SOURCES OF LINGUISTIC KNOWLEDGE IN THE SECOND LANGUAGE ACQUISITION OF

ENGLISH ARTICLES

Tania Ionin, Maria Luisa Zubizarreta, Salvador Bautista Maldonado

1. Introduction
Learners of a second language (L2) can in principle draw on at least three sources of
linguistic knowledge as they acquire the target language: (1) the L2-input (which may be
naturalistic and/or classroom-based); (2) the structures of their native language (L1); and
(3) innate linguistic knowledge not obviously traceable to either L1-transfer or L2-input.
No one disputes the existence of the first source (one can hardly acquire a language
without exposure to it). Most L2-researchers today acknowledge the relevance of the
second source, since L1-transfer has been shown to play a role across a variety of
linguistic domains (see, e.g., Dechert & Raupach 1989, Odlin 1989, Gass & Selinker
1992, Schwartz & Sprouse 1994, 1996, Schwartz 1998, among many others). The
existence of the third source, innately guided knowledge, is the one most subject to
debate. On the domain-specific view of L2-acquisition, Universal Grammar (UG) is
accessible by L2-learners (directly and/or through the L1); on the domain-general view,
L2-acquisition proceeds through mechanisms such as statistical learning, without
reference to language-specific knowledge.
In the present paper, we address the relevance of the above sources – L2-input, L1-
transfer, and UG-based knowledge – in the domain of article choice. This domain is
particularly well-suited for such an examination because it involves form/meaning
mappings: L2-English learners, for instance, have to map particular meanings to the and
a. We ask first, whether form-meaning mappings are transferred from the L1; and second,
whether form-meaning mappings can be deduced from L2-input alone or require innate
linguistic knowledge. Our conclusion will be that all three sources – L2-input, L1-
transfer, and UG-based knowledge – play a role in the acquisition of English articles. We
will argue that the meaning of English articles cannot be deduced from the input alone,
without help from L1-transfer and/or UG-provided knowledge.
This paper is organized as follows. Section 2 gives a brief overview of article
semantics, and summarizes previous findings that were relevant to the present study. In
Sections 3 and 4 we present the methodology and results, respectively, of a preliminary
study with L1-Russian and L1-Spanish learners of English. Section 5 considers the larger
questions of UG-based vs. input-driven learning, and Section 6 concludes the paper.

2. Background
The acquisition of English articles is a notoriously difficult process for second
language (L2) learners. Many studies (e.g., Huebner 1983; Master 1987; Parrish 1987;
Thomas 1989; Murphy 1997; Robertson 2000; and Leung 2001, among many others)

Acknowledgements: we would like to thank our undergraduate research assistants, Erin Bardales and
Anna Bokarius, for their help with the data collection and analysis. We are grateful to William Rutherford
for allowing us the use of his cloze test for measuring L2-learners’ proficiency. The research reported here
is supported by NSF grant # BCS-0444088 (PI: Maria Luisa Zubizarreta) and by a University of Southern
California undergraduate research grant. We are grateful to two anonymous reviewers for their comments.

1
have found that L2-English learners make errors omitting and/or misusing English
articles. Such errors appear to be particularly prevalent among L2-learners whose native
languages (L1s) lack articles. Studies which have compared article choice among L2-
learners with different L1s (e.g., Larsen-Freeman 1975a, 1975b; Thomas 1989; Parodi et
al. 1997; Murphy 1997) generally agree that speakers of article-less L1s (e.g., Korean or
Japanese) omit English articles in obligatory contexts to a greater extent than speakers
whose L1s do have articles (e.g., Spanish). However, little is known concerning the effect
of L1-transfer on the patterns of article (mis)use (but see Hawkins et al. 2006 for some
recent findings). In this paper, we ask two questions concerning L2-English article use:
(1) does L1-transfer affect article semantics? and (2) in the absence of L1-transfer, do L2-
learners have access to universal semantic distinctions – and if so, how?

2.1. Article semantics: definiteness and specificity

Before we can discuss the acquisition of English articles, we need to understand the
form/meaning mapping behind English articles in more detail. The distinction between
the and a is one of definiteness. We adopt here a standard view of definites as
presuppositional and indefinites as quantificational expressions, as shown in (1) (for more
discussion of different views of definiteness, see Heim 1991).
(1) a. Definiteness (Fregean analysis)
[the ζ] ξ expresses that proposition which is
- true at index i, if there is exactly one ζ at i, and it is ξ at i,
- false at an index i, if there is exactly one ζ at i, and it is not ξ at i,
- truth-valueless at an index i, if there isn’t exactly one ζ at i. (Heim 1991:9)
b. Indefinites (quantificational analysis)
A sentence of the form [a ζ] ξ expresses that proposition which is true if there is at
least one individual which is both ζ and ξ, and false otherwise. (Heim 1991:26)
As an illustration of (1), consider the sentences in (2) below. In (2a), there is no prior
presupposition that a cup exists (since no cups have been previously mentioned).
Therefore, use of the is infelicitous; on the other hand, use of a is perfectly fine, since a
carries no presupposition – (2a) simply asserts that a cup exists that Sally broke. In (2b),
on the other hand, the presupposition of a unique cup has been established by prior
discourse, so use of the is now felicitous: (2b) presupposes that a cup exists, and asserts
that Sally glued it back together.
(2) a. Sally broke a cup…
b. …Then she glued the cup back together.
Another illustration is given in (3). Here, no prior discourse is necessary in order to
make use of the felicitous: the semantics of superlatives ensures that the existence of the
biggest cup is presupposed (since only one cup can be the biggest cup that Sally owns).
(3) Sally broke the/#a biggest cup that she owns.
The task facing L2-English learners, then, is to realize that the distinction between the
and a is that of presupposition: use of the is felicitous when uniqueness of the referent has
been established (by prior discourse, world knowledge, etc.), while use of a is felicitous
when this is not the case (for more details on the relationship between the and a, see

2
Heim 1991). This distinction is in a large part discourse-based: appropriate use of the vs.
a requires an evaluation of the discourse situation in order to determine whether the
uniqueness presupposition on the has been satisfied. Crucially, this presupposition must
be shared by speaker and listener – the speaker cannot use the in (2a) simply because she
is aware of a unique cup; it is necessary that the hearer share this knowledge.
Definiteness is not the only distinction that articles in a language may in principle
encode. Ionin (2003, 2006) argues that a different discourse-based distinction also affects
article use in language. This distinction is specificity, with the semantics in (4).
(4) Specificity
A sentence of the form [sp α] ζ expresses a proposition only in those utterance
contexts c where the following felicity condition is fulfilled: the speaker of c
intends to refer to exactly one individual xc in c, and there exists a property ϕ which
the speaker considers noteworthy in c, and xc is both α and ϕ in c. When this
condition is fulfilled, [sp α] ζ expresses that proposition which is true at an index i
if xc is ζ at i and false otherwise. (Ionin 2003:56)
An illustration of specificity comes from indefinite use of this in colloquial / spoken
English (see also Prince 1981). Consider (5). Here, use of this is fine upon first mention
of ‘cup’, showing that this on such a use is an indefinite determiner, like a. Unlike a,
however, this carries the felicity condition in (4): it is necessary that the speaker intend to
refer to a particular cup, which has some property (such as being Sally’s favorite) that
makes it noteworthy from the speaker’s point of view. For more discussion of the
semantics of specificity, see Ionin (2003, 2006).
(5) Sally broke this beautiful cup… it was her favorite!
While English marks specificity only in the colloquial / spoken register, other
languages do so across registers. Ionin (2006) argues that specificity is marked by articles
in Samoan (see Mosel and Hovdhaugen 1992) as well as post-nominal particles in Sissala
(see Blass 1990). Importantly, the specificity distinction cross-cuts the definiteness
distinction: it is possible for both definites and indefinites to be either specific or non-
specific. In Samoan, for instance, one article (le) marks specific definites and specific
indefinites, while another one (se) marks non-specific definites and indefinites, as
illustrated in (6), from Mosel and Hovdhaugen 1992 (pp. 259-261). In (6a) and in (6c),
the speaker wishes to attract attention to a specific couple/child/woman/etc., so le is used,
regardless of whether the DP is indefinite or definite. On the other hand, the identity of
the lady in (6b) and the family in (6d) is irrelevant to the discourse or unknown, so se is
used. See Ionin (2003, 2006) for more discussion.
(6) a. use of ‘le’ with a specific indefinite
‘O le ulugāli’i, fānau l=a lā tama ‘o
PRES ART couplegive birth ART=Poss 3.du. child PRES
le teine ‘o Sina
ART girl PRES Sina
“There was a couple who had a child, a girl called Sina.”

