From Interlanguage grammar to target grammar in L2 processing of definiteness as uniqueness

Abstract

This study investigated the processing of English articles by second language (L2) speakers whose first language (L1) is Korean. Previous studies in L2 English article use had some issues unresolved such as using offline tasks, conflating definiteness with real-world knowledge, and operationalizing definiteness and relevant constructs in ways that participants can be primed or get metalinguistic cues. To revisit such issues, the construct ‘definiteness’ was operationalized as unique identifiability, a self-paced reading task was used to collect data, and regression models were employed to analyse logarithm residuals of raw reading time data, which can detect subtle differences that are otherwise buried. The results show that L1 speakers show sensitivity to the use of definite and indefinite articles in response to given contexts and that both advanced and intermediate L2 speakers first resort to their non-target-like Interlanguage grammar, but the advanced group later revises their initial interpretation and eventually shows the effect of target grammar. The L2 behavior is discussed in terms of its theoretical implications.

Keywords

L2 definiteness processing L2 English article use presupposition accommodation self-paced reading

I Introduction

As subtle morphological cues as they may be, English articles are an important element that establishes common ground between interlocutors. They cue a listener either to locate an existing, pre-mentioned referent within the discourse, or to bring in a new referent into the discourse (Heim, 1982). English articles can also indicate that either there is a uniquely identifiable referent agreed upon between participants of a conversation or there are multiple, non-unique referents expected in the given context (Hawkins, 1978; Russell, 1905). This important grammatical element, however, is rarely mastered by those who speak English as a second language (L2) when their first language (L1) does not have an equivalent system (Butler, 2002; Ionin, 2006; Ionin et al., 2012, 2004, 2008, 2009; Jarvis, 2002; Kim and Lakshmanan, 2009; Ko et al., 2010; Master, 1997; Robertson, 2000; Snape et al., 2006; Trenkic, 2007, 2008; Trenkic and Pongpairoj, 2013). Even at a highly advanced level, L2 speakers struggle with the correct use of the functional morphemes (see Trenkic et al., 2014).

The construct ‘definiteness’, the distinguishing feature of definite and indefinite articles, is not invariably defined even among researchers of L1 semantics (Donnellan, 1966; Hawkins, 1978, 1991; Heim, 1982; Lyons, 1999; Russell, 1905), and a great many number of L2 researchers have looked into the issue from various perspectives (Butler, 2002; Ionin et al., 2004; Jarvis, 2002; Kim and Lakshmanan, 2009; Master, 1997; Robertson, 2000; Snape et al., 2006; Trenkic et al., 2014). Studies on L2 English article use varied not only in the relevant linguistic constructs but also in the way the constructs are operationalized and in the experimental methods through which they were tested. The variety of studies resulted in sometimes common but other times vastly different, if not contradictory, outcomes. Because their research foci, methods, and outcomes all presented different stories about L2 speakers’ English article use, this study is yet another attempt to probe what leads to their non-target-like use of articles in L2 English. However, it will not be feasible nor practical to test in one study all of the factors previously investigated that were known to influence L2 English article use.

Among many questions that have been asked to investigate the topic, the current study focuses on one question: Do L2 speakers react to the appropriate and inappropriate use of definite and indefinite NPs the same way L1 speakers do in real-time language processing? To explore this topic, unique identifiability (Hawkins, 1991; Russell, 1905) was chosen as the defining construct of definiteness for methodological reasons. When definiteness is defined in terms of givenness or familiarity (where a referent is introduced as an indefinite NP and is re-mentioned by a definite NP later) (Heim, 1982), definite and indefinite NPs cannot be in psycholinguistically equal conditions within the context of an experiment. That is, if the experimental task is off-line, repeated definite NPs and first-mention indefinite NPs will provide participants with metalinguistic cues. Even if an on-line experimental method is employed, the definite condition will always involve a priming effect (due to repetition) whereas the indefinite condition will not. This problem will be prevented by operationalizing definiteness in terms of unique identifiability, which allows a first-mention NP to be either definite or indefinite.

With the main research focus specified, the current study uses a self-paced reading task, which taps into the online processing of English articles by participants. A real-time processing study is not anything new in L2 research; especially, self-paced reading is one of the more common methods these days. Analysis methods employed in this study, however, are much more advanced than the conventional analyses of raw reading time (RT) data and have not been commonly used in L2 studies. The fine-tuned data analyses reveal an intriguing developmental pattern in definiteness processing of advanced and intermediate L2 speakers, which was not observed in L1 controls. Both proficiency groups of L2 speakers first read definite NPs faster regardless of the (non-) uniqueness of the referents, which suggests that L2 English speakers resort to non-target-like Interlanguage grammar; however, advanced L2 speakers also show the effect of target grammar, if delayed. I interpret the results as evidence that L2 processing can be explained by multiple constrains competing over limited attentional resources.

In the background section, previous L2 studies on English article use are reviewed as well as relevant semantic constructs and definitions. The review is followed by processing studies on L1 English article use. The methods section describes participants, materials, experimental procedures involved, and statistical analyses. The results section reports the analysis results by native language and proficiency, followed by the discussion that explores the implications of the patterns observed.

II Background

1 L2 English article use

The acquisition and use of English articles have long intrigued many linguists and applied linguists alike. Some studied how they are acquired and used by L1 children (Marastos, 1976; Wexler, 2011) and adults (Clifton, 2013; Murphy, 1984; Schumacher, 2009) when others focused on how they are acquired and used by L2 children (Zdorenko and Paradis, 2008, 2011) and adults (Butler, 2002; Huebner, 1983; Ionin et al., 2004). Just as many studies as there have been, their research foci have also been of great diversity, and various methods of investigation have been employed from observations to fill-in-blanks to eye-tracking and electroencephalography (EEG). More diverse than the methods of data collection were the results of the studies and their implications, especially in L2 studies. Some showed that L2 speakers have comparable knowledge with regard to the English article system to that of L1 speakers (Trenkic et al., 2014) while others argued that L2 speakers’ use of English articles is not target-like but diverged in terms of what such discrepancy results from (Ionin, 2006; Ionin et al., 2012, 2004, 2008; Jarvis, 2002; Ko et al., 2010; Martin et al., 2013; Trenkic and Pongpairoj, 2013).

Among the studies of various foci are a set of studies that investigated L2 English article use based on the noun classification of definiteness and specificity. Although the terminology and notions were used not invariably among various researchers, the constructs have been the bases of many L2 English article studies. The beginning was Bickerton’s (1981) classification of incidents where English articles (or article-like determiners) are used in Hawaii Pidgin English (HPE) and Hawaii Creole English (HCE). He argued that HPE speakers’ use of English articles was not based on any apparent rules whereas that by HCE speakers showed clear patterns; da (‘the’ in HCE) was used for definite NPs, and wan (‘one’ or ‘a’ in HCE) was used for nondefinite (but specific) NPs while no article was used for nonspecific NPs. This distinction of definiteness and specificity was adopted by many subsequent studies that used similar categorizations of NPs. Huebner’ (1983) case study on a Hmong speaker classified NP types into four different types using the binary constructs of Hearer’s Knowledge (±HK) and Specific Reference (±SR) and reported that the Hmong speaker overused the in [–HK, +SR] NPs, where in English grammar a is required. This overuse of the in a specific but indefinite NP has triggered many discussions from diverse perspectives (Butler, 2002; Ionin, 2006; Ionin et al., 2004; Ko et al., 2010; Master, 1997).

One of the most prolific researchers using the constructs is Ionin and her colleagues (Ionin, 2006; Ionin et al., 2012, 2004, 2008, 2009). Unlike previous studies that dealt with comprehensive noun classes such as singular and plural nouns and count and non-count nouns, their studies mainly focused on singular count nouns (see Ionin and Montrul, 2010). Also, they used the notions of definiteness and specificity instead of hearer’s knowledge and specific reference, in order to account for the L2 phenomenon within the framework of semantic universals.

