Definiteness Across Languages and in L2 Acquisition: Bert Le Bruyn
Definiteness Across Languages and in L2 Acquisition: Bert Le Bruyn
Definiteness Across Languages and in L2 Acquisition: Bert Le Bruyn
We present evidence suggesting that articleless languages are not created equal
and that this influences how native speakers of these languages acquire article lan-
guages like English. The evidence suggested that Mandarin learners of English are
not unequivocally bearing out the predictions of the Fluctuation Hypothesis, unlike
learners of English with e.g. Korean, Russian and Japanese as an L1. We propose
a research program that approaches articles as a syntax/semantics interface phe-
nomenon. The program considers the syntax/semantics interface of definiteness
in its entirety and makes no a priori assumptions about how it is best analysed.
Rather, it adopts a data-driven comparative approach with multiple L1s that al-
lows us to give a fine-grained answer to the question how L1 influence plays out
for definiteness.
1 Introduction
The L2 acquisition of the definite article has already played an important role
in the debate on L1 influence. It is one of the morphemes that – according to
the original morpheme studies (e.g. Dulay & Burt 1974) – is acquired by all L2 Not
learners at the same time. The work of Ionin and colleagues (e.g. Ionin et al. listed in
2004; Ionin & Montrul 2010) has however shown that L1 influence distinguishes manuscript.
between learners with an L1 that has articles and those with an articleless L1. We Revision
argue that the time has come to probe further and look into whether L1 influence pending.
is identical for all learners with an articleless L1.
We briefly sketch the SLA literature on L2 article acquisition by learners with Not
an articleless L1 (§2), argue that L1 influence from articleless L1s is not uniform listed in
manuscript.
Revision
pending.
\lsCollectionPaperCitationText.
Bert Le Bruyn
(§3, §4), and propose a research program that allows us to investigate this in
detail (§5).
2
Definiteness across languages and in L2 acquisition
Secretary: She is meeting with a student, but I don’t know who it is.
(Ko et al. 2010)1
+definite -definite
+specific the a
-specific the a
The difference between the two systems lies in the fact that the Samoan ‘defi-
nite’ article is also used for specific indefinites. Ionin and colleagues hypothesize
that L2 learners need to determine which of the two article systems applies in the
languages they are learning, leading them to fluctuate between the two systems
and sometimes overproduce definite articles in specific indefinite contexts. This
1
We provide examples taken from Ko et al. (2010). These represent the most recent instance of
Ionin’s (2003) paradigm by Ionin and colleagues.
2
In this paper, we do not separately report on native speaker controls but refer to Ionin et al.
(2004) and Le Bruyn & Dong (2017a,b) for the relevant data. Native speakers in these studies
performed at ceiling on providing indefinite articles both in the indefinite specific and indefi-
nite non-specific condition.
3
Bert Le Bruyn
+definite -definite
+specific le le
-specific le se
the
Non-specific 8% (4/48)
Specific 50% (24/48)
Even though we lack some details about the studies and cannot report the
data fully in parallel, the general picture is clear: Japanese learners appear to
3
To present the cleanest possible picture, we restrict ourselves to data from experimental items
without scopal interactions and data that focus – as in Ionin’s original experiment – on the
singular.
4
Definiteness across languages and in L2 acquisition
a the
Non-specific 94% 4%
Specific 70% 29%
be sensitive to specificity and their production of English articles bears out the
predictions of the Fluctuation Hypothesis. Both in Hawkins et al. (2006) and Reid
et al. (2006), Japanese learners overproduce definites in the specific indefinite
condition but not in the non-specific indefinite condition.
The data from Mandarin learners that Snape et al. (2006) report on are taken
from Ting (2005). They are summarized in Table 5:
Table 5: Percentage of a and the responses by 8 Mandarin respondents
in the specific and non-specific conditions in Ting (2005).
a the
Non-specific 94% 4%
Specific 92% 3%
The contrast between the Mandarin and the Japanese learners is striking: whereas
Japanese learners overproduce definites in 29 to 50% of specific indefinite con-
texts, Mandarin learners seem to behave like native speakers in only overproduc-
ing definites in 3% of the same contexts.
Snape et al. (2006) conjecture that the contrast between Mandarin and Japanese
learners might be explained by the fact that Mandarin is in a more advanced stage
of developing an article system parallel to that of English. It would be grammati-
calizing the numeral yi (‘one’) as the indefinite and the demonstrative nei (‘that’)
as the definite. L1 transfer could then explain why Mandarin learners perform
more native-like.
