Native and second language processing of quantifier scope ambiguity

Abstract

The present study investigates native (L1) and second language (L2) processing of scope ambiguities in English sentences containing the universal quantifier every in subject NP and negation. Previous studies in L1 and L2 processing of scope ambiguities have found speakers to generally employ a ‘minimal effort’ principle that highly prefers the surface scope reading regardless of contextual support because accessing the inverse scope reading incurs significant processing cost. The present study compared L1 and L2 scope judgments and processing strategies of sentences such as Every horse didn’t jump over the fence and examined whether the two groups differ in their speed and manner of analysis. Thirty native English speakers and 42 Korean learners of English participated in a self-paced reading/interpretation task that has context (ambiguous vs. unambiguous) and scope reading (surface vs. inverse) as variables. The results revealed significant differences in scope endorsement rates with L2 learners arriving at the surface scope as the dominant reading and L1 learners’ judgments being highly dependent on contextual ambiguity. Moreover, L1 vs. L2 differences in processing strategies were found: L2 learners exhibited a strong tendency to arrive at the most economical interpretation while L1 speakers consulted detailed syntactic and semantic rules of computation.

Keywords

good-enough (GE) processing L1 processing L2 processing quantifier scope ambiguity scope interpretation

I Introduction

Research on second language (L2) sentence processing has primarily centered on the nature of native (L1) vs. non-native differences, which has remained controversial. Some report that while L2 learners can reliably integrate lexical, semantic, and pragmatic information (Hoshino et al., 2010; Juffs and Harrington, 1996; Roberts and Felser, 2011), they often exhibit difficulty integrating morphosyntactic information and show characteristic differences from native speakers in the processing of various linguistic phenomena such as relative clauses, filler-gap dependencies, inflected morphology, and temporary ambiguity resolution (Clahsen et al., 2015; Felser and Roberts, 2007; Jiang, 2004; Kirkici and Clahsen, 2013; Marinis et al., 2005).

Substantial differences between L1 and L2 processing are assumed in the Shallow Structure Hypothesis (SSH) (Clahsen and Felser, 2006), which holds that L2 learners rely heavily on lexical cues and less on structurally based parsing strategies leading to syntactic representations that are incomplete and less detailed than those of L1 speakers. Other studies, however, have challenged the SSH and have claimed that L1 and L2 sentence processing are qualitatively similar but differ with respect to peripheral factors such as memory retrievals and limited processing resources (Cunnings, 2017; Hopp, 2006, 2010; McDonald, 2006). These studies conclude that although L2 processing is fundamentally identical to native processing, it is more susceptible to interference and individual differences in proficiency (Hopp, 2006; Jackson, 2008; Van Hel and Tokowicz, 2010), crosslinguistic influence (Sabourin and Stowe, 2008; Tolentino and Tokowicz, 2011), or the secondary task (Jackson and Bobb, 2009; Leeser et al., 2011; Lim and Christianson, 2013b).

In fact, recent studies (Lee and Shin, 2016; Lim and Christianson, 2013a, 2013b) have tried to explain L2 sentence processing in terms of the Good Enough (GE) processing approach (Christianson and Luke, 2011; Christianson et al., 2010; Ferreira and Patson, 2007) that has been proposed for language comprehension in L1 processing. The GE processing approach suggests that comprehenders compute representations that are just ‘good enough’ for understanding the meaning using the heuristic parser rather than build a complete detailed structure using the syntactic parser. GE processing has since been supported by numerous studies in the L1 processing literature that examined active/passive structures (Christianson et al., 2010; Ferreira, 2003) and garden-path sentences (Christianson et al., 2006).

Lim and Christianson (2013a, 2013b) draw parallels between GE and the ‘shallow’ representations in L2 processing and point out that the mechanisms of GE processing are similar to those of the SSH differing only in the degree of accessibility to the syntactic parsing route. While the SSH proposes that L2 learners cannot build accurate syntactic structures because the syntactic route is impaired, the GE account assumes that the full parsing route is intact and available but may not necessarily be integrated in the ultimate representation when the output from the heuristic route is ‘good enough’. It is predicted that the syntactic operations are relatively more ‘fragile’ in L2 processing and are thus more susceptible to semantic/pragmatics-based heuristics or individual factors such as proficiency level. They tested whether L2 processing can be characterized as GE processing by examining Korean learners’ online comprehension of plausible or implausible English active/passive sentences (2013a) and subject/object relative clauses (2013b) using self-paced reading tasks and/or translation tasks. The results showed that L2 learners have control over morphosyntactic information like the native speakers and that L2 processing can be viewed as GE processing that is qualitatively similar to native processing. Similarly, a number of recent studies that have examined L2 comprehension using the GE approach have found their results for L1 and L2 processing to be largely compatible with the predictions of GE processing (Fujita and Cunnings, 2020; Lee and Shin, 2016; Tan and Foltz, 2020). Although L2 speakers exhibited greater difficulty recovering from the initial misinterpretation and displayed lower accuracy scores for comprehension questions than L1 speakers overall (Jacob and Felser, 2016; Pozzan and Trueswell, 2016), they did not lack detailed syntactic representations of ambiguous sentences and were able to reanalyse them successfully with increasing proficiency (Hopp, 2006).

Although much research on L1 and L2 sentence processing has been conducted using filler-gap dependencies (in relative clauses or wh-questions), inflected morphology, and garden-path sentences, previous works have looked at a very limited set of sentence structures, and there has been a lack of structural varieties examined. Little is known about how sentences with quantifier scope ambiguity are processed, and the differences between L1 and L2 processing for such sentences have not been systematically investigated. The present study thus investigates real-time comprehension of sentences with quantifier scope ambiguity in English. As a phenomenon that lies at the interface between form and meaning, quantifier scope ambiguity can address whether human sentence processing can compute covert structure automatically in real time the way it computes surface syntactic relations. It can provide a window into how the readers’ knowledge of linguistic structure interacts with conceptual knowledge in on-line interpretive decisions when there is a mismatch between the surface syntactic structure and the semantic structure. We seek to further illuminate the nature of the differences between L1 and L2 processing by examining whether L1 and L2 comprehenders consult detailed levels of linguistic analysis or compute the most economic scope configuration that incurs the least processing cost when interpreting sentences with quantifier scope ambiguity. If the comprehenders construct ‘good enough’ representations with the most economic Logical Form structure, can seeing an unambiguous paraphrase in the preceding context affect how scope is computed? Several studies in psycholinguistics have investigated processing of quantifier scope ambiguity in L1 acquisition (Anderson, 2004; Frazier, 1999; Lee, 2010; O’Grady and Lee, 2006), but there has been little discussion on how L2 processing differs from L1 processing when interpreting such sentences. Therefore, the present study examines L1 and L2 processing of English sentences containing a universally quantified subject NP and negation by comparing the processing strategies and scope interpretations of native speakers of English and Korean L2 learners of English.

II Background

1 Quantifier scope in English

The phenomenon of quantifier scope ambiguity arises when quantificational expressions (e.g. every, some, all, two) interact with one another or with negation (Horn, 1989; Jackendoff, 1972). In English, a sentence with a universally quantified noun phrase in the subject position and negation (every-negation sentence, henceforth) like Every horse didn’t jump over the fence can have two interpretations as shown in (1) below.

(1) Every horse didn’t jump over the fence.

a. Surface scope: ∀x [horse (x) → ¬ jump over the fence (x)]

b. Inverse scope: ¬∀x [horse (x) → jump over the fence (x)]

The two readings reflect the relative positions of the two scope-bearing elements at Logical Form (LF) (Fox, 1998). The surface scope reading interprets the phrase every horse outside of the scope of negation (every > not), which can be paraphrased as Every horse is such that it did not jump over the fence (i.e. none of the horses jumped over the fence). On the other hand, the inverse scope reading interprets every horse within the scope of negation (not > every) resulting in the interpretation that Not every horse jumped over the fence (i.e. some horses jumped over the fence). This inverse scope reading involves the phenomenon of covert displacement (Heim and Kratzer, 1998) in which every horse is interpreted back in its VP internal subject position, and there is a mismatch between the surface syntactic structure and the semantic structure as can be seen in (2).

