b. *Ben early [VP finishes all his homework assignments].

In Cinque’s system, these facts are interpreted as follows. Well and fast/early fill the
specifiers of functional projections that dominate the VP underlyingly. The surface order
is derived by obligatory movement of VP past these functional projections (cf. fn. 85).87
The height to which VP must raise could presumably be subject to crosslinguistic varia-
tion.
With this in mind, let us consider what, on Cinque’s hypothesis, should happen in a
clause containing both well and fast/early. The underlying order is well > fast/early > VP.
VP must precede fast/early on the surface, so it raises past it, deriving the order well >
VP > fast/early. The VP must also come to precede well. This can be brought about by
raising the new constituent [XP VP fast/early] past well, deriving [[XP VP fast/early]
well].88

























































87
It is not clear what drives this movement; the same is true of the optional movement of VP past the higher
completely (the one in [Spec,AspSgCompletive(I)P]) discussed right after (108) in §II.A.26.
88
As a matter of fact, it seems to me that there is a problem with these sorts of raising derivations. Consider
the case where the underlying structure [fast VP] is deformed by moving VP past fast, creating the structure
[XP VP fast]. Where exactly, on the Cinquean view, has VP moved to? It cannot be in the head position
above fast (by hypothesis, Voice°), on the assumption that a phrase cannot move to a head position. It can-
not be in [Spec,Aspcelerative(II)P], since that’s where fast is. And it cannot be in the complement position of
the head above fast (Voice°), since that position is already being filled by Aspcelerative(II)P. This problem
could be solved by postulating even more positions in the clausal hierarchy. But syntactic positions devoid
of semantic content whose only function is to host moved phrases (some of which move only optionally)
would clearly be an embarrassment to the theory. Alternatively, perhaps a head can have multiple specifi-
ers, in which case the moved phrase would be easy to accommodate anywhere. This is a point in need of
clarification.

49

 Consider another possibility. Once again, the underlying order is well > fast/early >
VP. Suppose VP moves past both adverb positions in one fell swoop. Then the surface
order will be VP > well > fast/early.
If both of these derivations are available—i.e., if one-fell-swoop movement is not
banned by some locality condition—then there are two surface orders compatible with
the Cinque hierarchy: VP > fast/early > well and VP > well > fast/early. And if we only
consider linear order, this seems to be exactly what we get:

(114)0 a. Alice can [VP paint pictures] well fast.

b. Alice can [VP paint pictures] fast well.

(115)0 a. Most of the kids do their homework poorly the first time around and only
do it well rather late,89 but Sean [VP does his homework] well early.
b. Sean [VP plans for things] early well.

However, there is a problem. Cinque (2010:6-7, esp. fn. 17) clarifies that he takes the
output of Merge to reflect relative scope relations. If this view is correct, and if the output
of Merge always obeys the Cinque hierarchy, it follows that well will always outscope
fast/early. But this is falsified by (114a) and (115a). In (114a), what Alice can do fast is
[paint pictures well], and in (115a), what Sean does early is [do his homework well]. In
fact, all the adverbs in (111-113) display the relative scopes we would expect if their sur-
face orders were base-generated and they were merged with VP from the inside out. In
other words, in each of these sentences, the adverb farther from VP on the surface out-
scopes the one closer to VP on the surface. Preliminary result: 

II.A.28. Aspcelerative(II)° (fast/early) > Asprepetitive(II)° (again)

Before we test the predictions of this ordering, there is a problem I should point out.
If, as Cinque maintains, the VP obligatorily raises past well and fast/early (see §II.A.26),
and these adverbs are higher than again2, then a fortiori, the VP must raise past again2 as
well. But the following sentence suggests that this prediction is false, assuming that the
lower of two agains in the same clause is again2:

(116) Jack again has [dropped his pen and then AGAIN [VP dropped his pen]].

Although this already indicates that there is something wrong with this region of the
Cinque hierarchy, let us test the predictions of the ordering statement at hand anyway.
Because most of the structures we will need in this subsection and the next are
somewhat elaborate, let us derive them very explicitly.90 This will make clearer the in-
tended relative adverb scopes, and hence the intended meanings of the sentences. The


























































89
Late as in close to the due date.
90
In the derivations in this thesis, a syntactic structure in row n, column 1, undergoes the operation in row n,
column 2, producing the syntactic structure in row n + 1, column 1, etc.

50

derivations used are bottom-up.91 For ease of exposition, they only show overt material
(so adverbs are included, but their corresponding silent functional heads are not).
The derivation in (117) merges again (i.e., again2) before fast, and the one in (118)
does the reverse. Therefore, the Cinque hierarchy predicts that the output of (117) should
be acceptable and that of (118) unacceptable.

(117) Syntactic structure Operation

a. [VP asked a question that she had asked before] Merge again
b. again [VP asked a question that she had asked before] Move VP over
again
c. [XP [VP asked a question that she had asked before] again] Merge fast
d. fast [XP [VP asked a question that she had asked before] Move XP92 over
again] fast
e. [YP [XP [VP asked a question that she had asked before] …
again] fast]
f. Miranda again [YP [[asked a question that she had asked
before] again] fast].

(118) Syntactic structure Operation

a. [VP ate] Merge fast
b. fast [VP ate] Move VP over
fast
c. [XP [VP ate] fast] Merge again
d. again [XP [VP ate] fast] Move XP over
again
e. [YP [XP [VP ate] fast] again] …
f. Miranda again [ZP ate fast and then ate fast again].

Contrary to the aforementioned prediction, both (117f) and (118f) are fine.
A parallel situation obtains in the case of early and again2.

(119) a. Joel again [[[proofread his homework and then proofread it again] early],
but handed it in late].
b. Joel again [got up early on Monday and then [[got up early] again] on
Tuesday].

Even without going through the movements that, on a Cinquean view, (119a-b) necessar-
ily involve, it is clear that the relevant core VP has merged with again2 before early in
(119a) and with early before again2 in (119b). The Cinque hierarchy therefore predicts
that only the former should be acceptable, but in fact they both are. Preliminary result: 


























































91
As far as I can tell, nothing hinges on this, and left-to-right derivations could have been used instead (as
they necessarily are, at least to some extent, in parsing and production).
92
In these derivations, XP, YP, and ZP are used as ad hoc labels for constituents in which movement has
taken place; there is no relation to the ZPs of Stowell (2007), which are arguments of T that denote times.

51

II.A.29. Asprepetitive(II)° (again) > Aspfrequentative(II)° (often)

The derivation in (120) merges often2 before again2, and that in (121) does the re-
verse. Therefore, the Cinque hierarchy predicts that only the former should produce an
acceptable sentence.

(120) Syntactic structure Operation

a. [VP goes to the beach] Merge often
b. often [VP goes to the beach] Move VP over
often
c. [XP [VP goes to the beach] often] Merge again
d. again [XP [VP goes to the beach] often] Move XP over
again
e. [YP [XP [VP goes to the beach] often] again] …
f. Mike again often, having gone to the beach often one year,
[YP goes to the beach often again] the next year.93

(121) Syntactic structure Operation

a. [VP does things that she’s done before] Merge again
b. again [VP does things that she’s done before] Move VP over
again
c. [XP [VP does things that she’s done before] again] Merge often
d. often [XP [VP does things that she’s done before] again] Move XP over
often
e. [YP [XP [VP does things that she’s done before] again] often] …
f. Liz again often [YP [XP [VP does things that she’s done be-
fore] again] often].

Contrary to the aforementioned prediction, both (120f) and (121f) are fine. Preliminary
result: 

II.A.30. Aspfrequentative(II)° (often) > AspSgCompletive(II)° (completely)

I will not be able to test the predictions of this ordering, because of the difficulties
involved in reliably distinguishing the two completelys (see the discussion following
(109) in §II.A.26). Preliminary result: —

II.A.31. Domain adverbs > Moodspeech act° (honestly)

Before we conclude this section, we should say a few words about domain adverbs,94
such as the politically in (117):


























































93
This is sentence (i) from fn. 9.
94
For a summary of the various terms that have been used to denote this class of adverbs, see Cinque
(1999:174, n. 33).

52

(122) Politically, there are frankly no grounds for being merry.
(based on Cinque 1999:175, n. 41, (ia))

Cinque (ibid., n. 41) notes that these adverbs “provide the relevant frame within which to
evaluate the truth (or appropriateness) of a given speech act” (cf. Delfitto 2006:90), and
posits that they are generated in a left-peripheral topic position. For this reason, he largely
ignores them in his 1999 study, which is primarily concerned with adverb positions in the
IP domain.
The hypothesis that domain adverbs are left-peripheral topics predicts that they will
precede all the adverbs in the Cinque hierarchy sensu stricto ((18)), since these latter ad-
verbs are all in the IP domain. Cinque (1999:175, n. 41, (i-vi)) provides evidence that, in
Italian, the domain adverb politicamente ‘politically’ does indeed obligatorily precede the
speech-act adverb francamente ‘frankly’, the evaluative adverb purtroppo ‘unfortu-
nately’, the epistemic adverb probabilmente ‘probably’, the temporal adverb ora ‘now’,
and the adverbs forse ‘perhaps’ (which Cinque locates in [Spec,ModirrealisP]) and saggia-
mente ‘wisely’ (on its subject-oriented reading).
In English, however, the situation appears to be more complex. In the following ex-
amples, the (a) sentences show that the domain adverb politically can indeed precede the
English counterparts of all the aforementioned high adverbs, plus the evidential adverb
evidently. The (b) sentences, however, show that, in each case, the opposite order is also
possible. In these cases, it appears that the domain adverb is still interpreted as a topic,
with its sister serving as the corresponding comment. This is brought out by the continua-
tions in the (b) sentences, which create a contrastive topic structure. Lastly, the (c) sen-
tences show that, apparently, the domain adverb politically can also occur in a much
lower position.

(123) a. Politically, that honestly couldn’t have gone worse.

b. Honestly, that politically couldn’t have gone worse, but morally was a real
victory.95
c. Honestly, that couldn’t have gone worse politically.96

(124) a. Politically, that unfortunately couldn’t have gone worse.

b. Unfortunately, that politically couldn’t have gone worse, and morally was
pretty awful as well.
c. Unfortunately, that couldn’t have gone worse politically.

(125) a. Economically, it couldn’t have gone worse, but politically it evidently was
a smashing success.
b. Evidently, [it economically couldn’t have gone worse, but politically was
a smashing success].
c. Evidently it was a smashing success politically.


























































95
This example is modeled closely on one that was suggested to me by Amy Zhou (05/2012).
96
This example is due to Anna Tchetchetkine (p.c., 05/2012).

53

(126) a. Politically, that probably couldn’t have gone worse.
b. That probably [politically couldn’t have gone worse, but morally was a
real victory].
c. That probably couldn’t have gone worse politically.

(127) a. But politically, they now could really use some help.
b. But now, they politically could really use some help, but economically are
doing really well.
c. They now could really use some help politically.

(128) a. Politically, that perhaps couldn’t have gone worse.

b. Perhaps [that politically couldn’t have gone worse, but morally was a real
victory].
c. That perhaps could not have gone worse politically.

(129) a. Politically, they wisely have been exercising caution, but economically
they’ve been being quite reckless indeed.
b. Wisely, they politically have been exercising caution and economically
have been consulting with their trustworthiest advisors.
c. Wisely, they have been exercising caution politically.

This is not to say that, contra Cinque, domain adverbs cannot be generated as left-
peripheral topics. On the contrary, this seems quite plausible for sentences like (122) and
the (a) sentences above. The (b) and (c) sentences merely show that domain adverbs at
least apparently have a wider distribution than Cinque acknowledged, a fact that the
Cinquean system should be revised to accommodate. If it turns out that politically does in
fact occupy distinct syntactic positions in some of the (a), (b), and (c) sentences above,
then it will be incumbent upon practitioners of the cartographic approach to show that
these differently placed politicallys differ in meaning somehow (cf. (20d)).

II.B. Conclusions

The empirical findings we have just laid out are summarized in the following table.97

(130) Putative order of functional heads (adverbs) Correct pre-

dictions for
English?
(a. Moodspeech act° (honestly) > Moodevaluative° (unfortunately) no
(b. Moodevaluative° (unfortunately) > Moodevidential° (allegedly) yes
(c.) Moodspeech act° (honestly) > Moodevidential° (allegedly) yes
(d. Moodevidential° (allegedly/clearly) > Modepistemic° (probably) no
(e. Modepistemic° (probably) > T(Past)° (once) (no)
(f. T(Past)° (once) > T(Future)° (then) no

























































97
Parentheses around a lowercase letter (as in “(c.)”) indicate that the corresponding pair of adverbs are not
adjacent in the Cinque hierarchy ((18)). Parentheses around a verdict (yes or no) indicate that I consider it
not fully clear, or particularly open to dispute (although, of course, anything can be disputed).

54

 (g. T(Future)° (then) > Modirrealis° (perhaps) no
(h.) Modepistemic° (probably) > Modirrealis° (perhaps) (yes)
(i.) Modepistemic° (probably) > Modirrealis° ([almost] certainly) no
(j. Modirrealis° (perhaps) > Modnecessity° (necessarily) no
(k. Modirrealis° (perhaps) > Modnecessity° (not necessarily) (no)
(l. Modnecessity° ([not] necessarily) > Modpossibility° (possibly) no
(m. Modpossibility° (possibly) > Asphabitual° (usually) yes
(n. Asphabitual° (usually) > Asprepetitive(I)° (again) no
(o. Asprepetitive(I)° (again) > Aspfrequentative(I)° (often) (yes)
(p.) Asphabitual° (usually) > Aspfrequentative(I)° (often) no
(q. Aspfrequentative(I)° (often) > Modvolitional° (intentionally) no
(r. Modvolitional° (intentionally) > Aspcelerative(I)° (quickly) (no)
(s. Aspcelerative(I)° (quickly) > T(Anterior)° (already) no
(t. T(Anterior)° (already) > Aspterminative° (no longer) no
(u. Aspterminative° (no longer) > Aspcontinuative° (still) (no)
(v. Aspcontinuative° (still) > Aspperfect(?)° (always) (no)
(w. Aspperfect(?)° (always) > Aspretrospective° (just) yes
(x. Aspretrospective° (just) > Aspproximative° (soon) no
(y. Aspproximative° (soon) > Aspdurative° (briefly) yes
(z. Aspdurative° (briefly) > Aspgeneric/progressive° (characteristically) no
(aa. Aspgeneric/progressive° (characteristically) > Aspprospective° (al- yes
most)
(bb. Aspprospective° (almost) > AspSgCompletive(I)° (completely) yes
(cc. AspSgCompletive(I)° (completely) > Voice° (well) —
(dd. Voice° (well) > Aspcelerative(II)° (fast/early) no
(ee. Aspcelerative(II)° (fast/early) > Asprepetitive(II)° (again) no
(ff. Asprepetitive(II)° (again) > Aspfrequentative(II)° (often) no
(gg. Aspfrequentative(II)° (often) > AspSgCompletive(II)° (completely) —
(hh. domain adverbs > {Moodspeech act° (honestly), Moodevaluative° no
(unfortunately), Moodevidential° (evidently), Modepistemic°
(probably), temporal adverbs (now), Modirrealis° (perhaps),
subject-oriented adverbs (wisely)}

II.C. Discussion

At this point, the following question arises immediately: what accounts for the data
in (130)?

II.C.1. A possible explanation: the Direct Attachment Hypothesis

One possibility is that the Cinque hierarchy is correct, but this fact is obscured by the
ability of certain adverbs to ATTACH DIRECTLY to various constituents.
Cinque (1999:30-31, §1.6), citing other researchers, notes that focus-sensitive ad-
verbs such as only can apparently attach to a wide variety of constituents, such as those
labeled in (131a-e) and “various clausal functional projections” ((131f-i)). Although

55

some readers may disagree with some of the category labels below, the observation cer-
tainly seems correct:

(131) a. ?He likes only [DP himself]. (judgments from

b. ?The success was only [AdjP partial]. Cinque ibid., (125))
c. ?He solved the problem only [AdvP partially].
d. ?He will have been beaten only [PP by John].
e. ?He will have been only [VP beaten by John].98
f. ?He will have only been beaten by John.
g. ?He will only have been beaten by John.
h. ?He only will have been beaten by John.99
i. ?He said only [CP that he doesn’t like it].