5
Bert Le Bruyn
falls short of providing sufficient motivation at two levels: (i) it does not provide
any comparative data that would support the difference in grammaticalization
between Mandarin and Japanese, (ii) it provides no systematic way of linking the
alleged difference to the performance of L2 learners.
In the remainder of this paper, we will do two things. The first is to provide
data from two further small-scale studies that lend support to the idea that Man-
darin learners of English do not unequivocally bear out the predictions of the
fluctuation hypothesis (§4). The second is to present a new methodology that
allows us to systematically study L1 influence in acquisition (§5).
4
Trenkic (2008) however does not agree with the interpretation of the data. See Trenkic (2008)
and Ionin et al. (2009) for discussion.
5
The specific indefinite items we used were items 25, 26, 27 and 28 from Ionin et al. (2004). For
6
Definiteness across languages and in L2 acquisition
a the ø
Non-specific 84% (118/140) 9% (13/140) 6% (9/140)
Specific 88% (123/140) 9% (13/140) 3% (4/140)
The data of the non-specific and the specific condition are almost fully paral-
lel. We did run a mixed effects model with item and participant as random fac-
tors. Given that the selection of ø gives us no insight into whether subjects con-
sider the item indefinite or definite, we modelled these responses as missing data.
As expected, there was no overall effect of condition and pairwise comparisons
showed no difference between the two conditions (F(1, 165) = 0.002, p=0.963).
We interpret the data in Table 6 as indicating that Mandarin L2 learners are
unlikely to be sensitive to specificity in the way it is operationalized by Ionin
et al. (2004). As we indicated before, we are aware of the fact that few to no
the non-specific indefinite items, we used items 37, 38, 39 and 40. These non-specific items were
control items in the original study but do not contain the explicit statement of lack of speaker
knowledge criticized in Trenkic (2008). Ionin et al. (2009) indicate that this explicit statement of
lack of speaker knowledge is not a crucial part of the operationalization of non-specificity and
Ionin et al. (2004) found that their indefinite control items pattern with non-specific indefinite
items: there is a significant difference in the responses with the specific indefinite test items
(p<0.001) but not with the non-specific indefinite test items.
7
Bert Le Bruyn
conclusions can be drawn on the basis of a null result, but we did consider it
relevant to at least check whether the null result found in Ting (2005) is not
merely due to its small sample size.
(3) Have I already told you about the scariest moment of my life? Well, one
day I saw a girl on top of a building… All of a sudden, she starts to dance,
slips on a brick and falls off the building! Fortunately she landed on some
cardboard boxes and didn’t get hurt…
The girl is the protagonist in (3): after her introduction, she is immediately
picked up as the subject of the next sentence and she remains the main character
of the story throughout. The brick is a secondary character: it is introduced but
never referred back to. We made 8 stories following the setup of the one in (3):
(i) introduction of the protagonist, (ii) story about actions of the protagonist, (iii)
optional introduction of a secondary inanimate character, (iv) continuation of
the story about the protagonist. A further 8 stories were created as fillers and
had a freer structure.6
Ionin et We adopted the forced choice setup used in Ionin et al.’s specificity paradigm
al. 2004 but limited the answer possibilities to the definite and the indefinite article. For
or 2009? 6
To keep the processing cost of the task as low as possible we decided not to increase the
number of fillers beyond 8.
8
Definiteness across languages and in L2 acquisition
the experimental items, an article had to be selected for the DP introducing the
protagonist (four items) or the secondary character (four items), thus leading to
our two experimental conditions. For the fillers, the relevant DPs concerned col-
locationally and/or grammatically enforced definites (four items) and indefinites
(four items):
(4) You’ll never guess what happened today! I’ve seen a woodpecker for the
first time in my life.
(5) You’ll never guess what happened today! You know I have a sister? Well,
she came to visit me for the first time in 10 years!
Wherever possible, the similarity between the stories across the two condi-
tions was maximized. We took care though to create sufficient variation to pre-
vent subjects from inferring answers. One way of doing so was to use the posses-
sive my daughter in the experimental item based on (3) when asking participants
to fill in the blank for the backgrounded character.
To create a communicative context for the stories, we inserted them in a pub
context in which one guy tells them to another. This was done pictorially as in
Figure 1 and Figure 3.