3
 b. use of ‘se’ with a non-specific indefinite
Sa fesili mai se tamaitai po=o ai
PAST ask DIR ART(nsp.sg.) lady Q-PRES who
l=o ma tama.
ART=Poss 1.exc.du. father
‘A lady asked us who our father was.’
c. use of ‘le’ with a specific definite
Māsani ‘o le tamāloa e usua’i=ina lava ia…..
used PRES ART man GENR get up early=ES EMPH 3sg
’ae nonofo ‘o le fafine ma l=a=na
but stay(pl.) PRES ART woman and ART=POSS=3.sg
tama i le fale
child LD ART house
“It was the man’s practice to get up early and… while the woman stayed at
home with her child.”
d. use of ‘se’ with a non-specific definite
Alu i se tou aiga e moe. Pe se
go LD ART(nsp.sg.) 2.pl. family GENR sleep. Q ART(nsp.sg.)
tama a ai!
boy POSS who
“Go to your family – whoever that may be – and sleep! [I wonder] whose boy
you might be!” [said to a boy who is selling necklaces at night in front of a
hotel]
In English, specificity with indefinites is marked via indefinite this, as in (5). While
such use of this is not possible with definites (see Ionin 2003, 2006, for an account of
why this is the case), English definites may in principle be specific or non-specific. As an
illustration, consider (7). In (7a), the speaker intends to refer to a particular individual
(her neighbor) who also happens to be the owner of this store; this context satisfies the
conditions on specificity. In (7b), on the other hand, the identity of the owner is
completely irrelevant, and the referent is less likely to be construed as specific. This is
reminiscent of Donnellan’s (1966) referential/attributive distinction for definites.1
(7) a. I want to talk to the owner of this store – she is my neighbor, and I have an
urgent message for her.
b. I want to talk to the owner of this store, whoever that is. I am going to
complain about the quality of the produce!
To sum up, specificity, like definiteness, appears to be a cross-linguistic semantic
universal underlying article choice. A given context in any language may or may not
satisfy the conditions on specificity, just as it may or may not satisfy the conditions on
definiteness. Where languages differ is in which of these distinctions they mark

1
Unlike Donnellan, however, we do not propose that the is ambiguous between specific and non-
specific readings. While the always has the presuppositional analysis in (1), the context containing the-NPs
may or may not satisfy the conditions on specificity. See Ionin (2003, 2006) for more detailed discussion.

4
morphologically. While standard/written English marks only the definiteness distinction,
other languages, notably Samoan, marks the specificity distinction instead. This is
illustrated pictorially in (8).
(8) Article grouping by definiteness Article grouping by specificity
(English) (Samoan)
+definite -definite +definite -definite

+specific +specific le
the a
-specific -specific se

2.2. Previous findings: speakers of article-less L1s

Previous studies by Ionin, Ko and Wexler (2003, 2004) examined the acquisition of
English articles by adult speakers of Russian and Korean. Since these two languages lack
articles, these learners faced the task of acquiring English articles without any help from
their L1.2 It was found that while these learners made many errors of article (mis)use in
English, their errors were not random. Rather, errors came in two types: overuse of the
with specific indefinites and overuse of a with non-specific definites. Article use on
specific definites and non-specific indefinites was accurate. This is summarized in (9).
(9) Patterns of article use from Ionin et al. (2003, 2004)
[+definite]: target the [-definite]: target a
[+specific] correct use of the overuse of the
[-specific] overuse of a correct use of a

Crucially, article misuse on the two problematic categories was never 100%: on the
categories of specific indefinites and non-specific definites, L2-learners from both L1s
went back and forth between use of the and a; this fluctuation occurred at the individual
as well as the group level. Article misuse (i.e., overuse of the with specific indefinites and
overuse of a with non-specific definites) was rarely above 50%.
The explanation advanced by Ionin et al. was that in the absence of L1-transfer, L2-
learners access both semantic universals – definiteness and specificity – provided by UG,
but do not (at least initially) know which of these semantic universals is relevant for the
choice of the vs. a in English. Hence, learners fluctuate between the two options: some of
the time, they treat the as marking definiteness (and a indefiniteness), and some of the
time they treat the as marking specificity (and a non-specificity). As a result, they
perform accurately on the categories of specific definites and non-specific indefinites,
where the two options give the same results. When definiteness and specificity are in
conflict, on the other hand – on the categories of specific indefinites and non-specific
definites – learners use the and a interchangeably, as shown in (9). Those learners who
are able to recover from the fluctuation converge on the target-like definiteness pattern,
since this is the pattern indicated by the English input. Learners are not expected to ever

2
See Ionin (2003), Ionin et al. (2004) for evidence that indirect L1-transfer from a domain other than
articles in these languages is also highly unlikely.

5
converge on the specificity pattern, since it is not favored by the input; specificity is an
option learners employ, alongside the definiteness option, until they come to realize that
English articles encode definiteness rather than specificity.

2.3. Hypothesis and predictions

The studies of Ionin et al. looked only at L2-learners whose native languages lack
articles, and left open the question of what happens with L2-learners whose L1s have
articles, such as Spanish. In principle, there are two possibilities: (1) fluctuation overrides
transfer; or (2) transfer overrides fluctuation. According to the first possibility, all L2-
learners, regardless of their L1, should fluctuate between definiteness and specificity.
According to the second possibility, learners whose L1 has articles should transfer article
semantics from their L1 to their L2; since the Spanish article system is like the English
one where definiteness and specificity are concerned, this should result in accurate article
use. The two competing hypotheses, with their corresponding predictions, are
summarized in (10) and (11), respectively.
(10) Possibility 1: Fluctuation overrides transfer
All L2-learners should fluctuate between definiteness and specificity in their L2-
article choice.
a. Both L1-Spanish and L1-Russian L2-English learners should exhibit the
pattern in (9), showing interchangeable use of the and a on non-specific definites
and specific indefinites.
(11) Possibility 2: Transfer overrides fluctuation
L2-learners whose L1 has articles transfer article semantics from their L1 to their
L2. L2-learners whose L1 lacks articles exhibit fluctuation.
a. L1-Russian L2-English learners should exhibit the pattern in (9).
b. L1-Spanish L2-English learners should exhibit accurate use of the in all
definite categories and accurate use of a in all indefinite categories, with no effects
of specificity.
The main goal of the present paper is to test the hypotheses in (10) and (11) in order
to determine what role L1-transfer plays in the acquisition of article semantics. On a
more conceptual level, we will also address the question of where fluctuation comes
from; we will argue that it is a result of UG-access coupled with difficulties processing
the input. Our ultimate goal, therefore, is to examine the corresponding roles that L1-
transfer, UG-access, and input processing play in L2-acquisition of article semantics.