In addition to the syntactic knowledge of calculating scope relations based on c-command relations and covert displacement, the linguistic phenomenon in question also involves semantic knowledge of truth conditions based on entailment relations between different interpretations. There is an entailment pattern where the ‘none’ (every > not) interpretation entails the ‘not all’ (not > every) interpretation: If none of the horses jumped over the fence, it follows that not all horses jumped over the fence but not vice versa as shown in (3). The interpretation that entails the other interpretation is the ‘strong’ or ‘subset’ reading, and the interpretation that is entailed is the ‘weak’ or ‘superset’ reading.

(3) ∀x[¬P(x)] → ¬[∀x [P(x)]] (none → not all)

¬[∀x[P(x)]] → #[∀x [¬P(x)]] (not all → #none)

According to Musolino (2006), such entailment patterns may also involve pragmatic knowledge as they give rise to a class of conversational inferences called scalar implicatures that assumes that the speaker will be as maximally informative and relevant to the exchange as s/he can (Grice, 1975). Thus, when the speaker uses a weaker term on the scale such as some (‘not every’), the listener will assume that the speaker does not mean to offer the informationally stronger statement that can be interpreted as none. In sum, the interpretation of quantification in the scope of negation can be seen as a phenomenon that involves syntax-semantics-pragmatic levels of linguistic analysis: (1) structural configuration of logical operators (every, not) based on syntactic principles such as c-command and covert displacement, (2) semantic knowledge of truth conditions and entailment relations, and (3) the pragmatic ability to make scalar implicature calculation.

Most previous psycholinguistic studies on quantifier scope ambiguity examine the processing of sentences containing two quantifiers such as Every kid climbed a tree (Anderson, 2004; Dwivedi, 2013; Frazier, 1999; O’Grady and Lee, 2006). There can be either one or multiple trees, but readers in English were found to prefer the surface scope reading (every > a) with multiple trees in which the first quantifier takes logical scope over the second. Frazier (1999) proposed that the human processing system abides by the ‘minimal lowering hypothesis’ that moves phrases from their position in the surface structure to other positions in LF only if necessary. Readers highly prefer surface scope interpretations due to the economic LF structure of surface scope configurations, and while the inverse scope reading can be accessed when unambiguous contextual support is given, it nevertheless incurs more processing cost than surface scope as evidenced by delays in reading times (Anderson, 2004; O’Grady and Lee, 2006). The additional movement of either quantifier raising or quantifier lowering adds to the syntactic complexity of the inverse scope interpretation causing processing difficulty. Anderson (2004) thus concludes that processing quantifier scope depends crucially on the grammatical operation and proposes the principle of Processing Scope Economy in which the processor is sensitive to syntactic complexity as stated in (4).

(4) Processing Scope Economy

The human sentence processing mechanism prefers to compute a scope configuration with the simplest syntactic representation (or derivation). Computing a more complex configuration is possible but incurs a processing cost (Anderson, 2004: 46).

Little is known about the processing of English every-negation sentences, but the limited evidence available in Lee (2010) suggests that, as found for doubly quantified sentences, calculating the inverse scope is dispreferred and incurs a higher processing cost. The inverse scope is seen to be computationally costly because the initial interpretation that follows the unidirectional operation of the processor must be ‘undone’ and reconstructed. Lee (2010) conducted a self-paced reading task in which adult English speakers first read preceding contexts that favored either the surface or inverse scope reading and then read experimental every-negation sentences in a region-by-region, non-cumulative moving window fashion. After reading the final region, the participants had to decide the truth value of the ambiguous sentence. The results revealed that 71% of responses adhering to the surface scope reading were accepted as ‘True’ in contrast to 37% of acceptance for the inverse scope. Slower reading and response times for the inverse scope reading also revealed the relative difficulty of calculating the inverse scope in real time when compared to the surface scope interpretation.

Considering the higher processing cost of computing the inverse scope, it is somewhat surprising then that the every-negation sentences were almost always given the inverse scope interpretation in an interview with adult English speakers (Musolino et al., 2000) and that spontaneous uses of every-negation sentences in an informal corpus were ‘invariably used on a “not all” interpretation’ (Musolino and Lidz, 2006: 842). Moreover, results of earlier studies that examine L1 interpretations of every-negation sentences using offline truth-value judgment tasks in which stories were acted out with puppets and props indicate that adult native speakers of English can equally access both scope interpretations but overwhelmingly endorse the ambiguous every-negation utterance as a fitting description for stories that are compatible with the inverse scope reading (Musolino, 1998; Musolino and Lidz, 2006; Musolino et al., 2000). More recently, Attali et al. (2021) also found a general preference for the inverse scope interpretation using a paraphrase-endorsement methodology where participants had to rate paraphrases of every-negation sentences on a sliding scale for different scenarios. The participants’ ratings indicated that the ‘not all’ paraphrases were better than ‘none’ paraphrases in the majority of the target sentences although some were more ambiguous than others. While the results of such offline studies differ in methodology and do not directly inform us about online processing difficulty, they suggest that the inverse scope interpretation, despite its greater computational cost, may be the more prevalent and even preferred interpretation of every-negation sentences by native speakers of English.

Lee (2010) explains the discrepancies in performance of time-sensitive online tasks and offline judgment tasks using the processing-based account by O’Grady et al. (2009), which predicts a greater processing burden for the inverse scope reading. According to this account, the operation of an efficiency-driven processor will strongly prefer the scope that is easier to access especially when pressured for time in online tasks. It can be pointed out, however, that studies using offline judgment tasks have also found native speakers of English to prefer the surface scope reading of the sentences in question (Chung, 2009; Wu and Ionin, 2019). In offline tasks that were similar in format to the task in Lee (2010) where participants first read a biasing context and then had to judge the truth-value of every-negation sentence, L1-English speakers in both Wu and Ionin (2019) and Chung (2009) were significantly more likely to answer ‘True’ to every-negation sentences after reading a context that favors the surface scope than a context that favors the inverse scope. Given such evidence, it is unclear whether scope interpretations in Lee (2010) are indeed driven by processing efficiency or whether other factors and processing mechanisms are at play. As it is, the exact workings of the processes involved in parsing every-negation sentences still remain largely unknown, and further investigations using online tasks are needed to better understand how L1-English speakers process and interpret sentences with universal quantifier-negation scope interactions.

2 Quantifier scope in Korean

Because this study examines native speakers of Korean, it is important to understand how scope interactions between negation and quantified arguments work in Korean. Korean is an SOV language which has two types of negation: Short-form negation, where the negative morpheme (an/ani) is placed before the verb, and long-form negation, where the negative morpheme follows the stem of the verb and the nominalizer suffix -ci. The negative morpheme ani in long-form negation is attached to the dummy verb ha- (‘do’) resulting in the contraction anh-. The surface forms of Korean sentences with the two types of negation are shown in (5).

(5) a. Motun ai-ka ppang-ul an mek ess ta

every child-nom bread-acc neg-eat-pst-decl

‘Every child did not eat bread’ (short-form negation)

b. Motun ai-ka ppang-ul mek-ci an-h -ass -ta

every child-nom bread-acc eat-nom neg do-pst-decl

‘Every child did not eat bread’ (long-form negation)

As in English, the meaning of sentences in (5) can have two readings: The sentences can mean that none of the children ate bread when the quantifier takes scope over negation (every > neg) or that only some children ate bread when negation takes scope over quantifier (neg > every). Experimental studies investigating the interaction between quantifier and negation in Korean generally report the surface scope to be the dominant reading regardless of negation type, but most of these studies have been conducted with the quantified noun phrase in the object position (Lee, 2009; Han et al., 2007; O’Grady et al., 2009). In fact, as of yet, the only study that has examined L1 processing of sentences with quantified subject and negation is Lee (2017). Similar to what has been found for sentences with quantified object and negation, native speakers of Korean exhibited a strong preference for the surface scope interpretation in both types of negation even when preceded by discourse contexts that were heavily biased toward inverse scope interpretation. Response times were significantly longer when accessing the inverse scope, and reading times at the critical region (i.e. the negated verb) and the following regions were also significantly slower in the inverse scope reading leading Lee to conclude that the results provide evidence for the efficiency-based processing account (O’Grady et al., 2009) in which the processor computes the simplest scope.