Cinque further notes that some other adverbs can do the same thing, for instance the
probabilmente in (132a) and the forse in (132b). The English translations given for
(132a-b) display the same phenomenon and are fully acceptable.

(132) a. ‘Lo avrà già detto [probabilmente a tutti].

‘it(DO) will.have already said [probably to everyone
‘He will have already said that probably to everybody.’
(Cinque 1999:32, (134a))

(127) b. ‘Non legge più romanzi [forse proprio per questo].

‘not reads more novels perhaps just because.of this
(Adv)
‘He no longer reads novels perhaps precisely for this reason.’
(Cinque 1999:32, (134c))

Critically, (132a-b) appear to violate the Cinque hierarchy’s predictions concerning ad-
verb orders, according to which probably should precede already and perhaps should
precede no longer (and likewise, mutatis mutandis, in Italian). But these “violations” are
illusory, because in each of these sentences, the two adverbs are not part of the same F-
structure. In each case, the linearly later boldfaced adverb is DIRECTLY ATTACHED to a


























































98For me, this sentence requires some prosodic and information-structural manipulation to be well formed.
This requirement is satisfied if only is preceded by a brief pause and a subconstituent of the VP is marked
as a contrastive focus:

(xx) a. He will have been - only BEATEN by John, not humiliated by him as WELL.
b. He will have been - only beaten by JOHN, not beaten by Lucy as WELL.


99
For me this sentence is perfect provided that only has a focus to associate with:

(xxi) a. He only will have been BEATEN by John.

b. He only will have been beaten by JOHN.

56

PP.100,101 I will refer to PPs, nominals, and other constituents not lying along the “spine”
of the clause as NONSPINAL CONSTITUENTS, or NONSPINALS for short.
We have just seen that there are some sentences where the Cinque hierarchy superfi-
cially appears to be disobeyed (contrary to fact) because an adverb A higher in the
(clausal) hierarchy than another adverb B can attach directly to a nonspinal that is lower
than B. Now, let us return to (130), the table summarizing which adverb ordering state-
ments made correct predictions for English. Specifically, consider the cases where an or-
dering statement AdvA > AdvB apparently failed to make correct predictions.
Suppose that these cases are exactly the cases in which AdvA can attach directly to
nonspinals. If an adverb is “promiscuous” enough to be able to attach to a wide variety of
nonspinals, then it would not be surprising if it could also directly attach to many differ-
ent projections within the clausal hierarchy itself (cf. (131) and preceding discussion).
Therefore, it would be possible to analyze the seemingly Cinque-noncompliant orders
AdvB > AdvA as instances of AdvA attaching directly to a clausal functional projection
lower than AdvB. Therefore, the following would be entirely plausible:

(133) Hypothetical scenario

a. The Cinque hierarchy is correct.
b. It is often possible to construct sentences that obscure the Cinque hierarchy’s
effects via direct attachment.

In fact, the exact correspondence between the cases in which AdvB can precede AdvA and
the cases in which AdvA is independently known to be a “direct attacher”102 would be so
striking that it would count as good evidence for (133).
Because this would be a very interesting conclusion, we will test the relevant hy-
pothesis now. But first, let us state it explicitly one more time:


























































100
To be more precise, the forse in (132b) is directly attached to a PP to which proprio ‘just’ has directly
attached already.
101
We might well ask whether Cinque’s invocation of “direct attachment” (albeit not in those words) is not
just a reintroduction of adjunction in disguise, despite his view that it would be desirable to eliminate ad-
junction altogether (Cinque 1999:44; cf. fn. 10). The answer seems to be not necessarily: Cinque (ibid:31)
briefly endorses the view that direct attachment of an adverb to an XP is Adv° taking the XP as a comple-
ment. Naturally, this would appear to create problems for the locality of selection: how can a higher head H
c-select the head X° of that XP in a local head–complement configuration if Adv° is in the way? Further-
more, surely H° never imposes selectional restrictions on a “directly attached” Adv°, even though this
would now appear to be the head of its complement, when present. But these are problems for cartographic
analyses generally: consider how many heads, on Cinque’s view of clause structure, C° must somehow
bypass when c-selecting T°. I leave this issue aside.
102
Cinque (1999:30, §1.6) uses the term “FOCUSING” ADVERB, since these adverbs display the same ability
to attach to a wide variety of constituents as focus-sensitive adverbs such as only. I will refer to Cinque’s
“focusing” adverbs as DIRECT ATTACHERS, as I do not wish to come across as making any particular claim
about how these adverbs interact with focus.

57

(134) Direct Attachment Hypothesis
a. The Cinque hierarchy is correct.
b. A Cinquean adverb ordering A > B can be circumvented by direct attachment
of A to a clausal functional projection lower than B.
c. The circumvention described in b. is possible just in case A can directly at-
tach to nonspinals.

II.C.2. Testing the Direct Attachment Hypothesis

In order to test the Direct Attachment Hypothesis, we must 1) determine which

Cinque adverbs can directly attach to nonspinals, and then 2) verify whether they are the
same adverbs that can unexpectedly appear lower than other adverbs (unexpectedly, that
is, from the point of view of the Cinque hierarchy).

II.C.2.a. What nonspinals to use?

To execute step 1) of our plan, we need to decide what kind of nonspinals we want to
use. One possibility would be small clause predicates. However, we cannot be sure that
small clauses are really that small. Admittedly, as pointed out to me by Edwin Williams,
they cannot host auxiliaries or Tense. But for all we know, they could obey the Cinque
hierarchy as faithfully as finite clauses do.103 If this is so, then “directly attaching” ad-
verbs to small clause predicates could actually involve slotting them into a rich functional
structure—essentially, investigating the Cinque hierarchy all over again.
A second possibility would be appositive nominals:

(135) a. #Ethan, a brilliant college student and [probably the fastest skier in the
#world], was mobbed by fans on the street.
b. #Samantha, the smartest kid on the math team and [often the friendliest], was
#interviewed for the school newspaper.

But consider the following:104

(136) a. #You’re not going? Probably the fastest skier in the WORLD is probably go-
#ing to be there!
b. #Often the friendliest kid on the math team was interviewed for the school
#newspaper.

In (136a), the linearly earlier probably is part of the nominal [probably the fastest skier in
the world]. It is unlikely to be the clausal probably, because that’s what the linearly later

























































103
Preliminary support for this hypothesis comes from the following contrast, in which a small clause ap-
pears to obey the Cinque ordering allegedly > often:

(xxii) a. *If three people have alleged that Blake is often a pain, then SURE I’d consider him allegedly
*often a pain.
b. *If people often allege that Blake is a pain, then SURE I’d consider him often allegedly a pain.
104
I am grateful to Edwin Williams for enlightening discussion of this point.

58

probably appears to be. Omitting some details, this analysis predicts that (136a) should
be paraphrasable as It is probable that [the person who is probably the fastest skier in the
world] is going to be there, which is correct. In other words, the string probably the fast-
est skier in the world in (136a) is interpreted as denoting a person. In (136b), on the other
hand, the string often the friendliest kid on the math team cannot be interpreted as denot-
ing a person. In fact, it is not a semantic constituent: the often is a clausal modifier.
Let us take stock. On the basis of (135a-b), we might have concluded that the brack-
eted strings in these sentences represented nominals to which an adverb had been directly
attached. However, the fact that one of these putative nominals was unable to appear in
subject position in (136b) suggests that perhaps the predicative “appositive nominal” in
(135b), at least, is actually embedded in a predicative functional structure not unlike a
small clause. So we don’t want to pick “appositive nominals” for our nonspinals either:
like small clause predicates, they could turn out to lie at the bottom of a rich F-structure
like that of finite clauses.
With the foregoing considerations in mind, I have picked for my nonspinals primar-
ily PPs.

II.C.2.b. Data and discussion

In the following sentences, all the Cinque adverbs have been attached to nonspinals
that can be.105 As it turns out, that’s the vast majority of them:

(137) a. I broke up with her [honestly for several reasons].

(138) She talked to me [fortunately about really interesting things], but [UNfortu-
nately when I didn’t really have time for it].

(139) Seth talked to Hannah [allegedly for nine hours straight].

(140) Seth talked to Hannah [probably for nine hours straight].

(141) Mia, [once a great musician], had lost much of her former talent.106

(142) Owen—[then for only a few reasons], but afterwards for many more rea-
sons—strongly opposed the death penalty.


























































105
We will not be concerned here with whether the adverb we are attaching to a nonspinal is (for instance)
often1 or often2. This is because, on Cinque’s view, they are really the same adverb (often), which acquires
the semantic idiosyncrasies of often1 and often2 based on which of the two relevant specifiers it is base-
generated in (cf. fn. 64). These two distinct positions (and their interpretive peculiarities) are part of the
functional structure of the clause, and are therefore presumably irrelevant for direct attachment.
106
The data in (37) and fn. 34 suggested that once must be licensed by past tense or perfect aspect, and these
data are fully compatible with the hypothesis that the licensor must be in the same clause as once. In (141),
however, this putative requirement is not satisfied. Perhaps the once in this sentence is licensed by the past
tense and/or by the perfect aspect in the matrix clause. Pursuing this issue here would take us too far astray.
Note that the bracketed constituent in (141) is an “appositive nominal”—uniquely in this data set.

59

(143) a. Seth talked to Hannah [perhaps for nine hours straight].
b. Seth talked to Hannah for [perhaps nine hours straight].

(144) And then, for a short period of time, although [necessarily for more than a
millisecond], Seth and Hannah discussed their plans for the future.

(145) And then, for a short period of time, although [not necessarily for less than a
minute], Seth and Hannah discussed their plans for the future.

(146) Seth talked to Hannah [possibly for nine hours straight].

(147) On Sundays, Seth talks to Hannah [usually for nine hours], but sometimes for
only eight.

(149) On Sundays, Seth talks to Hannah usually for nine hours, but [often for only
eight].

(150) a. Mia referred to Owen’s behavior as [intentionally perverse].

b. Perverse, his behavior undoubtedly was. [INTENTIONALLY perverse], it was
not.
c. [Intentionally perverse] though his behavior undoubtedly was, …

(151) a. He has drunk [already several beers].107

b. For the fifth-straight year—and [already for the second time this season]—
South Carolina will host a Thursday night ESPN game…108

(152) For many good reasons, and [no longer for any bad reasons], Mia is a politi-
cal independent.

(153) For some good reasons, but [still for many bad reasons], Mia is a political
independent.

(154) On Mondays, and for some reason [always on Mondays], Seth is in an in-
credibly foul mood all day.

(155) Because of Mia, and [briefly because of Chloe], Owen was completely un-
able to concentrate in class.

(156) On Mondays, and [characteristically on Tuesdays as well], Seth is in a terri-

ble mood.

(157) For pretty bad reasons, and indeed [almost for totally perverse reasons], Mia
decided not to vote in the election.


























































107
Cinque (1999:31, (129b)).
108
http://bleacherreport.com/articles/260204-five-reasons-the-gamecocks-will-shock-ole-miss

60

(158) Mia became a political independent for [completely the wrong reasons].

There are five adverb classes whose representative members are not shown directly
attached to nonspinals in the above data set, because they cannot be. The adverbs in ques-
tion are quickly, retrospective just, soon, well, and fast/early. The following data set
shows that these adverbs cannot be directly attached to any of the following kinds of non-
spinals: nominals, AdjPs, AdvPs, PPs, or CPs (cf. (131)).109,110 The reader should keep in
mind that the judgments on the following sentences are relativized strictly to interpreta-
tions on which the boldfaced adverb is directly attached to the bracketed constituent and
modifies it. The sentences should be read with no unusual pauses.

(159) a. ?*Zoe discussed quickly [DP herself].

b. ?*Quickly [AdjP victorious] though Zoe was, …
c. ?*Quickly [AdvP ecstatically], Zoe began talking.111
d. ?*Quickly [PP with great enthusiasm], Zoe greeted the guests.112
e. A:1 ?*What did Zoe say?
B: ?*Quickly [CP that she didn’t like it].113

The judgments in (160) are relativized to the retrospective meaning of just:

(160) a. ?*Zoe discussed just [DP herself].

b. ?*Just [AdjP happy] though Zoe had been, …
c. ?*Just [AdvP energetically], Zoe inexplicably became tired.
d. ?*Just [PP with a lot of energy], Zoe inexplicably became tired.
e. A:1 ?*What did Zoe say?
B: ?*Just [CP that she didn’t like it].

(161) a. ?*Zoe will discuss soon [DP herself].

b. ?*Soon [AdjP victorious] though Zoe {*is / ?*will be}, …
c. ?*Soon [AdvP ecstatically], Zoe {is / will be} eating ice cream.
d. ?*Soon [PP with great enthusiasm], Zoe {is / will be} eating ice cream.
e. A:1 ?*What will Zoe say?
B: ?*Soon [CP that she doesn’t like it].

The adverbs in the following sentences (well and fast/early) can modify VP (or the ex-
tended projection thereof), but as we have discussed, they must follow the core VP on the
surface, yielding the linear order VP > Adv. Therefore, when investigating whether they
can directly attach to nonspinal XPs, we want to test both possible surface orders: Adv >
XP (cf. I talked to [probably Andrew] yesterday) and XP > Adv (cf. She [VP said that]
well, which does not relevantly involve a nonspinal).


























































109
Of course, CP is not really a nonspinal constituent, since it does lie along the clausal spine—specifically,
at the top. That CPs should pattern with true nonspinals is therefore surprising.
110
Some of the sentences in this data set are based on sentences in (126), which is taken from Cinque.
111
Cf. Having quickly become ecstatic, Zoe began talking.
112
Cf. Having quickly been filled with great enthusiasm, Zoe greeted the guests.
113
Cf. Only [CP that she didn’t like it].

61

(162) a. ?*Zoe will discuss well [DP herself].114
b. ?*{Well [AdjP intelligent] / Intelligent well}115 though Zoe is, …
c. ?*{Well [AdvP intelligently] / Intelligently well}, Zoe explained her
?*proposal.
d. ?*{Well [PP with skill] / With skill well}, Zoe repaired the hard drive.
e. A: ?*What will Zoe say?
B1: ?*Well [CP that she thinks our government should be reformed].
B2: ?*[CP That she thinks our government should be reformed] well.

(163) a. ?*Zoe will discuss fast [DP herself].

b. ?*{Fast [AdjP victorious] / Victorious fast} though Zoe was, …
c. ?*{Fast [AdvP ecstatically] / Ecstatically fast}, Zoe began talking.

d. ?*{Fast [PP with great enthusiasm] / With great enthusiasm fast},

?*Zoe greeted the guests.
e. A: ?*What will Zoe say?
B1: ?*Fast [CP that she thinks our government should be reformed].
B2: ?*[CP That she thinks our government should be reformed] fast.

(164) a. ?*Zoe will finish early [DP her chores].

b. {*Early [AdjP victorious] / ?*Victorious early} though Zoe was, …
c. ?*{Early [AdvP happily] / Happily early}, Zoe began talking.
d. ?*{Early [PP with great enthusiasm] / With great enthusiasm early},
?*Zoe greeted the guests.

e. A: ?*What will Zoe say?

B1: ?*Early [CP that she thinks our government should be reformed].
B2: ?*[CP That she thinks our government should be reformed] early.

There are two more adverbs we should mention in this connection: once and almost.

(165) a. A:1 ?*He ruled once [DP France].

b. ?*Once [AdjP powerful] though the king {*was / ?*had been}, …
c. ?*Once [AdvP powerfully], the former king was living on the streets.
d. ?*Once [PP in a palace], the former king was living on the streets.