9
Bert Le Bruyn
a PowerPoint presentation with one slide for the instructions and one slide for
each test story. Participants were asked to indicate for each story whether they
preferred the version with the indefinite (Option 1) or the definite article (Op-
tion 2). A small language biography survey was orally carried out by the student
assistant to check for the potential influence of stays abroad or other languages.
No student had spent time in an English-speaking country or mastered another
article language than English. The participants were given no time limit but all
of them completed the experiment in under five minutes.
Table 7 summarizes the results of our 22 participants on the test items. L2
learners are at ceiling in the specific/foregrounded condition but produce 31% of
definites in the non-specific/backgrounded condition.
Table 7: Percentage of a and the responses by 22 Mandarin respondents
in the foregrounded and backgrounded conditions in Le Bruyn & Dong
(2017b).
a the
Non-specific 69% (53/88) 31% (27/88)
Specific 95% (75/88) 5% (5/88)
10
Definiteness across languages and in L2 acquisition
(t(174)=4.576, p<0.001).
The results indicate that our participants were likelier to produce a definite
for non-specific referents than for specific referents. This is exactly the opposite
of what we would expect based on the fluctuation hypothesis. In combination
with the data from Ting (2005) and the data we presented in §3.1, we conclude in
that evidence is accumulating that suggests Mandarin learners of English are manuscript:
different from learners with other articleless L1s in that they do not unequivocally sec 2.1;
bear out the predictions of the fluctuation hypothesis. In section 4, we propose Revision
a research program that aims at establishing L1 influence in article acquisition pending.
for learners with an articleless L1. We approach articles as a syntax/semantics
interface phenomenon. The setup of the program allows it to be adapted to study
L1 influence for other phenomena at the syntax/semantics interface.
11
Bert Le Bruyn
The example from Snape et al. (2006) shows us a realistic picture of cross-
linguistic syntax/semantics. Too often, two simplifying assumptions are made:
(i) things that superficially look the same are the same (e.g. numerals, demon-
stratives), (ii) languages either make the same distinctions or are underspecified
(definiteness) without there being a (combined) role for other expressions. These
simplifications are a limitation in cross-linguistic syntax/semantics. The first
challenge a systematic study of L1 influence in article acquisition faces is thus
to force a paradigm shift in cross-linguistic syntax/semantics that gives transfer
research the groundwork it needs (the comparative challenge).
The example from Snape et al. (2006) is also indicative of another challenge
the field faces. Transfer research at the SSI is too often synonymous with L2
morpheme studies. This is a reductionist view in two respects. The first is that
the SSI is not a mere sum of morphemes but a system in which all morphemes
interact. The second is that the SSI of L2 learners can only be properly understood
if we model it as a system in which the SSIs of the learner’s L1 and TL come
together. We need methodology that allows us to do justice to the full complexity
of the SSI of L2 learners (the L2 interface challenge). Meeting this challenge allows
us to compare the SSIs of L2 learners from the same L1 background and across
not in learner groups (intragroup homogeneity, intergroup heterogeneity) while at the
italics in same time comparing them to the L1s and TL of the learners (cross-linguistic
manuscript congruity).
5.1.1 Data
ITM uses translation corpora to generate networks of translation equivalents
across languages.7 E.g., one takes a and the as seed words, looks up their uses in
the English source texts and matches their translations. These can be demonstra-
tives, specific word orders, case configurations, etc. As a second step, one looks
up all uses of the translations of a and the in the source and target texts and
matches the translations of these in all the languages of the corpus. The first step
creates one-way contrastive analyses focusing on how English nominal definite-
ness is rendered in the other languages. The second step creates a many-to-many
contrastive analysis that gives access to the paradigms of nominal definiteness
cross-linguistically with an equal weight for the different languages.
The output of the data collection is a set of contexts with – for every lan-
guage – an indication of the markers of definiteness. Multi-Dimensional Scal-
ing (MDS) automatically generates clusters of contexts by maximizing the dis-
tances between contexts in which (individual) languages use different markers
and minimizing the distances between contexts in which the same markers are
used (Hamming distance). Based on Analyses of Similarities (Clarke 1993; Oksa-
nen et al. 2017), we determine the significance of these clusters. The combination
of the clusters and the contexts that appear in them is an inductively construed
semantic map (Haspelmath 1997), the basis for our cross-linguistic analyses. It
furthermore allows us to shift the focus of transfer research from morphemes to
the full SSI.
ITM introduces iterations in the Translation Mining technique (TM) we de-
signed with Henriëtte de Swart and Martijn van der Klis (van der Klis et al. 2017).
