Reanalysing object gaps during non-native sentence processing: Evidence from ERPs

Abstract

The present study used event related potentials (ERPs) to investigate how native (L1) German-speaking second-language (L2) learners of English process sentences containing filler-gap dependencies such as Bill liked the house (women) that Bob built some ornaments for __ at his workplace. Using an experimental design which allowed us to dissociate filler integration from reanalysis effects, we found that fillers which were implausible as direct objects of the embedded verb (e.g. built the women) elicited similar brain responses (an N400) in L1 and L2 speakers when the verb was encountered. This confirms findings from behavioral and eye-movement studies indicating that both L1 and L2 speakers immediately try to integrate a filler with a potential lexical licensor. L1/L2 differences were observed when subsequent sentence material signaled that the direct-object analysis was in fact incorrect, however. We found reanalysis effects, in the shape of a P600 for sentences containing fillers that were plausible direct objects only for L2 speakers, but not for the L1 group. This supports previous findings suggesting that L2 comprehenders recover from an initially plausible first analysis less easily than L1 speakers.

Keywords

English ERPs filler-gap dependency L2 processing reanalysis

I Introduction

Determining which aspects of non-native language (L2) processing are native-like and which are different is crucial for developing adequate models of L2 processing (see, for example, Clahsen and Felser, 2006; Cunnings, 2017; Kaan, 2014). For sentence-level processing, both non-canonically ordered sentences and sentences containing global or temporary ambiguities can provide useful test cases. Previous research has shown that proficient L2 comprehenders can link displaced constituents to potential lexical licensors when processing non-canonically ordered sentences (e.g. Boxell and Felser, 2017; Felser et al., 2012; Marinis et al., 2005; Williams et al., 2001) but often show measurably greater difficulty than native (L1) comprehenders correcting an initial misanalysis of ambiguous input (e.g. Dussias and Piñar, 2010; Gerth et al., 2017; Hopp, 2006; Jacob and Felser, 2016; Roberts and Felser, 2011). The current study seeks to provide converging evidence from event-related potentials (ERPs) for the hypothesis that L2 speakers use an active gap-filling strategy, consistent with prior work, and to examine whether L2 speakers have more difficulty than L1 speakers revising a gap prediction that turns out to be wrong. We investigate L2 speakers’ processing of non-canonically ordered, temporarily ambiguous sentences that require reanalysis, which allows us to examine different aspects of sentence processing within the same experimental design, thereby controlling for potentially confounding participant-level factors.

Sentences containing displaced constituents (henceforth, ‘fillers’) tend to incur greater processing costs than canonically ordered sentences as they require a dependency to be computed between the filler and its lexical licensor for the filler to be properly integrated into the emerging sentence representation. Encountering a fronted wh-phrases such as which books in (1), for example, triggers an active search for a suitable lexical licensor while the parser must simultaneously process and integrate new incoming words or phrases.

(1) The teacher asked which books her students had read __ last year.

In example (1), the only potential lexical licensor is the transitive verb read. As the verb’s canonical direct object position is indeed vacant (with underscores indicating a corresponding ‘gap’), a legitimate filler-gap dependency can be established between which books and read here. According to Clifton and Frazier’s (1989) Active Filler Strategy (AFS) comprehenders try to link a filler to the first potential gap they come across during real-time processing. The AFS is a processing economy strategy that helps ensure fast and efficient processing, but also carries the risk of comprehenders’ postulating gaps prematurely. In a sentence such as (2), for instance, the AFS will lead readers or listeners to postulate a gap when encountering the verb read, which however turns out to be a ‘false’ gap (indicated by parentheses) later on, when it becomes clear that the filler is in fact licensed by the preposition about.

(2) The teacher asked which books her students had read (__) about __ last year.