(166) a. A:1 ?*Zoe had finished almost [DP her homework].

b. ?*Almost [AdjP happy] though Zoe was, …
c. ?#Almost [AdvP energetically], Zoe became tired.
d. ??Almost [PP with great energy], Zoe became tired.


























































114
The XP > Adv variant of this sentence would be linearized as Zoe will discuss herself well. This string is
fine, but presumably well is modifying (the extended projection of) VP, not herself.
115
Note that the “*” is outside the braces, and hence applies to both variants. This will be important to keep
in mind in connection with the ecstatically fast variant of (163c), which is unacceptable on the intended
interpretation (with fast modifying ecstatically).

62

(166c) is acceptable, although its meaning is quite strange from the point of view of our
world knowledge. It can be paraphrased as follows:

(167) a. In an almost energetic manner, Zoe became tired.

b. In a manner that was almost energetic, Zoe became tired.

In other words, (166c) asserts that Zoe became tired in a way that comes close to being
adequately describable as energetic. It seems—for the sake of giving this kind of inter-
pretation a name—that the almost in (166c) has an APPROXIMATIVE reading. This is dif-
ferent from the PROSPECTIVE interpretation that almost receives in the following sen-
tences:116

(168) a. When his parents get home, Scott characteristically has almost finished his
homework. (= (106a))
b. They had almost gotten out the door when the phone rang.

It seems, then, that the sentences in which we attempted to directly attach almost to non-
spinals ((166a-d)) are all either degraded, deviant, or acceptable only on an approxima-
tive interpretation of almost.117 Because the almost in the Cinque hierarchy is prospective
almost, this adverb must be added to the list of Cinque adverbs that cannot attach directly
to nonspinals. (Note that the directly attached almost in (157) is clearly approximative
almost, not prospective almost.)

II.C.2.c. Why are some of the Cinque adverbs unable to attach to nonspinals?

Having examined the facts in some detail, we can now attempt to account for why a
handful of the Cinque adverbs cannot attach directly to nonspinals. Again, these adverbs
are the following:

(169) Cinque adverbs that cannot attach directly to nonspinals

a. once
b. quickly
c. just (retrospective reading)
d. soon
e. well
f. fast/early
g. almost (prospective reading)

























































116
For some discussion of the semantics of prospective aspect, see Cinque (1999:99, §4.22).
117
(166b) can clearly be interpreted as containing approximative almost. Can the almost in this sentence
also get a prospective interpretation? The question is not easy to answer, because the two meanings of al-
most can at times be hard to pry apart. If Zoe is about to qualify for the description “happy” (prospective),
then it is likely that she already comes close to being adequately describable as happy (approximative). This
means that the facts could be as follows. Only approximative almost, not prospective almost, can attach
directly to nonspinals. However, the description [almost happy] (involving direct attachment of approxima-
tive almost) can often be truthfully predicated of people who are about to be happy (prospective), obscuring
the difference between the two interpretations of almost.

63

Before we proceed, it will be helpful to introduce two terms:

(170) a. Direct attacher

An adverb that can be directly attached to a constituent.118

b. Spinal adverb
An adverb that can lie on the clausal spine (in the extended projection of V),
but cannot be directly attached to a nonspinal constituent.

First, let us consider once. The data in (37) and fn. 34 suggested that once must be li-
censed by past tense or perfect aspect. The only category whose extended projection can
host either of these is V. Therefore, once can directly attach to the “spinal” category
VP,119 but not to any other category. Thus its status as a spinal adverb is derived.120
Now let us consider almost:

(171) .[Context:
Logan is a member of a particular club, but is in no way in a posi-
#tion of power within it. However, the presidency of the club will automati-
#cally be transferred to him in two days.]
a. #Logan is almost [DP the president of the club].
b. #Logan is almost [NP president of the club].

c. .[Context:Logan is quite stupid, but a very advanced machine from outer

#space will render him intelligent in two days.]
#He’s almost [AdjP intelligent]!

d. .[Context:Logan is quite content, but a very advanced machine from outer

#space will make him spontaneously become furious in two minutes.]
#He’s almost [AdjP furious]!


























































118
As opposed to being generated in a specifier in the constituent’s extended projection.
119
A natural question at this point is how direct attachment to VP interacts with the Cinque hierarchy. This
issue will be taken up in §II.C.2.e.
120
The data in (165) suggest that direct attachment of once to an XP other than VP results in deviance no
matter what, even if the clause containing the [once XP] structure is marked for past tense or perfect aspect.
This suggests that the licensor of once must be highly local. But if this is so, then it is not clear why (141)
(repeated here as (xxiii)) is acceptable:

(xxiii) Mia, [once a great musician], had lost much of her former talent.

Recall that we have some preliminary evidence that these predicative “appositive nominals” may come
with a rich functional structure (see (135-136) and following discussion). If this is so, then we may want to
consider the possibility that this structure can host a covert counterpart of past tense or perfect aspect—but
such a claim would of course require independent evidence in order to have any content.

64

 e. .[Context: Logan is on his way to swim in the ocean. He has driven 3,000
#miles, and is finally just a few dozen meters from the water. He is still in
#his jeep.]
#Logan is almost [PP in the ocean].121

The sentences in (171) are all well formed. However, they are infelicitous in the contexts
given, which favor the prospective reading of almost over its approximative reading. This
means that prospective almost is for some reason unavailable in these sentences. Why
should this be?
Note that the problem cannot have anything to do with direct attachment, because the
surface forms of (171a-e) in no way force a direct attachment analysis. Each of these sen-
tences consists of a finite clause. Therefore, (clausal) prospective almost should in princi-
ple be available, as it is in (168a-b), repeated here as (172a-b):

(172) a. When his parents get home, Scott characteristically has almost finished his
homework.
b. They had almost gotten out the door when the phone rang.

There is, however, a critical difference between (171a-e) and (172a-b). The first group of
sentences have main predicates of a variety of categories: DP, NP, AdjP, and PP. In the
second group of sentences, the main predicate is a VP: [finished his homework] in (172a)
and [gotten out the door] in (172b).122 It would seem, then, that prospective almost can
only modify VPs (or constituents along their projection line). This generalization derives
two facts about the distribution of prospective almost:

(173) Prospective almost…

a. cannot directly attach to constituents other than VP.
b. cannot occupy its normal position in the clausal hierarchy when the main
predicate of the clause is a constituent other than VP.

The same facts hold for quickly, well, and fast/early. As we have seen, these adverbs
can certainly appear in clauses where the main predicate is a VP. But if we attempt to use
one of these adverbs in a clause whose main predicate is not a VP, the result is degrada-
tion to various degrees, often to the degree of full-on deviance:


























































121
This judgment is delicate, but I do not find (171e) fully felicitous in the context given. Regardless, how-
ever, there is a problem. If Logan is walking from the grass to the ocean via the beach (with the intention of
swimming), and is five meters away from the waterline, the following seems quite felicitous:

(xxiv) Logan is almost [PP in the ocean].

This suggests that, to the degree that (171e) is infelicitous in the context given for it, this is because, in our
conceptualization of the world, sitting in one’s jeep a few dozen meters from the water simply doesn’t
count as being almost in the ocean. (xxiv), meanwhile, seems like just a flagrant exception to the generali-
zation being developed in the text. I have no suggestions to offer.
122
I show the surface forms of the VPs, after evacuation of the subject (on the VP-Internal Subject Hy-
pothesis).

65

(174) a. ?*Logan was quickly [DP the president of the club].
b. ?*Logan was quickly [NP president of the club].
c. ?*Logan was quickly {[AdjP intelligent] / furious}.
d. ?*Logan was quickly [PP on the porch].

(175) a. ?*Logan was [DP the president of the club] well.123

b. ?*Logan was [NP president of the club] well.124
c. ?*Logan was {[AdjP intelligent] / furious} well.125
d. ?*Logan was [PP in a state of enlightenment] well.126

(176) a. ?*Logan was [DP the president of the club] fast.

b. ??Logan was [NP president of the club] fast.
c. ??Logan was {*[AdjP intelligent] / ?*furious} fast.
d. ??Logan was [PP on the porch] fast.127

The following judgments on early-sentences are relativized to the interpretation of Logan

was XP early on which it can be paraphrased by Logan was XP earlier than expected.

(177) a. ??Logan was [DP the president of the club] early.

b. ??Logan was [NP president of the club] early.
c. ??Logan was {*[AdjP intelligent] / ?furious} early.
d. ??Logan was [PP on the porch] early.128

The judgments above are in no way the end of the story. (177b), for instance, can be
made perfect by adding a PP: Logan was [NP president of the club] early in his career.


























































123
Cf. Logan was the president of the club, and he was GOOD at being the president of the club.
124
Cf. Logan was president of the club, and he was GOOD at being president of the club.
125
One can create contexts where [T was furious well] (with the stage-level predicate furious, as opposed to
the individual-level predicate intelligent) improves somewhat:

(xxv) [Context: Logan is an actor.]

a. ?*We asked Logan to be intelligent, and he was intelligent well.
b. ??We asked Logan to be furious, and he was furious well.

Here well is probably modifying the predicate be furious (plausibly a VP before be raises to T), with be
‘act’.
126
Cf. Logan was in a state of enlightenment, and he was GOOD at being in a state of enlightenment.
127
In the context of the generalization we are developing, it is surprising that this sentence is acceptable (as
is Dinner was [PP on the table] fast). It may be relevant in this connection that fast seems to interact differ-
ently with eventive and with stative predicates. When modifying an eventive predicate, as in (xxvi) below,
fast conveys that the event takes place in a short period of time. When it modifies a stative predicate (as in
(176c) and the aforementioned dinner-sentence), fast seems to convey that it only takes a short time for the
relevant state to obtain. In other words, what is short-lived is not the state (as one might expect in light of
the eventive case), but the run-up to the state.

(xxvi) Naomi painted the wall fast.

128
This unexpectedly acceptable sentence seems to be another as of yet unexplained counterexample (cf.
(xxiv) in fn. 121).

66

 Now let us consider just. As usual, the following judgments are relativized to the ret-
rospective reading of this adverb:

(178) a. ?*Logan just ate. [past]

b. ?*Logan had just eaten. [past perfect]
c. ?*Logan just eats. [present]
d. ?*Logan has just eaten.129 [present perfect]
e. ?*Logan will just eat. [future]
f. ?*Logan will have just eaten. [future perfect]

Just seems to pattern like once in requiring licensing by past tense or perfect aspect.
Therefore, our argument concerning once applies to just as well. The only category
whose extended projection can host either past tense or perfect aspect is V. Therefore,
just can directly attach to VP, but not to any other (i.e., to any nonspinal) category.
Last but not least, let us consider soon:

(179) a. ?*Logan soon grew tired of the food there. [finite clause, past]
b. ?*Soon I depart for Toronto. [finite clause, present]
c. ?*Logan will soon grow tired of the food there. [finite clause, future]
d. ?*I expect Logan to soon become impatient. [infinitival clause]
e. ?*I will see Logan [soon becoming impatient]. [active participial clause]
f. ?*We will see Logan soon criticized mercilessly. [passive participial clause]
g. ?*I will see Logan soon become impatient. [small clause]

Whatever it is that licenses soon, it must be fairly specific: it must be an element present
in the syntactic environments listed in the box in (179), and not in those listed outside the
box. Presumably this element is not internal to VP (since VPs can appear in all the envi-
ronments listed in (179)). It seems inevitable that it is somewhere in the clausal hierarchy.
If this is so, then soon is like once and just. Because it is licensed by some element in
the functional structure above VP, the only category to which it can directly attach is V.
This explains why it cannot directly attach to other (i.e., nonspinal) constituents.

II.C.2.d. Conclusions

Let us take stock. We have shown (this claim will be qualified shortly) that the
Cinque adverbs that cannot directly attach to nonspinal constituents are subject to the fol-
lowing restrictions:


























































129
This sentence would be felicitous in the description of the context of a scene in a script, for instance.

67

(180) Spinal adverb Restriction
a. once must be licensed by past tense or perfect aspect
b. quickly must modify a VP or something on its projection
line
c. just (retrospective read-
must be licensed by past tense or perfect aspect
ing)
d. soon must be licensed by some element in the F-
structure above VP
e. well
f. fast
must modify a VP or something on its projection
g. early line
h. almost (prospective
reading)

These restrictions prevent the adverbs in (180) from directly attaching to any constituent
that is not a VP (or on the projection line of a VP). But crucially, they do not only apply
in direct attachment contexts: they apply across the board, including in finite clauses.
The generalization that emerges from the data is the following:

(181) Spinal Adverb Generalization

If an adverb cannot directly attach to a nonspinal constituent, it cannot modify
that constituent in any context (for instance, in a finite clause).

At this point, some qualification is essential. Although the generalization in (181) seems
quite robust, it at present has a few counterexamples (see fn. 121, fn. 128, and the discus-
sion immediately following (177); cf. fn. 127). Indeed, there may be more.
Our investigation of the restrictions on spinal adverbs was far from complete; a
number of questions remain unanswered. If the difficulties currently posed by the coun-
terexamples can be overcome, then we will be able to make the following generalization:

(182) a. Direct Attachment Generalization

All Cinque adverbs are direct attachers.

Now, let us return to the Direct Attachment Hypothesis:

(183) Direct Attachment Hypothesis (= (129))

a. The Cinque hierarchy is correct.
b. A Cinquean adverb ordering A > B can be circumvented by direct attachment
of A to a clausal functional projection lower than B.
c. The circumvention described in b. is possible just in case A can directly at-
tach to nonspinals.

If (182) is correct, then (183) cannot be, for the following reason. Let A > B be a
Cinquean adverb ordering. (183b-c) together state that if A can directly attach to nonspi-
nals, it can directly attach to a clausal functional projection lower than B, and thereby ap-
pear to disobey the Cinque hierarchy. (182) entails that all Cinque adverbs can directly

68

attach to nonspinals. Therefore, (182) and (183) both being correct would predict that any
Cinquean adverb ordering could be subverted. But this is not the case ((130)).
To reiterate, we have just seen that (182) being true would mean that (183) is false.
Now, suppose (182) turns out not to be true. In that case, (183) will still be false. We
know this because of cases like the following. Unfortunately is a direct attacher:

(184) She talked to me [fortunately about really interesting things], but [UNfortunately
when I didn’t really have time for it]. (= (133))

Hence, (183) predicts that, for a Cinquean adverb ordering unfortunately > B, unfortu-
nately should be able to appear lower than B, thereby appearing to subvert the hierarchy.
But this is not true for B = allegedly:

(185)0 *Allegedly Wayne unfortunately committed a murder. (= (25b))

The Direct Attachment Generalization ((182)) at present seems quite robust, and is
hence interesting in its own right. But we have just seen that regardless of whether it
turns out to be correct, the Direct Attachment Hypothesis ((183)) cannot be. This means
that the Cinque hierarchy cannot be rescued from the numerous counterexamples in (130)
by appealing to direct attachment.