Ionin et al. (2004) proposed the Fluctuation Hypothesis, reporting that L1-Korean and L1-Russian speakers of L2 English fluctuated between definiteness and specificity. Their participants did not show any difficulty using definite or indefinite articles when an NP is either [+definite, +specific] or [–definite, –specific]; however, their accuracy in deciding to use definite or indefinite articles decreased when an NP is either [+definite, –specific] or [–definite, +specific]. They argued that the accuracy of L2 English article use is unstable in such cases because L2 speakers’ access to semantic universals fluctuates between the two different parameters definiteness and specificity. Their theoretically solid analysis of L2 English article use, however, left a few important issues to be addressed.

The first was how the construct ‘specificity’ was defined and operationalized in their experiments. In their definition, a [+specific] NP is considered by a speaker to possess some noteworthy property. This definition resulted in [+specific] items including sentences like ‘I know who she is’ or ‘I know him very well’ and all [–specific] items including ‘I don’t know who she is’ or ‘I don’t know who it is.’ This visible contrast between specific and non-specific conditions could have misled participants to use it as a metalinguistic cue.

Another issue is related to whether there is a natural language that has an article system that uses specificity as a distinguishing feature between two articles (so as to argue that specificity is part of the language universals). To show that specificity is a semantic universal, they originally argued that Samoan has two different articles le [+specific] and se [–specific] distinguished by specificity. However, Ionin et al. (2009) reported that the Samoan article system is not solely based on specificity¹ and their argument lost its ground that L2 speakers fluctuate between definiteness and specificity because of their fluctuating access to semantic universals.

The other issue was a methodological one: They collected data using ‘forced-choice elicitation’, which was a paper-and-pencil type test where participants were asked to fill in blanks with either the or a(n) or ∅. This off-line task could not have been conclusive evidence that reflects the participants’ real-time language use. This methodology, along with the explicit and rather misleading operationalization of specificity mentioned above, could have caused the fluctuating accuracy in [+definite, –specific] and [–definite, +specific] conditions.

Although Ionin et al. (2004) left some questions to be addressed, it led to several subsequent studies of importance that delved into other factors such as L1 transfer, L2 input, and explicit strategies (Ionin et al., 2008, 2009) as well as the role of presuppositionality of existence (Ko et al., 2010). Especially, the theoretical grounds of Ko et al.’s (2010) work is worth noting. They argued that presuppositionality of existence affects L2 English article use. Their data were from L2-English speakers whose L1 was Korean, which, again, doesn’t have an article system. In addition, L1 children are also affected by presuppositionality of existence. Based on findings from their own research and from previous studies, they argued that the effect found in their study came from both groups’ fluctuating access to semantic universals, which, this time, was expanded to include presuppositionality.

According to Ko et al. (2010), semantic universals consist not only of definiteness and specificity, but also of presuppositionality of existence, and L2 speakers’ access to the three factors fluctuates, which causes the non-target-like behavior that has been well documented. Ko et al. (2010) define definiteness and indefiniteness as (1) and (2) (p. 218), respectively, which reflects the Fregean view of definiteness based on Heim (1991). Also, they define presuppositionality as (3) (Ko et al., 2010: 220), based on Wexler (2003), which refers back to Heim (1991).

(1) Definiteness: A sentence of the form [def α] ζ presupposes that that there exists at least one individual which is α and that there exists at most one individual which is α, and it asserts that the unique individual which is α is also ζ.

(2) Indefiniteness: A sentence of the form [indef α] ζ asserts that there is at least one individual which is α and ζ.

(3) Presuppositionality: A sentence of the form [pres α] ζ presupposes that there exists at least one individual which is α and asserts that there exists at least one individual which is both α and ζ.

The definitions above show that definiteness is defined by the presuppositions of uniqueness and existence whereas indefiniteness asserts only the existence of an entity. The distinguishing factor between definite and indefinite NPs should be the presupposition of uniqueness. Then, what is Ko et al.’s ground that L2 learners and L1 children mistake presuppositionality (of existence) for the distinguishing factor between definite and indefinite NPs? They present as evidence a striking similarity between the contexts in which definite NPs are used and those in which wh-phrases can denote only those presupposed to exist (Pesetsky, 1987; 2000, as cited in Ko et al., 2010).

(4) Definiteness established through mutual world knowledge

The moon is very bright tonight.

Honey, can you get the car while I’m locking the door?

Can you pass me the salt, please?

(5) wh-phrases denoting referents presupposed to exist

Which month will be most convenient for you?

(Among the new students) Which student did the professor talk to?

(Pointing to a bookshelf) Which book would you like to read?

Each of the definite NPs in (4) presuppose that there is a uniquely identifiable referent, and the presupposition is accommodated through knowledge shared by the members of the world (4a), a smaller community (e.g. family) (4b), or a specific spatial and temporal context (4c). Sentences in (5) sound natural only when there is a set of referents known to exist to both the speaker and the listener. This indicates that the set of referents is established through interlocutors’ mutually shared world knowledge, similarly to the case of (4). The world knowledge could be something that is shared by most people (5a), something they share as members of the same community (5b), or something established through shared temporal and spatial contexts (5c).

Despite the striking similarity between the environments definite NPs occur and those where wh-phrases of presupposed existence occur, the underlying semantic principles for the two grammatical forms are distinct; the former presupposes both existence and unique identifiability whereas the latter presupposes only existence. Ko et al. (2010) postulated that the resemblance of environments in which the two occur leads L2 speakers to mistake presupposition of existence for definiteness and that, because of this inaccurate concept of definiteness, L2 speakers overuse the in the place of an indefinite article if it denotes a [+presupposed (to exist)] referent.

Ko et al. (2010) operationalized presuppositionality of existence via overt partitivity, arguing that ‘phrases such as one of the books and two of the books presuppose existence without presupposing uniqueness’ (p. 221). Examples (6) and (7) are borrowed examples from their study (p. 236).² Both are conversations between Elissa and Robert, where they talk about how Robert’s nephew got a new pet. In the beginning of (6), a set of pets are mentioned as candidates for the nephew and one of them is selected ([+partitive]) while, in (7), the pet that finally gets selected has nothing to do with the context given in the beginning of the conversation ([–partitive]).

(6) [–definite, +partitive] extensional contexts, partitivity established explicitly:

Elissa: How is your nephew Aaron doing? He is such a nice little boy!

Robert: He has some good news; his parents finally allowed him to get a pet—just one! So last week, he went to our local pet shop. This pet shop had five puppies and seven kittens, and Aaron loved all of them. But he could get only one!

Elissa: Oh, so what did he do?

Robert: Well, it was difficult for him to make up his mind. But finally, he got (a, the, –) puppy. Aaron went home really happy!

(7) [–definite, –partitive] extensional contexts:

Elissa: How is your nephew Joey doing? He is such a nice boy!

Robert: Well, he was a bit depressed the last few days. So, his parents decided to get him a pet. So last week, he went to our local pet shop.

Elissa: Oh, so did he buy some animal there?

Robert: No, he did not like the puppies in the pet shop, in fact. But then he was walking home, and he found (a, the, –) kitten in the street! So now he has a new pet after all!

Both conditions sampled above require an indefinite article; however, if the definiteness distinction in L2 grammar is mistaken for presupposition of existence [±presupposed], L2 speakers will erroneously use definite articles more in [+partitive] conditions than in [–partitive] conditions, as proxies for [+presupposed] and [–presupposed], respectively. Ko et al. (2010) argue that this hypothesis was verified through a forced-choice elicitation task.

However, their study still had the same issue as their earlier work did: The forced-choice elicitation task is a paper-and-pencil type fill-in-the-blank task. All participants saw the same items, and they were allowed to go back and forth between items, which makes it difficult to hide the patterns of the items. This nature of paper-and-pencil tasks cannot be viewed lightly because Koreans are the most trained in test-taking skills. The washback effect of standardized English tests in Korea has been well known, and it was also reported that exam scores often do not do justice to test-takers’ actual English proficiency in Korea (Choi, 2008). The suspicion is supported by the following facts.