Using time-course sensitive experimental techniques can reveal at what point during processing comprehenders attempt to link a filler to a potential licensor, and many studies have used a plausibility manipulation as an experimental diagnostic. Garnsey et al. (1989) used ERPs to investigate whether readers are sensitive to the semantic plausibility of a fronted wh-phrase as a direct object of a potentially licensing verb. Comparing L1 English speakers’ brain responses to sentences like (3) which contained either a plausible (which customer) or implausible (which article) direct object filler, they observed an enhanced negative-going brain response (N400) at the verb (called) for sentences which contained an implausible filler.

(3) The business man knew which customer/article the secretary called __ at home.

The N400 is an ERP component elicited by meaningful words, and decreases when the context aids their processing (Kutas and Hillyard, 1984). A larger N400 amplitude therefore indicates increased processing difficulty in comparison to a control condition.¹ In example (3) the presence of a larger N400 in the implausible compared to the plausible condition indicates that comprehenders tried to construe the filler as the object of the verb call, encountering relatively higher processing difficulty at the verb in the implausible condition due to the verb’s low predictability and/or its semantic incompatibility with the filler. Garnsey et al.’s (1989) findings thus confirm the AFS and show that plausibility manipulations provide an effective diagnostic for filler-gap dependency formation. Other studies have shown that for sentences such as (2), where an erroneous direct object analysis must later be corrected, an initially plausible direct object analysis is more difficult to undo than an implausible one (e.g. Pickering and Traxler, 1998).

Reading-time studies investigating L2 comprehenders’ processing of filler-gap dependencies have also frequently used plausibility as a diagnostic for dependency formation. Data from self-paced reading (Williams, 2006; Williams et al., 2001) indicate that L2 learners of English from typologically different language backgrounds all apply the AFS, although depending on the nature of the task, plausibility effects may be delayed in L2 in comparison to L1 processing (Williams, 2006). However, as Williams and colleagues only examined effects of filler plausibility on reanalysis difficulty, their experimental design does not allow for effects of initial filler integration to be disentangled from reanalysis effects. This is also the case in Dussias and Piñar’s (2010) self-paced reading study examining how the plausibility of an initial direct-object analysis affects reanalysis difficulty in L1 Chinese-speaking learners of English. Their results indicate greater reanalysis difficulty in L2 compared to L1 processing.

Methods that provide a more fine-grained temporal resolution than self-paced reading may allow for different kinds of mental processes to be empirically dissociated more easily. Using eye-movement monitoring during reading, Felser et al. (2012: experiment 1) found that native German-speaking L2 learners of English showed the predicted plausibility effects at potential direct object gaps even earlier than their L1 English-speaking controls, suggesting that during more natural reading, the semantic fit between the filler and its potential licensor was evaluated immediately in L2 processing.

L2 comprehenders’ ability to link a filler to its licensor during real-time processing may also be influenced by proficiency. Dallas et al. (2013) used ERPs to investigate L1 and L2 processing of sentences like (4).

(4) The umpire asked which player/football the coach threatened __ before the game.

For sentences containing an implausible filler (e.g. which football) they found an N400 effect at the verb threatened in their L1 group but no between-condition differences in the L2 group, which was comprised of native Chinese-speaking, proficient learners of English. Only when splitting up their L2 participants into ‘high’ and ‘low’ proficiency subgroups did the authors find a small N400 effect for the most proficient participants. The lack of an N400 effect in the L2 group as a whole is surprising considering that behavioral and eye-movement results indicate that L2 comprehenders do apply the AFS, and calls for further investigation.

The above studies indicate that L2 speakers’ sensitivity to plausibility information when computing filler-gap dependencies, and their ability to recover from an erroneous initial analysis or interpretation, may be affected by various participant-level factors as well as by the experimental task, which often makes comparisons between individual studies difficult. In this research note we report the results from an ERP experiment investigating L2 speakers’ processing of filler-gap dependencies, using stimulus sentences which require reanalysis at a later point, thus allowing for integration and reanalysis effects to be empirically dissociated. Building on and extending the research discussed above, the present study examines how plausibility information affects filler integration and reanalysis in L2 in comparison to L1 sentence processing.