In sum, Korean native speakers were found to strongly prefer the surface scope interpretation regardless of the type of negation in Korean every-negation sentences. If there is interlinguistic influence, Korean learners of English are predicted to experience difficulty in accessing the inverse scope interpretation in equivalent sentences in English. However, studies have shown that L1 influence may not always be present in L2 processing (Gerth et al., 2017; Marinis et al., 2005; Roberts et al., 2008; Sato and Felser, 2010), and it remains to be seen whether L1 effects can be found in Korean learners’ processing of English sentences with quantified subject and negation.

3 Quantifier scope in L2 processing

There has not been much research on L2 acquisition of every-negation scope interactions, and the few studies that exist mostly used offline tasks that were not time constrained. Studies that examined L2 acquisition of English sentences with universally quantified objects and negation (i.e. ‘subject didn’t V every N’) (Chung, 2013; Kim, 2010; O’Grady et al., 2009) and with universally quantified subjects and negation (Chung, 2009; Wu and Ionin, 2019) all found L2 learners to strongly prefer the surface scope reading especially in the beginning stages of acquisition. While the inverse scope reading may become increasingly accessible as proficiency develops, the studies agree that the surface scope is generally the dominant reading by L2 learners. In their investigation of L1-Mandarin L2-English Speakers’ interpretation of English sentences with every/all and negation, Wu and Ionin (2019) found that Mandarin learners considered the inverse scope reading to be more acceptable in English than in Mandarin, but the inverse scope interpretation of English sentences in question was a difficult property for even advanced learners.

Comparatively little is known about how L2 learners process sentences with universally quantified subjects and negation in real-time, and to the best of our knowledge, Lee (2018) is the only study that has examined this phenomenon with L2 learners from a psycholinguistic perspective. Forty Korean learners of English participated in a self-paced reading task in English where they first read a discourse context that favored either scope reading and then the ambiguous every-negation sentence. After reading the experimental sentence, they answered a True/False question, and their reading and response times were recorded. L2 learners in the study strongly preferred the surface scope reading regardless of proficiency level, and the inverse scope interpretation required longer reading and judgement times. The slowdown in reading times was found to occur after the negated verb at the direct object. Lee discusses these reading time patterns within the principle of Processing Scope Economy in (4) and argues that L2 learners naturally prefer the reading in which the processor computes the simplest scope as has been found for native speakers of English in Lee (2010).

Given the conflicting evidence in the L1 literature and the lack of L2 processing studies that examine sentences containing every in subject NP and negation, the present study further investigates L1 and L2 processing of the sentences in question by examining native speakers of English and L1-Korean L2-English learners’ performance in a self-paced reading/interpretation task. More specifically, we examine whether the two groups consult detailed levels of linguistic analysis when interpreting sentences with quantifier scope ambiguity or compute the most economic scope configuration that incurs the least processing cost as found in Lee (2010, 2018) and how exactly the two groups differ in their speed and manner of analysis. Additionally, we examine how the presence of unambiguous quantifiers in the preceding context or in the target sentence affect scope preferences and reading times, which could provide a window into whether an unambiguous context can bias readers towards a certain reading and how other quantifiers affect scope interpretation.

III Method

1 Participants

Forty-two Korean L2 learners of English (mean age 22.7; range 19–29; 30 female and 12 male) and 30 native English speakers (mean age 29.3; range 27–33; 14 female and 16 male) participated in the study. The L2 learners were all students at a university in Seoul and were paid KRW 10,000 for their participation. They were high-intermediate to advanced learners whose TOEIC scores were above 800 points (mean score 928.5; range 800–990¹). The native control group was recruited via Amazon Mechanical Turk, a crowdsourcing web platform, and only native speakers of English in US locations were allowed to participate. They were compensated USD $3 for their participation.

2 Materials

The main experiment of the study was a self-paced reading/interpretation task that asks for participants’ interpretation of the sentences in question and measures response and reading latencies. The materials for the self-paced reading task comprised a total of 64 sentences, including four practice items, 24 experimental sentences, and 36 filler sentences. The four experimental conditions came in a design with context (ambiguous vs. unambiguous) as within-participant factor and with scope reading (surface vs. inverse) as within-participant factor nested in each level of context. The experimental sentences were divided into four segments in a non-cumulative moving-window procedure, and the stimulus sentences were presented in a segment-by-segment fashion as shown in (6a)–(6d). The participants were asked to judge the sentence as ‘True’ or ‘False’ after the last segment of the sentence.

(6) a. Ambiguous context, Unambiguously surface scope sentence (AmbS):

Context: ‘Last week there was a horse race at the stadium, and the horses had to jump over a very high fence. However, every horse didn’t jump over the fence.’

Sentence: Last week/ none of the horses/ jumped over the fence/ at the race.

b. Ambiguous context, Unambiguously inverse scope sentence (AmbI):

Context: ‘Last week there was a horse race at the stadium, and the horses had to jump over a very high fence. However, every horse didn’t jump over the fence.’

Sentence: Last week/ some of the horses/ jumped over the fence/ at the race.

c. Unambiguously surface scope context, Ambiguous sentence (UnambS):

Context: ‘Last week there was a horse race at the stadium, and the horses had to jump over a very high fence. However, none of the horses jumped over the fence.’

Sentence: Last week/ every horse/ didn’t jump over the fence/ at the race.

d. Unambiguously inverse scope context, Ambiguous sentence (UnambI):

Context: ‘Last week there was a horse race at the stadium, and the horses had to jump over a very high fence. However, only some of the horses jumped over the fence.’

Sentence: Last week/ every horse/ didn’t jump over the fence/ at the race.

As can be seen in the examples of experimental conditions above in (6a)–(6d), either the context or the experimental sentences were unambiguously surface or inverse scope and biased the readers towards a particular scope reading. In conditions (a) and (b), the context contained the ambiguous every-negation sentence which was followed by an experimental sentence that was either unambiguously surface (‘none’) or inverse (‘some’) scope.² Conversely, the context in conditions (c) and (d) contained a sentence with an unambiguous quantifier that indicated surface (‘none’) or inverse (‘some’) scope respectively which was followed by the ambiguous experimental sentence. By having the two context conditions, we could examine whether or not having a preceding context with an unambiguous quantifier can facilitate the participants’ access to the inverse scope reading in comparison to conditions with the ambiguous sentence in the preceding context. Furthermore, the ambiguous or unambiguous experimental sentences could reveal the manner and ease by which participants access the surface and inverse scope readings.

There were three dependent variables in this task: (1) the True or False response, (2) the response times, and (3) the reading times for each segment in the experimental sentences. Since the segment length in conditions (a) and (b) was not directly comparable to that of conditions (c) and (d), comparisons were made within the same context conditions, which varied only in scope reading. The critical region was segment 3 in which the verb appears after ‘none’ or ‘some’ for conditions (a) and (b) and in which negation makes its appearance for conditions (c) and (d). Segment 4 was also examined for spillover and wrap-up effects.

Two counterbalanced presentation lists were constructed with each list containing six test items of each condition with no context appearing twice in the same list. That is, each list contained an equal number of inverse scope and surface scope contexts, but only one version of each test item (i.e. a context that appeared in condition (a) on one list was presented in condition (b) on the other and vice versa). Each presentation list also included 36 fillers presented in a random order for each participant. Of the 36 filler sentences that were of similar length and format, half was unambiguously ‘True’ and the other half was unambiguously ‘False’. The fillers contained unrelated quantifiers (e.g. several, most, all, few, many, two) that did not give rise to quantifier scope ambiguity. Participants were assigned to one of the two presentation lists randomly and tested individually. Each presentation list had the same number of participants.

3 Procedure

The self-paced reading experiment was conducted using the online IBEX platform (Drummond, 2012). Recent works in psycholinguistics have used the IBEX web interface for self-paced reading experiments (Fine and Jaeger, 2016; Myslín and Levy, 2016), and studies have shown that results obtained in lab-based experiments can be replicated in experiments administered over the web (Demberg, 2013; Enochson and Culbertson, 2015). The participants filled out a consent form and were given specific instructions and sample questions as a practice session. In the self-paced reading/interpretation task, the participants were given sufficient time to read the context of each story and were instructed to press the space bar when they are ready for the experimental sentence. Once they pressed the space bar, the context disappeared and a cross appeared on the computer screen. They were asked to press the space bar again, and the first segment of the experimental sentence appeared. The participants were instructed to press the space bar at a rate that allowed them to read at their normal speed as they read each segment as carefully as possible. The end of each sentence was indicated by a period after the last word of the final segment, which was followed by the same question ‘Is this sentence True or False?’ for every sentence. The participants were asked to press either the number 1 key for True or 2 for False. After the response, the next story appeared on the screen. The task took about 20 minutes for each participant.

