L2 processing of sentences with or without filler-gap dependencies: an ERP evidence

Abstract

The present study investigated how L2 learners process sentences with filler-gap (object relative clauses) or without filler-gap dependencies (appositive clauses) via ERP experiments. By recording 26 Chinese EFL learners’ electrophysical responses in comprehending two types of clauses differing in one segment, the study revealed varied EEG responses in the time window of 300–500 ms. The results showed that in comparison with object relative clauses, appositive clauses elicited a more negative N400. It indicates that in comparison with clauses with filler-gaps, clauses with relativizers but no filler-gaps elicit higher difficulty in meaning processing and possible integration. However, the two types of clauses do not display significant differences in other ERP components, i.e., P600. The findings suggest that L2 learners adopt the Direct Association Strategy in complex sentence processing, and meanwhile the difficulty led by the non-filler-gap sentence is attributed to the semantic integration, rather than the syntactic unexpectation. It thus contributes to the understanding of L2 learners’ processing strategies and the underlying reasons from separated semantic and syntactic levels.

Introduction

In the domain of L2 complex sentence processing, there have been mixed results: some found that L2 learners can effectively use L1-like strategies, such as Active Filler Strategy (AFS for short, e.g., Williams et al., 2002); some argue that L2 learners can only process shallow structures (Clahsen & Felser, 2006a & b), and are lack of capabilities in deep structure processing. These studies mainly involved sentences with long-distance filler-gap dependencies, such as subject relative clauses (SRCs for short) and object relative clauses (ORCs for short), via self-paced-reading experiments. So far, studies of appositive clause (AC for short) processing are scantier compared to that of relative clauses (RCs for short). The reason may partly be attributed to the linguists’ over-generalization of some similar structures. For example, Jacobs and Rosenbaum (1968: 199) classified ACs, RCs and nominal clauses as “embedded sentences”. Similarly, Radford (1981) regarded both RCs and ACs (“noun complements” in his expressions) as cases of complex NPs. The prevailing tendency to conflate RCs and ACs in existing research has resulted in insufficient examination of their distinct online processing mechanisms, particularly regarding how pseudo versus explicit filler-gap dependencies are resolved during real-time comprehension. In spite of rare explorations to the processing of ACs, syntacticians have provided detailed syntactic descriptions, which reveal the pseudo filler-gap dependency structure of ACs. For instance,

1. The possibility that worries him was discarded. (SRC)

2. The idea that the earth revolves around the sun was quite revolutionary. (AC)

3. *That worries him was the discarded possibility. (restructured SRC)

4. That the earth revolves around the sun was quite a revolutionary idea. (restructured AC). (Aldwayan, 2012: 96)

Aldwayan (2012) listed an SRC and an AC (e.g., 1a, b), and further analyzed the semantic relation between the head word and the following clause, that is, the grammatical and logical relation between them. As above, an AC can be reconstructed by the structure of “that + the clause + be + complement” (e.g., 1d) while the relative clause can’t be restructured as this (e.g., 1c). The equivalent syntactic alternative reveals that the clause of an AC is an independent and complete sentence, which denies the filler-gap dependency.

As for the processing of filler-gap dependency, the AFS (Frazier, 1987) has assumed that the parser will predict a gap when processing (potentially) RCs, and the filler which is retained in working memory will be later reactivated to create dependency when the gap is encountered along the linear processing. The strategy claims that the parser attempts to assign it to the linearly closest gap since the sooner the dependency is established, the sooner the cognitive resource is released. Namely, in RCs, when people identify a filler (e.g., head noun as a filler) and a relativizer (e.g., that), they immediately predict the gap position and then construct the filler-gap association once the gap position is caught. Whereas, ACs show the same structure of “head noun + that + NP + V.” but a pseudo filler-gap dependency, and the co-occurrence of candidate nouns (filler vs. object of the clause verb) which contradicts the initial prediction, will lead to “surprise effect” and reanalysis. Thus, the contrast of the processing of ACs and ORCs reflects the processing difference between explicit and pseudo filler-gap dependency structures. Although AFS gives some explanations for filler-gap dependency processing, it doesn’t make it clear whether or how semantic cues are integrated to resolve the filler-gap dependency. To this question, the Direct Association Hypothesis (DAH for short, Gorrell, 1993) provides an alternative account. The hypothesis claims that the filler-gap dependency establishment is a lexically driven process triggered by the automatic mental reconstruction of the subcategorizer’s argument structure when the verb is encountered (Pickering, 2001; Pickering & Barry, 1991; Sag & Fodor, 1994). That is to say, the dependency is directly associated with the verb rather than the gap; therefore, there is little necessity to analyze the deep structures. To sum up, the DAH believes that filler-gap dependency processing is lexically-driven while the AFS is syntactically-driven.