II Participants

Our non-native group consisted of 21 native speakers of German with English as an L2 (10 females, mean age: 25.4 years, range: 20–34 years). Our non-native participants were recruited from the student population of the University of Potsdam (Germany) and surrounding area. The L2 speakers’ general level of English grammar was assessed using the second part of the Oxford Placement Test’s (OPT; Allan, 2004) grammar section.² They achieved a mean score of 39.9 out of a possible maximum of 50 (range: 29–47), which places our non-native participants within the B2–C1 (high intermediate to advanced) proficiency range according to the Common European Framework for Reference for Language (CEFR). Our L1 German-speaking participants had started learning English at a mean age of 10.3 years (range: 6–13 years), with none of them reporting to have grown up bilingual. Twenty native speakers of English (13 females, mean age: 29.0 years, range: 18–59 years) served as our L1 control group. Fifteen of the participants lived and worked in Potsdam, and five were students at Potsdam University. All participants were right-handed, had normal or corrected to normal vision, and were offered a small fee for their participation.

III Materials

Forty-eight minimal sentence pairs were constructed as exemplified by (5a,b) which contained a filler (a relativized noun phrase) that was either plausible (5a) or implausible (5b) as a direct object of the verb in the relative clause.

(5) a. Plausible-at-V

Bill liked the house that Bob built (__) some ornaments for __ at his workplace.

b. Implausible-at-V

Bill liked the women that Bob built (__) some ornaments for __ at his workplace.

All items contained a relative clause headed by the complementizer that which modified the object of a transitive matrix verb, such as the house in (5a) or the women in (5b). All experimental sentences were globally plausible.

Manipulating the filler’s plausibility as a direct object allows us to examine whether both L1 and L2 speakers apply the AFS, and at what point filler integration is first attempted. Following earlier findings (e.g. Garnsey et al., 1989) we expect implausible direct object fillers to elicit an N400 at the verb, in comparison to plausible fillers, reflecting the processing difficulty incurred by trying to integrate the filler semantically with its lexical licensor. If L2 speakers do not attempt filler integration at the verb, they should show no N400 effects at this point (compare Dallas et al., 2013).

The next word (the quantificational determiner some in all items) still allows for a direct object analysis to be maintained as it could be part of an adverbial phrase such as some time ago. Encountering the head noun of a following noun phrase (e.g. ornaments), however, signals that the initial direct object analysis was incorrect, which is subsequently confirmed by the preposition for, the filler’s real licensor. Our critical analysis regions were (i) the embedded verb (e.g. built), (ii) the object noun (e.g. ornaments), and (iii) the preposition for, which was the same for all items. Reanalysis difficulty should be reflected in reversed plausibility effects, with sentences containing initially plausible fillers such as (5a) incurring greater processing difficulty than (5b) at the direct object NP and/or at the preposition. We expect reanalysis difficulty to be indexed by a late positive-going response (P600), an ERP component which has previously been linked to reanalysis difficulty during sentence comprehension (e.g. Osterhout and Holcomb, 1992; for review, see Gouvea et al., 2010). If initiating reanalysis takes more time, or revising an initial misanalysis is more effortful in L2 compared to L1 processing, then our non-native participants might show delayed and/or larger P600 effects than the native controls.

To help ensure that our plausibility manipulation was as controlled as possible across the two conditions, the frequencies of the manipulated nouns were taken from the British National Corpus (2007): Implausible mean = 141.63, SD = 225.34; Plausible mean = 133.52, SD = 207.84. A paired-sample t-test showed no significant between-condition difference (t(94) = 0.1832, p = 0.86). The nouns were also matched as closely as possible for length in terms of number of letters: Implausible mean = 5.62, SD = 1.63; Plausible mean = 5.44, SD = 1.65. There was no significant between-condition difference in length, either (t(94) = 0.5598, p = 0.58). To control for possible confounding effects of noun animacy, half of our experimental items contained an implausible animate and a plausible inanimate filler, and half contained an implausible inanimate and a plausible animate one.