4 Predictions

It was predicted that native speakers and the learners would show significant differences in scope preferences as well as speed and manner of analysis. Unlike Lee (2010), we predicted that native speakers would not only build complete detailed syntactic structures but also consult various factors in processing of every-negation sentences rather than solely relying on the operations of an efficiency-driven processor. In conditions (a) and (b) in which the context does not contain an unambiguous quantifier, the speakers were predicted to strongly endorse the inverse scope, which has been found to be the dominant scope interpretation in English for L1 adults (Attali et al., 2021; Musolino, 2006; Musolino and Lidz, 2006). We speculated that the reading and response times of the two scope interpretations will not differ significantly for L1 speakers in these conditions because the costlier LF structure of the inverse scope will be balanced by the benefits of being the more prevalent interpretation and will be just as easily accessed as the surface scope. However, when there is an unambiguous quantifier in the preceding context, L1 speakers will be facilitated by the biasing context and calculate the scope that matches the context. This predicts a higher number of ‘True’ responses in the unambiguous context conditions than ambiguous context conditions. If the speakers are sensitive to semantic calculations of entailment relations, the surface scope reading (every > not) which is the strong reading that entails the inverse scope (not > every) will receive a higher number of ‘True’ responses than the inverse scope reading. If, however, pragmatic inferences of scalar implicature and expectations of maximum informativity trump other calculations, we can expect to see an increase in the number of ‘False’ responses and response times irrespective of scope conditions because speakers would find it pragmatically infelicitous to read a scopally ambiguous sentence after its unambiguous counterpart in the preceding context.

As for L2 learners, they were predicted to follow the ‘minimal effort’ principle and strongly prefer the surface scope reading in all conditions. They would show increased reading and response times for the inverse scope relative to the surface scope reading due to the economic LF structure of surface scope configurations and L1 transfer of scope preferences as found in previous studies (Chung, 2009; Lee, 2018). L2 learners were predicted to construct ‘good enough’ representations with the most economic LF structure regardless of context ambiguity and be reluctant to calculate semantic entailment relations or scalar implicatures, which could be costly and require extra time during real-time language comprehension especially for L2 learners (De Neys and Schaeken, 2007).

IV Results

Reading times for critical regions, response latencies, and the response of ‘True’ or ‘False’ were compared between L1 and L2 speakers. Comparison of reading times were made separately for different context conditions: Conditions (a) and (b) were compared separately from conditions (c) and (d). Reading times between surface scope and inverse scope sentences were compared in the critical regions as well as the following segments to capture spillover effects. The dependent variables were analysed with logit mixed-effect models (Baayen et al., 2008; Jaeger, 2008) using the ‘lme4’ package (Bates et al., 2015) in the R environment (R Core Team, 2019). In all models, Context (ambiguous vs. unambiguous; effect-coded as −.5 and .5), Scope (inverse vs. surface; effect-coded as −.5 and .5), Group (L1 vs. L2; effect-coded as −.5 and .5), and their interactions were fixed effects (Jaeger, 2008). We employed the maximal random effects structure that would converge (Barr et al., 2013) by conducting a maximum likelihood ratio model comparison. The maximal random effects structure of the models did not converge, and we eliminated random slopes for random effects until the models converged. The final models included (1) only the by-participant and by-item random intercepts for True-False responses, (2) a random intercept and a random slope for Context as by-participant random effect and a random intercept and a random slope for Scope as by-item random effects for response latencies, and (3) a random intercept and a random slope for Context as by-participant random effect and a random intercept and random slope for Group as by-item random effects for reading times.

1 Truth-value judgments

An examination of fillers revealed an accuracy rate of 85.92% for L1 speakers and 84.40% for L2 speakers, which indicated that participants were paying attention to the experiment. Table 1 and Figure 1 illustrate the mean value of True/False responses (True = 1, False = 0) by L1 and L2 speakers in the four experimental conditions. The most notable difference between the two groups could be found in ambiguous context conditions (a) and (b) with native speakers exhibiting a higher mean for the inverse scope and L2 learners the surface scope. In unambiguous context conditions (c) and (d), the two groups showed similar interpretative patterns in which the surface scope had a higher mean than the inverse scope. The results revealed a main effect of Context (β = −.75, SE = .13, z = −5.63, p < .001) with ambiguous context conditions receiving fewer ‘True’ responses than unambiguous context conditions overall. There was also a main effect of Scope (β = −1.89, SE = .13, z = −13.99, p <.001) as surface scope conditions were more likely to receive ‘True’ responses. A significant interaction between Context and Scope (β = 3.58, SE = .27, z = 13.28, p < .001) indicated that scope interpretations were strongly influenced by the ambiguity of the context. Group did not have a significant main effect (β = −.26, SE = .16, z = −1.62, p = .11), but significant interactions were found between Scope and Group (β = −2.11, SE = .27, z = −7.75, p < .001) as well as Context, Scope, and Group (β = −3.57, SE = .55, z = −6.55, p < .001).

Table 1.

Mean value of True/False responses (standard deviation in parentheses).

	Ambiguous context		Unambiguous context
	Surface	Inverse	Surface	Inverse
L1	0.31 (.46)	0.78 (.41)	0.93 (.26)	0.30 (.46)
L2	0.67 (.47)	0.28 (.45)	0.93 (.26)	0.24 (.43)

Notes. True = 1. False = 0.

Figure 1.

Mean value of Truth (= 1) responses.

To better understand the interactions between Group and other factors, separate analyses were conducted for each group. In native speakers’ responses, there was a main effect of Context (β = −.64, SE = .21, z = −3.06, p = .002) and Scope (β = −.67, SE = .21, z = −3.19, p = .001) as well as a significant interaction between Context and Scope (β = 5.80, SE = .44, z = 13.26, p < .001). The surface scope received a higher number of ‘True’ responses than the inverse scope, but these responses were significantly modulated by context ambiguity: the inverse scope was endorsed at a higher rate than the surface scope in the ambiguous context condition, whereas the surface scope was more likely to be endorsed than the inverse scope in the unambiguous context condition. Similarly, L2 learners’ responses were also significantly modulated by Context (β = −.83, SE = .17, z = −4.78, p < .001) and Scope (β = −2.72, SE = .18, z = −15.39, p < .001), and there was a significant interaction between Context and Scope (β = 2.08, SE = .35, z = 5.98, p < .001). The learners exhibited a higher number of ‘True’ responses for the surface scope than the inverse scope regardless of context ambiguity but were significantly more ambivalent in their answers in the ambiguous than unambiguous context conditions. That is, the likelihood of accepting the inverse scope and rejecting the surface scope was significantly greater in ambiguous than unambiguous context condition.

2 Response latencies

The response times of True or False presses are suggestive of how difficult it is to access a particular scope reading. To correct for outliers, extreme response times below 200 milliseconds and above 5,000 milliseconds and 2.5 standard deviations above or below the group’s mean were removed prior to analyses (Dussias and Piñar, 2010; Simmons et al., 2011), which affected 12% of L1 data and 4% of L2 data. Table 2 and Figure 2 that illustrate the mean value of True/False response times show slower response times by L2 learners compared to native speakers in all four conditions. There was a main effect of Group (β = .25, SE = .08, t = 3.09, p = .003) and a significant interaction between Context and Group (β = .15, SE = .05, t = 3.11, p = .003). When separate analyses were conducted for each group, there was only a main effect of Context (β = −.14, SE = .04, t = −3.57, p = .002) for native speakers. No other factors or interactions were significant for either group. Native speakers exhibited significantly longer response times in unambiguous than ambiguous context conditions, but L2 speakers’ response times were not affected by context ambiguity. In order to examine whether accepting the inverse scope incurred extra processing cost in response times than accepting the surface scope, we conducted a separate analysis of only ‘True’ responses and found no significant differences in response times between inverse and surface scope conditions for both groups in both ambiguous (L1 β = .004, SE = .10, t = .04, p = .969; L2 β = −.01, SE = .08, t = .19, p = .853) and unambiguous (L1 β = .06, SE = .09, t = .66, p = .515; L2 β = .03, SE = .08, t = .42, p = .677) context conditions.