Previous studies have shown that L2 learners have difficulty in distinguishing ACs and RCs due to similarity in the surface structure (Deng, 2013: 209; Fan & Wei, 2002: 84; Zheng & Zhou, 2018: 54; Tang & Wu, 2025). Cross-linguistic investigations (Deng, 2013; Fan & Wei, 2002) to ACs mainly revolved around the contrast between ACs and RCs, such as the translation of them based on the syntactic analyses. The elucidation for L2 learners’ processing mechanism of ACs is obviously insufficient. It is noteworthy that Zheng and Zhou (2018) compared the processing of ACs and ORCs, finding that Chinese EFL learners showed a processing advantage for ORCs with less response times but the limitation is that they also found significant lexical differences between the noun phrases and the preposition phrases (the critical segments for ACs and ORCs, respectively), which means the differences between processing speed might be attributed to the lexical factors rather than syntactic differences. In general, studies using online measures suggest that L1 and L2 speakers may show similar processing patterns to filler-gap sentences, yet it is still unclear about the underneath rationale. Therefore, it is worth exploring what L2 learners’ neuro-mechanism is when processing sentences with or without filler-gap dependencies.

Brain activities in processing filler-gap dependencies

Filler-gap dependency, in essence, reflects syntactic displacement in language processing. Some research has investigated how syntactic displacements are processed by native speakers in different languages such as English (Phillips & Wagers, 2007), Japanese (Ueno & Kluender, 2009), German (Fiebach, Schlesewsky & Friederici, 2001) and Chinese (Hsiao & Gibson, 2004; Liu & Jiang, 2016). These studies looked into diverse structures with syntactic displacements such as passive voice (Liu & Jiang, 2016), relative clauses (Fiebach, Schlesewsky & Friederici, 2001; Hsiao & Gibson, 2004; Phillips & Wagers, 2007) and wh-dependency sentences (Felser et al., 2003; Gouvea et al., 2010; Phillips et al., 2005).

In L1 processing of filler-gap dependencies, the Left Anterior Negative (LAN for short) is sometimes bilaterally distributed (therefore sometimes “AN” for short) and elicited at 300–500 ms after the onset of the displaced constituent or its following position. This component is reported to be associated with the storage of a filler in working memory and its subsequent retrieval of filler-gap assignment (Kluender & Kutas, 1993), and the evidence comes from the contrast in electrophysical responses of L1 speakers for sentences such as (2a, b).

1. wh-question: What _i have you forgotten _ti if he dragged her to that weekend?

2. yes/no-question: Have you forgotten if he dragged her to a movie that weekend?. (Kluender & Kutas, 1993: 198)

In this pair, “you” in (2a) elicited an enhanced LAN in comparison with the equivalent pronoun in (2b), and the difference is attributed to storing of fronted wh-phrase (e.g., “what” in 2a) in working memory. The LAN was also found in Chinese passive voice such as (3a), in which the object (e.g., “客户” as a filler) was fronted so that the long-distance filler-gap dependency was generated. Liu and Jiang (2016) found that both the VP₁ (e.g., “指使”) and VP₂ (e.g., “赶走了”) incur a larger negativity at bilateral anterior sites (referred as AN) relative to the gap-free control sentences, and the former AN is interpreted as a result of incorrect assignment of the filler and the VP₁ while the latter one demonstrates the reactivation of the filler in working memory. A similar finding has been reported for filler-gap dependency processing (Fleser et al., 2003; Hestvik et al., 2007).