Finally, 108 distractor items were added to the critical ones, plus five practice items. The distractor sentences included 24 grammatical and 24 ungrammatical filler-gap sentences which were part of a different study, and 60 further distractor sentences. 38 of these were grammatical and plausible and represented a range of different structures, 12 were grammatical but globally implausible (e.g. Reese saw the apartment that Mandy and her sister grew last month.), and 10 were ungrammatical filler-gap sentences. Overall there were 36 implausible sentences, 34 ungrammatical sentences, and 86 grammatically and semantically well-formed sentences. The materials were pseudo-randomized to ensure that items from the same experimental condition did not occur adjacently. They were distributed across two lists using a Latin Square design such that participants saw an equal number of trials (n = 24) from each condition and each condition was seen an equal amount overall.

IV Procedure

Participants were tested individually in a dedicated laboratory room. The experiment was presented on a 61cm wide computer monitor using Presentation version 14.9 (Neurobehavioral Systems). Sentences were presented word-by-word at a rate of 550 ms per word in black letters against a white background. Each participant saw 156 experimental sentences in total.

Overall, 52 of the trials were followed by a ‘yes–no’ comprehension question presented in green letters against white background (40-point size font), which remained on the screen until the participant answered the question by a button press. Participants received feedback either in terms of a smiley face for correct or a blank screen for incorrect answers, each presented for 1,000 ms in the center of the screen. The appearance of the questions was fully randomized between items and participants. Each item ended with a blank screen presented for 500 ms until the next item started with a fixation cross.

Participants were first given five practice trials, all followed by comprehension questions. They were asked to relax before each trial and advised to avoid eye and muscle movements until each trial was over, and were offered four breaks throughout the experiment. The experiment took about 30 minutes to complete, and an entire session normally lasted for up to 90 minutes including electrode preparation and clear-up.

1 Electrophysiological recording

The EEG was recorded using Brain Products Vision Recorder software from 31-electrode sites (FP1, FP2, F7, F3, Fz, F4, F8, FC5, FC1, FC2, FC6, T7, C3, Cz, C4, T8, TP9, CP5, CP1, CP2, CP6, TP10, P7, P3, Pz, P4, P8, PO9, O1, O2, PO10) according to the international 10–20 system using active electrodes embedded in an elastic cap (ActiCap, Brain Products). One additional electrode beneath the right eye was applied for blink detection (vertical electro-oculogram). Signals were recorded continuously with an on-line band-pass filter between 0.016–70 Hz and digitized at 2,500 Hz. Electrode impedances were kept below 20 kΩ (in accordance with the guidelines for using ActiCaps). The on-line reference electrode was FCz but electrodes were re-referenced offline to the average of the left and right mastoids. Recordings were filtered between 0.1 Hz and 30 Hz offline, and down-sampled to 250 Hz. The EEG data were processed with Brain Vision Analyser 2.

For segment analysis, first a large epoch was extracted from 200 ms before the presentation of the embedded verb (first critical word) until 1,200 ms after the presentation of the preposition for (last critical word), resulting in a 3,110 ms epoch (–200 ms to 2.910 ms). We applied a 200 ms pre-stimulus baseline correction to this time window (Tanner et al., 2016). We then segmented this larger epoch into smaller ones from 200 ms before until 1,200 ms after the presentation of each of the critical words, resulting in three 1,400 ms epochs (plus an additional epoch after the determiner some). No further baseline corrections were applied prior to the analysis of these interest regions.³

2 Data cleaning

Two participants, one from each group, had to be excluded due to technical problems. Another native German speaker was excluded because of a high rate of artifacts. To remove typical muscle and eye movement artifacts from the scalp recordings, an independent component analysis (ICA) algorithm (Infomax) was applied to the data. Epochs containing additional artifacts were identified with the semi-automatic rejection option in the Analyser, and rejected after visual inspection. In sum, 98% of all trials from the remaining participants were included in the statistical analysis.