Table 2.

Mean value of True/False response times in milliseconds (SD in parentheses).

	Ambiguous context		Unambiguous context
	Surface	Inverse	Surface	Inverse
L1	541.84 (331.41)	563.20 (398.23)	620.36 (394.76)	674.19 (442.97)
L2	737.28 (393.80)	756.04 (425.16)	732.07 (347.04)	721.07 (315.47)

Figure 2.

Mean value of True/False response latencies (ms).

3 Reading times

For analysis of reading times, the segment length of critical regions in conditions with an ambiguous context was not directly comparable to that of conditions with an unambiguous context, and comparisons were made within the same context conditions which varied only in scope reading. To correct for outliers, extreme reading times below 200 ms and above 5,000 ms or 2.5 standard deviations above or below the group’s mean were excluded. This correction affected around 6% of the data for the critical region and 4% of the data for the spillover region. The trimmed data was then log-transformed to satisfy the normal distribution assumption. The mean reading times of critial and spillover regions are shown in Table 3 and Figure 3.

Table 3.

Mean reading times of critical and spillover regions in milliseconds (SD in parentheses).

	Ambiguous context		Unambiguous context
	Surface (a)	Inverse (b)	Surface (c)	Inverse (d)
Critical region:	VP + Obj NP	VP + Obj NP	didn’t VP . . .	didn’t VP . . .
L1	489.00 (223.91)	476.68 (196.87)	466.47 (228.77)	514.12 (308.62)
L2	778.86 (387.56)	767.37 (361.56)	850.78 (369.70)	798.36 (408.93)
Spillover region:	PP	PP	PP	PP
L1	541.32 (429.74)	544.86 (331.93)	563.62 (329.01)	605.67 (368.98)
L2	690.82 (431.36)	706.95 (431.81)	686.87 (442.95)	670.29 (424.56)

Figure 3.

Mean reading times of critical and spillover regions (ms).

In ambiguous context conditions (a) and (b), there was a main effect of Group for both critical and spillover regions, which revealed that L2 learners’ reading times were significantly longer than those of L1 speakers. No other main effect or significant interactions were found in the reading times for these conditions (see Table 4 for parameter estimates). In the critical region of unambiguous context conditions (c) and (d), there was a main effect of Group as well as a marginally significant effect of Scope. To examine whether the effect of Scope in the critical region was significant for each group, separate analyses were performed. The results revealed that both groups’ reading times were significantly modulated by Scope: native speakers took significantly longer to process the critical region in the inverse than the surface scope condition (β =.07, SE = .03, t = 2.59, p = .01), whereas L2 learners took significantly longer to process the surface than the inverse scope (β = −.09, SE = .03, t = −2.53, p = .012). In the spillover region of unambiguous context conditions, there were no significant group differences, but there was a significant interaction between Group and Scope. When further analysis was conducted to investigate this interaction, we found a significant main effect of Scope in native speakers’ reading times (β =.07, SE = .04, t = 2.01, p = .045); the native speakers were slower to read the inverse scope condition than that of the surface scope condition, similar to what has been found in the critical region. However, no main effect of scope was found in the spillover region for L2 learners (β = −.03, SE = .04, t = −.76, p = .446).

Table 4.

Fixed effects estimate for reading times (critical and spillover regions).

	Fixed effects	β	SE	t	p
Ambiguous context (a) and (b):
Critical	Intercept	6.35	.03	188.68	<.001
	Scope	−.03	.02	−1.26	.210
	Group	.44	.06	7.33	<.001
	Scope × group	.01	.04	.29	.774
Spillover	Intercept	6.30	.04	155.91	<.001
	Scope	−.01	.03	−.32	.750
	Group	.24	.07	3.43	<.001
	Scope × group	−.01	.05	−.21	.837
Unambiguous context (c) and (d):
Critical	Intercept	6.37	.05	118.34	<.001
	Scope	−.04	.02	−1.71	.088
	Group	.52	.09	6.03	<.001
	Scope × group	−.03	.06	−.05	.596
Spillover	Intercept	6.33	.04	146.30	<.001
	Scope	.02	.03	.64	.521
	Group	.06	.07	.93	.358
	Scope × group	−.16	.07	−2.22	.027

Note. SE = standard error.

In sum, the L2 learners’ reading times were significantly slower than those of the native speakers in the critical region of all four conditions and the spillover region of ambiguous context conditions. Reading times were significantly modulated by Scope only in unambiguous context conditions: native speakers were significantly slower processing the inverse scope than the surface scope in both the critical and the spillover regions, whereas the L2 learners were significantly slower processing the surface scope than the inverse scope only in the critical region. Spillover and wrap-up effects were evident in the native speakers’ reading times as they exhibited longer reading times in the spillover than critical regions despite the shorter length of the spillover region. Such effects were not present in L2 learners’ reading times, and it took longer for them to read the critical region than the spillover region in all four conditions.

V Discussion

The present study compared native speakers of English and Korean L2 learners of English in the processing of English sentences containing a universally quantified subject NP and negation and examined the effect of context ambiguity on scope assignment and response/reading times.

The results indicate that the native speakers show a general tendency to assign the inverse scope reading when processing the sentences in question as previously found (Musolino, 1998; Musolino and Lidz, 2006; Musolino et al., 2000) but consult detailed syntactic and semantic rules of computations when interpreting the sentence after a highly biasing context. Contrary to previous claims that the greater processing cost of the inverse scope is due to its structural complexity (Anderson, 2004; Lee, 2010), our results suggest that the delays in response and reading times could be due to processing mechanisms in L1 that are sensitive to not only structural but also semantic considerations. The L2 results, on the other hand, revealed different processing mechanisms at work: L2 learners exhibited a dominant preference for surface scope as found in previous L2 studies (Chung, 2009; Kim, 2010; Lee, 2018) irrespective of the ambiguity of the preceding context and showed little indication of detailed processing of the sentences in question. L2 scope preferences and reading times revealed the learners’ strong tendency to arrive at the most economical interpretation, which could reflect ‘good enough’ processing and/or L1 transfer.

1 Native speakers

Native speakers’ scope endorsement rates were significantly modulated by context ambiguity: when they saw the every-negation sentence in the preceding context and then read an unambiguous target sentence, they were more likely to accept the inverse scope than the surface scope, but when they read an unambiguous sentence in the context and then the ambiguous target sentence, they considered the every-negation sentence to be more acceptable in the surface scope context than the inverse scope context. Response latencies were also modulated by context ambiguity as the native speakers were significantly slower to respond in conditions with an unambiguous context than conditions with an ambiguous context. Longer response latencies in conditions (c) and (d) suggest that processing the ambiguous sentence after a highly biasing context required more effort than when processing the ambiguous sentence without contextual support. While the sentences in question were readily given the inverse scope reading in the absence of contextual support, the presence of a biasing context appears to have forced them to process the sentence differently rather than facilitating scope interpretation. It is possible that these findings are suggestive of deep processing mechanisms in L1 that are sensitive to not only structural but also semantic considerations. In light of the fact that the two scope readings are available to the native speakers although the inverse scope is more salient (Musolino, 1998; Musolino and Lidz, 2006; Musolino et al., 2000), we speculate that the presence of an unambiguous quantifier in the preceding context invoked semantic knowledge of quantifiers, thereby leading comprehenders to calculate both scopes and their semantic entailment relations. When the preceding context biased them towards the surface scope reading, which semantically entails the inverse scope, they answered ‘True’ to the ambiguous sentence because the context supports both scope readings. However, in the case of an unambiguously inverse scope context, which does not entail the surface scope, the readers would be more reluctant to respond ‘True’ to the ambiguous sentence because the context only supports one scope interpretation. It is not surprising then that the unambiguous context conditions incurred a greater processing cost because the speakers were weighing the semantic relations of both scope readings in light of the preceding context. Another possible reason for the delayed response times in the unambiguous context conditions may be due to pragmatic reasons as speakers would find it infelicitous to read an ambiguous sentence after its unambiguous counterpart in the preceding context. Such pragmatic effects are highly likely, but a higher endorsement rate of the surface scope in these conditions indicates that the calculation of semantic entailment must have had a greater effect in the native speakers’ behavior.