II.C.2.e. Direct attachment and (or vs.?) the Cinque hierarchy

Given that adverbs can directly attach to various types of nonspinals—in other
words, that they are “promiscuous”—it would be very strange if this promiscuity (ability
to attach directly) were suspended precisely in the case of VPs and other constituents on
their projection line. It would seem that in order for the theory to incorporate such a ban,
it would be necessary to stipulate it.
Therefore, by hypothesis, there is no such ban, and adverbs can directly attach to
VPs, and probably to functional projections dominating them too. But now we should
address the question we raised in fn. 119: how does direct attachment to VP and its “ex-
tensions” interact with the Cinque hierarchy?
We have just seen that we cannot assume the Cinque hierarchy to be correct and then
allow the orderings it predicts to be subverted via direct attachment (the Direct Attach-
ment Hypothesis, (183)), because this makes incorrect predictions ((184-185) and sur-
rounding discussion).
We have good evidence that direct attachment exists, whatever exactly it is—
whether adjunction ([XP AdvP XP]) or complementation ([Adv Adv° XP]). By contrast, the
Cinque hierarchy currently seems to be on less solid ground, in light of its considerable
empirical shortcomings with regard to English ((130)). So if direct attachment and the
Cinque hierarchy are in conflict, perhaps it is the latter that must yield, and direct attach-
ment can take care of everything.
This, of course, is nothing but the Adjunction Hypothesis (see §I.B.1). But because
free adjunction overgenerates, proponents of this hypothesis must give an account of how
the rigid adverb orders that do exist come about. The general intuition (or hope) is that

69

the source of these orders is semantic. I will now briefly mention two arguments against
this approach.
First, Richard Kayne (p.c., 01/2012) has argued that semantic explanations would be
unsatisfying in this domain. His argument is roughly as follows: take a rigid adverb or-
dering A > B, *B > A. The acceptable order A > B has an interpretation, but in many
cases the deviant order *B > A will be impossible to make sense of. In these cases, it is
tempting to attribute the rigid adverb ordering to semantics. But the absence of an inter-
pretation for *B > A is precisely what we want to explain. In other words, perhaps the
possibility of that order having an interpretation is not ruled out in principle, but we sim-
ply don’t have access to what the interpretation could be because the order is ill formed.
Secondly, Cinque (2004:3, fn. 5) makes the following point:

In the absence of a complete understanding of the semantics of each adverb class, from
which its scope with respect to the other adverb classes can be made to follow, a claim such
as Ernst’s (2002: 130–133) that, as a consequence of their lexicosemantic properties,
speaker- and subject-oriented adverbs have a rigid ordering while quantificational and aspec-
tual adverbs can have a variable ordering (with meaning differences), and participant PPs
have a free ordering (with no meaning differences), essentially restates the question rather
than explaining it.

It is certainly true that semantic accounts of the syntax of adverbs must be explicit, and
must show how the semantic machinery they invoke predicts the distribution of adverbs.
I cannot now say anything definitive to contribute to this longstanding debate one
way or the other. My primary aim in this chapter has been to gauge the success of the
Cinque hierarchy in accounting for the facts of adverb ordering in English—noting its
correct predictions, on the one hand, and, on the other, calling attention to the empirical
inadequacies it must address in order to remain a viable hypothesis concerning the syntax
of adverbs.

III. When adverbs are interpreted higher than they appear

III.A. Introduction

III.A.1 Semantics and syntax

In this chapter, we will investigate a phenomenon in English in which an adverb ap-

pears to be attached to a constituent within an embedded clause, but is interpreted as
modifying a predicate in a higher clause.
Take the following sentence:

(186) John considers Mary probably a murderer.

In (186), probably appears to be inside the small clause [Mary probably a murderer],130
which occurs as complement to the verb considers. All else being equal, this predicts that


























































130
I abstract away from the question of how the subject of the small clause gets Case, and whether this in-
volves it moving into the matrix clause.

70

probably should be interpreted within the small clause—i.e., should contribute to the in-
terpretation of the propositional attitude complement.
This is indeed possible: (186) has a reading that can be paraphrased as John’s
evaluation of Mary is “She’s probably a murderer.” On this reading of (186), the proba-
bly tells us (among other things) that John is not certain that Mary is a murderer.
If we adopt the traditional view that a complement is the sister of its head, we will
analyze the considers in (186) as c-commanding probably. Just as probably occurs lower
than considers syntactically on this view, we can also say that it is interpreted lower than
considers semantically, because it is interpreted as modifying the embedded clause,
rather than the matrix verb considers or the clause built on top of it.
Because (at least on the surface) probably occurs lower than considers syntactically
in (186), and on the reading at hand it is also interpreted lower than considers semanti-
cally, I will refer to this reading of (186) as its SURFACE SCOPE READING (or INTERPRETA-
TION).
Now, as it turns out, (186) also has another interpretation, on which probably appears
to modify the extended projection of consider. This interpretation can be brought out by
embedding (186) in the following context:

(187) A: What does John consider Mary to be?

B: John considers Mary probably a murderer—but to be honest I’m not really
SURE what John considers Mary to be.

The prosodic characteristics of this interpretation will be discussed shortly, but first let us
say a few words about the interpretation itself. In (187B), if the clause John considers
Mary probably a murderer is interpreted on its surface scope reading—with probably
modifying the small clause—then the matrix assertion is not being qualified by that
probably or any other epistemic adverb.131 Therefore, the matrix clause is interpreted as
making an assertion that the speaker is confident of the truth of.132 But this inference is
incompatible with the continuation of (187B) (from but to the end).133
The fact that this continuation eliminates the surface scope reading of the first con-
junct of (187B) allows us to more easily consider its INVERSE SCOPE reading, on which
probably modifies the matrix clause. On this reading, (188) is paraphrasable as follows:


























































131
Cf. Cinque (1999:86, §4.7): “epistemic modality expresses the speaker’s degree of confidence about the
truth of the proposition (based on the kind of information he/she has).”
132
This is presumably due to the second submaxim of Grice’s (1975:46) Maxim of Quality: “Do not say
that for which you lack adequate evidence.” If the speaker did not have adequate evidence for the claim
John considers [Mary probably a murderer], but only for a hedged claim such as John probably considers
[Mary probably a murderer], then by the aforementioned submaxim, (s)he would have to make the latter
claim, assuming that the Cooperative Principle was being followed (ibid:45).
133
In order for (187B), on the surface scope reading of the first conjunct, to not be contradictory, it would
have to include an explicit corrective marker such as actually. Unlike actually, but is not sufficient for a
self-correction, as shown by the following contrast:

(xxvii) a. #Layla is in France right now—actually, she’s not in France right now.
b. #Layla is in France right now—but she’s not in France right now.

(xxviib) is only felicitous if the speaker is exploiting its contradictoriness for rhetorical effect.

71

(188) A: What does John consider Mary to be?
B: John probably considers Mary a murderer—but to be honest I’m not really
SURE what John considers Mary to be.

III.A.2 The prosodic characteristics of the inverse scope reading

The inverse scope reading of the clause in question favors the pronunciation of
probably at a higher pitch than it would have otherwise (i.e., on the surface scope read-
ing). It seems to me that, at least in a context that strongly favors the inverse scope inter-
pretation (such as that in (187)), this pitch accent is not absolutely required, but it is more
natural to have it than not to. Including the pitch accent makes it quite easy to get the in-
verse scope reading of the clause even in the absence of any context:

(189) John considers Mary ↑probably a murderer.

The two intonation contours under discussion can be seen in the spectrograms in
(190) and (191), which I made by recording myself in Praat. The first of these, (190), was
made by recording the sentence John considers Mary probably a murderer on its surface
scope interpretation. The second, (191), was made by recording the “same” sentence on
its inverse scope interpretation.

(190)

72

(191)

By comparing the fundamental frequency (F0) lines in the two spectrograms, we can see
that the probably in (191) is pronounced with a marked rise in pitch compared to the
probably in (190).
As mentioned above, (190) was made by recording the relevant sentence on its sur-
face scope reading and (191) was made by recording it on its inverse scope reading. It is
important to note, however, that the pitch accent on the adverb is neither necessary to get
nor sufficient to force the inverse scope reading. It is not necessary (at least for me) be-
cause, as mentioned previously, for me the inverse scope reading is available even with-
out the pitch accent at least when the context favors it. The pitch accent is also not suffi-
cient to force the inverse scope reading because its presence is in fact compatible with
either reading.
It will be useful for us, in the rest of this chapter, to be able to avail ourselves of the
following term:

(192) Interpretive escape (or semantic displacement)

The phenomenon wherein an adverb apparently located in one clause is inter-
preted as modifying the predicate of a higher clause.

I have chosen this term because an adverb affected by this phenomenon seems as though
it’s escaping from the clause it’s in and moving to a higher clause for purposes of inter-
pretation—and all this with no effect on phonetic form. This, however, is not an analysis
but an impressionistic description. Some possible explanations for the phenomenon will
be considered in §III.B.

III.A.3 Do the surface scope and inverse scope readings differ in information structure?

We have established that (193a) below has an inverse scope reading on which it can
be paraphrased by (193b):

73

(193) a. John considers Mary ↑probably a murderer.134
b. John probably considers Mary a murderer.

If these two sentences are paraphrases of each other, a natural question is whether they
are in any way semantically or pragmatically different, or whether they are in true free
variation. This question is particularly interesting in light of Bolinger’s (1968:127) claim
that “[a] difference in syntactic form always spells a difference in meaning.”135 If we are
looking for differences between (193a) and (193b), a natural domain to look in is infor-
mation structure.
I have not as of yet probed in detail the topic–comment and focus–presupposition ar-
ticulations of these sentences to determine whether they differ along those dimensions.
But (193a) and (193b) definitely do differ in their information packaging in at least one
respect, which is worth mentioning even though it is rather obvious. They differ in sur-
face word order, and hence in order of presentation of information. In (193b), probably
occurs in the default position for epistemic adverbs. In (193a), it occurs unusually late, in
linear terms. The consequences of this are as follows. A hearer processing (193a) will
begin comprehending it as an epistemically unmarked136 sentence about what John con-
siders Mary to be. (S)he will later have to revise this semantic representation to reflect the
fact that the sentence is not epistemically unmarked, but rather hedged to a ‘probably’
level of certainty.
This difference in information packaging will not be reflected in the paraphrases
provided for interpretive escape sentences, for instance ‘John probably considers Mary a
murderer’ for John considers Mary - ↑probably a murderer.

III.B. Some possible explanations for interpretive escape

III.B.1. Escaped adverbs as parentheticals

III.B.1.a. A first attempt

Norvin Richards, in personal communication (03/2012), proposed to me the follow-

ing (I have fleshed it out a bit):

(194) The parenthetical analysis of interpretive escape

a. Suppose that the escaped adverb is really a parenthetical.
b. Assume that parentheticals always receive a matrix interpretation.
c. Then, the escaped adverb’s status as a parenthetical explains why it receives a
matrix interpretation, despite appearing to be in an embedded clause.

Support for the assumption in (194b) comes from sentences like the following:


























































134
I include an indication of the marked pitch accent because this accent helps to bring out the inverse scope
reading, and this is particularly helpful when there is no context to favor this reading. Recall, though, that
the pitch accent is not necessary to get the inverse scope reading (§III.A.2).
135
For some discussion of this claim (Bolinger’s Dictum), see Zyman (MS, 2011:16).
136
On the marked and unmarked values of his Modepistemic° head, see Cinque (1999:128-129).

74

(195) And then Chris said that Audrey—and this was truly shocking—was getting mar-
ried.

In (195), the parenthetical [and this was truly shocking] appears to occur inside the em-
bedded clause [that Audrey was getting married], which is the complement of said. But
the parenthetical cannot be interpreted as conveying anything that Chris said. Rather, it is
obligatorily interpreted as a comment by the speaker of the whole sentence. The paren-
thetical’s inability to contribute to the denotation of the speech report (the embedded
clause) would be explained if parentheticals were unable to receive embedded interpreta-
tions in general.
As pointed out by Richards (ibid.), the parenthetical account of interpretive escape as
formulated in (94) predicts that adverbs should only be able to “escape” into the root
clause for purposes of interpretation, never into an intermediate clause. But this does not
seem to be the case. Consider the following:

(196) [John said [that Bill believes [that Mary likes - probably Fred]]].

In both my and Richards’s judgment, (196) can be interpreted with probably modifying
the intermediate clause, given the right prosody (on which more soon). On this reading,
(196) can be paraphrased as follows: [John said [that Bill probably believes [that Mary
likes FRED]]]. Call this the MEDIAL INTERPRETATION of (196), since it involves interpret-
ing probably in the intermediate clause.
For me, (196), on this interpretation, is naturally uttered with the prosodic character-
istics visible in the following spectrogram:

(197)

Impressionistically, these characteristics are as follows:

75

(198) Prosodic characteristics of medial interpretations (with special reference to
(196-197))
a. Probably is realized with the salient High pitch accent (henceforth, H*137) that
facilitates “escaped” readings of adverbs quite generally.
b. Probably is also preceded by a pause, or, more likely, a lengthening of the final
[s] of likes: apparently, giving probably Fred some autonomy as a prosodic unit
also helps the escaped reading of probably.
c. The string Bill believes that Mary likes is temporally compressed. This may, like
the features in b., be a strategy for giving probably Fred some prosodic inde-
pendence and thereby facilitating the escaped reading of probably.
d. John said is preferentially separated to some degree from the rest of the sentence.
If this is a strategy for turning [that Bill believes that Mary likes probably Fred]
into a sort of derived root clause, this may be reason to believe that there is a
preference for interpreting escaped adverbs in (basic or derived) root clauses.
Critically, however, the partial separation of John said is not obligatory, indicat-
ing that it is possible to interpretively displace adverbs into intermediate clauses
even without prosodically rendering them derived root clauses.

III.B.1.b. A second attempt

We have seen that the parenthetical analysis of interpretive escape as formulated in

(194) incorrectly predicts that medial interpretations should be impossible. The reason it
predicts this is that it incorporates the assumption ((194b)) that parentheticals always re-
ceive a matrix interpretation.
Let us revise this assumption in order to allow parentheticals to receive medial inter-
pretations. How can we do this? One possible revision would be the following:

(199) Condition on Parentheticals

A parenthetical cannot be interpreted in the minimal clause in which it appears to
occur (henceforth, its HOST CLAUSE).

This, however, will not work. Consider the following:

(200) [Lauren believes—and this is truly shocking—[that Alex is about to eat twelve
donuts]].

The condition in (199) predicts that, in (200), it should only be possible to interpret [and
this is truly shocking] in the embedded clause. But this gets the facts exactly backwards.
The parenthetical must be interpreted as a comment by the speaker on Lauren’s belief. It
cannot contribute to the interpretation of the attitude complement. In other words, (200)
cannot mean ‘Lauren believes that [Alex is about to eat twelve donuts, a state of affairs
she finds truly shocking]’.138

























































137
The asterisk indicates that this pitch accent is aligned with the stressed syllable of the word on which it is
realized.
138Another problem with (199) is that it predicts that parentheticals should be barred from monoclausal

sentences, contrary to fact: Alex—and this is truly shocking—ate twelve donuts yesterday.


76

 This suggests that a parenthetical cannot be interpreted in a clause lower than its host
clause. Therefore, let us revise (199):

(201) Condition on Parentheticals (revised)

A parenthetical cannot be interpreted lower than its host clause.

Now, let us use this generalization to revise our parenthetical analysis of interpretive es-
cape:

(202) The parenthetical analysis of interpretive escape (revised)

a. An escaped adverb is really a parenthetical.
b. A parenthetical cannot be interpreted lower than its host clause.
c. By a. and b., an escaped adverb’s status as a parenthetical explains why it can
be interpreted in any clause (matrix or intermediate) higher than its host
clause.139

The impetus behind the revision in (202) was the desire to predict that adverbs can
be interpretively displaced into intermediate clauses, not just matrix clauses. (202) does
predict this, but the analysis crucially relies on the assumption (implicit in (202b); cf. fn.
139) that parentheticals can receive medial interpretations (in addition to matrix interpre-
tations). Is this assumption correct? Apparently not:

(203) And then [Chris said [that Audrey believed [that Justin—and this was truly
shocking—was getting married]]].

In (203), the parenthetical [and this was truly shocking] cannot be interpreted as contrib-
uting to the denotation of what Chris said. In other words, (203) cannot mean ‘And then
Chrisi said that [Audrey believed that Justin was getting married], a state of affairs hei
found truly shocking’. Once again, the parenthetical must instead be interpreted as a
comment by the speaker.
Although we cannot here embark on a deeper investigation of parentheticals, the em-
pirical picture right now suggests that they cannot be interpreted in embedded clauses
((195), (203), and following discussion). A parenthetical must instead be interpreted as a
comment by the speaker. Perhaps it is inappropriate to call this a matrix interpretation,
but it is definitely not an embedded interpretation. This suggests that we cannot analyze
escaped adverbs as parentheticals, because these adverbs can be interpreted in embedded
clauses—in the case of medial interpretations ((196) and following discussion).