V Results

For statistical analysis 25 electrodes were grouped into five regions of interest (ROIs): left frontal: F7, F3, FC3, FC1, C3, T7; right frontal: F8, F4, FC4, FC2, C4, T8; left posterior: CP5, CP1, P7, P3, PO9; right posterior: CP6, CP2, P8, P4, PO10; midline: Fz, Cz, Pz. We examined local ERPs at the verb, at the object noun (as the earliest point indicating that reanalysis is necessary), and at the preposition where the dependency could ultimately be resolved. We additionally examined participants’ brain responses to the determiner some, to see whether there were any signs of reanalysis already starting at this ambiguous region. Time windows and regions for statistical analysis following the critical words were chosen based on visual inspection and a 50 ms time-line analysis. ANOVAs were used for the statistical analysis, and we also calculated Cohen’s d for effect sizes.

1 Behavioral results

Participants’ average accuracy for their end-of-trial responses to comprehension questions was 83.33% (range 63.5%–92.31%) for the L1 group and 82.53% (range 65.87%–90.38%) for the L2 group. This indicates that participants were attending to the stimulus materials and actively read them for comprehension.

2 ERP results

We first examined ERP responses time-locked to the embedded verb. As can be seen in Figures 1 and 2, an enhanced negativity, most pronounced around 400 ms after word onset, was seen in the implausible condition in both groups. This negativity appeared smaller and more short-lived in the L2 compared to the L1 group and was differently distributed. Whilst in the L1 group, the negativity was distributed left-frontally, in the L2 group it was most pronounced across right-frontal electrodes. We used one-way ANOVAs with the factor Condition (plausible vs. implausible) to examine differences within individual ROIs in each group for a time window from 400–550 ms after word onset. These showed a significant between-condition difference in the left frontal ROI for the L1 group F(1,18)=4.57, p=.046 [d=0.69] and in the right frontal ROI for the L2 group F(1,18)=5.01, p=.038 [d=0.73], plus a marginal difference in the left frontal ROI in the latter group: F(1,18)=3.04, p=.09 [d=0.57].⁴ The between-group difference in the negativity’s distribution was confirmed by significant two-way interactions of ROI x Group: F(4,144)=4.56, p=.002 and ROI x Condition: F(4,144)=2.85, p=.03, and a marginal three-way interaction of Condition x ROI x Group: F(4,144)=1.99, p=.09.

Figure 1.

Grand average event related potentials at the critical verb for the second-language group for four regions of interest.

Figure 2.

Grand average event related potentials at the critical verb for the first-language group for four regions of interest.

The analysis of the ambiguous ‘buffering’ region (the determiner some) also revealed a between-condition difference from 600–850 ms after word onset. This was significant in the left frontal ROI for the L2 group: F(1,18)=6.39, p=.02 [d=0.82], with a more negative going wave form for the implausible-at-V condition. For the L1 group we found no significant between-condition differences at the determiner.

Visual inspection of the two groups’ ERPs responses to the other two critical words further downstream (the object noun and the preposition), which signaled the need for reanalysis, indicated that the expected condition effects appeared at a time window from 600–750 ms after the onset of the preposition for in the right posterior region for the L2 group only: F(1,18)=5.17, p=.036 [d=0.74], with a more positive going wave form for the plausible-at-V condition (see Figure 3). No significant differences were found for the L1 group (e.g. right posterior: F(1,18)=0.02, p=.89). There were no condition or group effects at the object noun preceding the preposition.

Figure 3.

Grand average event related potentials at the preposition for the second-language group for four regions of interest.