3. Methodology
The study reported here is part of a larger-scale study comparing the acquisition of
English articles by child as well as adult speakers of Russian and Spanish, all of whom
are acquiring English as a second language. We report here preliminary data from adult
L1-Russian and L1-Spanish participants; more adult data, as well as child data, are in the
process of being collected.

6
3.1. Participants
The participants in this study were 23 adult speakers of Russian and 24 adult speakers
of Spanish, as well as a control group of six adult native English speakers. The L1-
Russian group and the control L1-English group were tested in southern California; the
L1-Spanish group was tested in Mexico.
The Russian speakers ranged in age from 22 to 72 (mean age 43, median age 40).
They were living in the U.S. at the time of the testing, and had arrived in the U.S. as
adults. Age of arrival in the U.S. ranged from 19 to 60 (mean age of arrival 35, median
age of arrival 34). While all but three of the subjects had started studying English before
arrival in the U.S., many as children, intensive exposure to English did not begin until
arrival in the U.S., past puberty. Several of the subjects were bilingual in Russian and
another language (Yiddish for one, Byelorussian for one, and Armenian for six), but all
spoke Russian as their dominant and primary language. The L1-Russian participants lived
in southern California at the time of the study, and were recruited through advertising.
Most of the participants reported using English at work and/or in daily life; only three
reported that they were taking courses in English as a second language at the time of the
study.
The Spanish speakers were all speakers of Mexican Spanish. They ranged in age from
19 to 28 (mean age 22, median age 21). All of them were studying English in a classroom
setting at a Mexican university. All of them had been studying English in school since
childhood (age 13 or younger), and only one of the participants had ever lived in an
English-speaking country (for one year).
The native English speakers ranged in age from 19 to 25 (mean age 21). They were
all students at a university in southern California. Two were bilingual in English and
another language (Spanish for one, Gujapatni for the other), but spoke English as their
dominant and primary language.

3.2. Tests
All of the participants took two written tests: an elicitation test of English article use
and a cloze test of L2-proficiency. The L2-learners also took part in an oral production
task, not reported here.

3.2.1. Cloze test

The cloze test, created by William Rutherford, consisted of four passages. Each
passage began with a paragraph of uninterrupted text, followed by another page of text in
which every tenth word had been replaced by a blank. There were 30 blanks in each
passage, for a total of 120 blanks in the test. Participants were given 25 minutes to
complete as much of the test as they could, filling in each blank with an appropriate
word. While the native English speakers finished the entire test in 25 minutes, most of the
L2-learners finished two or three of the four passages. The test was corrected using an
acceptable-word criterion rather than an exact-word criterion, since the former has been
reported as yielding a more accurate assessment of subjects’ proficiency (Butler 1980).

3.2.2. Elicitation test

The elicitation test consisted of 60 short dialogues each designed to elicit a particular
target word. This task was modeled after elicitation tests used by Ionin, Ko and Wexler

7
(2003, 2004), but with two important changes. First, while the previous studies’ tests
gave the participants a choice of target words (e.g., the and a), the present task contained
a blank in the target sentence; the participants were instructed to fill in the blank with any
word they considered appropriate for the context (including a dash if no word was
necessary). Second, unlike the previous tests, the present test included fillers targeting
items other than articles – in particular, prepositions and pronouns (some filler items had
‘dash’ as the target response). There were 36 target items and 24 filler items.
The 36 target items aimed at eliciting articles with singular NPs: there were three
categories of six items each aimed at eliciting the, and three categories of six items each
aimed at eliciting a. We report four of the six target categories here – the categories most
relevant to the definiteness / specificity distinction (the other two categories examine the
effects of previous mention on use of definites and indefinites).3 The four categories are
exemplified below, with the target article filled in. Within each category, half of the items
included scope interactions with an intensional/modal operator, while half of the items
contained no intensional/modal operators (the examples below fall into the latter
category). Previous work by Ionin et al. (2003, 2004) showed that specificity, not scope,
had an effect on article use in L2-English (see also the discussion in Section 5.1.2).
(12) [+definite, +specific]
Louise: Where’s your mother?
Julie: She is meeting ___the_____ principal of my brother’s elementary school.
He is a very nice man. He is talking to my mother about my brother's grades.
(13) [+definite, -specific]
Ruby: It’s already 4pm. Why isn’t your little brother home from school?
Angela: He just called and told me that he got in trouble! He is talking to
___the_____ principal of his school! I don’t know who that is. I hope my brother
comes home soon.
(14) [-definite, +specific]
Grandfather comes for a visit
Grandfather: Where is my little granddaughter Beth? Is she home?
Father: No… She is not going to be back till late. She is having dinner
with ___a_____ girl from class – her name is Angie, and Beth really likes her.

3
An anonymous reviewer asks whether our test items in fact test specificity (i.e., presence of a
noteworthy property) or a different property, namely explicitly stated vs. denied familiarity with the
referent. The most common way of establishing specificity is in fact to state something that the speaker
knows about the referent (and conversely, denial of all knowledge helps establish lack of specificity),
which makes the two properties (noteworthy property vs. familiarity) quite difficult to distinguish.
However, it is possible to construct test items in which a noteworthy property is explicitly stated but
speaker knowledge of the referent is denied: instead, this knowledge is attributed to somebody other than
the speaker. Such items were tested in a previous study, reported in Ionin (2003, Chapter 6). It was found
that L1-Russian and L1-Korean L2-English learners in fact do overuse the in such items significantly more
often than in items which contain truly non-specific indefinites. This suggests that in fact the relevant factor
is presence of a noteworthy property (a hallmark of specificity) rather than direct speaker knowledge of the
referent.

8
(15) [-definite, -specific]
Mother comes home
Mother: How did Peter spend the day at his grandmother’s?
Father: He had a good time. He did his homework for tomorrow. Then he went
outside and played with ____a____ little girl – I don’t know who it was. Then he
came back inside; and then I came and took him home.
Participants were given 45 minutes to complete the elicitation task, and all of them
finished the test within that time limit. Participants took the elicitation and cloze tests in a
single session, and were reimbursed at the end of the testing. Participants in the U.S. were
reimbursed monetarily, while participants in Mexico were reimbursed with small gifts.4
Testing in Mexico was conducted by one of the investigators (Salvador Bautista
Maldonado), a native speaker of Mexican Spanish. Testing of Russian-speaking subjects
in the U.S. was conducted in some cases by another investigator (Tania Ionin), and in
other cases by a student assistant; both the investigator and the student assistant are native
speakers of Russian. Testing of English-speaking subjects in the U.S. was conducted by
student assistants.

4. Results and discussion

We first report the learners’ breakdown by proficiency, and then proceed to a detailed
report of the elicitation test results.

4.1. Cloze test results

The L2-learners were divided into three proficiency groups (beginner, intermediate,
and advanced) on the basis of their cloze test results; the division into three levels was
done via a K-means cluster analysis in SPSS. The cloze test scores for the three
proficiency levels, as well as the native controls, are given in (16).
(16) Cloze test results
# subjects range of scores
L1-Russian L2-English learners
beginners 8 2 - 27
intermediate 10 30 - 56
advanced 5 67 - 100
L1-Spanish L2-English learners
beginners 15 11 - 28
intermediate 8 31 - 57
advanced 1 67

control L1-English speakers 6 81-110

4
Monetary reimbursement of students taking the test in a university setting was considered by the
investigator in Mexico to be culturally inappropriate.