III.B.2. The two-fragment analysis

We have just seen that interpretive escape apparently cannot be analyzed as arising
from the use of adverbs parenthetically. So how does it arise?

























































139
This follows because, if the only restriction on where parentheticals can be interpreted is that a paren-
thetical cannot be interpreted lower than its host clause ((186b)), then nothing prevents a parenthetical from
being interpreted in any clause (matrix or intermediate) higher than its host clause. Of course, it is also the
case that, at present, nothing prevents it from being interpreted in its host clause.

77

 Let us return to (189), repeated here (with some annotations) as (204):

(204) ‘John considers [Mary ↑probably a murderer].

‘John probably considers Mary a murderer.’

One possible analysis of (204) is that it actually consists of two separate sentences, the
first incomplete. This analysis is shown in (205), where the two sentences—henceforth
FRAGMENTS—are bracketed, and labeled F1 and F2 for convenience.

(205) [F1 John considers Mary—] [F2 probably a murderer].140

The essence of this analysis is that F1 is a structure that is never built to completion.
This analysis makes interpretive escape sentences precisely parallel to exchanges like
(206), which might occur when A does not know what small clause predicate would
make his/her utterance true, and so the missing information is helpfully supplied by B.

(206) A: John considers Mary…

B: Probably a murderer.

On the most salient reading of (206B), the probably is interpreted as modifying the clause
built on top of considers (the matrix clause, begun by A). However, the probably can also
be interpreted within the small clause, which is distributed over the two conversational
turns.
The appeal of the TWO-FRAGMENT ANALYSIS ((205)) is that it gives a unified treat-
ment of two apparently separate phenomena: interpretive escape ((204)) and exchanges
like (206). On this analysis, (204), despite seeming like a single sentence, actually con-
sists of two syntactically unconnected fragments, just like (206). Therefore, the ambiguity
of (204) can be ascribed to whatever it is that makes (206B) ambiguous.
Now, there is a possible problem with this analysis. In the cases that look like a sin-
gle sentence (e.g., John considers Mary probably a murderer), the more salient interpre-
tation is the surface scope reading (with probably interpreted within the small clause),
unless probably is pronounced with a prominent H* or the context strongly favors inter-
pretive escape. In the dialogue case, on the other hand—(206)—the more salient interpre-
tation is the inverse scope reading, with probably interpreted as modifying the matrix
clause rather than the small clause. However, it is hard to know what to make of this dif-
ference in the absence of a fuller understanding of the various prosodic factors at play.
A more serious shortcoming of the two-fragment analysis of interpretive escape is
that it does not obviously add to our understanding of the phenomenon it is supposed to
explain. This is because it is not obvious why “promoting” the [Adv XP] structure (part
of a small clause) to the status of a separate sentence should give rise to a new interpreta-
tion, making it possible to interpret the adverb not only in situ but also in a higher


























































140
F1 in (205), with different prosody, can in fact stand alone as a full sentence, but then it gets a radically
different interpretation. (205) as a whole makes an assertion about John’s evaluation of Mary, but the sen-
tence John considers Mary means roughly ‘John {thinks about / reflects on} Mary’.

78

clause.141 It would seem, then, that we currently have no clear basis for choosing between
the one-sentence ((204)) and two-sentence ((205)) structures in analyzing interpretive es-
cape.

III.B.3. Summary

Our brief investigation of parentheticals suggested that these constituents cannot be

interpreted in embedded clauses. Escaped adverbs can, when they are interpreted not in
the matrix clause but in an intermediate clause (the so-called “medial interpretation”).
This suggests that escaped adverbs cannot be analyzed as parentheticals.
The two-fragment analysis of interpretive escape, on the other hand, may be correct.
For the time being, I will set this analysis aside, and treat interpretive escape structures as
single sentences. In other words, I will assume the structure in (204) and not that in (205).
The issue of whether an interpretive escape structure is a single sentence ((204)) or a
paratactic fragment pair ((205)) will be taken up again in §III.D.

III.C. Testing the limits of interpretive escape

We have just engaged in some preliminary exploration of interpretive escape, and

examined some possible explanations for it. We will return to the theoretical issues raised
by the phenomenon later in this chapter. But if we want to gain a better understanding of
interpretive escape, it will be useful for us to subject it to further empirical investigation.
It is to this project that we now turn.
This section is entitled “Testing the limits of interpretive escape” because we will be
investigating how widely available interpretive escape is, in a couple of senses:

(207) Research questions

a. Which adverbs can be semantically displaced?
b. How big a structure can an adverb semantically escape from?

The first of these questions is self-explanatory. The adverbs we will be using to an-
swer it are the set of adverbs we have been using to exemplify the Cinque hierarchy,
since this set contains representatives of a wide variety of adverb classes.
The second question ((207b)) requires a little more elaboration. All the instances of
interpretive escape we have seen so far involve adverbs semantically escaping from small
clauses. Small clauses are considered small—i.e., less rich in syntactic structure than
other types of clauses—because they cannot host auxiliaries or Tense (see §II.C.2.a). On
the assumption that small clauses indeed are smaller than other clauses, a natural question
that arises concerning interpretive escape is whether it can take place out of larger
clauses. The types of (putatively) larger clauses we will use to answer this question are
infinitivals and finite clauses. We will also investigate whether interpretive escape is un-
bounded.


























































141
One could perhaps argue, on a somewhat functionalist note, that a full sentence/utterance is the kind of
thing that can host high adverbs such as probably, but this would need to be elaborated in detail to qualify
as a serious proposal.

79

 The primary “articulation” of this chapter will be according to the second question
((207b)), with the investigation proceeding from the putatively smallest structures to the
putatively largest (so the order will be small clauses, infinitivals, finite clauses, multi-
clausal structures). Within each of these subinvestigations, we will attempt to interpre-
tively displace the full range of Cinque adverbs.

III.C.1. Interpretive escape out of small clauses

We already know that adverbs can semantically escape from small clauses ((204)).
The next question, then, is which adverbs can do this. And the answer, apparently, is all
of them. (There are a couple of exceptions, but I will argue that they involve interference
from independent factors.) Let us begin to examine the data.

(208) ‘I find his opinions - ↑honestly frightful.142

‘I honestly find his opinions frightful.’

(208) is shown with a paraphrase of its escaped reading. The honestly in this sentence can
also be interpreted within the small clause it appears to be in, yielding an interpretation
paraphrasable as follows: I find [his opinions to be such that I could honestly say that
they were frightful] (note that here honestly is not modifying the matrix clause). Because
what we are testing is the (un)availability of escaped readings, the sentences to be exam-
ined will be shown with paraphrases of their escaped readings only (as in (208)).
Proceeding to the next adverb, we have:

(209) ‘She finds me - fortunately funny, but UNfortunately unprincipled.

‘Fortunately, she finds me funny, but UNfortunately, she finds me unprincipled.’

In (209), the escaped adverbs are contrastive topics, and the corresponding comments are
contrastive foci (funny and unprincipled). The contrastive topic structure seen in (209) is
shared by most of the interpretive escape sentences we will be analyzing, so it will be
worthwhile to note the prosodic characteristics of this kind of structure. These can be
seen in the following spectrogram:


























































142
On the up arrow, see fn. 134. The hyphen preceding the adverb is an ad hoc notation for a brief pause. It
is used here because, as discussed in (198b), setting off the [Adv XP] structure with a brief pause gives it
some prosodic autonomy, which apparently helps bring out the escaped reading of the adverb (just as the
pitch accent represented by the arrow does).

80

(210)

The relevant intonational characteristics of (209-210) may be informally schematized as

follows:

(211) ‘She finds me - fortunately funny, but UNfortunately unprincipled.

H* ↓H* H*
H*
H*
H*

The linearly first escaped adverb (fortunately) receives a very high pitch accent,143 and its
corresponding focus (funny) also receives a pitch accent, but this one is much lower. The
second escaped adverb (unfortunately) also receives quite a high pitch accent (one higher
than the one on the first focus), but this pitch accent is lower than the one on the first ad-
verb, presumably due to the overall downstep characteristic of declaratives. Finally, the
linearly second focus also receives a pitch accent, but this one is the lowest one of all.
For me, it is not essential that the pause (or lengthening) preceding fortunately be
long. But other speakers may find that, for them, the pause/lengthening must be long in
order for interpretive escape to be available.
I will continue to represent sentences like (208) with up arrows. However, for conci-
sion’s sake, I will not reproduce the schematic in (211) under every interpretive escape
sentence with a contrastive topic structure. The reader should nevertheless keep the pro-
sodic characteristics of sentences like (211) in mind.
Continuing:

(212) ‘John considers Mary - ↑allegedly a murderer.

‘John allegedly considers Mary a murderer.’


























































143
The H accent that often accompanies an escaped adverb can be realized on the adverb’s stressed syllable
alone (H*) or over the entire word. For me, an adverb pronounced in the latter manner is particularly ame-
nable to being interpreted as escaped.

81

(213) ‘John considers Mary - ↑probably a murderer.
‘John probably considers Mary a murderer.’

(214) ‘I considered John - once a great musician, but last week a lousy one.
‘I once considered John a great musician, but last week considered him a lousy
‘one.’

(215) ‘I considered John - then a great musician, but last week a lousy one.
‘I at that point considered John a great musician, but last week considered him a
‘lousy one.’

(216) ‘John considers Mary - ↑perhaps a murderer.

‘Perhaps John considers Mary a murderer.’

(217) ‘John considers Mary - possibly a pain, but necessarily a huge bore.
‘John possibly considers Mary a pain, but necessarily considers her a huge bore.’

(218) ‘John considers Mary - possibly a pain, but not necessarily a huge bore.
‘John possibly considers Mary a pain, but doesn’t necessarily consider her a huge
‘bore.’

(219) ‘I consider John - usually an idiot, but occasionally a genius.

‘I usually consider John an idiot, but occasionally consider him a genius.’

Edwin Williams has suggested to me that, when the small clause contains a quantifica-
tional adverb such as usually and the matrix subject is a generic plural ((220)), it may be
especially easy to get a (particular type of) escaped reading: one in which the adverb not
only gets a matrix interpretation but, indeed, outscopes the matrix subject.

(220) ‘People consider John - usually an idiot, but occasionally a genius.

As it turns out, (220) has at least three interpretations, which are listed in (221). The in-
terpretation mentioned by Williams is (221c).

Description of interpretation Paraphrase

(221) a. adverbs interpreted in situ ‘People’s evaluation of John is “He’s usually
‘an idiot, but occasionally a genius.” ’
b. adverbs interpreted as es- ‘People usually consider John an idiot, but
caped, but under people ‘occasionally consider him a genius. They
‘always all agree.’
c. adverbs interpreted as es- ‘Most people consider John an idiot, but
caped and outscoping people ‘some people consider him a genius.’

Moving on:

82

(222) ‘I considered John - first an idiot, then a genius, and then again an idiot.
‘I first considered John an idiot, then considered him a genius, and then again
‘considered him an idiot.’

In (222), interpreting the adverbs as escaped is actually obligatory. This is presumably

because first and then must be licensed by either tense or aspect in the same clause, and
neither of those licensors is present in the small clause in (222).144 This explanation is
supported by the following facts. A first and then apparently located in an embedded
clause can be interpreted as modifying this clause when it is marked for past tense
((223a)), perfect aspect ((223b)), or both ((223c)). (The relevant paraphrases are provided
beneath the sentences.) In (223d), the embedded clause is marked for neither past tense
nor perfect aspect, and the apparently embedded first and then are obligatorily interpreted
as modifying the matrix predicate.

(223) a. ‘I believe that [John was first an idiot and then a genius].
‘I believe the following: “John was first an idiot and then a genius.” ’

b. ‘I consider John [to have been first an idiot and then a genius].
‘I hold the following opinion: “John was first an idiot and then a genius.” ’

c. ‘I believe that [John was acting first like a lion and then like a tiger].
‘I believe the following: “John was first acting like a lion and then acting like
‘a tiger.” ’

d. ‘I consider John [first an idiot and then a genius].

‘I (habitually) [first consider John an idiot and then consider him a gen-
‘ius].’145

Continuing:

(224) ‘I consider John - often an idiot, but occasionally a genius.

‘I often consider John an idiot, but occasionally consider him a genius.’146


























































144
We can tell that again is not subject to the same condition, since it can be interpreted in embedded
clauses not marked for tense or aspect. This is shown by the following sentence (on the interpretation
given):

(xxviii) ‘I consider [John again an idiot].

‘I hold the following opinion: [John is again an idiot].’
145
Consider the following sentence, on the interpretation given:

(xxix) ‘I expect [John to act first like a lion and then like a tiger].
‘I have the following expectation: [John will first act like a lion and then act like a tiger].’

If the analysis of first/then licensing in the text is correct, then the infinitival complement in (xxix) must be
marked for tense or aspect covertly. The relevant marking—which has to yield a future-oriented interpreta-
tion—is presumably selected for by the verb expect.

83

(225) ‘The playwright made the character - intentionally a bore, but accidentally a
‘genius.
‘The playwright intentionally made the character a bore, but accidentally made
‘him/her a genius.’

(226) ‘I could imagine John - quickly a violent hoodlum, but only much more slowly a
‘distinguished professor.
‘I could quickly imagine John a violent hoodlum, but I could only much more
‘slowly imagine him a distinguished professor.’

(227) ‘I considered John - already brilliant, but not yet a full-blown genius.
‘I already considered John brilliant, but didn’t yet consider him a full-blown gen-
‘ius.’

(228) ‘I considered John - still brilliant, but no longer a full-blown genius.

‘I still considered John brilliant, but no longer considered him a full-blown gen-
‘ius.’

(229) ‘I consider John - sometimes a pain, but always a huge bore.

‘I sometimes consider John a pain, but always consider him a huge bore.’147

The following judgment is relativized to retrospective just, as usual:

(230) **I imagined John - habitually an idiot, but just a genius.

* ‘I habitually imagined John an idiot, but just imagined him a genius.’

(231) ‘I predict that I’m going to consider John - soon an idiot, but eventually a genius.
‘I predict that I’m soon going to consider John an idiot, but eventually going to
‘consider him a genius.’


























































146
Changing the matrix subject to a generic plural introduces a three-way ambiguity parallel to the one in
(220-221):

(xxx) People consider John - often an idiot, but occasionally a genius.

a) ‘People’s evaluation of John is “He’s often an idiot, but occasionally a genius.” ’
b) ‘People often consider John an idiot, but occasionally consider him a genius. They always all
agree.’
c) ‘Many people consider John an idiot, but some people consider him a genius.’

As we will see, all the quantificational adverbs we are examining display this property. The relevant data
will be provided in footnotes.
147
Cf.:

(xxxi) People consider John - sometimes a pain, but always a huge bore.
a) ‘People’s evaluation of John is “He’s sometimes a pain, but always a huge bore.” ’
b) ‘People sometimes consider John a pain, but always consider him a huge bore. They always all
agree.’
c) ‘Some people consider John a pain, but everyone considers him a huge bore.’

84

In (231), the adverbs must be interpreted as escaped. In the case of soon, this is because
this adverb must be licensed by an element that cannot appear in a small clause (cf. (179)
and following discussion).

(232) ‘I considered John - briefly an idiot, but subsequently a genius.

‘I briefly considered John an idiot, but subsequently considered him a genius.’

(233) ‘I find John - characteristically boring, but occasionally enthralling.

‘I characteristically find John boring, but occasionally find him enthralling.’148

In (233), it is very natural to interpret characteristically within the small clause it appears
in, but the escaped reading is also possible, though the H accent on this adverb and the
pause preceding it may have to be made more prominent to bring this out.