First of all, the items were created to have a clear distinction between conditions: In [+partitive] conditions, a set of referents was mentioned either to explicitly state or to imply that the set includes the referent denoted by the critical NP while no such contexts were given in [–partitive] conditions. Secondly, each item was composed of a rather long conversation (around seven to nine sentences) in a total of 80 critical items with no fillers.³ Participants would have felt tired to focus and could have decided to use test-taking strategies. Finally, participants were asked to choose among three options of a definite article, an indefinite article, and no article in all of the 80 trials. This all makes it very likely that they would have developed metalinguistic strategies to find answers. Therefore, the pattern observed in Ko et al. (2010) cannot be conclusive evidence that presuppositionality of existence accounts for the L2 behavior. These methodological issues, however, should not be deemed grounds to dismiss their systematic and theoretical endeavors altogether. Examining their argument via a psycholinguistically sound online method will provide more precise insight into this population’s English article use.

There have been recent attempts to investigate L2 English article use via more psycholinguistically advanced methods. Trenkic et al. (2014) used visual world paradigm eye tracking to observe Chinese speakers’ English article use. Although their methods were sound enough, the results of the study should be reconsidered in that they replicated Chambers et al. (2002), which was designed to probe participants’ use of real-world knowledge, not grammatical knowledge (i.e. English articles). That is, they manipulated the contexts for definite and indefinite NPs not by the number of referents but by the suitability of referents. Since the authors of the original study intended to emphasize the role of situation-specific non-linguistic information in processing linguistic information, one should be careful in interpreting the findings of Trenkic et al. (2014) as indicators of target-like grammar processing by L2 speakers. Also, we made a few observations that led us to suspect that L2 speakers in the study might be interpreting a as a singular marker ‘one’.

First of all, L2 participants were slower when they had to choose ‘a referent’ in the two-referent condition than when having to identify ‘the referent’ in the one-referent condition, which is quite the opposite from L1 counterparts. Secondly, the proportion of looks to solo-standing distractors was higher in the indefinite condition. Lastly, the interaction between uniqueness and definiteness showed quite different patterns in L1 and L2 speakers. L1 participants showed a larger fixation time difference between definite and indefinite NPs in the two-referent condition than in the one-referent condition. On the other hand, in L2 participants, the fixation time difference between definite and indefinite NPs was much larger in the one-referent condition than in the two-referent condition. This difference between the two groups suggests that the L1 and L2 speaker groups might be resorting to different grammars resulting in a seemingly similar but, in reality, different interaction of definiteness and uniqueness.

The studies discussed so far are not the exhaustive list of studies on L2 English article use but they are representative of some of the most systematic and theoretically oriented investigations of the topic. The investigations, however, vary not only in their research methods but also in results and implications. The primary cause of the different results might not be due only to the extent to which the methods of investigation are rigorous from psycholinguistic perspectives. English articles are rarely salient morphemes and their processing mechanisms are hard to depict in invariable terms even in L1 population. Below we go over a few online studies that looked into the processing of definiteness by L1 speakers of English that are most relevant to the current study.

2 L1 research in definiteness processing

As was in L2 research on English articles, L1 research in definiteness processing also had various foci, which resulted in various outcomes and findings. Some reported that definite NPs were processed faster than indefinite NPs (Murphy, 1984) while others argued that discourse contexts play an important role in processing definite and indefinite NPs (Burkhardt, 2006, 2008; Clifton, 2013; Schumacher, 2009).

The history of L1 definiteness processing research dates back to Murphy (1984), a sentence-level reading time study that compared definite and indefinite NPs. Examples in (8) are from Murphy (1984: 489). In the contextual sentence (8a), a referent was introduced (a truck), and retrieving the same referent from the mental model of a discourse was reported to be easier and faster (the truck) than introducing a new referent (a truck) in the critical sentence (8b). As the definite and indefinite NPs represent a previously introduced referent and a newly introduced referent, respectively, the reading time difference was explained via a mental discourse model, in which readers experience different levels of ease/difficulty in comprehending sentences with old and new referents.

(8) a. Though driving 55, Steve was passed by a truck.

b. Later, George was passed by the/a truck, too.

More recently, researchers aimed to probe the processing of definiteness from different perspectives. Clifton (2013) showed that the (in)definiteness of an NP per se did not lead to different reading times when the presupposition of either a unique vs. non-unique referents was ‘accommodated’ by the readers. Accommodation means updating a prior belief one thought was shared on the common ground between speaker and listener (Stalnaker, 2002; Von Fintel, 2008). Sentences in (9) and (10) exemplify this (Clifton, 2013: 490). In (9), the sentence is contextualized in the kitchen. A typical kitchen has only one stove, and the definiteness of the NP the stove (9a) can be accommodated more easily than the indefiniteness of the NP a stove (9b). On the other hand, in the context where no unique referent is presupposed, a definite NP cannot have the kind of reading time advantage observed in Murphy (1984). That is, in the appliance store context (10), the indefinite condition (10b) was observed to be read faster than the definite condition (10a). In this way, the beginning part of the sentence allows readers to update the situational context of the discourse, which leads to the presupposition of unique or non-unique referents.

(9) a. In the kitchen, Jason checked out the stove very carefully.

b. In the kitchen, Jason checked out a stove very carefully.

(10) a. In the appliance store, Jason checked out the stove very carefully.

b. In the appliance store, Jason checked out a stove very carefully.

Schumacher’s (2009) focus was on probing the effects of discourse contexts in comparing the processing of definite and indefinite NPs in German in an Event-Related brain Potential (ERP) study. As in (11),⁴ the three contextual sentences created three different conditions: given, inferred, and new (11a–11c). The critical sentence (11d) had either a definite or an indefinite NP.

(11) a. Peter has recently visited a speaker in Munich. (given)

b. Peter has recently visited a lecture in Munich. (inferred)

c. Peter has recently visited Hannah in Munich. (new)

d. He said that the/a speaker had been very nice.

The three contexts (11a–11c) and two articles (11d) created six different conditions, and ERP evidence showed the following. First, the discourse contexts were reflected in N400, a negative deflection of a brain’s electrical activity peaking around 400 ms past the onset of the critical noun. Second, the (in)definiteness of the critical NP had a significant main effect on Late Anterior Negativity (LAN), which appears between 400 ms and 700 ms mostly on anterior sites of the brain. Lastly, a pronounced Late Positivity (i.e. P600), a positive deflection that peaks around 500 ms after the onset of the critical noun, was observed around 550 ms and 700 ms when a new discourse entity was introduced either by an indefinite NP or by contextual cues; that is, all conditions but the given-definite condition showed significantly higher Late Positivity peaks.

The most relevant finding for the current discussion is the similarity observed between Murphy (1984) and Schumacher (2009) that given-indefinite NPs are more costly to process than given-definite NPs. The effect was observed via Late Positivity in Schumacher (2009). The definite-indefinite pair for the given condition (11a) is most relevant for Murphy’s (1984) comparison between definite and indefinite conditions (8b). The difference between given-definite and given-indefinite conditions in Schumacher was significant with the definite condition showing lower Late Positivity than its indefinite counterpart. Although it will be difficult to directly translate pronounced and deflected electrical activities in brain into longer and shorter reading/reaction times, the two studies can be viewed to dovetail with each other, given that both pronounced electrical activities in brain and longer reading times reflect higher processing demands.

A direct comparison between Clifton’s (2013) and Schumacher’s (2009) studies is not as simple. The inferred condition (11b) in Schumacher (2009), the only condition relatable to Clifton (2013), has only a unique referent condition whereas Clifton’s kitchen and appliance store conditions represent unique vs. non-unique referent conditions, respectively. In addition, Schumacher reported no significant difference in the pairwise comparison of definite and indefinite NPs in the context of (11b) whereas Clifton (2013) does not report the results of the pairwise comparison between (9a) and (9b).