9
4.2. Elicitation test results
We now turn to the results of the elicitation test. Before examining results from the
target items, we looked at learners’ performance on the filler items. Performance on filler
items gives us an indication of whether learners have the level of English necessary to
understand the test items. We excluded from further analysis those L2-learners who gave
an acceptable response on fewer than two thirds of the filler items (16 out of 24). This
resulted in the exclusion of four L1-Russian participants and four L1-Spanish
participants. All of the excluded participants had scored as beginners on the cloze test (in
fact, the excluded L1-Russian participants had the four lowest cloze test scores in the
study, ranging from 2 to 7).

4.2.1. Group results

Turning to the results on the target items, we divided participants’ responses on each
category into four types: use of the; use of a; use of a dash to indicate a ‘no article’
response; and an ‘other’ response, when participant supplied another word, such as some,
one, this, etc. The tables below report the percentages of all four response types for the
three populations in the study.5 The target response for each category is highlighted.
(17) Elicitation test results: L1-English speakers (N=6)
[+definite]: target: the [-definite]: target: a
[+specific] 100% the 0% the
0% a 94.4% a
0% -- 0% --
0% other 5.6% other
[-specific] 100% the 0% the
0% a 100% a
0% -- 0% --
0% other 0% other
(18) Elicitation test results: L1-Russian speakers (N=19)
[+definite]: target: the [-definite]: target: a
[+specific] 93.0% the 23.7% the
5.3% a 74.6% a
0% -- 0% --
0.9% other 1.8% other
[-specific] 86.0% the 3.5% the
14.0% a 95.6% a
0% -- 0% --
0% other 0.9% other

5
In principle, the four percentages given in each table cell should add up to 100%. While this is the
case for the native English speakers, L2-learners’ responses sometimes add up to a little less than 100% as
a result of skipped responses (cases where the learners failed to supply any answer). These were extremely
rare, however.

10
(19) Elicitation test results: L1-Spanish speakers (N=20)
[+definite]: target: the [-definite]: target: a
[+specific] 87.5% the 1.7% the
0.8% a 92.5% a
8.3% -- 0% --
2.5% other 5.0% other
[-specific] 96.7% the 4.1% the
0.8% a 91.7% a
1.7% -- 0.8% --
0.8% other 3.3% other

The native English speakers performed exactly as expected, supplying the with all
definites and a with (nearly) all indefinites (the only exceptions were two ‘other’
responses on the specific indefinite category – a ‘one’ and a ‘my’). We turn to the L2-
learners’ results next.
The results of the two groups of L2-learners show two different patterns. The L1-
Russian speakers exhibit results similar to those of the L1-Russian (as well as L1-
Korean) speakers in the studies of Ionin et al. (2003, 2004): the two main error types are
overuse of the with specific indefinites and overuse of a with non-specific definites,
exactly as predicted.6 The learners are highly accurate on the other categories, specific
definites and non-specific indefinites.
The L1-Spanish speakers do not exhibit such a pattern: in fact, they are more accurate
on non-specific definites than on specific definites, contrary to the predictions. However,
the lowered accuracy on specific definites is not due to overuse of a but rather to
unexpectedly high article omission. Closer examination reveals that nearly every case of
article omission on this category was with a single item, before the target NP house of
Ben’s parents. This error is straightforwardly traceable to L1-transfer, since Spanish
requires article omission in this context (Fuí a casa, lit. ‘I went to house’). Except for this
one error type, the L1-Spanish speakers are quite accurate on both definites and
indefinites. Like native English speakers, they show no particular patterns of article
misuse, and crucially no effects of specificity.

4.2.2. Fluctuation effects and proficiency

We next examine the role that proficiency plays in the degree of fluctuation. Here,
only the results of the L1-Russian speakers are relevant, since the L1-Spanish speakers do

6
As mentioned in Section 3.1, six of the L1-Russian participants were bilingual in Russian and
Armenian. While Armenian does not have English-type articles, it does have an indefinite article (the
numeral ‘one’) as well as a post-nominal morpheme traditionally analyzed as a definite article. However,
Megerdoomian (2002) analyzes the latter as an Accusative Case marker and links it to specific/partitive
readings, similarly to Enç’s (1991) analysis of Accusative-Case markers in Turkish. A detailed
investigation of definiteness and specificity markers in Armenian, and how they might bear on L2-
acquisition, is beyond the scope of this paper. Looking at the results of the six bilingual Russian/Armenian
speakers, we find that three of them were among the participants excluded from data analysis due to poor
performance on fillers, while the other three showed very accurate article use. We leave this issue open for
further investigation.

11
not exhibit fluctuation in their article use (see section 4.2.4 for evidence that accuracy in
article use holds for L1-Spanish speakers at the individual as well as the group level). For
the L1-Russian speakers, on the other hand, we would expect that fluctuation should
decrease as proficiency increases. According to our hypothesis, fluctuation occurs until
learners receive sufficient input to recognize that English has the definiteness rather than
the specificity setting. Learners with greater English proficiency have presumably
received more input and/or attended better to the input than learners with lesser
proficiency; therefore, we would expect the more proficient learners to exhibit less
fluctuation and be more likely to converge on the definiteness setting. In order to
examine this prediction, we report the results of the L1-Russian speakers broken down by
proficiency level.
The results for percentages of the and a use by category are given in (20) and (21),
respectively. These show that as proficiency increases, use of the goes up with definites
and down with indefinites, while the reverse pattern is attested for use of a. Furthermore,
proficiency effects are most pronounced in the [+definite, -specific] and [-definite,
+specific] categories – the categories where fluctuation between the definiteness and
specificity options manifests itself. We see that this fluctuation indeed goes down as
proficiency increases.
(20) L1-Russian speakers: %the use by proficiency level
[+def, +spec] [+def, -spec] [-def, +spec] [-def, -spec]
Beginners 88% 79% 29% 8%
Intermediate 93% 87% 27% 3%
Advanced 97% 90% 13% 0%

(21) L1-Russian speakers: %a use by proficiency level

[+def, +spec] [+def, -spec] [-def, +spec] [-def, -spec]
Beginners 8% 21% 67% 92%
Intermediate 5% 13% 72% 95%
Advanced 3% 10% 87% 100%

However, while we see proficiency effects numerically, none of the differences

between groups reach statistical differences, for any category. Thus, while these results
are promising, they should be interpreted with caution. It is possible that the differences
would reach significance with a larger sample of participants. We leave the issue open to
further research.

4.2.3. Statistical test results

We conducted repeated-measures ANOVAs on the L2-learners’ responses in order to
test for effects of definiteness and specificity; proficiency level (beginner, intermediate,
advanced) was the between-subjects variable. Separate ANOVAs were conducted for the
Russian and Spanish groups; since there were four possible responses on each category,
we conducted separate ANOVAs on use of the and on use of a. The p-values are given in
the tables below for the two L1-groups (‘level’ corresponds to proficiency level).
Significant p-values (p < .05) are highlighted.