VI Discussion

The current study sought to investigate the presence and timing of filler integration and reanalysis effects in L1 vs. L2 sentence comprehension. Using a plausibility manipulation as an experimental diagnostic, we observed both similarities and differences between the two groups’ brain responses to sentences containing filler-gap dependencies.

Let us first consider effects of initial filler integration. Both groups patterned similarly in that implausible direct object fillers elicited a negative-going wave at the embedded verb within the time window typical of N400 effects. There was little difference between the two groups regarding the point in time at which this negativity started to emerge. Clearer differences between the participant groups were seen in the negativity’s distribution and duration, however: It was located left frontally and lasted from 300–650 ms after word onset in the L1 group, and was most pronounced right frontally, and in a somewhat shorter time window from 400–550 ms after word onset, in the L2 group.

Although the scalp distribution was slightly different for the L1 and the L2 group, this difference does not seem substantial enough to conclude that the negativities we observed reflect different ERP components or index different mental processes. Despite its somewhat unusual distribution (especially for the L1 group), we interpret this negativity as an N400. Note that Dallas et al. (2013) also found a more frontally located N400 for implausible objects in their second set of experimental items, containing whether instead of which NP (e.g. The referee asked whether the team threatened the player/football before the event). Other studies report neural activity in middle and superior temporal areas, inferior temporal areas, and prefrontal areas accompanying the occurrence of the N400 (see Kutas and Federmeier, 2011, for review). Finding plausibility effects at the verb in both participant groups replicates and extends Garnsey et al.’s (1989) original finding by showing that both native and non-native speakers of English evaluate the filler’s semantic fit immediately after encountering a potential subcategorizer as predicted by the AFS (Clifton and Frazier, 1989). The effects we observed at the verb are also consistent with previous ERP studies that have found native-like N400 brain responses to semantic anomalies in L2 sentence processing (e.g. Ardal et al., 1990; Hahne and Friederici, 2001) and with the results from the behavioral and eye-movement studies on filler-gap dependencies discussed above (e.g. Felser et al., 2012; Williams, 2006; Williams et al., 2001).

Unlike Dallas et al. (2013), who found evidence of an N400 response to implausible object fillers in sentences containing filler-gap dependencies to be restricted to a highly proficient subset of their L2 participants, we observed this effect in our L2 group as a whole. One possible reason for this discrepancy is the fact that our L2 group was comprised of native German speakers, whose L1 is closely related to English typologically and also has overt wh-movement, whereas Dallas et al. (2013) tested native speakers of Chinese, a typologically more distant wh-in-situ language. An explanation in terms of L1 influence is called into question, however, by Williams et al.’s (2001) finding of L1 Chinese, Korean and German-speaking learners of English all showing similar reading profiles for locally plausible vs. implausible filler-gap sentences (see also Cunnings et al., 2010). Instead, it seems possible that our L2 participants were somewhat more proficient than those tested by Dallas and colleagues.⁵

The effect we found at the post-verbal determiner some for the L2 group is likely to be a spillover effect from the N400 effect that we observed at the critical verb. Like the N400, and unlike the positivity that we saw later on at the preposition (the filler’s real lexical licensor), this effect had a frontal rather than a posterior distribution. A closer look at the brain waves (Figure 1) seems to confirm this assumption: The negativity for the implausible-at-V condition, although only marginally significant in the left frontal region, is still visible even at 1,200 ms after the onset of the verb, which is approximately 650 ms after the onset of the determiner. This continued negativity might indicate that L2 speakers need more time to process the semantic incongruency between the filler and its presumed lexical licensor.

Regarding effects at the disambiguating region, only the L2 group showed the expected reverse plausibility effects at later sentence regions, where the real licensor of the filler was presented, in the shape of a positive-going wave in a time window typical of the P600 component. The L1 group did not show any effects beyond the N400 at the critical verb reported above.