The survey of the three studies above suggests that definite and indefinite NPs might differ in terms of processing load if the former refers back to an existing referent and the latter introduces a new referent into the discourse. However, in terms of definite and indefinite NPs presupposing unique and non-unique referents, respectively, Clifton (2013) and Schumacher (2009) cannot be compared in equal terms. Clifton (2013) argues that the processing costs of definite and indefinite NPs should not differ as long as their presuppositions are accommodated; however, Schumacher’s (2009) ERP research argues for the different brain activities only when given NPs are repeated as definite. In sum, the studies introduced above diverge in their evaluations of the processing loads of definite and indefinite NPs because their research foci and methods differed from one another and because none of the three studies shared a consistent way of operationalizing definiteness.

3 Research questions and predictions

With various studies focusing on diverse research questions in both L1 and L2 definiteness processing, the study aimed to conduct an experiment that can probe definiteness defined in terms of unique identifiability. Despite many possible ways of investigating L2 English article processing, definiteness as unique identifiability was chosen for the following reasons. First, it does not involve the repetition of NPs in the definite condition, which is the case when definiteness is defined in terms of givenness. It is difficult to completely tease apart the effect of definiteness from that of priming when definite NPs are always repeated and indefinite NPs are always not. Also, the study focuses only on one factor because psycholinguistic experiments require a large number of data points for each condition. Including too many factors in one experiment might not lead to clear results, or it might require too many trials in one session, which is not ideal because participants’ fatigue can create noises in data.

The main research questions are 1) whether the presupposition of unique vs. non-unique referents by first-mention definite vs. indefinite NPs, respectively, is accommodated in given contexts in both L1 and L2 sentence processing, 2) whether such accommodation pattern will differ between L1 and L2 speakers, and 3) whether the accommodation pattern will display any relationship with the extent to which L2 proficiency is developed.

Not only L2 participants but also L1 participants are observed because it is necessary to examine whether the accommodation of presupposition is an observable phenomenon in native controls. Such accommodation will be manifested as an interaction of uniqueness and definiteness. That is, definite NPs should be read faster when unique referents are typically expected as in (9) while indefinite NPs will be read faster in contexts where non-unique referents are expected (10). Therefore, if participants accommodate the (in)definiteness of first-mention NPs based on the (non-)uniqueness of referents, definiteness and uniqueness should show an interaction in statistical analyses. Simply put, (9a) will be read faster than (9b) and (10b) faster than (10a).

III Method

1 Participants

a Participants

A total of 115 participants participated in the experiment. Fifty-four L1 English speakers participated as a native control group, and 61 L1-Korean L2-English speakers were recruited from Honolulu, Hawaii in the USA and Seoul, Korea.⁵ Course credit or monetary compensation was provided. Table 1 summarizes the demographics of the participants. The L2 participants were divided into advanced and intermediate groups by their C-test scores. The native language of L2 participants was limited to Korean, a language that does not have an article system. This was to ensure that learners will not resort to the article system in their first language to process English articles.

Table 1.

Participant demographics.

Group	n	LOR in years (s.d.)	C-test (s.d.); 40 maximum	Age in years (s.d.)
L1	54 (f = 35)	Not applicable	35.7 (2.3)	25.7 (7.3)
L2 advanced	28 (f = 21)	m = 5.2 (4.9)	35.2 (1.7)	25.5 (4.8)
L2 intermediate	33 (f = 25)	m = 2.7 (3.9)	25.2 (2.6)	26.2 (6.2)

Note. LOR = Length of residence.

b Proficiency

A C-test, borrowed from Schultz (2006), was used to measure L2 speakers’ proficiency but was also administered to L1 speakers for a comparison purpose. The test was made up of two paragraphs, each of which had 20 blanks. The blanks were placed on the latter half of every other content words. Both L1 and L2 speakers were given 15 minutes maximum to work on the C-test; the statistics of each group’s C-test scores are summarized along with other demographic information in Table 1.

As Table 1 shows, the proficiency test scores of the L1 group and the advanced L2 group were comparable whereas the intermediate L2 group showed a lower mean score compared to the former two groups. A one-way ANOVA test and a Tukey post-hoc analysis were conducted to see if the differences were meaningful. The difference between the L1 group and the advanced L2 group was not significant (p = .23) while the differences between the L2 intermediate group and the other two groups were (ps < .001).

2 Materials and procedure

a Self-paced reading

A non-cumulative (moving-window style) self-paced reading task was used to measure participants’ sensitivity to the relationship between the definiteness of an NP and the uniqueness of its referent.

b Stimuli

Most of the critical items were borrowed from Clifton’s (2013) study. Because some of Clifton’s items included cultural knowledge that L2 participants might not be familiar with, a few items were replaced altogether, and others were modified. For example, Staples may not be recognized as the name of a stationery store by L1-Korean L2 learners of English whose length of residence in the USA was not long enough. Such items were excluded, and to increase the number of observations, new items were created in the same pattern as Clifton’s for a total of 20 experimental items. Sentences as in (12) and (13) were shown region by region, each of which ranged from one word to a few words, which, in turn, resulted in 8 to 10 regions per sentence. Each item was followed by a comprehension question as in (14) to ensure that participants were paying their full attention. The complete list of critical items is found in Appendix 1 with region boundaries marked with an underscore (_).

(12) Unique

a. In_the_kitchen,_Jason_checked out_the stove_very carefully_

to make sure_it was safe.

b. In_the_kitchen,_Jason_checked out_a stove_very carefully_

to make sure_it was safe.

(13) Non-unique

a. In_the_appliance store,_Jason_checked out_the stove_very carefully_

to make sure_it was safe.

b. In_the_appliance store,_Jason_checked out_a stove_very carefully_

to make sure_it was safe.

Each item occurred in unique (12) and non-unique (13) contexts and was counterbalanced with definite and indefinite articles. Four lists were created for a Latin-square design. That is, each list included 20 items in total, with five items in each of the four conditions. No item was repeated in different conditions within the same list. Items were divided into regions such that critical NPs (bold face) always appeared in Region 6. Each participant’s reading time at each region was measured by the experimental software E-prime 2.0 (Psychology Software Tools, 2014).

(14) What was Jason checking out?

Fifty-two filler items were included as in (15). Their sentence structures were similar to those of the critical items, but the critical region did not include the (in)definite article contrast.

(15) On the bottom of the page, Jimmy wrote his name and number to join the volunteer group.

c Procedure

When participants arrived in the lab, they were briefed on the procedure and seated in a sound-attenuated booth. The self-paced reading task was given first and the c-test for English proficiency was given afterwards. The experiment included a recorded instruction and sample trials before the main experiment.

Clifton (2013) reported that his experiments resulted in significant results only when a simple arithmetic question was inserted between a sentence and its comprehension question. He explained that articles are such subtle morphemes, of which the effects of accommodation can be detected only when reading speed is slow enough and that reading times were much slower in experiments with intervening arithmetic tasks than in those without. To ensure that participants in this experiment will also slow down to read the sentences more carefully, the current study followed the same experimental procedure and included a simple math question (16) between each sentence and its corresponding comprehension question.

(16) 34 + 3 = ?

3 Analysis

a Log transformation and residual reading time analysis

Table 2 summarizes both raw reading time (RT) data and logarithm-transformed RT data. Because raw RT data were heavily skewed, the raw RTs were transformed into logarithms. The values of skewness and kurtosis show that the normality of the data was improved after logarithm transformation. The log RTs differed significantly in all pairs of groups (ps < .05). Due to the differences, further analyses will be carried out separately for L1, advanced L2, and intermediate L2 groups.⁶ When the mean values differ significantly among groups, calculating all data in one model could result in wiping off any meaningful effects that could appear in individual groups. Also, native speakers’ behavioral pattern needed to be examined independently. Therefore, the results reported below are from models fitted separately for each group.

Table 2.

Reading time data before and after logarithm transformation.

Group	Raw RT (ms)			Log RT
Group	Mean (s.d.)	Skewness	Kurtosis	Mean (s.d.)	Skewness	Kurtosis
L1	604 (315)	1.86	7.03	6.30(0.44)	0.61	3.10
L2 advanced	689 (376)	1.81	6.63	6.43(0.46)	0.62	2.93
L2 intermediate	783 (427)	1.53	5.16	6.54(0.48)	0.49	2.61

For the trimming of RT data and the residual calculation, the procedure from Hofmeister (2011) was followed. Any data points shorter than 200 ms or longer than 2,500 ms were eliminated from analysis, and those that were over or under 3 s.d. from the mean were also excluded from analysis.⁷ The data lost from trimming was 2.2%.