It seems pertinent to mention here that previous L1 studies that examined the ambiguous sentences in question have reported mixed results. Some have found the inverse scope to be the dominant interpretation by native speakers of English (Attali et al., 2021; Musolino, 1998; Musolino and Lidz, 2006; Musolino et al., 2000) while others have found the surface scope to be more prevalent (Chung, 2009; Lee, 2010; Wu and Ionin, 2019). While all of these studies asked participants to judge or rate sentences with every NP and negation as the target sentence, the every-negation sentence in the present study appeared either in the preceding context or the target experimental sentence, the presence of which alternated with unambiguous sentences with quantifiers ‘none’ or ‘some’. Our results show that the dominant scope interpretation by native speakers can change depending on how the experimental sentences are presented in relation to the preceding context. That is, the inconsistencies in the previous literature may not be due to differences between time-sensitive online tasks and offline judgment tasks as suggested by Lee (2010) but due to differences in how the context and the ambiguous sentence were presented. An examination of the methodologies in previous studies suggests that the presence of other quantifiers and/or explicit numbering in the preceding context could be an important factor that affects processing of the every-negation sentence: Studies that presented the biasing contexts using explicit quantifiers or by numbering each item/person in written discourses have found the surface scope to be the more dominant reading (Chung, 2009; Lee, 2010; Wu and Ionin, 2019), while studies that presented the biasing contexts using act-outs with puppets/props or pictures have found the inverse scope to be the more dominant reading (Attali et al., 2021; Musolino, 1998; Musolino and Lidz, 2006; Musolino et al., 2000). It could be that the comprehenders become more keenly aware of the semantic entailment relations between the quantifiers when they see other quantifiers and/or explicit numbering in the preceding context and show a higher rate of acceptance for the surface scope, not because the surface scope is easier to access, but because it semantically entails the inverse scope reading and is ‘True’ for both readings. That is, contrary to Lee’s (2010) explanation that the surface scope is preferred because the processor computes the simplest scope, it seems possible that the scope interpretation is triggered by other quantifiers and/or explicit numbering in the preceding context.

When L1 reading times were examined, no significant difference in reading times was found between the two scope readings in ambiguous context conditions (a) and (b). That is, after reading the scopally ambiguous sentence in the preceding context, the time taken to process the unambiguously surface scope target sentence was not significantly different from that of the unambiguously inverse scope target sentence. No main effect of Scope in the reading times and response latencies of ambiguous context conditions suggest that the inverse scope reading, the more dominant interpretation in these conditions, could be the ‘default’ reading of the ambiguous sentences in question by native speakers, which is consistent with the results of previous studies (Attali et al., 2021; Musolino, 1998; Musolino and Lidz, 2006; Musolino et al., 2000). While no significant difference in reading times was found between the two scope readings in ambiguous context conditions, reading times were significantly modulated by Scope in unambiguous context conditions. Native speakers took longer to process the inverse scope than the surface scope in both the critical and the spillover regions. These results are quite surprising considering the native speakers’ general preference for inverse scope in ambiguous context conditions. However, in accordance with our interpretation of the truth-value responses and response latencies in L1 data, we speculate that they took longer to process the ambiguous sentence after reading the unambiguously inverse scope context than the surface scope context because they had to reject their ‘default’ scope reading in this condition after calculating both scopes and taking semantic entailment into consideration. That is, there would have been two competing forces at work: whether to resort to the default scope reading or reject it to satisfy the semantic entailment properties. Following the conversational Principle of Charity (Grice, 1975), the assumption that a sentence will be accepted when at least one reading makes it true, it would have been easier to accept a reading that is not strongly preferred but nevertheless available than to reject a reading that is not only available but also strongly preferred. The slower reading times found for the inverse scope in the present study is consistent with the findings of Lee (2010). While Lee (2010) attributes the slower reading times to processing costs of computing a more complicated syntactic structure, we argue that the inverse scope does not necessarily incur more processing cost than the surface scope and that the delay in reading times is indicative of native speakers’ tendency to consult multiple levels of linguistic analysis.

2 L2 learners

While native speakers judgments were highly dependent on the ambiguity of the preceding context, L2 learners exhibited a stronger preference for the surface scope than the inverse scope in both ambiguous and unambiguous context conditions. Such preference was significantly stronger in conditions with unambiguous than ambiguous contexts. That is, the learners were facilitated by an unambiguous preceding context, showing a stronger tendency to accept the surface scope after an unambiguously surface context in (c) than after the ambiguous context in (a) and a stronger tendency to reject the inverse scope after an unambiguously inverse context in (d) than after the ambiguous context in (b). The learners’ ambivalence shown in ambiguous context conditions suggests that although the surface scope is the predominant reading, the inverse scope reading may be available when reading the sentences in question. However, Scope and Context had no main and interaction effects on the response times of True or False presses by the learners. Only a main effect of Group was found in response latencies, as the learners were significantly slower to respond than the native speakers. Such delays and lack of automaticity are suggestive of heavier processing load for L2 learners which is predicted to be characteristic of L2 processing (Segalowitz, 2003).

As for L2 learners’ reading times, they were also significantly slower than those of the native speakers in the critical region of all four conditions and the spillover region of ambiguous context conditions. L2 reading times were significantly modulated by Scope in unambiguous context conditions but not in ambiguous context conditions. The L2 learners took significantly longer to process the critical region (but not the spillover region) of the ambiguous sentence after reading an unambiguously surface scope context than an unambiguously inverse scope context. This is in fact puzzling, as it indicates a higher processing cost for the dominant scope reading that the learners generally prefer. It is also inconsistent with previous findings (Lee, 2018) in which the inverse scope, not the surface scope, required longer reading times. A possible explanation for this unexpected result may be attributed to factors related to pragmatic knowledge. Scalar implicature calculation may incur additional processing cost but is not a source of L2 difficulty because pragmatic competence is universal (Slabakova, 2010), and the learners may have found it pragmatically infelicitous to read the informationally stronger statement with the unambiguous quantifier none in the preceding context before reading the target sentence. Although the surface scope is their dominant interpretation of every-negation sentences, seeing the potentially ambiguous sentence after a more informative statement may go against the readers’ pragmatic expectations of maximal informativity. As for the inverse scope condition with the unambiguous quantifier some in the preceding context, the every-negation sentences were rejected without a delay in reading times because the dispreferred interpretation would not have invoked pragmatic calculations. It can be pointed out that both L1 and L2 groups exhibited unexpected behavior in unambiguous context conditions, which can be indicative of pragmatic effects in these conditions. Previous studies have found pragmatic factors such as world knowledge, plausibility, and expectations of informativity to affect scope preferences (Attali et al., 2021), and future works must thus test these speculations using a controlled research design that can tease apart the influences of pragmatic knowledge from other factors.

Another notable point of discussion in the present data is that the learners’ scope interpretations were assigned in the critical region; every NP was integrated into the learners’ scope interpretation in the critical region, and no indications of processing difficulty or forced reinterpretation of scope were found at the end of sentences. The learners took longer to read the critical than the spillover region in all four conditions and did not exhibit spillover or wrap-up effects after the negated verb unlike native speakers who exhibited longer reading times in the spillover than the critical region despite the shorter length of the spillover region. This indicates that the learners’ scope interpretations occurred immediately without delay making reanalysis of the initial interpretation unnecessary.

Overall, L2 scope preferences and reading times reveal the learners’ general tendency to prefer the surface scope reading with a more economic syntactic structure that incurs less processing cost as has been found in previous studies (Anderson, 2004; Lee, 2018; O’Grady and Lee, 2006). It could also be indicative of L1 transfer of scope interpretation (Lee, 2017). In accordance with the theory of cost-driven transfer (O’Grady, 2013), which assumes that L1 transfer is driven by processing efficiency and pressures to control the cost of using the L2, it seems only natural that the learners would resort to the surface scope reading that does not add to processing cost and transfers the scopal preference for Korean to English. However, it would be premature to assume that L2 learners lack detailed syntactic representations given that the inverse scope interpretation was available although not preferred in ambiguous context conditions. Despite the strong preference for surface scope interpretation, L2 learners were able to access the inverse scope reading, which suggests that they are not restricted to shallow processing. Although they can build detailed syntactic structures when required, they seem to have a strong tendency to create a structure that is just good enough to understand the meaning of the sentence (Lee and Shin, 2016; Lim and Christianson, 2013a, 2013b).