1. 客户i/被/老板/指使(VP₁)/员工/赶走了(VP₂) _ti 。

The customer/(passive marker)/the boss/ordered/the employee/was driven away.

The boss ordered the employee to get the customer driven away. (Liu & Jiang, 2016: 615).

During the time window of LAN, there is another ERP component which is known as N400. In spite of the overlapping on the time window for the two components, existing evidence showed that the N400 is typically largest over the right hemisphere and over posterior sites (Kutas & Federmeier, 2011). Beyond the conventional interpretation of N400 as an index of semantic violation, recent studies have associated the N400 effect with multiple cognitive processes, including lexical retrieval (Delogu, Brouwer, & Crocker, 2019), semantic integration (Brouwer et al., 2016), and cloze probability (DeLong, Urbach, & Kutas, 2005). It is now understood that all the content words elicit an N400 with its amplitude varying inversely with the cloze probability of the word (DeLong, Urbach & Kutas, 2005). In filler-gap dependency processing, the N400 indexes the cognitive effort of retrieval of word meaning from long-term memory (Friederici et al., 1993, 1996 & 2002). Phillips and his colleagues (2005) observed a larger N400 at the position of the intermediate verb (e. g. “hoped” in 4a) in long-distance wh-dependency sentences, and they argued that the N400 is associated with lexical predictive processes and priming. A sequence like the italic part in (4a) may create a strong expectation for an upcoming transitive verb, relative to the control condition that contains the clausal-object construction italicized in (4b).

1. The lieutenant knew which accomplice _i (the detective hoped that) the shrew witness would recognize _ti in the lineup.

2. The lieutenant knew (that the detectivehoped that) the shrew witness would recognize the accomplice in the lineup. (Phillips et al., 2005: 410)

A different ERP component, which is known as P600 (a positive deflection peaking at 500–700 ms after the stimulus onset), is also claimed closely related to the filler-gap dependency processing. Kann et al. (2000) compared the wh-dependency sentences and gap-free sentences (e.g., 5a, b), showing that P600 is observed at the verb position (e.g., “initiated”) which allows the completion of the dependency. Kaan et al. suggested that the P600 response elicited at the completion of a wh-dependency reflects the processing cost associated with syntactic integration of the wh-phrase (e.g., “which pop star”) and the verb, and therefore concluded the positive correlation between the amplitude of the P600 and the syntactic integration difficulty.

1. Emily wondered which pop star _i the performer in the concert had imitated_ti for the audience’s amusement.

2. Emily wondered whether pop star the performer in the concert had imitated for the audience’s amusement. (Kann et al., 2000)

Friederici et al.’s (1993, 1996 & 2002) 3-phase model for language comprehension provided a holistic picture for the filler-gap dependency processing. It is claimed that language comprehension include three phases: in the first phase, normally 100–200 ms after the phonetic analysis, early syntactic structure building is processed (i.e., word class recognition), and the violation occurred in this phase elicits eLAN (Friederici, Pfeifer & Hahne, 1993; Friederici, Hahne & Mecklinger, 1996); Next, within the time region of 300–500 ms, processes of lexical-semantic and verb-argument structure information take place. This phase manifests neurophysiologically as N400 for argument structure violations (Kutas & Federmeier, 2000; Lau, Phillips & Poeppel, 2008) and as LAN for morphosyntactic argument violations (Friederici, Pfeifer & Hahne, 1993); Finally, the third phase, in which the syntactic reanalysis or repair may be involved, closely associates with the P600 which is evoked by the phrase structure assignment errors or complex sentence structures such as sentences with garden path (Hagoort et al., 1993 & 1999; Osterhout & Holcomb, 1992 & 1993) and long-distance filler-gap dependencies (Kaan et al., 2000). This model can be mapped to specific ERP components evoked during the parsing operations of filler-gap dependencies. For instance, Hestvik et al. (2007) designed ungrammatical sentences by adding another object (e.g., 6a), in which an NP (“the camel”) is inserted in the gap. The 6a elicited an eLAN in the verb (kissed) when compared to the 6b, showing syntactic building in the early phase, which is attributed to the predicted gap. The absence of the N400 (e.g., “on the nose” in 6a) suggests that the filler-gap dependency is prematurely established, and the researchers resorted the absence of P600 in 6a to no available reanalysis (e.g., 6c) for ungrammatical sentences.