12
(22) Results of repeated-measures ANOVAs: L1-Russian speakers (N = 19)
use of the use of a
Definiteness p < .001*** p < .001***
Specificity p < .05* p < .05*
Definiteness X Specificity p = .05* p = .07
Definiteness X Level p = .40 p = .43
Specificity X Level p = .92 p = .82
Definiteness X Specificity p = .73 p = .82
X Level

(23) Results of repeated-measures ANOVAs: L1-Spanish speakers (N = 20)

use of the use of a
Definiteness p < .001*** p < .001***
Specificity p = .17 p = .95
Definiteness X Specificity p = .57 p = .95
Definiteness X Level p < .001*** p < .05*
Specificity X Level p = .10 p = .71
Definiteness X Specificity p = .83 p = .71
X Level

The statistical comparisons further highlight the differences between the Russian and
Spanish groups. For the Russian speakers, both definiteness and specificity have highly
significant effects on article use, with no effect of proficiency level. Definiteness and
specificity interact significantly when use of the is measured, marginally when use of a is
measured. This interaction is due to the fact that specificity effects are stronger with
indefinites than with definites. Paired sample t-tests for means show that the difference in
the use between specific and non-specific definites is not significant (p = .20); when use
of a is measured, the difference between specific and non-specific definites becomes
marginally significant (p = .076). On the other hand, the corresponding difference for
indefinites is highly significant whether use of the or a is measured (p < .01). Thus,
specificity effects are very clear with indefinites but marginal with definites.7
The Spanish speakers, on the other hand, are completely unaffected by specificity.
There is a significant effect of definiteness, as expected (since the learners clearly
distinguish definites from indefinites), and furthermore a significant interaction between

7
There is a possible confound with one of the test items that may bring down the specificity effect with
definites even further. One of the non-specific definite items included a context concerning a girls’ soccer
team, with the target sentence being: “I need to talk to ________ winner of this game – I don’t know who
she is, so can you please help me?”. This context was intended to be definite by virtue of there being only
one winner to a game. However, it was not realized until too late that a soccer match is not an appropriate
context for such a statement, since it is an entire team, not a single individual, who is the winner. Russian
speakers did show slightly more overuse of a in this context than in the other non-specific definite contexts,
and it is possible that they were interpreting the target sentence as talking about “a member of the winning
team”, in which case their use of a would have a different source – non-uniqueness rather than non-
specificity. On the other hand, neither the L1-Spanish group nor the control L1-English group overused a in
this context, which casts doubt upon this possibility.

13
definiteness and proficiency level. The interaction is due to the fact that accuracy of
article use with definites and indefinites increases with proficiency: one-way ANOVAs
show that use of the with definites improves significantly with proficiency (p < .01) as
does use of a with indefinites (p = .051). Thus, even though the Spanish speakers are
quite accurate on article use at all proficiency levels, they become even more accurate as
proficiency increases, as they become more attuned to the discourse context. The Russian
speakers, on the other hand, do not exhibit statistically significant effects of proficiency,
consistent with the findings reported in Section 4.2.2 (but see that section for evidence of
numerical proficiency effects).

4.2.4. Individual results

In order to determine whether the predicted fluctuation held at the level of individual
L1-Russian speakers, we looked at individual results for this group. A learner was
considered to make a specificity distinction with indefinites if he or she had at least one
more use of the, and at least one less use of a, with specific indefinites than with non-
specific indefinites. The same criteria applied to definites.
For the Russian speakers, four patterns were found: there were 6 L2-learners who
made a specificity distinction with both definites or indefinites (in two of these cases, the
distinction was stronger for indefinites); 3 L2-learners who made the specificity
distinction with indefinites only; 1 L2-learner who made the distinction with definites
only; six L2-learners who made no specificity distinction at all (i.e., performed in a fairly
target-like manner); and 3 L2-learners who made an unexpected reverse distinction with
definites, overusing a slightly more for specific than for non-specific ones. The
breakdown of these results by proficiency level is given in (24).
(24) Number of L1-Russian L2-English learners in each pattern, by proficiency level.
beginners intermediate advanced
specificity distinction with 2 3 1
both indefinites and definites
specificity distinction with 2 1
indefinites only
specificity distinction with 1
definites only
no specificity distinction at all 4 2
unexpected non-specificity 1 1 1
distinction with definites only

As seen in the above table, more L2-learners made the specificity distinction with
indefinites than with definites; there were both intermediate and advanced L2-learners
who behaved in an essentially target-like manner, not making the specificity distinction at
all. Importantly, all of the learners who made a specificity distinction made it only some
of the time – there was not a single L2-learner who consistently used the with all specific
indefinites, or used a with all non-specific definites. Those learners who are non-
targetlike show fluctuation between definiteness and specificity – they do not adopt
specificity as the pattern for English. This is to be expected, since the English input does
not favor the specificity pattern.

14
 In the case of the L1-Spanish speakers, individual results are much the same as the
group results: the vast majority of these L2-learners (17 out of 20) are target-like, making
no distinction between specific and non-specific indefinites, or between specific and non-
specific definites. The remaining three learners, all beginners, make an unexpected
‘reverse’ distinction – one using a slightly more with specific than non-specific definites,
two using the slightly more with non-specific than with specific indefinites. Since these
errors are very minor, and present only for beginners, they are likely to reflect lack of
attention to contextual information rather than any deep linguistic distinction. The same
explanation may hold for the three L1-Russian speakers who make the reverse
distinction.

4.3. Discussion
To sum up, the L1-Russian L2-English learners in this study behaved much like L1-
Russian (and L1-Korean) L2-English learners in the earlier studies by Ionin et al.; since
the test format differed in this study and the earlier studies, this provides convergent
evidence for the role of specificity in L2-English article choice. One difference is that the
Russian speakers in the current study showed only marginal specificity effects with
definites, whereas in the earlier studies specificity effects were seen with both indefinites
and definites (although generally being stronger with indefinites). It is necessary to
collect more learner data before a definitive conclusion about the status of specificity
effects with definites can be reached. It is possible that the contexts in the present study
did not clearly set up the lack of specificity for the category of non-specific definites. As
discussed in Ionin (2003, 2006), it is much easier to construe a definite context as specific
than to construe an indefinite context as specific: since definites by definition denote
unique referents, there could always be something noteworthy about the referent that
results in the DP being treated as [+specific]. In (13), for instance, the DP the principal of
his school is designed to be non-specific, since the speaker denies all knowledge of the
referent; however, the context could be construed as [+specific], with the referent bearing
a noteworthy property such as “the principal who is going to punish my brother”. We
come back to this issue briefly in Section 5.2.
Turning to the L1-Spanish L2-English learners, we find that they were highly
accurate in their article choice, on both definites and indefinites, providing support for the
“transfer overrides fluctuation” view in (11). We now have clear empirical evidence that
L1-transfer is operative at the level of article semantics.
It might be argued that the different performance of the Spanish and Russian groups
is due not to L1-transfer but to differences between the two subject populations in this
study: while the Russian speakers had been receiving naturalistic exposure to English in
the U.S., the Spanish speakers were confined to classroom instruction. Not surprisingly,
given this difference in exposure, the 19 L1-Russian speakers whose results are reported
above had higher cloze test scores, on average, than the 20 L1-Spanish speakers: 47.2 vs.
29.9; this difference was highly significant (p = .01).
Crucially, however, if proficiency were the main issue, these differences should lead
to the opposite pattern than the one we saw: the Russian speakers should be more
accurate than the Spanish speakers with respect to article use. In fact, in Ionin et al.
(2003), the Korean-speaking L2-learners were more proficient than the Russian-speaking
L2-learners, and were also more accurate on article use (although the two groups showed

15
the same pattern). In the present study, however, the Russian speakers were less accurate
than the Spanish speakers in their article use, despite having significantly higher
proficiency and more L2-exposure. This effect is best explained by L1-transfer.