The present study is not without limitations. The current research design did not allow us to tease apart the crosslinguistic influence of Korean scope preferences and the workings of an efficiency-driven processor when interpreting L2 data. In order to understand the intricate underpinnings of the L2 processor, our account of the results must be confirmed in future investigations, and other languages with different L1 scope preferences must also be examined. In addition, the current research design can be improved by using a Latin-square 2x2 design for a complete examination of the fixed effects instead of the current design with scope reading nested in each level of context. Also, the spillover region of target sentences should not be at the end of the sentence to distinguish spillover effects from wrap-up effects. Furthermore, as a reviewer has pointed out, future work should probe the present results by replacing ‘some/only some’ with ‘not all’ for the inverse scope interpretation, as the mismatch between the affirmative nature of ‘some’ and the negative nature of the every-negation sentence may have had an effect on the endorsement rates and the response/reading times.

In spite of these limitations, the current findings also contribute to our understanding of how semantic representations of quantifier scope ambiguity are constructed in L1 and L2 processing. It has been pointed out that semantic representations of sentences with quantifier scope ambiguity constructed in real-time could be underspecified such that scope interpretation is delayed unless explicitly required (Dwivedi, 2013; Dwivedi et al., 2010; Sanford and Sturt, 2002; Swets et al., 2008). In Dwivedi et al. (2010) and Dwivedi (2013), doubly quantified sentences in English (e.g. Every kid climbed a tree) were not deeply processed during interpretation in L1 processing unless explicitly demanded by a question task. In contrast, Radó and Bott (2011) found that scope computation in German doubly quantified sentences starts even before encountering the disambiguating information and that quantifier scope assignment occurs immediately without delay despite being instructed to delay scope interpretation until the reading is at all available. Our findings are consistent with the findings of Radó and Bott (2011) in which scope computation occurs immediately without delay. Scope interpretations were assigned without much delay and participants exhibited clear preferences for a certain scope reading when processing the ambiguous sentence. That is, if scope is underspecified and full computation of scope ambiguity occurs only when interpretation is explicitly required, participants would be more likely to respond ‘True’ than ‘False’ when reading the unambiguous target sentences in conditions (a) and (b) after having read the ambiguous sentence in the preceding context. However, there was a clear preference for a certain scope reading by both groups, and the time taken to respond ‘True’ to inverse scope was not significantly different from that of the surface scope. These results suggest that the ambiguous sentence in question is not underspecified but given full immediate interpretation without delay by both native speakers and L2 learners.

VI Conclusions

The results of the present study revealed significant differences between L1 and L2 processing of sentences with a universally quantified subject NP and negation especially with regard to how the ambiguity of the preceding context affected scope interpretation: an unambiguous context strengthened L2 preference for the surface scope whereas it added processing burden for native speakers. L2 learners consistently preferred the surface scope reading with a more economic structure that does not add a processing cost irrespective of contextual ambiguity. They seemed to depend on ‘good-enough’ processing to arrive at the most efficient interpretation even in pragmatically infelicitous conditions whereas the native speakers were sensitive to calculations of the scope relations between the universal quantifier and negation as well as the semantic knowledge of truth conditions of different interpretations and were able to build detailed structures. We found that the efficiency-based processing account (O’Grady et al., 2009) that is used by previous online studies (Lee, 2010, 2018) to explain processing of universal quantifier-negation scope interactions may apply to L2 processing but not to L1 processing. In sum, while both native speakers and L2 learners tend to assign full immediate interpretation without delay to sentences with a universally quantified subject NP and negation, they were different in that the learners exhibited a strong tendency to arrive at the most economical interpretation irrespective of the ambiguity of the preceding context whereas the native speakers consulted detailed syntactic and semantic rules of computation.

Footnotes

Acknowledgements

We would like to express deep gratitude to the anonymous reviewers for their valuable comments and suggestions.

Declaration of Conflicting Interests

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 was supported by the Global Research Network program through the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea (NRF-2017S1A2A2041367) to Jeong-Ah Shin.

ORCID iDs

Eun Seon Chung

Jeong-Ah Shin

Notes

References

Anderson

(2004) Context and real-time comprehension of scope ambiguity. Unpublished PhD thesis, Northwestern University, Evanston, IL, US.

Attali

Scontras

Pearl

(2021) Pragmatic factors can explain variation in interpretation preferences for quantifier-negation utterances: A computational approach. In: Proceedings of the 43rd annual meeting of the Cognitive Science Society. Vienna: Cognitive Science Society, pp. 917–23.

Baayen

Davidson

Bates

(2008) Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory and Language 59: 390–412.

Barr

Levy

Scheepers

Tily

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

Bates

Maechler

Bolker

Walker

(2015) lme4: Linear mixed-effects models using Eigen and S4: R package version 1.1-8 [software]. Available at: http://CRAN.R-project.org/package=lme4 (accessed February 2022).

Christianson

Luke

(2011) Context strengthens initial misinterpretations of text. Scientific Studies of Reading 15: 136–66.

Christianson

Luke

Ferreira

(2010) Effects of plausibility on structural priming. Journal of Experimental Psychology 36: 538–44.

Christianson

Williams

Zacks

Ferreira

(2006) Younger and older adults’ good enough interpretations of garden-path sentences. Discourse Processes 42: 205–38.

Chung

(2009) The interpretation of English sentences containing quantification and negation by L1-Korean L2-English adult learners. In: Bowles

Ionin

Montrul

Tremblay

(eds) Proceedings of the 10th Generative Approaches to Second Language Acquisition conference. Somerville MA: Cascadilla Proceedings Project, pp. 144–56.

10.

Chung

(2013) Sources of difficulty in L2 scope judgments. Second Language Research 29: 285–310.

11.

Clahsen

Felser

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

12.

Clahsen

Gerth

Heyer

Schott

(2015) Morphology constrains native and non-native word formation in different ways: Evidence from plurals inside compounds. The Mental Lexicon 10: 53–87.

13.

Cunnings

(2017) Parsing and working memory in bilingual sentence processing. Bilingualism: Language and Cognition 20: 659–78.

14.

Demberg

(2013) Integration costs on auxiliaries? A self-paced reading study using WebExp. In: Knauff

Pauen

Sebanz

Wachsmuth

(eds) Proceedings of the 35th annual meeting of the Cognitive Science Society. Austin, TX: Cognitive Science Society, pp. 2160–65.

15.

De Neys

Schaeken

(2007) When people are more logical under cognitive load: Dual task impact on scalar implicature. Experimental Psychology 54: 128–33.

16.

Drummond

(2012) Ibex: A web interface for psycholinguistic experiments. Available at: https://github.com/addrummond/ibex (accessed February 2022).

17.

Dussias

Piñar

(2010) Effects of reading span and plausibility in the reanalysis of wh-gaps by Chinese–English second language speakers. Second Language Research 26: 443–72.

18.

Dwivedi

(2013) Interpreting quantifier scope ambiguity: Evidence of heuristic first, algorithmic second processing. PLoS ONE 8: e81461.

19.

Dwivedi

Phillips

Einagel

Baum

(2010) The neural underpinnings of semantic ambiguity and anaphora. Brain Research 1311: 93–109.

20.

Enochson

Culbertson

(2015) Collecting psycholinguistic response time data using Amazon Mechanical Turk. PLoS ONE 10: e0116946.

21.

Felser

Roberts

(2007) Processing wh-dependencies in a second language: A cross-modal priming study. Second Language Research 23: 9–36.

22.

Ferreira

(2003) The misinterpretation of noncanonical sentences. Cognitive Psychology 47: 164–203.

23.

Ferreira

Patson

(2007) The ‘good-enough’ approach to language comprehension. Language and Linguistics Compass 1: 71–83.

24.

Fine

Jaeger

(2016) The role of verb repetition in cumulative structural priming in comprehension. Journal of Experimental Psychology: Learning, Memory, and Cognition 42: 1362–76.

25.

Fox

(1998) Economy and Semantic Interpretation – A study of scope and variable binding. Unpublished PhD thesis, Massachusetts Institute of Technology, Cambridge, MA, US.