The log-transformed data were then put into a linear mixed effects regression model (lmer) that factored into item types (critical vs. filler items), the length of the region, the order of the trial, the position of the region in a sentence, and the random effect of individuals (17) (Hofmeister, 2011). Residuals of each data point from the regression line were calculated and used for all subsequent analyses.

(17) $\begin{array}{l} Y_{s i} = β_{00} + γ_{0 s} + β_{1} * i t e m T y p e + β_{2} * r e g i o n L e n g t h + β_{3} * l o g (t r i a l O r d e r) + \\ β_{4} * r c s (r e g i o n P o s i t i o n) + e_{s i} \end{array}$

b Linear mixed effects regression (lmer)

All data analyses were performed using the R package lmerTest. In all data analyses, a full model was first considered with the maximal structure of random intercepts and slopes for both items and participants following Barr et al. (2013). Only when the maximal model did not converge, random slope elements were eliminated step by step until a model converged. To minimize the possibility that any effects obtained are due to RT differences of the previous regions, the lmer models also included the log RT values of up to three regions before the analysed region, of which the calculation was also carried out following Hofmeister (2011). The equation (18) shows the model with the maximal structure.

(18) $\begin{array}{l} Y_{s i} = β_{00} + γ_{0 s} + δ_{0 i} + (β_{1} + γ_{1 s} + δ_{1 i}) * X + (β_{2} + γ_{2 s} + δ_{2 i}) * W + \\ (β_{3} + γ_{3 s} + δ_{3 i}) * X W + e_{s i} + ω_{s i 1} + ω_{s i 2} + ω_{s i 3} \end{array}$

s = participant index; i = item index; β = estimates; γ = participant effect; δ = item effect;

X = uniqueness, W = definiteness; e = error; $ω$ = previous region effect

IV Results

Analysis results are reported from the four regions surrounding and including the critical region, where the critical NP is shown: The four regions, henceforth, will be referred to as the pre-critical (checked out), critical (the/a stove), spillover (very carefully), and post-spillover regions (to make sure). As was predicted, L1 speakers showed an interaction of uniqueness and definiteness in the spillover region, the region after the critical region, as was in Clifton (2013). Such interaction also appeared in advanced L2 speakers, albeit in the post-spillover region, which is one region later compared to their L1 counterparts. Both advanced and intermediate L2 speakers also showed a main effect of definiteness in the critical region. The delayed interaction effect observed in advanced L2 speakers, however, was not observed in intermediate L2 speakers.

1 L1 participants

The lmer model introduced above (18) converged with the maximal structure of random effects in all four regions analysed (Figure 1). However, the only region that showed any effect in the L1 group was the spillover region. An interaction effect of uniqueness and definiteness was found (ß = − 0.022, S.E. = 009, t-value = − 2.296, p = .024), as shown in Tables 3a and 3b. In the other three regions, neither any main effects nor any interaction effects were observed.

Figure 1.

Residuals⁸ of log reading time (RT) in critical and surrounding regions: First language (L1).

Table 3a.

Fixed effects of definiteness and uniqueness in the spillover region: L1 group.

	Estimate	Std. error	t value	Pr(>\|t\|)
(Intercept)	−0.051	0.221	−0.231	0.817
Definiteness	0.000	0.012	0.009	0.993
Uniqueness	−0.010	0.011	−0.909	0.371
definiteness:uniqueness	−0.022	0.009	−2.276	0.025*
previous region 1⁹	0.032	0.025	1.299	0.194
previous region 2	−0.032	0.025	−1.280	0.200
previous region 3	0.008	0.025	0.328	0.742

Table 3b.

Random effects of participant and item in the spillover region: L1 group.

Groups	Names	Variance	Std. dev
Participant	(Intercept)	0.004	0.062
	definiteness	0.000	0.018
	uniqueness	0.002	0.040
	definiteness:uniqueness	0.000	0.005
Item	(Intercept)	0.006	0.076
	definiteness	0.001	0.035
	uniqueness	0.000	0.010
	definiteness:uniqueness	0.000	0.010
Residual		0.086	0.293

Notes. Number of observations: 1,029. Number of participants: 54. Number of items: 20. Marginal R²: 0.008. Conditional R²: 0.138.

Only in the spillover region was the interaction effect of uniqueness and definiteness observed. With the significant interaction observed, a separate model was fit for pairwise comparisons. An lmer model was fit with condition (4 levels = 2 definiteness × 2 uniqueness) as a factor, to which a multiple comparison procedure was applied. No pairs of conditions showed a significant difference in log RT residuals (ps < 1). Despite the lack of significant differences in pairwise comparisons, one should pay attention to different amounts of RT differences that definite and indefinite NPs incur in unique vs. non-unique contexts. While the difference in L1 participants’ reaction to definite NPs in the two contexts is relatively larger, their reaction to indefinite NPs in the two contexts changes only very subtly (see Figure 2). This relative lack of reading time change in the indefinite condition is even clearer in the advanced L2 participants.

Figure 2.

Interaction of definiteness and uniqueness (context) in spillover: First language (L1).

2 Advanced L2 participants

As with L1 data, data from the advanced L2 group were analysed in all four regions surrounding and including the critical region. Lmer models with the maximal structure of random effects were fit in all regions except the post-spillover region, where the maximal model failed to converge, perhaps due to the relatively smaller sample size (n = 28). Unlike the L1 group that showed an interaction effect between uniqueness and definiteness only in the spillover region, the advanced group showed a main effect of definiteness in the critical region (ß = − 0.039, S.E. = 0.018, t-value = − 2.200, p = .037) and an interaction effect of uniqueness and definiteness in the post-spillover region (ß = − 0.030, S.E. = 0.014, t-value = − 2.122, p = .034). There were no effects observed in the pre-critical and the spillover regions.

As seen in Figure 3, definite NPs (the stove) was read faster than indefinite NPs (a stove) in the critical region, regardless of contexts. A pairwise comparison, however, did not show any significant difference between any pair of conditions.

Figure 3.

Residuals of log reading time (RT) in critical and surrounding regions: Second language (L2) advanced.

In the post-spillover region, the lmer model with the maximal random effects structure did not converge, and model complexity was reduced iteratively following Bates et al. (2015). The report on advanced L2 speakers is based on the model below (19).

(19) $\begin{array}{l} Y_{s i} = β_{00} + γ_{0 s} + δ_{0 i} + (β_{1} + γ_{1 s}) * X + β_{2} * W + β_{3} * X W \\ + e_{s i} + ω_{s i 1} + ω_{s i 2} + ω_{s i 3} \end{array}$

s = participant index; i = item index; β = estimates; γ = participant effect; δ = item effect;

X = uniqueness, W = definiteness; e = error; $ω$ = previous region effect

They showed an interaction of definiteness and uniqueness in the post-spillover region (Tables 4a and 4b), which was slower than L1 speakers, who showed the interaction effect in the spillover region. As Figure 4 shows, log RT residuals are shorter when the presupposed uniqueness of referents and the definiteness of referring NPs match than when they mismatch. A pairwise comparison showed that there was a marginal difference between the unique definite condition and the non-unique definite condition (ß = 0.126, S.E. = 0.050, z-value = 2.533, p = .068). As was the case in L1 participants, advanced L2 participants also showed different degrees of surprise for definite and indefinite NPs in different contexts. Figure 4 shows that RT clearly changed between unique vs. non-unique contexts when the NP was definite but the change was barely existent when the NP was indefinite.

Table 4a.

Fixed effects of definiteness and uniqueness in the post-spillover region: L2 advanced.

	Estimate	Std. error	t-value	Pr(>\|t\|)
(Intercept)	−0.080	0.339	−0.236	0.813
Definiteness	−0.005	0.014	−0.382	0.703
Uniqueness	−0.033	0.017	−1.961	0.061.
definiteness:uniqueness	−0.029	0.014	−2.122	0.034*
previous region 1⁹	−0.012	0.032	−0.364	0.716
previous region 2	0.000	0.031	−0.016	0.988
previous region 3	0.023	0.032	0.710	0.478

Table 4b.