5. Universal Grammar and input

In this paper, we have shown that L1-Spanish and L1-Russian learners of English
differ in their treatment of English articles. While Spanish speakers distinguish between
the and a on the basis of definiteness, Russian speakers fluctuate between distinguishing
them on the basis of definiteness vs. on the basis of specificity. We now address in more
depth the question of why this is the case. For Spanish speakers, the answer seems fairly
straightforward: these learners apparently transfer the semantics of Spanish onto English
articles. Since, in the domain under consideration, Spanish and English behave in the
same way (dividing articles on the basis of definiteness), L1-transfer allows the learners
to perform in a native-like fashion. It is not necessary to posit any additional factors, such
as direct UG-access, in order to account for the Spanish speakers’ performance.
In the case of Russian speakers, however, multiple possibilities should be considered.
There are in principle four factors that can account for these learners’ performance: (1)
L1-transfer (from a domain outside the article system); (2) explicit instruction; (3) input-
driven learning; and (4) direct access to semantic universals. We will not be concerned
with the first two factors here, but instead refer the reader to Ionin (2003), who shows
that neither L1-transfer nor explicit instruction can successfully account for article use
among L1-Russian L2-English learners: on the one hand, nothing in the grammar of
Russian can account for the pattern of the learners’ errors; on the other hand, textbooks of
English as a second language provide learners with multiple strategies for article use, and
there is no a priori reason to suppose that learners should choose the strategy that leads to
an association of the with specificity. We will be concerned here with the other two
possibilities: input-driven learning vs. access to semantic universals. We turn to this next.

5.1. Domain-general or domain-specific learning?

We ask first whether the pattern of article use exhibited by the learners in our study
can be accounted for via domain-general statistical-learning mechanism. Recent research
on various topics in language acquisition, including word segmentation and syntax, has
suggested that some aspects of language acquisition are accomplished by mechanisms of
statistical learning: mechanisms that compute various complex statistics concerning the
occurrence and distribution of linguistic forms (see, among others, Mintz, Newport and
Bever, 2002; Newport and Aslin, 2004; Saffran, Aslin and Newport, 1996; Saffran,
Newport and Aslin, 1996).

5.1.1. Statistical learning in L2 article use

While studies of statistical learning are typically concerned with child first language
learners, a recent paper by Hudson Kam and Newport (2005) has looked at second
language acquisition, comparing children and adults on their ability to learn a mini-
language in the laboratory. This mini-language consisted of unpredictable input:
determiners were sometimes used and sometimes dropped, for no apparent grammatical
reason. The study found that children were more likely than adults to overregularize the

16
input, imposing a systematicity on the language they were learning. Adults, on the other
hand, learned and reproduced the variability: the output closely resembled the input.
Hudson Kam and Newport suggest that this difference between children and adults is at
least in part due to domain-general rather than language-specific learning mechanisms:
unlike adults, children impose patterns on the input.
The subjects in Hudson Kam and Newport’s study were exposed to a deliberately
variable mini-language, where the use vs. omission of determiners had no grammatical
motivation. These learners simply mimicked the statistical distribution of present vs.
absent determiners; they didn’t acquire any deeper linguistic rule because there was none
to acquire. For instance, in one of the experimental conditions, the adults in Hudson Kam
and Newport’s study heard articles used 60% of the time and dropped 40% of the time,
and therefore concluded that the language required article use approximately 60% of the
time, without worrying about the grammatical reasons for article use.
On the other hand, the adults in our study were exposed to a natural language, in
which the choice of one determiner over another is semantically motivated. Suppose that
the learners have no innate linguistic knowledge to guide them as they acquire English
determiners, but are constrained to purely statistical learning. What would be the result?
We see two possibilities. The first is that our L2-learners blindly mimic the statistical
distribution of articles in the input, disregarding contextual considerations: e.g., if
learners’ input indicates that about 60% of all article uses are uses of the and 40% are
uses of a, the learners will subsequently use the about 60% of the time and a 40% of the
time, regardless of the context. Clearly, such random behavior does not take place:
learners do show consistency in their results, as we have seen, and their errors are
confined to particular types of definites and indefinites.
The other (and more plausible) possibility to consider is whether adult learners could
extract patterns of article use from the input without any innate linguistic mechanisms to
guide them. Suppose that the learners are driven purely by domain-general mechanisms:
that is, they are able to notice patterns in the input, but they have no language-specific
mechanisms telling them that these patterns should be built upon linguistic factors like
definiteness or specificity. What would be the result?

5.1.2. L2-learners’ hypothesis space

In the absence of any linguistic constraints, the learners’ hypothesis space should be
very large. They may decide that the means any number of things, such as “the is used
only at the start/end of the sentence”; “the is used only with singular/plural nouns”; “the
is used only when the noun is preceded (or not preceded) by an adjective”, “the is used
only with nouns denoting concrete objects”; “the is used only with nouns whose referents
are green in color”; and so on. The same virtually unlimited hypothesis space will exist
for a. Clearly, learners do not entertain the majority of these hypotheses. Of course, some
hypotheses (e.g., the one about green referents) are easily disqualified on the basis of just
a few examples in the input. However, many others still remain; for instance, a
hypothesis that links use of the to the presence of relative clause (RC) modification is a
fairly reasonable one (cf. expressions such as the student that I spoke to yesterday, which
is more likely to be uttered out of the blue than a simple the student). However, as
discussed in Ionin (2003), this strategy alone does not account for the patterns of L2-

17
English article use; the specific indefinite contexts in the present study also did not
contain RC-modification.
The learners’ data provide evidence that in deciding what the and a mean, learners
entertain only two hypotheses: definiteness and specificity. They do not appear to
entertain other hypotheses, whether plausible ones (involving wide scope or
modification) or implausible ones (involving the color of the referent). While the present
study was not set up to test every single alternative hypothesis (indeed, this would be
impossible), the results do provide evidence against at least two alternative hypotheses:
association of the with ‘assertion of existence’ and association of the with modification.
We discuss each in turn.
First of all, L2-learners do not appear to associate the with assertion of existence,
contrary to what has been claimed in previous L2-literature (e.g., Thomas 1989). As
mentioned in Section 3.2.2, half of the items within each category contained
intensional/modal operators and half did not. This meant that in the category of non-
specific indefinites, half of the target sentences (those without any operators) asserted
existence of the referent (e.g., the little girl in (15)), whereas half of the sentences did not
assert existence because the indefinite scoped under an operator (as in I am looking for a
warm hat, where no specific hat is asserted to exist). If the L1-Russian L2-English
learners equated the with ‘assertion of existence’, we would expect higher overuse of the
in the former case; however, overuse of the was in fact very low in both categories of
non-specific indefinites: 1.8% when existence was asserted, and 5.4% when it was not.
Overuse of the was largely confined to the category of specific indefinites, suggesting
that specificity, not assertion of existence, influences L2-English article choice.8 This
replicates earlier findings of Ionin et al. (2003, 2004).
Yet another hypothesis ruled out by the present study is that L2-learners use the
whenever they see some type of modification, e.g., adjectival modification. Both
[+specific] and [-specific] indefinite items in the test involved modification: in each
category, five of the six test items contained adjectival modification (as in a warm hat),
and the sixth item contained PP modification (as in a book about airplanes). If
modification contributed to overuse of the, we would expect it to do so both specific and
non-specific indefinites; however, this was not the case. Conversely, both [+specific] and
[-specific] definite items in the test involved PP-modification (as in winner of this game),
but overuse of a was largely confined to non-specific definites.
Thus, we can see that the learners do not entertain every logical possibility in
deciding what the and a mean; rather, their hypothesis space is constrained by the
concepts of definiteness and specificity. Note further that the learners’ performance
cannot be accounted for exclusively by the frequency of specific vs. non-specific DP-
types in English. Consider the following reasoning. Suppose that specific definites and
non-specific indefinites are more common in the input than specific indefinites or non-
specific definites. The learners hear enough specific definites to know that they require

8
Note that specific indefinites by definition always involve assertion of existence: in order to intend to
refer to a particular individual (the hallmark of specificity), the speaker must assert the existence of that
individual (see Ionin 2003, 2006 for more discussion). Thus, all specific indefinite items in the test asserted
existence, regardless of the presence or absence of intensional/modal operators. Overuse of the was 32%
for specific indefinites in sentences with operators and 16% for those in sentences without operators.