Reverse plausibility effects at disambiguating sentence regions are typically found where an initial misanalysis was highly plausible. This is likely due to the fact that comprehenders are more committed to an initially plausible analysis than to an implausible one, and thus less willing to abandon it in the face of new ‘evidence’ (e.g. Pickering and Traxler, 1998). Our L1 speakers appeared to be able to perform the necessary reanalysis in a highly automatized manner which did not give rise to any significant condition differences. The L2 group, in contrast, showed an enhanced positivity at posterior electrodes in a time window from 600–750 ms after the onset of the preposition for for the plausible-at-V compared to the formerly implausible condition. This finding is in line with previous findings reported by Jessen et al. (2017), who observed an enhanced (compared to native speakers) P600 response to filled indirect object gaps in L2 speakers of English.

Although P600 effects may reflect reanalysis processes, they have also been linked to other processes and to greater sentence complexity (for review, see Gouvea et al., 2010). Note that in the absence of systematic comprehension data we cannot be entirely sure what the P600 represents, or whether reanalysis was in fact accomplished successfully.

Together, these findings indicate that our L2 speakers found it more effortful to initiate or carry out reanalysis than did the native speakers. This is consistent with earlier findings showing that recovering from an initial misanalysis or misinterpretation is more difficult in L2 than in L1 comprehension (e.g. Dussias and Piñar, 2010). The fact that the initial plausibility effect was significant in a shorter time window in the L2 than in the L1 group at the verb raises the question of why our plausibility manipulation should nevertheless have had such a strong effect on reanalysis later on in the L2 group, however. Considering that the L2 group showed a possible spillover effect later (at the determiner), it is conceivable that the plausibility effect we saw in this group was not in fact more short-lived compared to native speakers’ brain responses, but longer lasting and with a delayed onset (see note 4). This would indicate higher processing costs for L2 speakers at both positions, the verb and the subsequent preposition.

On the assumption that the N400 may also index lexical prediction, the less pronounced and slightly delayed effect at the verb in the L2 group might be due to L2 speakers’ making less strong predictions about upcoming words during processing than L1 speakers (e.g. Martin et al., 2013). As a reduced prediction ability does not imply reduced sensitivity to pragmatic or semantic incongruity, however, it should be possible for comprehenders to show strong effects when forced to reanalyse what seemed a perfectly plausible initial analysis, regardless of how strongly they reacted to an unexpected verb before.

VII Concluding remarks

Using an experimental design that allowed for filler integration and reanalysis effects to be dissociated, we found ERP evidence showing that both L1 and L2 speakers evaluate the semantic fit between a filler and a potential lexical licensor as soon as the latter is encountered, but that L2 speakers find it harder than L1 speakers to revise an initially plausible misanalysis. Our results indicate that L1 and L2 speakers initiate the same kind of mental processes when parsing and reanalysing filler-gap dependencies but differ regarding the processing cost of reanalysis. Our findings also demonstrate that simply asking whether or not L2 filler-gap processing is native-like is misguided, since the answer to this question may differ depending on which particular aspect of processing is being investigated. Using highly time-course sensitive experimental techniques such as eye-movement monitoring or ERPs can help reveal subtle L1/L2 processing differences that may be obscured in behavioral tasks.

Footnotes

Acknowledgements

We thank Oliver Boxell for help with the stimulus creation, Julia Festman and Jennifer Meyer for programming assistance, and Caroline Beese for help with the data collection.

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 was supported by an Alexander-von-Humboldt Professorship awarded to Harald Clahsen.

Notes

References

Allan

(2004) Oxford Placement Test 2: Test pack. Oxford: Oxford University Press.

Ardal

Donald

Meuter

Muldrew

Luce

(1990) Brain responses to semantic incongruity in bilinguals. Brain and Language 39: 187–205.

Boxell

Felser

(2017) Sensitivity to parasitic gaps inside subject islands in native and non-native sentence processing. Bilingualism: Language and Cognition 20: 494–511.