5.1.2 Analysis
The way the analysis proceeds is close to the one in TM (e.g. de Swart et al.
2017). We illustrate with an example in which we apply TM and ITM to the same
(hypothetical) dataset. We restrict our attention to two languages (English and
Mandarin) and to a subset of the variation we expect to find.
The points in Figures 5 and 6 represent contexts from a translation corpus.
Their colours refer to the forms in English (upper), the coloured groupings to
the forms in Mandarin (lower). The clusters that emerge by crossing the form
variation in the two languages are numbered.
7
A reviewer correctly points out that the parallel methodology severely restricts the number
of languages that can be investigated. We hope this is however only a matter of time in the
sense that parallel corpora will hopefully become available for many more languages.
13
Bert Le Bruyn
Figure 3: Caption.
5.2 LOG-IT
LOG-IT (‘Logging Lexicons and OT Grammars in Translation’) is a data mining
technique that uses a custom-made high quality L1 to L2 translation corpus to
inductively study the SSI of individual learners at the same level of detail as the
output of ITM. It thus overcomes the L2 interface challenge. We use the output
14
Definiteness across languages and in L2 acquisition
of ITM in two ways. The clusters identified through ITM guide our selection of
contexts for the translation corpus. For the analysis, we use the ITM feature-
based lexicons and OT rules to generate all possible variations on the languages
involved. We compare these to the production of the learners and establish in-
dividual rankings of these variations. We establish similar rankings for the lan-
guages of the project based on our corpus and experimental data. The rankings
allow us to map and compare the SSI of individual learners, L1 groups and L1/L2s.
5.2.1 Data
We choose written L1 to L2 text translation as a data collection protocol for two
reasons:
(1) We need data that discriminate between possible rankings. (Semi)-free pro-
duction tasks cannot target all relevant data per learner.
(2) Unlike other high control tasks like Forced Choice Elicitation, translation can
focus on any level of production (DP/VP, sentence, discourse).
Relying on translation data comes with two risks. The first is a translation
bias: learners might be influenced by specific wordings in the source text or re-
sort to general translation processes like simplification. To address this bias, we
include two control tasks: a story rewrite task to control for influence from the
source text and L2 to L1 translation to control for translation styles. The second
risk is overinterpretation of the data: doubts of the learners are not visible in a
translation and learners might resort to a word-by-word or sentence-by-sentence
strategy while we hope to analyze all levels of production. To address this risk,
we exploit the potential of simultaneous key-stroke logging and eye-tracking
during translation. We use a combination of measures related to corrections,
eye-key spans (Timarová et al. 2011 and references therein), attention units (e.g.
Hvelplund 2016), etc. to establish a measure of reliability per data point. The
relevant experimental software goes under the name of TRANSLOG II and was
developed in the field of Translation Studies (Schwieter & Ferreira 2017).
5.2.2 Analysis
We use the semantic features and OT grammar constraints identified through
ITM to generate all possible lexical entries for the forms used by the learners and
all possible OT grammars. By crossing lexicons and grammars, we generate all
15
Bert Le Bruyn
possible variations on the languages involved and rank these per learner. Rank-
ings are based on how accurately the variations predict learner production and
corpus/experimental data. Accuracy is established as a measure of (weighted) in-
terrater reliability where the output of the learner and the variation are modeled
as raters.
We calculate the distances between learner/language rankings based on the
Damerau-Levenshtein distance and establish a dissimilarity matrix. This is the in-
put for Analyses of Similarities that statistically assess intra-group homogeneity/inter-
group heterogeneity for the L1 groups. We use MDS to graphically represent
similarities and differences between individual learners, learner groups and lan-
guages (Figure 4). In combination with the underlying rankings, the correspond-
ing graph is an inductively constructed map of L1 influence. The underlying data
allow us to establish cross-linguistic congruity.
Figure 4: Caption
6 Conclusion
We have presented evidence suggesting that articleless languages are not created
equal and that this influences how native speakers of these languages acquire ar-
ticle languages like English. The evidence suggested that Mandarin learners of
16
Definiteness across languages and in L2 acquisition
English do not unequivocally bear out the predictions of the Fluctuation Hypoth-
esis, unlike learners of English with e.g. Korean, Russian and Japanese as an L1.