26.

Frazier

(1999) On sentence interpretation. Dordrecht: Kluwer.

27.

Fujita

Cunnings

(2020) Reanalysis and lingering misinterpretation of linguistic dependencies in native and non-native sentence comprehension. Journal of Memory and Language 115: 104154.

28.

Gerth

Otto

Nam

Felser

(2017) Strength of garden-path effects in native and non-native speakers’ processing of subject–object ambiguities. International Journal of Bilingualism 21: 125–44.

29.

Grice

(1975) Logic and conversation. In: Cole

Morgan

(eds) Syntax and semantics. New York: Academic Press, pp. 41–58.

30.

Han

Lidz

Musolino

(2007) Verb-raising and grammar competition in Korean: Evidence from negation and quantifier scope. Linguistic Inquiry 38: 1–48.

31.

Heim

Kratzer

(1998) Semantics in generative grammar. Oxford: Blackwell.

32.

Hopp

(2006) Syntactic features and reanalysis in near native processing. Second Language Research 22: 369–97.

33.

Hopp

(2010) Ultimate attainment in L2 inflectional morphology: Performance similarities between non-native and native speakers Lingua 120: 901–31.

34.

Horn

(1989) A natural history of negation. Chicago, IL: University of Chicago Press.

35.

Hoshino

Dussias

Kroll

(2010) Processing subject–verb agreement in a second language depends on proficiency. Bilingualism: Language and Cognition 13: 87–98.

36.

Jackendoff

(1972) Semantic interpretation in generative grammar. Cambridge, MA: MIT Press.

37.

Jackson

(2008) Proficiency level and the interaction of lexical and morphosyntactic information during L2 sentence processing. Language Learning 58: 875–909.

38.

Jackson

Bobb

(2009) The processing and comprehension of wh-questions among second language speakers of German. Applied Psycholinguistics 30: 603–36.

39.

Jacob

Felser

(2016) Reanalysis and semantic persistence in native and non-native garden-path recovery. Quarterly Journal of Experimental Psychology 69: 907–25.

40.

Jaeger

(2008) Categorical data analysis: Away from ANOVAs (transformation or not) and towards logit mixed models. Journal of Memory and Language 59: 434–46.

41.

Jiang

(2004) Morphological insensitivity in second language processing. Applied Psycholinguistics 25: 603–34.

42.

Juffs

Harrington

(1996) Garden path sentences and error data in second language processing research. Language Learning 46: 286–324.

43.

Kim

(2010) Korean EFL learners’ interpretation of quantifier-negation scope interaction in English. English Language and Linguistics 16: 163–83.

44.

Kirkici

Clahsen

(2013) Inflection and derivation in native and non-native language processing: Masked priming experiments on Turkish. Bilingualism: Language and Cognition 16: 776–94.

45.

Lee

(2009) Interpreting ambiguity in first and second language processing: Universal quantifiers and negation. Unpublished PhD thesis, University of Hawaii, Manoa, HI, USA.

46.

Lee

(2010) Quantifier scope in English: The view from sentence processing. English Language and Linguistics 16: 163–86.

47.

Lee

(2017) Adults’ online processing of scopally ambiguous sentences. The Journal of Studies in Language 33: 245–64.

48.

Lee

(2018) Semantic processing of scope ambiguity by Korean L2 learners of English. The Journal of Studies in Language 33: 695–713.

49.

Lee

Shin

(2016) Syntactic reanalysis and lingering misinterpretations in L2 sentence processing. Linguistic Research 33: 53–79.

50.

Leeser

Brandl

Whiteglass

(2011) Task effects in second language processing research. In: Tromovich

McDonough

(eds), Applied priming methods to L2 learning, teaching and research: Insights from psycholinguistics. Amsterdam: John Benjamins, pp. 179–98.

51.

Lim

Christianson

(2013a) Second language sentence processing in reading for comprehension and translation. Bilingualism: Language and Cognition 16: 518–37.

52.

Lim

Christianson

(2013b) Integrating meaning and structure in L1-L2 and L2-L1 translations. Second Language Research 29: 233–56.

53.

Marinis

Roberts

Felser

Clahsen

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

54.

McDonald

(2006) Beyond the critical period: Processing based explanations for poor grammaticality judgement performance by late second language learners. Journal of Memory and Language 55: 381–401.

55.

Musolino

(1998) Universal Grammar and the acquisition of semantic knowledge: An experimental investigation into the acquisition of quantifier-negation interaction in English. Unpublished PhD thesis, University of Maryland, College Park, MD, USA.

56.

Musolino

(2006) On the semantics of the subset principle. Language Learning and Development 2: 195–218.

57.

Musolino

Lidz

(2006) Why children aren’t universally successful with quantification. Linguistics 44: 817–52.

58.

Musolino

Crain

Thornton

(2000) Navigating negative quantificational space. Linguistics 38: 1–32.

59.

Myslín

Levy

(2016) Comprehension priming as rational expectation for repetition: Evidence from syntactic processing. Cognition 147: 29–56.

60.

O’Grady

(2013) The illusion of language acquisition. Linguistic Approaches to Bilingualism 3: 253–85.

61.

O’Grady

Lee

(2006) The interpretation of inverse scope in Korean. In: O’Grady

Yoon

Han

, et al. (eds) Inquiries into Korean linguistics II. Seoul: Taehaksa, pp. 369–80.

62.

O’Grady

Lee

Kwak

(2009) Emergentism and second language acquisition. In: Ritchie

Bhatia

(eds) Handbook of second language acquisition. Bingley: Emerald Press, pp. 69–88.

63.

Pozzan

Trueswell

(2016) Second language processing and revision of garden-path sentences: A visual word study. Bilingualism: Language and Cognition 19: 636–43.

64.

R Core Team (2019) R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing. Available at: https://www.R-project.org/ (accessed February 2022).

65.

Radó

Bott

(2011) Underspecified representations of scope ambiguity? In: Aloni

Kimmelman

Roelofsen

, et al. (eds) Logic, language and meaning: The 18th Amsterdam colloquium. Springer, pp. 180–89.

66.

Roberts

Felser

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

67.

Roberts

Gullberg

Indefrey

(2008) Online pronoun resolution in L2 discourse: L1 influence and general learner effects. Studies in Second Language Acquisition 30: 333–57.

68.

Sabourin

Stowe

(2008) Second language processing: When are first and second languages processed similarly? Second Language Research 24: 397–430.

69.

Sanford

Sturt

(2002) Depth of processing in language comprehension: Not noticing the evidence. Trends in Cognitive Sciences 6: 382–86.

70.

Sato

Felser

(2010). Sensitivity to morphosyntactic violations in English as a second language. Second Language 9: 101–18.

71.

Segalowitz

(2003) Automaticity and second languages. In: Doughty

Long

(eds) The handbook of second language acquisition. Oxford: Blackwell, pp. 382–88.

72.

Simmons

Nelson

Simonsohn

(2011). False-positive psychology: Undisclosed flexibility in data collection and analysis allows presenting anything as significant. Psychological Science 22: 1359–66.

73.

Slabakova

(2010) Scalar implicatures in second language acquisition. Lingua 120: 2444–62.

74.

Swets

Desmet

Clifton

Ferreira

(2008) Underspecification of syntactic ambiguities: Evidence from self-paced reading. Memory and Cognition 36: 201–16.

75.

Tan

Foltz

(2020) Task sensitivity in L2 English speakers’ syntactic processing: Evidence for good-enough processing in self-paced reading. Frontiers in Psychology 11: 575847.

76.

Tolentino

Tokowicz

(2011) Across languages, space, and time: A review of the role of cross-language similarity in L2 (morpho)syntactic processing as revealed by fMRI and ERP methods. Studies in Second Language Acquisition 33: 91–125.

77.

Van Hel

Tokowicz

(2010) ERP reveals brain potentials and second language learning: Syntactic processing in late L2 learners at different L2 proficiency levels. Second Language Research 26: 43–74.

78.

M-J

Ionin

(2019) L1-Mandarin L2-English speakers’ acquisition of English universal quantifier-negation scope. In: Brown

Dailey

(eds) Proceedings of the 43rd Boston University Conference on Language Development. Somerville, MA: Cascadilla Press, pp. 716–29.