18
the and enough non-specific indefinites to know that they require a. However, they don’t
hear enough exemplars of the other two categories, which results in the pattern that we
found – fluctuation. While this may seem like an attractive explanation, it cannot work,
regardless of the actual frequencies of different DP types. This explanation presupposes
that learners are able to categorize DPs on the basis of definiteness and specificity: that
upon hearing an indefinite, for instance, they put it in the mental category of either ‘non-
specific indefinite’ (frequent) or ‘specific indefinite’ (infrequent). However, if the
learners are able to form such categories, this means that they have access to linguistic
categories and are relying on more than frequency (although frequency coupled with such
grammatical knowledge may play a role – more on this below).
It is not clear how L2-learners would settle on the precise patterns that they do
without innate linguistic knowledge to guide them. First of all, the input is
underdetermined: it does not tell learners that use of the corresponds to definiteness
rather than to some other distinction – as discussed above, the space of potential
hypotheses for what the might mean is virtually infinite. Even if learners could somehow
extract the definiteness pattern from the input by relying on purely statistical learning,
they should not be able to extract the specificity pattern as well, since this pattern does
not in fact exist in the input. Crucially, learners are producing an output that does not
quite correspond to the input, but that is nevertheless quite constrained (errors in only two
of four possible categories) and that, moreover, corresponds to article semantics in
natural language. The best explanation for this phenomenon is that learners are retrieving
this pattern directly from UG, by accessing the inventory of semantic universals.

5.2. The role of input

If the learners are in fact guided from UG in their acquisition of English article
choice, why do they make errors? Why don’t they just settle on the appropriate patterns
for English? Here, it is important to make the distinction between the state of the
learners’ grammar and the learning process. We have argued that the state of the learners’
grammar, as reflected by their article choice, is constrained by UG-provided semantic
universals. On our hypothesis, innate linguistic knowledge supplies the learners with
possible specifications of article use but does not tell them how to choose the actual
specifications for their L2. Crucially, this is where input begins to play an important role:
in order to decide that English articles are distinguished on the basis of definiteness rather
than specificity, L2-learners need to pay attention to the input triggers. We argue that it is
this learning from the input that is particularly difficult for L2-learners (much of the
following discussion can also be found in Ionin et al. (2004)).
The triggers related to article choice are discourse-based: they do not arise (at least
not obviously) from the syntactic configuration. Both specific and definite articles may in
principle appear in the same environments – e.g., in simple SVO sentences with no
intensional or modal operators or quantifiers. In order to determine whether the is
[+definite] or [+specific], the L2-learner needs to evaluate the discourse situation and
decide whether the is marking the presupposition of uniqueness (from the hearer’s
perspective) or the existence of a noteworthy property (from the speaker’s perspective).
Since definites are often specific, both hypotheses will be compatible with many
situations. The learner thus also needs to pay attention to use of a, and note that in

19
contexts which are [+specific] but [-definite], the is never used by native English
speakers.
The discourse triggers related to article choice are often ambiguous. For instance,
suppose that an L2-learner hears someone use a phrase like I talked to the new student
from Sweden this morning. This phrase is compatible with the hypothesis that the is
[+definite]: even if the learner has never heard of any student from Sweden, it is fairly
easy to accommodate the knowledge that there is a unique, salient new student from
Sweden that the speaker is acquainted with. The phrase is also compatible with the
hypothesis that the is [+specific]: the speaker may be wishing to attract attention to the
identity of the student in question.
Suppose next that an L2-learner hears the phrase I talked to a new student from
Sweden this morning. This phrase is compatible with the hypothesis that a is used with
indefinites: the speaker is not presupposing her listener to have knowledge of the student.
However, this phrase is also compatible with the hypothesis that a is used in the absence
of specificity: the speaker is simply choosing not to attach any importance to a particular
student’s identity. The same ambiguity persists across many contexts.9
One might then wonder how any learner can ever acquire the semantics of English
articles. The answer, we think, is generalization across individual instances. A single DP
might be ambiguous between definite and specific or indefinite and non-specific
readings. However, if the learner is consistently hearing the used only when the
presupposition of uniqueness has been met (even when the speaker attaches no
importance to the referent’s identity), and consistently hearing a used only when the
presupposition of uniqueness hasn’t been met (even when the speaker attaches
importance to the identity of the referent), the learner should generalize that the marks
definiteness rather than specificity. Given the subtlety of the discourse triggers related to
speaker and hearer knowledge, generalizing from them is likely to be a fairly long and
difficult process. This would result in the continued fluctuation between the two semantic

9
The fact that so many English DPs are both [+definite] and [+specific] brings up an interesting
alternative analysis of our results. It is possible that, instead of fluctuating between definiteness and
specificity, the L1-Russian L2-learners have correctly identified the as being [+definite]; however, at least
part of the time, they think that the also has a meaning on which it is [+specific]: i.e., they think that the is
ambiguous (much as, in reality, this is ambiguous between a demonstrative and a specific indefinite
reading). They might draw this conclusion on the basis of the fact that the vast majority of the-NPs in
English appear to be both [+definite] and [+specific]: as noted in Section 4.3, definites are easily
compatible with specificity. Furthermore, it is quite likely that [+specific] the+NPs are more frequent in
English than [+specific] a+NPs (although we would need a corpus analysis to check whether this is indeed
the case), a fact which would also lead learners to associate the with specificity as well as definiteness. This
‘ambiguity of the’ hypothesis would explain why the Russian speakers in this study tended to use the with
all definites (showing relatively little a overuse on [-specific] definites) while still overusing the with
specific indefinites. However, before we can draw such a conclusion, we would need more data concerning
L2-learners’ performance on different types of definites. For now, it remains a speculation. Crucially, like
the fluctuation hypothesis, this ambiguity hypothesis requires that learners are aware of the properties of
definiteness and specificity – an awareness that, as discussed above, can only come from access to semantic
universals. Frequency may play a role, but frequency alone is not enough.

20
universals on the part of otherwise advanced L2-learners. Target-like attainment is of
course not impossible: we saw that some of the L2-learners in the present study showed
target-like patterns of article use. However, it may be that (in the absence of L1-transfer)
acquisition of articles lags behind acquisition of other domains of the grammar, precisely
because it involves discourse-based triggers. The reason that L1-Spanish L2-English
learners are quite successful at acquiring English articles is that they can rely on L1-
transfer: they are able to converge on a target-like grammar as a result of transferring the
semantics of Spanish articles onto English, without having to pay attention to the
discourse-based triggers of English.
On a final note, the above discussion has interesting implications for age effects in
acquisition. Hudson Kam and Newport suggest that when the input is variable, child but
not adult L2-learners are able to impose grammatical patterns on this input. It is possible
that when the input is non-variable (as in the case of English articles), child L2-learners
are better than adult L2-learners at extracting grammatical patterns from the input. If this
is the case, then we would expect child L2-learners to converge on the definiteness
pattern for English articles with greater success than adult L2-learners from the same L1.
This prediction forms the basis of our currently ongoing study with child L1-Russian and
L1-Spanish L2-English learners.
To sum up, our account of article errors in adult L2-English depends on both UG-
access and input-driven learning: while UG tells the learners which article specifications
are possible, the input tells them which article specifications the target language actually
uses. We suggest that the learners have unimpaired access to the UG-provided
possibilities, but have problems generalizing from the input.