British National Corpus (2007) British National Corpus: version 3 (BNC XML edition). Distributed by Oxford University Computing Services on behalf of the BNC Consortium. Available at: http://www.natcorp.ox.ac.uk (accessed January 2018).

Clahsen

Felser

(2006) How native-like is non-native language processing? Trends in Cognitive Sciences 10: 564–70.

Clifton

Jr. Frazier

(1989) Comprehending sentences with long-distance dependencies. In: Carlson

Tanenhaus

(eds) Linguistic structure in language processing. Dordrecht: Kluwer, pp. 273–31.

Cunnings

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

Cunnings

Batterham

Felser

Clahsen

(2010) Constraints on L2 learners’ processing of wh-dependencies. In: VanPatten

Jegerski

(eds) Research in second language processing and parsing. Amsterdam: John Benjamins, pp. 87–112.

Dallas

DeDe

Nicol

(2013) An event-related potential (ERP) investigation of filler-gap processing in native and second language speakers. Language Learning 63: 766–99.

10.

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.

11.

Felser

Cunnings

Batterham

Clahsen

(2012) The timing of island effects in nonnative sentence processing. Studies in Second Language Acquisition 34: 67–98.

12.

Garnsey

Tanenhaus

Chapman

(1989) Evoked potentials and the study of sentence comprehension. Journal of Psycholinguistic Research 18: 51–60.

13.

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.

14.

Gouvea

Phillips

Kazanina

Poeppel

(2010) The linguistic processes underlying the P600. Language and Cognitive Processes 25: 149–88.

15.

Hahne

Friederici

(2001) Processing a second language: Late learners’ comprehension mechanisms as revealed by event-related brain potentials. Bilingualism: Language and Cognition 4: 123–41.

16.

Hopp

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

17.

Jacob

Felser

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

18.

Jessen

Festman

Boxell

Felser

(2017) Native and non-native speakers’ brain responses to filled indirect object gaps. Journal of Psycholinguistic Research 46: 1319–38.

19.

Kaan

(2014) Predictive sentence processing in L1 and L2. What’s different? Linguistic Approaches to Bilingualism 4: 257–82.

20.

Kutas

Federmeier

(2011) Thirty years and counting: Finding meaning in the N400 component of the event-related brain potential (ERP). Annual Review of Psychology 62: 621–47.

21.

Kutas

Hillyard

(1984) Brain potentials during reading reflect word expectancy and semantic association. Nature 307: 161–63.

22.

Lau

Holcomb

Kuperberg

(2013) Dissociating N400 effects of prediction from association in single word contexts. Journal of Cognitive Neuroscience 25: 484–502.

23.

Marinis

Roberts

Felser

Clahsen

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

24.

Martin

Thierry

Kuipers

Boutonnet

Foucart

Costa

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

25.

Molinaro

Carreiras

Duñabeitia

(2012) Semantic combinatorial processing of non-anomalous expressions. Neuroimage 59: 3488–3501.

26.

Osterhout

Holcomb

(1992) Event-related brain potentials elicited by syntactic anomaly. Journal of Memory and Language 31: 785–806.

27.

Pickering

Traxler

(1998) Plausibility and recovery from garden paths: An eye-tracking study. Journal of Experimental Psychology: Learning, Memory, and Cognition 24: 940–61.

28.

Roberts

Felser

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

29.

Tanner

Norton

Morgan-Short

Luck

(2016) On high pass filter artifacts (they’re real) and baseline correction (it’s a good idea) in ERP/ERMF analysis. Journal of Neuroscience Methods 266: 166–70.

30.

Williams

(2006) Incremental interpretation in second language sentence processing. Bilingualism: Language and Cognition 9: 71–88.

31.

Williams

Möbius

Kim

(2001) Native and non-native processing of English wh-questions: Parsing strategies and plausibility constraints. Applied Psycholinguistics 22: 509–40.