(234) *I imagined John - first an idiot, and later on almost a genius. I soon thought
*better of THAT, though.
‘I first imagined John an idiot, and later on almost imagined him a genius. …’

(234) is acceptable on the interpretation ‘I first imagined [John an idiot], and later on
imagined [him coming close to being a genius].…’ Note, however, that this interpretation
involves semantically displacing first and later on to the matrix clause, but not almost,
which is interpreted in the small clause it seems to be in. The reading given immediately
beneath (234)—on which almost is interpreted in the matrix clause—is impossible.
The unacceptability of (230) (with retrospective just) and (234) (with almost es-
caped) will be explained in §III.C.4.d. For now, however, we will return to our systematic
semantic displacement of the Cinque adverbs.
The next adverb to test is completely. It is difficult to find a small-clause-taking verb
such that the VP it heads can be idiomatically modified by completely. Therefore, we will
have to settle for a less-than-ideal test in this case; but this is preferable to not testing
completely at all. Consider the following dialogue:

(235) A: (?)?What could you imagine John as?

B: ()??I could partially imagine John a distinguished professor, but could com-
(?)?pletely imagine him a wildly successful punk rocker.149

























































148
Cf.:

(xxxii) People find John - characteristically boring, but occasionally enthralling.

a) ‘People’s evaluation of John is “He’s characteristically boring, but occasionally enthralling.” ’
b) ‘People characteristically find John boring, but occasionally find him enthralling. They always
all agree.’
c) ‘Most people find John boring, but some people find him enthralling.’
149
The second conjunct in the coordinate structure (with completely) superficially sounds far more accept-
able than the first—in fact, to me it sounds perfect. This is attributable to interference from the emphatic
affirmation reading of completely (fn. 84), even though this completely is colloquial and small-clause-
taking imagine (which occurs in the same sentence) is literary or educated. If we are careful to exclude the
emphatic affirmation reading of completely, the second conjunct under discussion becomes bizarre. If we
take this conjunct, “expand” it into a full sentence, and add the completely of emphatic affirmation, we

85

When a VP headed by small-clause-taking imagine is modified by partially or com-
pletely, the result is bizarre and degraded. But if these adverbs are fundamentally capable
of interpretive escape, then an interpretive escape sentence that paraphrases (235B)
should be approximately as acceptable as (235B). And this does seem to be the case:

(236) A: (?)?What could you imagine John as?

B: (?)?I could imagine John - partially a distinguished professor, but com-
(?)?pletely a wildly successful punk rocker.

If anything, (236B) actually seems slightly more acceptable than (235B). This is probably
a surface word order effect. In (235B), each of partially and completely is adjacent to the
head of the VP it is problematically modifying. In (236B), on the other hand, the de-
graded modification is to some degree obscured because the aforementioned adjacency
does not obtain. At any rate, the fact that (235B) and (236B) are approximately equal in
acceptability strongly suggests that completely (and partially) can in fact be semantically
displaced.
As for the last three Cinque adverbs—well and fast/early—we cannot test whether
they can be semantically displaced. This is presumably a reflex (somehow) of the fact
that they obligatorily follow the core VP on the surface, since it is unlikely to be a coinci-
dence that these two properties pick out exactly the same subset of Cinque adverbs.
The result of this section, then, is that all the Cinque adverbs can semantically escape
from small clauses except for just and almost. As mentioned above, these counterexam-
ples will be explained in §III.C.4.d. For now, however, we will turn our attention to in-
finitivals.

III.C.2. Interpretive escape out of infinitival clauses

In this section, we will show the following. First, interpretive escape is indeed possi-
ble from infinitival clauses. Secondly, all the Cinque adverbs can semantically escape
from infinitival clauses—except for just and almost. This precisely mirrors the results of
the previous section.

(237) ‘I find his opinions to be - ↑honestly frightful.

‘Honestly, I find his opinions to be frightful.’

(238) ‘She finds me to be - fortunately funny, but UNfortunately unprincipled.

‘Fortunately, she finds me to be funny, but UNfortunately, she finds me to be un-
‘principled.’

(239) ‘John considers Mary to be - ↑allegedly a murderer.

‘John allegedly considers Mary to be a murderer.’


























































force the lower completely to be interpreted as belonging to the same adverb class as partially. Unsurpris-
ingly, the result is degraded: ??I COMPLETELY could completely imagine John a wildly successful punk
rocker.

86

(240) ‘John considers Mary to be - ↑probably a murderer.
‘John probably considers Mary to be a murderer.’

(241) ‘I considered John to be - once a great musician, but last week a lousy one.
‘I once considered John to be a great musician, but last week considered him to
‘be a lousy one.’

In (241) (and (242)), a great musician and a lousy one are intended to be interpreted as
individual-level predicates.

(242) ‘I considered John to be - then a great musician, but last week a lousy one.
‘I at that point considered John to be a great musician, but last week considered
‘him to be a lousy one.’

(243) ‘John considers Mary to be - ↑perhaps a murderer.

‘Perhaps John considers Mary to be a murderer.’

(244) ‘John considers Mary to be - possibly a pain, but necessarily a huge bore.
‘John possibly considers Mary to be a pain, but necessarily considers her to be a
‘huge bore.’

(245) ‘John considers Mary to be - possibly a pain, but not necessarily a huge bore.
‘John possibly considers Mary to be a pain, but doesn’t necessarily consider her
‘to be a huge bore.’

(246) ‘I consider John to be - usually an idiot, but occasionally a genius.

‘I usually consider John to be an idiot, but occasionally consider him to be a gen-
‘ius.’150

(247) ‘I considered John to be - first an idiot, then a genius, and then again an idiot.
‘I first considered John to be an idiot, then considered him to be a genius, and
‘then again considered him to be an idiot.’

(248) ‘I consider John to be - often an idiot, but occasionally a genius.

‘I often consider John to be an idiot, but occasionally consider him to be a gen-
‘ius.’151


























































150
Cf.:

(xxxiii) ‘People consider John to be - usually an idiot, but occasionally a genius.

a) ‘People’s evaluation of John is “He’s usually an idiot, but occasionally a genius.” ’
b) ‘People usually consider John to be an idiot, but occasionally consider him to be a genius.
‘They always all agree.’
c) ‘Most people consider John to be an idiot, but some people consider him to be a genius.’
151
Cf.:

87

In the following two sentences, for some reason, the escaped reading is harder to bring
out than usual. This can, however, be done—by placing a pause before the linearly first
adverb and making the H* on this adverb quite prominent.

(249) ‘I declared John to be – involuntarily an idiot, but intentionally a pain.

‘I involuntarily declared John to be an idiot, but intentionally declared him to be a
‘pain.’

(250) ‘I declared John to be – quickly an idiot, and then slowly a genius.

‘I quickly declared John to be an idiot, and then slowly declared him to be a
‘genius.’

(251) ‘I considered John to be - already brilliant, but not yet a full-blown genius.
‘I already considered John to be brilliant, but didn’t yet consider him to be a full-
‘blown genius.’

(252) ‘I considered John to be - still brilliant, but no longer a full-blown genius.

‘I still considered John to be brilliant, but no longer considered him to be a full-
‘blown genius.’

(253) ‘I consider John to be - sometimes a pain, but always a huge bore.

‘I sometimes consider John to be a pain, but always consider him to be a huge
‘bore.’152

(254) **I declared John to be - traditionally an idiot, but just a genius.

* ‘I traditionally declared John to be an idiot, but just declared him to be a gen
* ‘ius.’

(255) ‘I predict that I’m going to consider John to be - soon an idiot, but eventually a
‘genius.
‘I predict that I’m soon going to consider John to be an idiot, but eventually going
‘to consider him to be a genius.’


























































(xxxiv) People consider John to be - often an idiot, but occasionally a genius.
a) ‘People’s evaluation of John is “He’s often an idiot, but occasionally a genius.” ’
b) ‘People often consider John an idiot, but occasionally consider him a genius. They always all
agree.’
c) ‘Many people consider John an idiot, but some people consider him a genius.’
152
Cf.:

(xxxv) People consider John to be - sometimes a pain, but always a huge bore.
a) ‘People’s evaluation of John is “He’s sometimes a pain, but always a huge bore.” ’
b) ‘People sometimes consider John to be a pain, but always consider him to be a huge bore. They
always all agree.’
c) ‘Some people consider John to be a pain, but everyone considers him to be a huge bore.’

88

(256) ‘I considered John to be - briefly an idiot, but subsequently a genius.
‘I briefly considered John to be an idiot, but subsequently considered him to be a
‘genius.’

(257) ‘I find John to be - characteristically boring, but occasionally enthralling.

‘I characteristically find John to be boring, but occasionally find him to be en-
‘thralling.’153

Again, it is tempting to interpret characteristically in situ, but it can be semantically dis-

placed. As before, this reading can be brought out by pronouncing the adverb with a more
prominent H* and lengthening the pause/segment preceding it.

(258) *I declared John to be - first an idiot, and then almost a genius. I soon thought
*better of THAT, though.
‘I first declared John to be an idiot, and then almost declared him to be a genius.
‘…’

The last adverb to be tested is completely. Consider the following test sentence:

(259) ‘I expected John to be - partially a spy, but completely an eccentric.

‘I partially expected John to be a spy, but completely expected him to be an ec-
‘centric.’

In (259), semantically displacing partially is unproblematic. Completely, however, seems

to me to resist semantic displacement for some reason, though I think the escaped reading
is not clearly impossible, given enough prosodic support (prominent H*s on both adverbs
and an audible pause before the first). At any rate, assuming that partially and completely
belong to the same adverb class (cf. Cinque 1999:10), the possibility of semantically dis-
placing the partially in (259) establishes that this adverb class is not immune to semantic
displacement.
As in the previous section, well and fast/early cannot be tested.
The result of this section, then, is exactly parallel to that of the previous section. In-
terpretive escape can take place out of infinitival clauses, and this can occur with all the
Cinque adverbs except for just and almost.

III.C.3. Interpretive escape out of finite clauses

In this section, we will see that interpretive escape is also possible out of finite
clauses. Although it is possible, it appears to be more effortful (a fact which will be dis-

























































153
Cf.:

(xxxvi) People find John to be - characteristically boring, but occasionally enthralling.

a) ‘People’s evaluation of John is “He’s characteristically boring, but occasionally enthralling.” ’
b) ‘People characteristically find John to be boring, but occasionally find him to be enthralling.
They always all agree.’
c) ‘Most people find John to be boring, but some people find him to be enthralling.’

89

cussed in §III.C.4.b). Therefore, the reader, in order to get the escaped readings of the
relevant sentences, may have to deploy one or both of the following strategies:

(260) Prosodic strategies for bringing out escaped readings

a. Pronounce the adverbs with more prominent High pitch accents.
b. Delay the pronunciation of the linearly first adverb by lengthening the pause
or the segment preceding it.

At times, the reader may find the linearly first adverb in a sentence straightforward to
semantically displace and the linearly second one more difficult to.154 Nevertheless, it
seems to me that, even in these more difficult cases, semantic displacement of the second
adverb is not impossible—except in the case of the usual outliers: just and almost.
The data follow:

(261) ‘I maintain that his opinions are - ↑honestly frightful.

‘Honestly, I maintain that his opinions are frightful.’

(262) ‘She maintains that I am - fortunately funny, but UNfortunately unprincipled.

‘Fortunately, she maintains that I am funny, but UNfortunately, she maintains that
‘I am unprincipled.’

(263) ‘John maintains that Mary is - ↑allegedly a murderer.

‘John allegedly maintains that Mary is a murderer.’

(264) ‘John maintains that Mary is - ↑probably a murderer.

‘John probably maintains that Mary is a murderer.’

(265) ‘I maintained that John was - once a great musician, but last week a lousy one.
‘I once maintained that John was a great musician, but last week maintained that
‘he was a lousy one.’

(266) ‘I maintained that John was - then a great musician, but last week a lousy one.
‘I at that point maintained that John was a great musician, but last week main-
‘tained that he was a lousy one.’

(267) ‘John maintains that Mary is - ↑perhaps a murderer.

‘Perhaps John maintains that Mary is a murderer.’

(268) ‘John maintains that Mary is - possibly a pain, but necessarily a huge bore.
‘John possibly maintains that Mary is a pain, but necessarily maintains that she’s
‘a huge bore.’


























































154
This is presumably related to the fact that interpretive escape is harder to process the greater the distance
it traverses (§III.C.4.b).

90

(269) ‘John maintains that Mary is - possibly a pain, but not necessarily a huge bore.
‘John possibly maintains that Mary is a pain, but doesn’t necessarily maintain that
‘she’s‘a huge bore.’

(270) ‘I maintain that John is - usually an idiot, but occasionally a genius.

‘I usually maintain that John is an idiot, but occasionally maintain that he’s a gen-
‘ius.’155

(271) ‘I maintain that John is - often an idiot, but occasionally a genius.

‘I often maintain that John is an idiot, but occasionally maintain that he’s a gen-
‘ius.’156

(272) ‘I announced that John was – involuntarily an idiot, but intentionally a pain.
‘I involuntarily announced that John was an idiot, but intentionally announced
‘that he was a pain.’

As in the previous section, the adverbs in (272) are noticeably more resistant to interpre-
tive displacement than most of the others. However, this resistance can be overcome by
deploying the prosodic strategies in (260). The result of this is shown in the following
spectrogram. Note in particular how prominent the H* on involuntarily is. Other speakers
may find it necessary, in order to get the escaped reading of (272), to lengthen the pause
before this adverb more than it has been in (273).


























































155
Cf.:

(xxxvii) People maintain that John is - usually an idiot, but occasionally a genius.
a) ‘People claim the following: “John is usually an idiot, but occasionally a genius.” ’
b) ‘People usually maintain that John is an idiot, but occasionally maintain that he’s a genius.
b) ‘They always all agree.’
c) ‘Most people maintain that John is an idiot, but some people maintain that he’s a genius.’
156
Cf.:

(xxxviii) People maintain that John is - often an idiot, but occasionally a genius.
a) ‘People claim the following: “John is often an idiot, but occasionally a genius.” ’
b) ‘People often maintain that John is an idiot, but occasionally maintain that he’s a genius. They
always all agree.’
c) ‘Many people maintain that John is an idiot, but some people maintain that he’s a genius.’

91

(273)

(274) ‘I announced that John was - quickly an idiot, and then slowly a genius.
‘I quickly announced that John was an idiot, and then slowly announced that he
‘was a genius.’

(275) ‘I maintained that John was - already brilliant, but not yet a full-blown genius.
‘I already maintained that John was brilliant, but didn’t yet maintain that he was a
‘full-blown genius.’

(276) ‘I maintained that John was - still brilliant, but no longer a full-blown genius.
‘I still maintained that John was brilliant, but no longer maintained that he was a
‘full-blown genius.’

(277) ‘I maintain that John is - sometimes a pain, but always a huge bore.
‘I sometimes maintain that John is a pain, but always maintain that he’s a huge
‘bore.’157

(278) **I declared that John was - traditionally an idiot, but just a genius.
* ‘I traditionally declared that John was an idiot, but just declared that he was a
* ‘genius.’


























































157
Cf.:

(xxxix) People maintain that John is - sometimes a pain, but always a huge bore.
a) ‘People claim the following: “John is sometimes a pain, but always a huge bore.” ’
b) ‘People sometimes maintain that John is a pain, but always maintain that he’s a huge bore.
They always all agree.’
c) ‘Some people maintain that John is a pain, but everyone maintains that he’s a huge bore.’



92

(279) ‘I predict that I’m going to maintain that John is - soon an idiot, but eventually a
‘genius.
‘I predict that I’m soon going to maintain that John is an idiot, but eventually go-
‘ing to maintain that he’s a genius.’

(280) ‘I maintained that John was - briefly an idiot, but subsequently a genius.
‘I briefly maintained that John was an idiot, but subsequently maintained that he
‘was a genius.’

(281) ‘I maintain that John is - characteristically boring, but occasionally enthralling.