Random effects of participant and item in the post-spillover region: L2 advanced.

Groups	Names	Variance	Std. dev
Participant	(Intercept)	0.003	0.054
	uniqueness	0.003	0.051
Item	(Intercept)	0.012	0.108
Residual		0.096	0.310

Notes. Number of Observations: 535. Number of participants: 28. Number of items: 20. Marginal R²: 0.018. Conditional R²: 0.167.

Figure 4.

Interaction of definiteness and uniqueness in post-spillover: Second language (L2) advanced.

3 Intermediate L2 participants

Lmer models with the same maximal structure of fixed and random effects (18) were fit for all four regions from the pre-critical to the post-spillover regions in the intermediate L2 group (Figure 5). However, only in the critical region was observed a significant main effect of definiteness (ß = − 0.043, S.E. = 0.014, t-value = − 3.096, p = .003). Furthermore, a pairwise comparison showed that a definite NP in the non-unique referent condition was read the fastest resulting in a significant difference from the indefinite non-unique condition (ß = 0.122, S.E. = 0.040, z-value = − 3.035, p = .014) and a marginal difference from the indefinite unique condition (ß = − 0.104, S.E. = 0.043, z-value = − 2.407, p = .080). The intermediate L2 group, however, did not show an interaction of uniqueness and definiteness in any of the regions.

Figure 5.

Residuals of log reading time (RT) in critical and surrounding regions: Second language (L2) intermediate.

V Discussion

The current study aimed to investigate whether L2 speakers showed sensitivity to grammar in the same way as L1 speakers with regard to the relationship between the unique identifiability of a referent and the definiteness of an NP. To probe the issue, a self-paced reading task was used to test whether L1 and L2 speakers accommodate the presupposition of unique and non-unique referents in first-mention definite and indefinite NPs, respectively. The prediction was that RTs will increase when the (in)definiteness of an NP cannot be accommodated based on the presupposition of (non-)uniqueness of a referent, which will be manifested via an interaction effect of uniqueness and definiteness at or after the critical NP region. The three groups tested in the study all showed different patterns. L1 speakers showed the predicted interaction between uniqueness and definiteness in the spillover region whereas both advanced and intermediate L2 speakers showed a main effect of definiteness in the critical region, which was not predicted. However, advanced L2 speakers also showed a delayed interaction between uniqueness and definiteness.

As was predicted, L1 participants did have grammatical sensitivity to the relationship between uniqueness and definiteness. They showed an interaction effect of uniqueness and definiteness in the spillover region, which is the region right after the critical NP is presented. However, according to a pairwise comparison analysis, no pairs of conditions showed any significant differences. This shows that the abnormality that native participants feel is very subtle when the presupposition of an NP cannot be easily accommodated in a given context. Also, one should pay attention to the amount of reading time differences that definite and indefinite articles incur when their presupposition of (non-)uniqueness cannot be easily accommodated. Figure 2 shows that the amounts of reading time change between contexts differed in definite and indefinite conditions.

The pair of conditions that showed the largest difference in estimates, if not statistically significant, was that of the unique/definite (12a) and the non-unique/definite (13a) conditions. The unique/definite condition elicited the shortest reading time of all conditions, which dovetails with the results of previous studies. Reading times (Clifton, 2013; Murphy, 1984), brain activities (Schumacher, 2009), and fixations on target referents (Trenkic et al., 2014) all point to the fastest (or easiest) processing of definite NPs in contexts where unique referents are typical. On the other hand, the non-unique/definite condition leads to a longer and more difficult processing, which is also supported by findings from earlier studies (Burkhardt, 2006; Clifton, 2013; Schumacher, 2009; Trenkic et al., 2014). Unlike definite NPs, indefinite NPs do not lead to a large difference in terms of processing difficulty based on whether the context presupposes a unique referent (12b) or a non-unique referent (13b). As is seen in Figure 2, the vertical distance between the unique/indefinite and the non-unique/indefinite conditions is not as great as that between the unique/definite and the non-unique/definite conditions. This discrepancy between the two pairs of conditions can be accounted for by looking at different types of presuppositions made by the sentences in (12) and (13) and how they are accommodated.

Let’s first consider the presuppositions that are more easily accommodated. The definite NP in (12a) the stove presupposes the existence of a uniquely identifiable stove, which is easily accommodated in the given context the kitchen. Likewise, the indefinite NP in (13b) a stove presupposes the existence of stoves, which need not be uniquely identifiable, which is also easily accommodated in the given context the appliance store.

Next are the cases in which the presupposition of (non-)uniqueness cannot be as easily accommodated. When the definite NP the stove is given in a context where it is difficult to identify a unique referent (the appliance store) as in (13a), it goes beyond a matter of accommodation. The reader will experience a difficult time understanding why a uniquely identifiable referent is presupposed, wondering what they have missed. On the other hand, when the indefinite NP a stove is used in the kitchen context (12b), one can accommodate the presupposition of multiple, non-unique stoves relatively easily, accepting the fact that a kitchen can have more than one stove, if not typical. As was mentioned earlier, accommodation is about updating one’s prior belief one thought was shared on the common ground between speaker and listener (Stalnaker, 2002; Von Fintel, 2008). This discrepancy indicates that not all violations of expectations are equally surprising. The difference between definite and indefinite NPs as to how they interact with unique vs. non-unique referents is observed in advanced L2 participants as well, which will be discussed shortly after the main effect of definiteness at the critical region is elaborated on.

A main effect of definiteness was observed at the critical region in both advanced and intermediate L2 data, and the effect is worth discussing in light of Ko et al.’s (2010) argument that L1-Korean L2-English speakers confuse definiteness with presuppositionality of existence. In the current experiment, referents are presupposed to exist in both the unique referent context (kitchen) and the non-unique referent context (appliance store), regardless of their unique identifiability. According to Ko et al. (2010), presuppositionality of existence can be one of the distinguishing features of definiteness in L2 Interlanguage grammar. Then, the faster reading time observed in the definite conditions regardless of unique identifiability might be argued to align with what can be predicted based on the presuppositionality of existence account.

However, there is one caveat in interpreting the main effect of definiteness as a proxy for that of presuppositionality of existence. Since the experiment was not intended to test the role of presuppositionality of existence, the conditions did not include non-existent contexts. That is, both the kitchen and the appliance store were contexts where the existence of a stove or stoves is typically expected; therefore, the analysis is only half complete without also testing such contexts where the same referents were not typically expected to exist. Hence, the current results cannot be concluded to support Ko et al. (2010).

The main effect of definiteness was observed at the critical region in both proficiency groups of L2 participants; however, the effect seems stronger and clearer in the intermediate L2 group. Not only were the statistic values of the main effect greater in the intermediate group (ß = − 0.043, S.E. = 0.014, t-value = − 3.096, p = .003) than in the advanced group (ß = − 0.039, S.E. = 0.018, t-value = − 2.200, p = .037), but also pairwise comparisons between conditions also showed that the intermediate group was more strongly affected by the definiteness effect than the advanced group. The advanced L2 group did not show any significant differences between conditions in any pairs (ps < 1) despite the significant main effect while the L2 intermediate group showed a significant difference between definite/non-unique and indefinite/non-unique conditions (ß = 0.122, S.E. = 0.040, z-value = − 3.035, p = .014) and a marginal difference between definite/non-unique and indefinite/unique conditions (ß = − 0.104, S.E. = 0.043, z-value = − 2.407, p = .080). The difference between the two groups might indicate the extents to which the two groups are resorting to the non-target grammar; the intermediate group is being affected more strongly by the Interlanguage grammar than the advanced group is.