We have proposed a research program that approaches articles as a syntax/semantics
interface phenomenon. The program considers the syntax/semantics interface of
definiteness in its entirety and makes no a priori assumptions about how it is best
analysed. Rather, it adopts a data-driven comparative approach with multiple
L1s that allows us to give a fine-grained answer to the question how L1 influence
plays out for definiteness.
Abbreviations
Acknowledgements
References
Bickerton, Derek. 1981. Roots of language. Ann Arbor: Karoma.
Clarke, K. Robert. 1993. Non-parametric multivariate analyses of changes in com-
munity structure. Austral Ecology 18(1). 117–143.
De Bot, Kees, Wander Lowie & Marjolijn Verspoor. 2007. A dynamic systems
theory approach to second language acquisition. Bilingualism: Language and
Cognition 10(1). 7–21.
de Swart, Henriëtte, Bert Le Bruyn & Martijn van der Klis. 2017. Lexical, compo-
sitional and dynamic semantic ingredients of the perfect. Manuscript.
Dulay, Heidi C. & Marina K. Burt. 1974. Natural sequences in child second lan-
guage acquisition. Language Learning 24(1). 37–53.
Hakuta, Kenji. 1976. A case study of a Japanese child learning English as a second
language. Language Learning 26(2). 321–351.
Haspelmath, Martin. 1997. Indefinite pronouns. Oxford: Clarendon Press.
Hawkins, Roger, Saleh Al-Eid, Ibrahim Almahboob, Panos Athanasopoulos, Rangsiya
Chaengchenkit, James Hu, Mohammad Rezai, Carol Jaensch, Yunju Jeon, Amy
Jiang, Yan-kit Ingrid Leung, Keiko Matsunaga, Martha Ortega, Ghisseh Sarko,
Neal Snape & Kalinka Velasco-Zárate. 2006. Accounting for English article in-
terpretation by L2 speakers. EUROSLA Yearbook 6(1). 7–25.
Hendriks, Petra, Helen De Hoop, Irene Krämer, Henriëtte de Swart & Joost Zwarts.
2010. Conflicts in interpretation. Sheffield: Equinox.
Huebner, Thom. 1983. A longitudinal analysis of the acquisition of English. Ann
Arbor: Karoma.
17
Bert Le Bruyn
18
Definiteness across languages and in L2 acquisition
Reid, J., P. Battaglia, M. Schuldt, E. Narita, M. Mochizuki & Neal Snape. 2006. The
article choice of learners of English as a second language. Unpublished term
paper. Colchester: University of Essex.
Schwarz, Florian. 2009. Two types of definites in natural language. Doctoral dis-
sertation. Amherst: University of Massachusetts.
Schwieter, John W. & Aline Ferreira (eds.). 2017. The handbook of translation and
cognition. Hoboken: John Wiley & Sons.
Snape, Neal, Yan-kit Ingrid Leung & Hui-Chuan Ting. 2006. Comparing Chi-
nese, Japanese and Spanish speakers in L2 English article acquisition: Evi-
dence against the fluctuation hypothesis? In Mary Grantham O’Brien, Chris-
tine Shea & John Archibald (eds.), Proceedings of the 8th Generative Approaches
to Second Language Acquisition Conference (GASLA 2006): The Banff Conference.
Sommerville, MA: Cascadilla Proceedings Project.
Tarone, Elaine & Betsy Parrish. 1988. Task-related variation in interlanguage: The
case of articles. Language Learning 38(1). 21–44.
Thomas, Margaret. 1989. The acquisition of English articles by first-and second-
language learners. Applied Psycholinguistics 10(3). 335–355.
Timarová, Sárka, Barbara Dragsted & Inge Gorm Hansen. 2011. Time lag in trans-
lation and interpreting: A methodological exploration. In Cecilia Alvstad, Adelina
Hild & Elisabet Tiselius (eds.), Methods and strategies in process research: Inte-
grative approaches in translation studies, 121–146. Amsterdam: John Benjamins.
Ting, Hui-Chuan. 2005. The acquisition of articles in L2 English by L1 Chinese and
L1 Spanish speakers. MA dissertation. Colchester: University of Essex.
Trenkic, Danijela. 2008. The representation of English articles in second language
grammars: Determiners or adjectives? Bilingualism: Language and Cognition
11(1). 1–18.
van der Klis, Martijn, Bert Le Bruyn & Henriëtte de Swart. 2017. Mapping the
PERFECT via Translation Mining. In Proceedings of the 15th Conference of the
European Chapter of the Association for Computational Linguistics, vol. 2, 497–
502.
19