‘I characteristically maintain that John is boring, but occasionally maintain that
‘he’s enthralling.’158

Again, the in situ interpretation of characteristically is very natural, but the escaped read-
ing can be brought out by deploying the prosodic support strategies in (260).

(282) *I declared that John was - first an idiot, and then almost a genius. I soon thought
*better of THAT, though.
‘I first declared that John was an idiot, and then almost declared that he was a
‘genius. …’

(283) ‘I was convinced that John was - partially a spy, but completely an eccentric.
‘I was partially convinced that John was a spy, but completely convinced that he
‘was an eccentric.’

In (283), semantically displacing completely is quite difficult—though not clearly impos-

sible, I think, given enough prosodic support—but, at any rate, the same reasoning about
adverb classes applies here that we used in the previous section (following (259)).
As usual, well and fast/early cannot be tested.

III.C.4. Discussion

III.C.4.a. Summary of findings

Our investigation of interpretive escape has shown the following. Interpretive escape
can occur out of small clauses, infinitivals, and finite clauses. In all these cases, it can ap-
parently affect all the adverbs we used to exemplify the Cinque hierarchy (with two ex-
ceptions, which will be discussed shortly).


























































158
Cf.:

(xxxx) People maintain that John is - characteristically boring, but occasionally enthralling.
a) ‘People claim the following: “John is characteristically boring, but occasionally enthralling.” ’
b) ‘People characteristically maintain that John is boring, but occasionally maintain that he’s
enthralling. They always all agree.’
c) ‘Most people maintain that John is boring, but some people maintain that he’s enthralling.’



93

 The relevant sentences differ in how much prosodic support (see (247)) they seem to
require to be interpretable on their escaped readings. In particular, it seems that generous
prosodic support is required in order to get the escaped readings of intentionally (and in-
voluntarily); quickly; and—in some of the cases in which it appears in a second con-
junct—completely.
But two adverbs (among the ones we have examined) clearly cannot semantically es-
cape no matter how much prosodic support they are given. These adverbs are just and
almost, and their exceptional behavior will be the focus of §III.C.4.d.

III.C.4.b. Interpretive escape and processing effort

Another finding of our investigation is that, as mentioned in fn. 154, interpretive es-
cape apparently becomes more effortful the greater the distance159 it traverses. Thus, it is
more effortful to semantically displace an adverb from a finite clause than from either an
infinitival or a small clause. I am not sure that I find semantic displacement more effort-
ful from infinitivals than from small clauses (as one would expect if the former are struc-
turally larger), but Edwin Williams tells me that he does. At any rate, his judgment on
this and mine can both be captured by the same scale:

(284) Effortfulness of Interpretive Escape Scale

small clauses ≤ infinitivals ≤ finite clauses

where “A ≤ B” means that interpretive escape out of A is less effortful than or as effortful
as interpretive escape out of B. (For both Williams and me, the “≤” between “infinitivals”
and “finite clauses” can be strengthened to a “<” without loss of empirical coverage.)
(284) suggests the following:

(285) The processing analysis of the effortfulness of interpretive escape

a. The more complex a structure is (but see fn. 159), the more effortful it is to
semantically displace an adverb from it.
b. However, there is no categorical threshold of complexity such that, if a struc-
ture exceeds it, an adverb cannot be semantically displaced from it.
c. Therefore, there is no categorical grammatical constraint that rules out inter-
pretive displacements that cross a certain amount of material.
d. Rather, the gradient in effortfulness we observe ((270)) is due to processing
factors.

III.C.4.c. Interpretive escape across multiple clause boundaries

The analysis in (285)—specifically, subpart (b)—predicts that interpretive displace-

ment should be possible across multiple clause boundaries (or, at least, that it should not
be impossible). This prediction is correct, as shown by (286B), which is fine on the inter-
pretation given:


























































159
It is currently not clear whether the notion of “distance” that is relevant here is defined in linear or hier-
archical terms (or both).

94

(286) A: ‘So, who does JOHN believe that Bill believes that Mary likes?
B: ‘[John believes [that Bill believes [that Mary likes - probably FRED, but
‘perhaps KYLE]]].
‘John probably believes that Bill believes that Mary likes FRED, but perhaps
‘believes that Bill believes that Mary likes KYLE.’

III.C.4.d. Explaining the counterexamples just and almost

III.C.4.d.i. Why just and almost cannot be semantically displaced

As we have mentioned, the two adverbs that cannot be semantically displaced are
(retrospective) just and almost. For convenience, the relevant data are repeated here:

(287) **I imagined John - habitually an idiot, but just a genius. (= (230))
* ‘I habitually imagined John an idiot, but just imagined him a genius.’

(288) **I declared John to be - traditionally an idiot, but just a genius. (= (254))
* ‘I traditionally declared John to be an idiot, but just declared him to be a genius.’

(289) **I declared that John was - traditionally an idiot, but just a genius. (= (278))
* ‘I traditionally declared that John was an idiot, but just declared that he was a
* genius.’

(290) **I imagined John - first an idiot, and later on almost a genius. I soon thought
* better of THAT, though. (= (234))
* ‘I first imagined John an idiot, and later on almost imagined him a genius. …’

(291) **I declared John to be - first an idiot, and then almost a genius. I soon thought
* better of THAT, though.
* ‘I first declared John to be an idiot, and then almost declared him to be a genius.
…’ (= (258))

(292) **I declared that John was - first an idiot, and then almost a genius. I soon
thought better of THAT, though.
* ‘I first declared that John was an idiot, and then almost declared that he was a
‘* genius.’ (= (282))

There are two things to explain here. First, why are these adverbs (and only these, as
far as we know) unable to be semantically displaced? Secondly, why is it that the almost-
sentences ((290-292)) do not sound as crashingly bad as the just-sentences ((287-289))?
It appears that the inability of just and almost to be semantically displaced is a con-
sequence of a lexical idiosyncrasy that they share:

(293) Licensing condition for just and almost160

Just and almost must be left-adjacent to their modifiee on the surface.

























































160
This is in addition to the licensing conditions for these adverbs noted in (180c) and (180h).

95

Support for (293) comes from the following contrasts:

(294) a. *I just [ate a slice of pizza].

b. *I [ate a slice of pizza] just.

(295) a. *I almost [finished the pizza].

b. *I [finished the pizza] almost.161

An escaped adverb in English is necessarily not left-adjacent to its modifiee. Therefore,

an adverb cannot both semantically escape into a higher clause and obey the condition in
(293). Because just and almost are subject to this condition, they cannot semantically es-
cape into a higher clause.

III.C.4.d.ii. Why the just-sentences are worse than the almost-sentences

Our second question was why the almost-sentences ((290-292)) do not sound as aw-
ful as the just-sentences ((287-289)). Although I do not think we can give a definitive an-
swer at this point, there are two facts I believe are relevant.
The first is that (290-292) are all fully acceptable if the almost is interpreted in situ
(as noted for (290) in the discussion following (234)):

(296) *I imagined John - first an idiot, and later on almost a genius.

‘I first imagined John an idiot, and later on imagined him coming close to being a
‘genius.’

(297) *I declared John to be - first an idiot, and then almost a genius.

‘I first declared John to be an idiot, and then declared him to come close to being a
‘genius.’

(298) *I declared that John was - first an idiot, and then almost a genius.
‘I first declared that John was an idiot, and then declared that he came close to
‘being a genius.’

This raises the possibility that these acceptable readings ((296-298)) are making the vio-
lations in (290-292) seem less egregious—in other words, contaminating the relevant
judgments. Now, in order for this to have a shot at explaining why the bad almost-
sentences do not seem as bad as the bad just-sentences, it must be the case that none of
the relevant justs can be interpreted in situ. This is correct. In (287-288), just cannot be
interpreted in situ because the embedded clause is not marked for past tense or perfect
aspect (see (178) and following discussion). In (289), just cannot be interpreted in situ
because it cannot modify the stative predicate be a genius: #Gabrielle was just a genius is
out on the retrospective reading of just.


























































161
This judgment assumes default prosody, which in this case would involve assigning almost the nuclear
stress. Almost can follow its modifiee if it is right-dislocated, in which case the prosody changes and the
nuclear stress is assigned to some other constituent: She was OTHERWORLDLY, almost.

96

 The other relevant fact is that, in general, the licensing condition in (293) seems to
apply less strictly to almost than to just. Consider the following two-line dialogue, in
which four possibilities are given for the second conversational turn:

(299) A: (?)Congrats, you completely finished the pizza!

B1: (?)I ALMOST finished it, I didn’t COMPLETELY finish it.
B2: ()*I finished it ALMOST, not COMPLETELY.
B3: (?)I finished it . [PC ALMOST, not COMPLETELY].
B4: ()?[PC I finished it . ALMOST], not COMPLETELY.

In (299B2), if I finished it ALMOST is pronounced in a fully prosodically integrated man-

ner, with no pause between it and ALMOST, it is unacceptable. This is predicted by the li-
censing condition in (293), since almost follows the core VP [finished it] and is hence not
left-adjacent to its modifiee. The responses in (299B3-4), by contrast, are much better.
Here, a sentence-internal period is used as an ad hoc notation for a pause accompanied by
a glottal stop, and the abbreviation PC is not a syntactic category label but rather marks
sequences that are (impressionistically speaking) processed as a chunk.
The reader may not get exactly the same intricate pattern of judgments for (299B2-4)
as me. The point, though, is that apparently there are circumstances where almost can
disobey the licensing condition in (293) and yet not incur much of a violation at all. By
contrast, just in the same environments is terrible ((200B3-4)):

(300) A: (?)I thought you finished the pizza a while ago!

B1: (?)I JUST finished it, I didn’t finish it a WHILE ago.
B2: ()*I finished it JUST, not a WHILE ago.
B3: ()*I finished it . [PC JUST, not a WHILE ago].
B4: ()*[PC I finished it . JUST], not a WHILE ago.

It seems reasonable to take (299-300) as evidence of an inherent lexical difference be-

tween almost and just: almost can disobey the licensing condition in (293) in a way that
just simply cannot. This difference may also play a role in why (290-292) above (with
almost) seem less glaringly deviant than (287-289) (with just).

III.D. Toward an account of unusually low adverbs

Now that we have investigated interpretive escape in detail, we can attempt to come
up with an account for it. The account should ideally also cover cases like I talked to
[probably Molly], which do not involve interpretive escape as we have defined it ((178)),
but share with it the property of involving an unusually low adverb. I will not here be
able to present a full formal analysis, and will therefore have to confine myself to some
informal and tentative suggestions.
Suppose that the semantics of the focus-sensitive adverb only involves a tripartite
structure, with only being an operator that has a restrictor and a nuclear scope, à la Partee
(1991):

97

(301) a. I only talked to MOLLY.
b. onlyx | I talked to x | I talked to Molly (based on Partee 1991:10, (11), (12b))

In (301b), the restrictor is I talked to x and the nuclear scope is I talked to Molly. On this
analysis of only, it is a fact of English that only can attach either to its restrictor, as in
(301a),162 or to (the focus of) its nuclear scope: I talked to only Molly.163
If the semantics of probably can also be captured by a Partee-style tripartite struc-
ture, as in (302a), then we can assimilate its syntax to that of only. We can say that
probably can be attached either to its restrictor, as in (302b), or to (the focus of) its nu-
clear scope, as in (302c).

(302) a. probablyx | I talked to x | I talked to Molly

b. I probably talked to Molly.164
c. I talked to probably Molly.

However, we cannot extend this analysis as is to adverbs like quickly (yielding se-
mantic representations like (303a)). This is because such an analysis would predict the
following: “We can attach quickly to the restrictor, yielding [quickly [Max destroyed…]],
and then move the subject out of the VP, yielding Max quickly destroyed the evidence.
We can also instead attach quickly to (the focus of) the nuclear scope, yielding Max de-
stroyed [quickly the evidence].” The first of these predicted surface forms is indeed ac-
ceptable ((303b)), but the second is not ((303c)).

(303) a. *quicklyx | Max destroyed x | Max destroyed the evidence

b. *Max quickly destroyed the evidence.
c. *Max destroyed quickly the evidence.

In (302c), probably has been directly attached to a nonspinal constituent (Molly, the
complement of to), and the sentence can be pronounced in a prosodically integrated man-
ner. But in (303c), where quickly has been attached to a nonspinal constituent (the direct
object), a prosodically integrated pronunciation is not possible. The string in (303c) only
becomes acceptable if the directly attached adverb (quickly) and the constituent it’s at-
tached to (the evidence) are construed as a topic-focus pair, explicitly ((304a)) or implic-
itly ((305b)) contrasted with another such pair:


























































162
Strictly speaking, this predicts that we should get Only I talked to Molly. In fact, this sentence can only
be interpreted with I as the focus (and must be pronounced with the corresponding prosody: Only I talked
to Molly). It cannot be interpreted as synonymous with (301a), because (at least roughly) only cannot occur
sentence-initially unless the focus is the subject. However, (301a), on the VP-Internal Subject Hypothesis,
is [only [I talked to Molly]] underlyingly, and only becomes I talked to Molly when I moves out of the VP
and into the clausal subject position.
163
I am grateful to Edwin Williams for clarificatory discussion of this point.
164
Once again, attaching probably to the restrictor would predict [probably [I talked to Molly]], and the VP-
Internal Subject Hypothesis allows us to derive I probably talked to Molly by moving I from within the VP
to the clausal subject position.

98

(305) a. [Max destroyed - quickly the evidence, but slowly his neighbor’s computer.
b. [Max destroyed - quickly the evidence.
[Implied: ‘But slowly something else.’]

This contrastive topic structure has the prosodic profile shown in the following spectro-
gram of (305b). The quickly in this sentence is pronounced with a marked H* and pre-
ceded by a brief pause.

(306)

Structures like (305) and interpretive escape sentences could perhaps be derived by a
syntactic movement rule. Such a rule would lower an adverb from its base-generated po-
sition as sister to its modifiee and make it the sister of the focus. (Of course, this would
have to be implemented differently in frameworks that do not permit downward move-
ment.) Such a rule would have to be constrained, since there are some elements that are
discourse-old (i.e., part of the presupposition or cofocus) that an adverb nevertheless can-
not lower past on its way to becoming the sister of the focus. One such element is the, as
shown by the ill-formedness of (307b):

(307) Everyone agreed that Keith and Maya’s relationship involved a genius and an id-
iot, but people disagreed over who was the genius and who was the idiot.
a. *Dylan considered Keith - probably [the genius], but perhaps [the idiot].
b. *Dylan considered Keith the - probably [genius], but perhaps [idiot].

It is unlikely that the inability of probably and perhaps to lower past the is exclusively
due to the’s status as (normally) a proclitic. This is because these adverbs are no more
able to lower past the demonstrative that, which is lexically stressed:

99

(308) Every class has one person who I’d call “that idiot” and one person who I’d call
“that genius.”
a. *In my chemistry class, Eva considers Austin - probably [that genius], but
*perhaps [that idiot].
b. *In my chemistry class, Eva considers Austin that - probably [genius], but
*perhaps [idiot].

It is surely relevant that the focus-sensitive adverb only cannot be generated under the or
that either. The following sentences are deviant on the relevant interpretations:

(309) a. *Sydney talked to the - only genius, not idiot.

‘Sydney only talked to the genius, not to the idiot.’
b. *Sydney talked to that - only genius, not idiot.165
‘Sydney only talked to “that genius,” not to “that idiot.” ’

If (some) escaped adverbs indeed do get where they are by syntactic lowering, then
the two-fragment analysis of interpretive escape (§III.B.2) cannot be correct, because
movement cannot cross sentence boundaries:

(310) a. *Julian bought that screwdriver on Sunday.

b. *Wheni did Julian buy that screwdriver __i?

(311) a. *Julian bought that screwdriver. He did it on Sunday.

b. *Wheni did Julian buy that screwdriver. He did it __i?

This would mean that John considers Mary probably a murderer is a single sentence on
both its escaped and nonescaped readings. We would naturally want to extend the same
analysis to cross-speaker cases. Hence we would analyze (312) as a single sentence dis-
tributed over two conversational turns, just like (313).