The story does not end here; the advanced L2 group also showed a delayed interaction effect of uniqueness and definiteness. Unlike the L1 group, who showed the interaction in the spillover region, the advanced L2 group showed the interaction effect in the post-spillover region. When an experimental manipulation is made in a critical region, it is typical that the effect of the manipulation spills over and appears in the next region, which was the case in the L1 group. The plausible explanation for the delayed effect in the advanced L2 group will be the main effect of definiteness shown in the critical region. It looks like they are affected by their Interlanguage grammar (the main effect of definiteness) first, and they later recall that definite and indefinite NPs presuppose unique and non-unique referents, respectively. From the current data analysis, one can cautiously speculate that the spillover region is where the competition occurs between their regressive Interlanguage grammar and target grammar.

In sum, the data seem to show a developmental pattern in English article use in L1-Korean L2-English speakers. They first resort to their Interlanguage grammar, which leads them to process definite NPs faster regardless of context, but as their proficiency level goes up, they seem to be able to also use target grammar, which manifests via the interaction of definiteness and uniqueness at a later region.

VI Conclusions

The results of the study confirmed the previous findings that unique identifiability does play a role in L1 definiteness processing, and they also revealed a developmental pattern in L2 English article use by speakers whose L1 does not have an equivalent to the English article system. The role of unique identifiability in L1 definiteness processing may not be anything novel, nor is the effect of non-target-like Interlanguage grammar in L2 English article research. However, the results reported in the present study qualify findings from previous studies.

Data from the L1 group in this study implies that the accommodation of presupposition may not be a matter of dichotomy but that of gradience. When the presuppositions made did not align with L1 speakers’ expectation, the amount of RT increase differed by what was presupposed. The degree of surprise was much greater when they had to accommodate the presupposition of a uniquely identifiable referent in a context where multiple non-unique referents were typical than when a uniquely identifiable referent was referred to as indefinite.

Data from the L2 group in this study suggests that it is possible that many L2 studies report no effects in real-time sentence processing not because L2 processing is merely slower than L1 processing but because L2 speakers’ Interlanguage grammar is qualitatively different from the target grammar. The developmental pattern observed in the current study seems compatible with two theoretical accounts of language acquisition that are on the opposite ends of the horizon.

Ionin and colleagues (Ionin, 2006; Ionin et al., 2012, 2004; Ko et al., 2010) have proposed the Fluctuation Hypothesis (FH) that L2 speakers’ access to semantic universals fluctuate between different parameters (i.e. definiteness, specificity, presuppositionality of existence). Under the speculative assumption that the main effect of definiteness in the current study could be viewed as the effect of presuppositionality of existence, the L2 data did show faster reaction to definite NPs than to indefinites in all contexts where referents were typically expected to exist. Along with the facts that children are also susceptible to presuppositionality of existence (Wexler, 2011), and that there is a striking similarity between environments where definite NPs occur and those where wh-phrases presupposed to exist occur, nativists might argue that the pattern observed at the critical region should be real-time evidence for the FH, and more importantly, innate knowledge.

On the other hand, researchers who view L2 learning as an outcome of multiple constraints’ competing against each other over limited attentional resources (Ellis, 2006; MacWhinney, 2018; O’Grady, 2015) need not stipulate universal features across languages.¹⁰ Unlike the nativists’ account, the observed behavior in L2 adults can be explained without resorting to innate knowledge. As was seen in the definitions of definiteness and indefiniteness in (1) and (2) in the beginning of this article, existence was part of the conditions in both definiteness and indefiniteness. When definite NPs are given in input to L2 learners, the referents denoted by them will always be existent, which might mislead L2 learners to conflate definiteness with presuppositionality of existence. The condition that the unique identifiability of the referent must also be established between interlocutors requires a more advanced level of cognitive process, which makes it difficult for L2 speakers to have a full grasp on the construct initially. Wexler (2011) argues that the egotistic account cannot be applicable for adults because it is unlikely that adults will have a hard time taking others’ perspectives into account. However, when the rule is not noticed, they are not being egotistic, they are simply being unaware of the rule.

One will have to remain agnostic about whether the main effect of definiteness observed in the critical region was in fact presuppositionality of existence at work in L2 interlanguage grammar and even more so about whether presuppositionality of existence is a semantic universal. However, regardless of the way in which one interprets the main effect in the critical region, one thing is clear. L2 participants first used their (non-target-like) Interlanguage grammar to make sense of what they were reading, and advanced participants, who also had knowledge of the target grammar, revised their initial interpretation; hence, the delayed effect. Even if the main effect of definiteness were evidence for L2 speakers’ access to semantic universals, that would not undermine the argument that L2 processing can be accounted for by the competition of multiple grammars. After all, there have been recent proposals that the two approaches to (second) language acquisition, traditionally considered incompatible, complement each other in describing developmental paths (MacWhinney, 2017; Shirai and Juffs, 2016).

This is a very exciting finding in that one could not only observe a delayed effect but also had the opportunity to take a peek at why the effect was delayed. Many studies reported that L2 sentence processing is shallower, slower, and not predictive (Clahsen and Felser, 2006; Grüter et al., 2017; Marinis et al., 2005; Martin et al., 2013), but they should rethink such claims that are made based on the absence of evidence, which is not always the evidence of absence. Dekydtspotter et al. (2006) also criticized studies that argued for qualitative differences between L1 and L2 sentence processing based on observations of no effects in L2 data (Clahsen and Felser, 2006; Marinis et al., 2005). The data in the current study show that the target grammar is there in advanced L2 English albeit with the Interlanguage grammar interfering on the way.

The findings reported here should benefit from further research that investigates whether the main effect of definiteness observed is truly the disguise of presuppositionality of existence in the Interlanguage grammar of L1-Korean L2-English speakers or it was due to another mechanism that could not be detected in this article. For such an investigation, one should first conduct an experiment that includes non-existent conditions unlike the current study that had only those where referents were presupposed to exist. If the faster RT in the definite condition was observed in non-existent conditions as well, the effect observed cannot be translated as presuppositionality of existence. Also, one should explore L2 speakers with other L1s than Korean that do not have an article system equivalent to that of English. If L2 speakers of different L1s do not show the same pattern of using presuppositionality of existence as their incorrect grammar for distinguishing definite and indefinite articles, those who argue for fluctuating access to semantic universals will have to answer why L2 speakers of some L1s fluctuate between different parameters and those of others do not.

Footnotes

Appendix 1 Acknowledgements

I would like to thank my dissertation supervisor, Dr William O’Grady at the University of Hawai‘i at Mānoa for his support and guidance while I was working on this project (as part of my dissertation). I also feel greatly indebted to the Department of English Language and Literature at Seoul National University for their help with my data collection in Seoul. I also greatly appreciate an anonymous reviewer’s thorough and careful review of the earlier draft of this article. All errors are mine.

Declaration of Conflicting Interest

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work is part of the author’s dissertation work from the University of Hawai’i at Mānoa, which was supported by a Language Learning dissertation grant and a field research scholarship fund from the East-West Center in Honolulu, HI.

ORCID iD

Hyunah Ahn

Notes

References

Barr

Levy

Scheepers

et al. (2013) Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language 68: 255–78.

Bates

Kliegl

Vasishth

et al. (2015) Parsimonious mixed models. arXiv: 1506.04967. Available at: https://arxiv.org/abs/1506.04967 (accessed July 2019).

Bickerton

(1981) Roots of language. Ann Arbor, MI: Karoma.

Burkhardt

(2006) Inferential bridging relations reveal distinct neural mechanisms: Evidence from event-related brain potentials. Brain Lang 98: 159–68.

Burkhardt

(2008) Two types of definites: Evidence for presupposition cost. In: Grønn

(ed.) Proceedings of Sinn und Bedeutung 12. Oslo: Department of Literature, Area Studies and European Languages, University of Oslo, pp. 66–80.

Butler

(2002) Second language learners’ theories on the use of English articles. Studies in Second Language Acquisition 24: 451–80.

Chambers

Tanenhaus

Eberhard

et al. (2002) Circumscribing referential domains during real-time language comprehension. Journal of Memory and Language 47: 30–49.

Choi

I-C

(2008) The impact of EFL testing on EFL education in Korea. Language Testing 25: 39–62.