6. Conclusion
In this paper, we have examined English article use among adult L1-Russian and L1-
Spanish learners of English. We have found that three factors are relevant for the
acquisition of English articles: L1-transfer, input, and Universal Grammar. Speakers of
Spanish, coming from an L1 with articles, are able to transfer article semantics from their
L1 to their L2. Speakers of Russian, which lacks articles, are not able to do this, so their
learning proceeds through a combination of UG-access and input processing. UG tells
learners what possible patterns of article choice in a natural language may look like.
Learners fluctuate between these patterns – definiteness and specificity – until the input
tells them that only one of these patterns is appropriate for English. We have argued that
it is at this stage that L2-learners may experience particular difficulties: given the subtlety
of discourse-based triggers for article choice, even fairly advanced learners exhibit
fluctuation.
Importantly, the patterns of article use that we found cannot be explained via
statistical learning alone: there is no a priori reason to expect definiteness and specificity
to be the patterns that learners (especially those in the earlier stages of acquisition) should
extract from the input if they have no domain-specific knowledge to guide them. On the
domain-specific view of L2-acquisition, on the other hand, there is a reason to expect
precisely these patterns, since they are instantiated in natural language.

21
References
Blass, R. 1990, Relevance Relations in Discourse. A Study with Special Reference to
Sissala, Cambridge University Press, Cambridge.
Butler, Frances 1980. Cloze as an alternative method of ESL placement and proficiency
testing. Research in Language Testing, eds. John Oller and Kyle Perkins, 121-128,
Boston, MA: Newbury House.
Dechert, H. & M. Raupach (eds.). 1989. Transfer in language production. Norwood, NJ:
Ablex.
Donnellan, K. 1966. ‘Reference and Definite Descriptions’, The Philosophical Review
75, 281-304. Reprinted in D. Steinberg and L. Jakobovits (eds.), (1971) Semantics:
An Interdisciplinary Reader in Philosophy, Linguistics, and Psychology, pp. 100-114,
Cambridge University Press, Cambridge.
Enç, M. 1991. The semantics of specificity. Linguistic Inquiry, 22, 1-25.
Gass, S. & L. Selinker (eds.). 1992. Language transfer in language learning. Amsterdam:
John Benjamins.
Hawkins, R., S. Al-Eid, I. Almahboob, P. Athanasopoulos, R. Chaengchenkit, J. Hu, M.
Rezai, C. Jaensch, Y. Jeon, A. Jiang, Y.-K. Leung, K. Matsunaga, M. Ortega, G.
Sarko, N. Snape, & K. Velasco-Zárate. 2006. ‘Accounting for English article
interpretation by L2 speakers’, EUROSLA Yearbook.
Heim, I. 1991. ‘Articles and Definiteness,’ published in German as ‘Artikel und
Definitheit,’ in A. V. Stechow and D. Wunderlich (eds.), Semantics: An International
Handbook Of Contemporary Research, De Gruyter, Berlin.
Hudson Kam, C. and E. Newport. 2005. Regularizing unpredictable variation: The roles
of adult and child learners in language formation and change. Language Learning and
Development, 1(2), 151-195.
Huebner, T. 1983. A longitudinal analysis of the acquisition of English. Ann Arbor:
Karoma.
Ionin, T. 2003. Articles Semantics in Second Language Acquisition, Ph.D. dissertation,
Massachusetts Institute of Technology. Distributed by MITWPL.
Ionin, T. 2006. ‘This is definitely specific: specificity and definiteness in article systems’.
Natural Language Semantics, 14, 175-234.
Ionin, T., H. Ko & K. Wexler. 2003. ‘Specificity as a grammatical notion: Evidence from
L2-English article use’, In G. Garding and M. Tsujimura (eds.), Proceedings of
WCCFL 22, Cascadilla Press, 245-258.
Ionin, T., H. Ko and K. Wexler: 2004, ‘Article semantics in L2-acquisition: the role of
specificity,’ Language Acquisition 12, 3-69.
Larsen-Freeman, D. 1975a. The acquisition of grammatical morphemes by adult learners
of English as a second language. Ph.D. dissertation, University of Michigan, Ann
Arbor.
Larsen-Freeman, D. 1975b. The acquisition of grammatical morphemes by adult ESL
students. TESOL Quarterly, 9, 409-430.
Leung, Y.-K. (2001) “The Initial State of L3A: Full Transfer and Failed Features?,” in X.
Bonch-Bruevich, W.J. Crawford, J. Hellermann, C. Higgins and H. Nguyen, eds., The
Past, Present and Future of Second Language Research: Selected Proceedings of the
2000 Second Language Research Forum, Cascadilla Press, Somerville, MA, 55-75.

22
Master, P. (1987) A Cross-Linguistic Interlanguage Analysis of the Acquisition of the
English Article System, Doctoral dissertation, University of California, Los Angeles.
Megerdoomian, K. 2002. Beyond words and phrases: A unified theory of predicate
composition. Ph.D. dissertation, University of Southern California.
Mintz, T. H., E. L. Newport, and T.G. Bever. 2002. The distributional structure of
grammatical categories in speech to young children. Cognitive Science, 26, 393–425.
Mosel, U. and E. Hovdhaugen. 1992. Samoan Reference Grammar, Scandinavian
University Press, Oslo.
Murphy, S. 1997. Knowledge and production of English articles by advanced second
language learners. Ph.D. dissertation, University of Texas at Austin.
Newport, E. L., & R. N. Aslin. 2004. Learning at a distance: I. Statistical learning of non-
adjacent dependencies. Cognitive Psychology, 48, 127–162.
Odlin, T. 1989. Language transfer: Cross-linguistic influence in language learning.
Cambridge: Cambridge University Press.
Parodi, T., B.D. Schwartz and H. Clahsen. 1997. On the L2 acquisition of the
morphosyntax of German nominals. Essex Research Reports in Linguistics, 15, 1-43.
Parrish, B. 1987. A new look at methodologies in the study of article acquisition for
learners of ESL. Language Learning, 37, 361-383.
Prince, E.: 1981, ‘On the Inferencing of Indefinite-This NPs,’ in A.K. Joshi, B.L. Webber
and I.A. Sag (eds.), pp. 231-250, Elements Of Discourse Understanding, Cambridge
University Press, Cambridge, UK.
Robertson, D. 2000. Variability in the use of the English article system by Chinese
learners of English. Second Language Research, 16, 135-172.
Saffran, J. R., R. N. Aslin, and E. L. Newport. 1996. Statistical learning by 8-month old
infants. Science, 274, 1926–1928.
Saffran, J. R., E. L. Newport, and R. N. Aslin. 1996. Word segmentation: The role of
distributional cues. Journal of Memory and Language, 35, 606-621
Schwartz, B.D. 1998. The second language instinct. 1998. Lingua, 106, 133-160.
Schwartz, B.D. and R.A. Sprouse. 1994. Word order and nominative case in non-native
language acquisition: A longitudinal study of (L1 Turkish) German interlanguage. In
T. Hoekstra & B. D. Schwartz (eds.), Language Acquisition Studies in Generative
Grammar: Papers in Honor of Kenneth Wexler from the 1991 GLOW Workshops.
Amsterdam: John Benjamins, 317-368.
Schwartz, B. D. and R. A. Sprouse. 1996. L2 cognitive states and the Full Transfer/Full
Access model. Second Language Research, 12, 40-72.
Thomas, M. 1989. The acquisition of English articles by first- and second-language
learners. Applied Psycholinguistics, 10, 335-355.

23