(312) A: John considers Mary… (= (206))

B: Probably a murderer.

(313) A: Yesterday, Caitlin bought…

B: A hammer and a saw.

As we have seen, a syntactic lowering rule that derives interpretive escape sentences,
and perhaps also direct attachment sentences such as (305), would have to be formulated

























































165
Admittedly, we have not excluded the following possibility:

(xxxxi) a. “That idiot” and “that genius” in the sense of (308) are names.
b. Names are words (X°s).
c. The Lexicalist Hypothesis is correct: words are syntactic atoms, and hence syntactic opera-
tions cannot add material inside them or remove material from inside them.
d. Therefore, a syntactic movement rule cannot insert an adverb inside “that idiot” or “that
genius”.

100

in a way that does not run afoul of facts like those in (307-308). The task of doing this
explicitly is, unfortunately, one I must leave for future work.

IV. Conclusion

This thesis and its main findings may be summarized as follows.

IV.A. The Cinque hierarchy faces substantial difficulties in accounting for English ad-
verb orders

The Cinque hierarchy (Cinque 1999) is a prominent hypothesis concerning adverb

syntax and clause structure. On this hypothesis, each of the various semantically defined
classes of adverbs has its own position in the structure of the clause. These positions are
the specifiers of functional projections. When an AdvP occurs in one of these specifiers,
it checks semantic features against the corresponding head. Therefore, there is a perfect
matchup between the hierarchy of adverbs and the hierarchy of their corresponding se-
mantically contentful functional heads, which are realized in the world’s languages as
verbal suffixes, auxiliaries, and “particles.” These functional heads—and therefore, their
specifiers—are rigidly ordered universally. On this view, the structure of the clause is
much richer than it is often taken to be.
The Cinque hierarchy makes strong predictions concerning which adverbs will pre-
cede which other adverbs. A number of these predictions—mostly concerning adverbs
that are adjacent in the hierarchy—were tested in English. The result of this investigation
was that the Cinque hierarchy faces considerable empirical difficulties in accounting for
the facts of English adverb ordering.
It was suggested that perhaps the Cinque hierarchy is correct, and the counterexam-
ples arise as follows. When the Cinque hierarchy predicts a rigid adverb ordering A > B,
but the ordering B > A is in fact acceptable, this is because A has directly attached to a
clausal functional projection lower than B. A can do this because, in addition to having a
fixed position in the Cinque hierarchy, it is also “promiscuous”: it can directly attach to
constituents of many different types, as revealed by its ability to attach to constituents
such as nominals and PPs that do not lie on the clausal spine. This hypothesis predicts
that there should be a perfect correspondence between the adverbs that can directly attach
to nonspinals and the adverbs that can appear low enough to disobey Cinquean adverb
orders. This prediction is not correct, since virtually every Cinque adverb can directly
attach to nonspinals, but not every Cinque adverb can appear low enough to disobey
Cinque orders.
In fact, we argued that, fundamentally, all Cinque adverbs can directly attach to non-
spinals (the DIRECT ATTACHMENT GENERALIZATION). There are a few adverbs that appear
to only be able to attach to VP or a constituent on its projection line (once, quickly, retro-
spective just, soon, well, fast/early, and prospective almost). But we argued that this re-
striction is in place not because the Direct Attachment Generalization is false, but be-
cause these adverbs must be licensed by elements that can only occur in the extended
projection of V. This argument was bolstered by data indicating that these adverbs are
barred even in finite clauses—where they could in theory occur in their Cinque positions
rather than directly attached to anything—unless the relevant licensing elements are pre-

101

sent. (It currently appears, however, that that last generalization has a few counterexam-
ples.)
Direct attachment appears to be widespread, whereas the Cinque hierarchy in its cur-
rent form is very far from making only correct predictions concerning adverb orders in
English. This state of affairs suggests that perhaps we should try to reduce all adverbial
modification to direct attachment, and filter out illicit orders on semantic grounds. This
proposal is similar in spirit to, or perhaps ultimately identical to, the Adjunction Hy-
pothesis. But Cinque is quite right to point out that it is one thing to propose this and
quite another to actually work it out explicitly, showing how exactly the semantics rules
the illicit orders out. Regardless of the ultimate outcome of the “specifier vs. adjunct” de-
bate, I hope to have shed light on some of the Cinque hierarchy’s current empirical defi-
ciencies, and to have brought to the table some facts (about English adverb ordering) that
any hypothesis concerning adverb syntax should be able to predict.

IV.B. Findings regarding interpretive escape

We then turned our attention to a phenomenon we termed INTERPRETIVE ESCAPE,

which involves an adverb appearing to be in an embedded clause, but being interpreted in
a higher clause. After some preliminary discussion of the prosody, semantics, and syntax
of interpretive escape, we evaluated a couple of possible analyses of the phenomenon.
The hypothesis that escaped adverbs are parentheticals was shown to make incorrect pre-
dictions. This is because parentheticals apparently must be interpreted as comments by
the speaker and cannot be interpreted as contributing to the denotation of an embedded
clause, whereas this is not true of escaped adverbs. The two-fragment analysis, by con-
trast, did not suffer from any obvious empirical problems, but did not seem to shed any
more light on interpretive escape than a one-sentence analysis, making it impossible at
that point to choose between them.
We then embarked on a detailed investigation of interpretive escape to determine
how widely available it is. The results were as follows.
Interpretive escape can occur out of small clauses, infinitivals, and finite clauses, and
can even occur across multiple clause boundaries. It appears that interpretive escape be-
comes more effortful the greater the distance it traverses. (Whether this distance is de-
fined in linear or structural terms or both is not yet clear.) However, there is no categori-
cal threshold of complexity Tc such that interpretive escape is impossible from structures
that exceed Tc. This suggests that the effortfulness of long-distance semantic displace-
ments is due to processing factors rather than a categorical grammatical constraint.
Although a few of the Cinque adverbs require more prosodic support than the rest in
order to be interpretable as escaped, the only ones that clearly cannot semantically escape
to a higher clause are almost and retrospective just. This was attributed to a lexical prop-
erty of these adverbs, namely that they must be left-adjacent to their modifiee on the sur-
face. Attempting to interpret retrospective just as displaced results in markedly greater
deviance than doing the same thing with almost. It was suggested that this could be ex-
plained in either of the following ways: 1) almost can perfectly well be interpreted in situ
in the relevant sentences, and perhaps this contaminates the escaped reading and makes it
seem not so bad; 2) the aforementioned surface left-adjacency constraint seems to apply
less strictly to almost than it does to just.

102

 Finally, it was suggested that perhaps interpretive escape could be explained in the
following ways. Some adverbs can be given a Partee (1991)–style treatment as operators
that semantically occur in a tripartite structure and that syntactically can attach to either
their restrictor or (the focus of) their nuclear scope. Other adverbs can only appear low
and be interpreted high when they are one of the topics in an implicit or explicit contras-
tive topic structure; these structures have a distinctive prosodic profile. Interpretive es-
cape sentences involving these adverbs can be derived by a syntactic movement rule that
lowers an adverb from its base position as sister to its modifiee and makes it the sister of
the focus. Such a rule would have to be constrained in order to avoid lowering adverbs
past certain elements in the cofocus, such as the and the demonstrative that. An explicit
formulation of these proposals must await further work.
It is my hope that the investigations here presented will contribute to our understand-
ing of adverbs, clause structure, and modification in English and—even if indirectly—in
the languages of the world at large.

103

 References

Aboh, Enoch. 1993. Théorie syntaxique et syntaxe du Gun. Mémoire de Maîtrise, Univer-
sity of Geneva. Cited in Cinque (1999).

Aboh, Enoch. 1996. “A propos de la syntaxe du Gungbe.” Rivista di Grammatica Gen-

erativa 21: 3-56. Cited in Cinque (1999).

Alexiadou, Artemis. 1997. Adverb Placement: A Case Study in Antisymmetric Syntax.

Amsterdam: J. Benjamins.

Baker, Mark. 1985. “The Mirror Principle and Morphosyntactic Explanation.” Linguistic
Inquiry 16: 373-415.

Bobaljik, Jonathan David. 1999. “Adverbs: The Hierarchy Paradox.” Glot International
4.9/10: 27-28.

Bolinger, Dwight L. 1968. “Entailment and the Meaning of Structures.” Glossa 2.2: 119-
127.

Bowers, John. 1993. “The Syntax of Predication.” Linguistic Inquiry 24: 591-656. Cited
in Cinque (1999).

Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT. Cited in
Delfitto (2006).

Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspec-

tive. New York: Oxford University Press.

Cinque, Guglielmo. 2004. “Issues in Adverbial Syntax.” Lingua 114.6: 683-710.

Cinque, Guglielmo. 2005. “Deriving Greenberg’s Universal 20 and Its Exceptions.” Lin-
guistic Inquiry 36.3: 315-332.

Cinque, Guglielmo. 2010. “Word Order Typology. A Change of Perspective.” MS. Lear |
Linguistic Electronic Archive. Università Ca’ Foscari. Web. <lear.unive.it/bitstream/
10278/1517/1/TADWO_Cinque.13.08.10.pdf>.

Comrie, Bernard. 1976. Aspect: An Introduction to the Study of Verbal Aspect and Re-
lated Problems. Cambridge: Cambridge University Press. Cited in Cinque (1999).

Dahl, Östen. 1985. Tense and Aspect Systems. Oxford: Blackwell. Cited in Cinque
(1999).

104

Dékány, Éva. 2011. “A Profile of the Hungarian DP: The Interaction of Lexicalization,
Agreement and Linearization with the Functional Sequence.” Diss. University of
Tromsø. LingBuzz. <http://ling.auf.net/lingBuzz/001418>.

Delfitto, Denis. 2006. “Adverb Classes and Adverb Placement.” The Blackwell Compan-
ion to Syntax. Ed. Martin Everaert and Henk van Riemsdijk. Malden, MA, USA:
Blackwell Pub. 83-120. Blackwell Handbooks in Linguistics.

Ernst, Thomas. 2002. The Syntax of Adjuncts. Cambridge, UK: Cambridge University
Press.

Fox, Danny, and David Pesetsky. 2005. “Cyclic Linearization of Syntactic Structure.”
Theoretical Linguistics 31: 1-45.

Grice, H. P. 1975. “Logic and Conversation.” Syntax and Semantics: Speech Acts. Ed.
Peter Cole and Jerry L. Morgan. Vol. 3. New York: Academic Press. 41-58.

Grohmann, Kleanthes K. “Adverbs in Modern Syntactic Theories.” 11 Nov. 2005. Class

notes for ENG 575: Comparative Syntax. University of Cyprus, Nicosia, Cyprus.
<http://www.punksinscience.org/kleanthes/courses/UCY05F/CS/Adverbs_in_Moder
n_Syntactic_Theories.pdf>.

Haider, Hubert. 2004. “Pre- and postverbal adverbials in OV and VO.” Lingua 114.6:
779-807.

Hendrick, Randall. 1991. “The Morphosyntax of Aspect.” Lingua 85: 171-210. Cited in
Cinque (1999).

Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT.

Ladd, D. Robert. 2008. Intonational Phonology. 2nd ed. Cambridge: Cambridge Univer-
sity Press. Cambridge Studies in Linguistics.

Laenzlinger, Christopher. 1998. Comparative Studies in Word Order Variation: Adverbs,

Pronouns, and Clause Structure in Romance and Germanic. Amsterdam: J. Benja-
mins.

Lewis, David. 1975. “Adverbs of Quantification.” Formal Semantics of Natural Lan-

guage. Ed. Edward Keenan. Cambridge: Cambridge University Press. 3-15.

Mahootian, Shahrzad, and Beatrice Santorini. 1996. “Code Switching and the Comple-
ment/Adjunct Distinction.” Linguistic Inquiry 27: 464-79. Cited in Cinque (1999).

Manninen, Satu. 2005. Rev. of Adverbs and Functional Heads: A Cross-Linguistic Per-
spective. Journal of Linguistics 41.2: 452-57. JSTOR. Web. 16 Apr. 2012.
<http://www.jstor.org/stable/4176946>.

105

Nida, Eugene. 1974. Morphology. Ann Arbor, MI: University of Michigan Press. Cited in
Cinque (1999).166

Nilsen, Øystein. 2003. “Eliminating Positions: Syntax and semantics of sentence modifi-
cation.” Diss. Universiteit Utrecht, 2003. LingBuzz. <http://ling.auf.net/lingBuzz/
000014>.

Partee, Barbara H. 1991. “Topic, Focus and Quantification.” Proceedings of SALT I. Ed.
Steve Moore and Adam Wyner.

Pollock, Jean-Yves. 1989. “Verb Movement, Universal Grammar and the Structure of
IP.” Linguistic Inquiry 20: 365-424.

Postal, Paul M. 1998. Three Investigations of Extraction. Cambridge, MA: MIT.

Pullum, Geoffrey K. 2010. “The Truth about English Grammar: Rarely Pure and Never
Simple.” A New Look at Language Teaching & Testing: English as Subject & Vehi-
cle. Ed. Tien-en Kao and Yao-fu Lin. Taipei, Taiwan: Language Training and Test-
ing Center. 16-39. Available at <http://ling.ed.ac.uk/~gpullum/LTTCpaper.pdf>.

Rizzi, Luigi. 1997. “The Fine Structure of the Left Periphery.” Elements of Grammar: A
Handbook of Generative Syntax. Ed. Liliane Haegeman. Dordrecht: Kluwer. 281-
337.

Smith, Carlota S. 1991. The Parameter of Aspect. Dordrecht: Kluwer. Cited in Cinque
(1999).

Sohn, Ho-min. 1994. Korean. London: Routledge. Cited in Cinque (1999).

Stowell, Tim. 2007. “The Syntactic Expression of Tense.” Lingua 117.2: 437-63.

Tenny, Carol. 2000. “Core Events and Adverbial Modification.” Events as Grammatical
Objects. Ed. Carol Tenny and James Pustejovsky. Stanford: CSLI Publications. 285-
334.

Thomason, Richmond, and Robert Stalnaker. 1973. “A Semantic Analysis of Adverbs.”

Linguistic Inquiry 4: 195-220. Cited in Cinque (1999).

Vendler, Zeno. 1984. “Adverbs of Action.” Papers from the Parasession on Lexical Se-
mantics. Chicago: Chicago Linguistic Society. 297-307. Cited in Cinque (1999).

Wenger, Neven. Cinque’s Functional Hierarchy. 09 Feb. 2009. Syntax Reading Circle
(SRC) notes. Universität Heidelberg, Heidelberg, Germany.

























































166
I have here reproduced the bibliographic information given in Cinque (1999:248), but as far as I can tell,
the full title of the work is Morphology: The Descriptive Analysis of Words (2nd ed.), and it was published
in 1976, not 1974.

106

Williams, Edwin. 1982. “Another Argument that Passive Is Transformational.” Linguistic
Inquiry 13.1: 160-163.

Williams, Edwin. LIN 302: Syntax. Princeton University, Princeton, NJ. Fall 2009. Class
lectures.

Williams, Edwin. LIN 412: Advanced Syntax. Princeton University, Princeton, NJ. Fall
2010. Class lectures.

Wilson, Stephen, and Ayşe Pınar Saygın. 2001. “Adverbs and Functional Heads in Turk-
ish: Linear Order and Scope.” Proceedings of WECOL 2001: Western Conference on
Linguistics. Seattle, WA, USA. Available at <crl.ucsd.edu/~saygin/papers/WECOL
paper.pdf>.

Zagona, Karen. 1988. Verb Phrase Syntax: A Parametric Study of English and Spanish.
Dordrecht: Kluwer. Cited in Cinque (1999).

Zyman, Erik. 2011. “Remarks on the Relationship Between Information Structure and
Syntax in P’urhépecha.” MS (junior paper). Princeton University, Princeton, NJ.

107