Clahsen

Felser

(2006) Grammatical processing in language learners. Applied Psycholinguistics 27: 3–42.

10.

Clifton

Jr. (2013) Situational context affects definiteness preferences: Accommodation of presuppositions. Journal of Experimental Psychology: Learning, Memory and Cognition 39: 487–501.

11.

Dekydtspotter

Schwartz

Sprouse

(2006) The comparative fallacy in L2 processing research. In: O’Brien

Shea

Archibald

(eds) The 8th Generative Approaches to Second Language Acquisition. Sommerville, MA: Cascadilla Proceedings Project, pp. 33–40.

12.

Donnellan

(1966) Reference and definite descriptions. The Philosophical Review 75: 281–304.

13.

Ellis

(2006) Language acquisition as rational contingency learning. Applied Linguistics 27: 1–24.

14.

Grüter

Rohde

Schafer

(2017) Coreference and discourse coherence in L2: The roles of grammatical aspect and referential form. Linguistic Approaches to Bilingualism 7: 199–219.

15.

Hawkins

(1978) Definiteness and indefiniteness: A study in reference and grammaticality prediction. London: Croom Helm.

16.

Hawkins

(1991) On (in)definite articles: Implicatures and (un)grammaticality prediction. Journal of Linguistics 27: 405–42.

17.

Heim

(1982) The semantics of definite and indefinite noun phrases. Department of linguistics. Amherst, MA: University of Massachusetts.

18.

Heim

(1991) Artikel und Definitheit. In: Stechow

Wunderlich

(eds) Semantik: Ein internationales Handbuch der zeitgenössischen Forschung. Berlin: De Gruyter, pp. 487–535.

19.

Hofmeister

(2011) Representational complexity and memory retrieval in language comprehension. Language and Cognitive Processes 26: 376–405.

20.

Huebner

(1983) A longitudinal analysis of the acquisition of English. Ann Arbor, MI: Karoma.

21.

Ionin

(2006) This is definitely specific: Specificity and definiteness in article systems. Natural Language Semantics 14: 175–234.

22.

Ionin

Montrul

(2010) The role of L1 transfer in the interpretation of articles with definite plurals in L2 English. Language Learning 60: 877–925.

23.

Ionin

Wexler

(2004) Article semantics in L2 acquisition: The role of specificity. Language Acquisition 12: 3–69.

24.

Ionin

Zubizarreta

Maldonado

(2008) Sources of linguistic knowledge in the second language acquisition of English articles. Lingua 118: 554–76.

25.

Ionin

Zubizarreta

Philippov

(2009) Acquisition of article semantics by child and adult L2-English learners. Bilingualism: Language and Cognition 12: 337–61.

26.

Ionin

Baek

Kim

et al. (2012) That’s not so different from the: Definite and demonstrative descriptions in second language acquisition. Second Language Research 28: 69–101.

27.

Jarvis

(2002) Topic continuity in L2 English article use. Studies in Second Language Acquisition 24: 387–418.

28.

Kim

Lakshmanan

(2009) The processing role of the article choice parameter: Evidence from L2 learners of English. In: Garcia Mayo

MdP

Hawkins

(eds) Second language acquisition of articles: Empirical findings and theoretical implications. Amsterdam: John Benjamins, pp. 87–113.

29.

Ionin

Wexler

(2010) The role of presuppositionality in the second language acquisition of English articles. Linguistic Inquiry 41: 213–54.

30.

Lyons

(1999) Definiteness. Cambridge: Cambridge University Press.

31.

MacWhinney

(2017) Are these approaches incompatible? Linguistic Approaches to Bilingualism 7: 730–33.

32.

MacWhinney

(2018) A unified model of first and second language learning. In: Hickmann

Veneziano

Jisa

(eds) Sources of variation in first language acquisition: Languages, contexts, and learners. Amsterdam: John Benjamins, pp. 287–312.

33.

Marastos

(1976) The use of definite and indefinite reference in young children. Cambridge: Cambridge University Press.

34.

Marinis

Roberts

Felser

et al. (2005) Gaps in second language sentence processing. Studies in Second Language Acquisition 27: 53–78.

35.

Marsden

Thompson

Plonsky

(2018) A methodological synthesis of self-paced reading in second language research. Applied Psycholinguistics 39: 861–904.

36.

Martin

Thierry

Kuipers

J-R

et al. (2013) Bilinguals reading in their second language do not predict upcoming words as native readers do. Journal of Memory and Language 69: 574–88.

37.

Master

(1997) The English article system: acquisition, function and pedagogy. System 25: 215–32.

38.

Murphy

(1984) Establishing and accessing referents in discourse. Memory and Cognition 12: 489–97.

39.

O’Grady

(2015) Processing determinism. Language Learning 65: 6–32.

40.

Pesetsky

(1987) Wh-in-situ: Movements and unselective binding. In: Reuland

ter Meulen

(eds) The representation of (in)definites. Cambridge, MA: MIT Press, pp. 98–129.

41.

Pesetsky

(2000) Phrasal movement and its kin. Cambridge, MA: MIT Press.

42.

Psychology Software Tools (2014) E-Prime 2.0 [software]. Pittsburgh, PA: Psychology Software Tools. Available at: http://www.pstnet.com (accessed July 2019).

43.

Roberts

Felser

(2011) Plausibility and recovery from garden paths in second language sentence processing. Applied Psycholinguistics 32: 299–331.

44.

Robertson

(2000) Variability in the use of the English article system by Chinese learners of English. Second Language Research 16: 135–72.

45.

Russell

(1905) On denoting. Mind 14: 479–93.

46.

Schultz

(2006) Wh-scope marking in English interlanguage grammars: Transfer and processing effects on the second language acquisition of complex wh-questions. Honolulu, HI: Department of Second Language Studies, University of Hawaii at Manoa.

47.

Schumacher

(2009) Definiteness marking shows late effects during discourse processing: Evidence from ERPs. In: Devi

Branco

Mitkov

(eds) Anaphora processing and applications: Lecture notes in artificial intelligence. Heidelberg: Springer, pp. 91–106.

48.

Shirai

Juffs

(2016) Introduction: Convergence and divergence in functional and formal approaches to SLA. Second Language Research 33: 3–12.

49.

Snape

Leung

Y-kI

Ting

H-C

(2006) Comparing Chinese, Japanese, and Spanish speakers in L2 English article acquisition: Evidence against the Fluctuation Hypothesis? In: O’Brien

Shea

Archibald

(eds) Proceedings of the 8th Generative Approaches to Second Language Acquisition Conference. Somerville, MA: Cascadilla Proceedings Project, pp. 132–39.

50.

Stalnaker

(2002) Common ground. Linguistics and Philosophy 25: 701–21.

51.

Trenkic

(2007) Variability in second language article production: Beyond the representational deficit vs. processing constraints debate. Second Language Research 23: 289–327.

52.

Trenkic

(2008) The representation of English articles in second language grammars: Determiners or adjectives? Bilingualism: Language and Cognition 11: 1–18.

53.

Trenkic

Pongpairoj

(2013) Referent salience affects second language article use. Bilingualism: Language and Cognition 16: 152–66.

54.

Trenkic

Mirkovic

Altmann

GTM

(2014) Real-time grammar processing by native and non-native speakers: Constructions unique to the second language. Bilingualism: Language and Cognition 17: 237–57.

55.

Von Fintel

(2008) What is presupposition accommodation, again? Philosophical Perspectives 22: 137–70.

56.

Wexler

(2003) Maximal trouble: The 16th Annual CUNY Conference on Human Sentence Processing. Cambridge, MA: MIT.

57.

Wexler

(2011) Cues don’t explain learning. In: Gibson

Pearlmutter

(eds) The processing and acquisition of reference. Cambridge, MA: The MIT press.

58.

Zdorenko

Paradis

(2008) The acquisition of articles in child second language English: fluctuation, transfer or both? Second Language Research 24: 227–50.

59.

Zdorenko

Paradis

(2011) Articles in child L2 English: When L1 and L2 acquisition meet at the interface. First Language 32: 38–62.