1. *The zebra that the hippo kissed the camel on the nose ran far away.

2. The zebra said that the hippo kissed the camel on the nose ran far away.

3. The zebra _i that the hippo kissed on the nose for _ti ran far away. (Hestvik et al., 2007)

However, in L2 processing, it is unclear if L2 learners have the similar brain activities in processing filler-gap dependency. Some studies (Dallas et al., 2013; Wang et al., 2015) compared the ORCs and SRCs & with the focus on the effect of filler-gap distance on processing. Wang et al. (2015) compared the electrophysical responses of three segments of ORCs and SRCs which are partitioned by slashes in the brackets (e.g., 7a, b), finding that the greater P600 was elicited by ORCs. The results suggested that L2 learners have processing advantage for SRCs, which is consistent with L1 processing performance. However, the processing difficulty differentiating between such filler-gap dependency structures is far from clear since the structures differ at least in two aspects: first, they contain a different argument order; second, the filler-gap distance for ORCs is longer than that for SRCs (Fiebach, Schlesewsky & Friederici, 2001). Jessen and Felser (2019) investigated how English native speakers and German EFL learners process sentences with filler-gap dependencies (7c), and found that the implausible direct object condition (built some women) elicited a larger N400 in L1 and L2 speakers, suggesting that L2 learners may immediately integrate a filler with the potential gap as native speakers. However, it is noted that, compared to L1 speakers, the N400 found in L2 learners is most pronounced right frontally, and within a somewhat shorter time window from 400 to 550 ms after the word onset. There are also differences in P600, that is, L2 learners showed an enhanced P600 after the onset of disambiguating preposition (for), indicating that it is harder for L2 learners to revise an initially plausible misanalysis (built the women). However, these studies did not involve a comparison of processing mechanism through the contrast of filler-gap dependency structures and gap-free structures.

1. The banker_i that _ti  (irritated/ the lawyer/ met) the priest and talked a lot.

2. The banker_i that (the lawyer/ irritated _ti / met) the priest and talked a lot. (Wang et al., 2015: 10)

3. Bill like the house/women that Bob built some ornaments for at his workplace. (Jessen & Felser, 2019)

To sum up, both in the studies of L1 processing and that of L2 processing, (e) LAN, N400 and P600 have been considered to reflect the filler-gap dependency processing procedure. With the limited amounts of neuro studies on L2 processing of filler-gap dependencies, most previous literature found that L2 brain activities differ from that of L1 speakers to some extent, reflected in ERP components. However, these studies mainly focused on the plausibility of syntax or semantics within the filler-gap sentences, rather than a direct comparison of filler-gap dependency structures and pseudo filler-gap dependency sentences. The current study aims to fill this research gap by investigating Chinese EFL learners’ processing of ORCs and ACs via ERP experiments, and explore the components, LAN, N400 and P600.

The current study

The current study applies an ERP experiment to record Chinese EFL learners’ brain activity in online processing of ORCs and ACs, aiming to explore how they process the filler-gap dependency structure. To this end, this paper adopts pairs such as (8a, b), which are split into 7 segments (S1-7) according to the syntactic functions. For the stimuli sentences, the frequency and difficulty of lexicons were carefully controlled by a selection of vocabulary from the National full-time senior high school English syllabus (Revised) (2004). The word length in each segment was also controlled, including S5. At this critical segment, there are prepositional phrases (PPs) in the ORCs but noun phrases (NPs) in the ACs. The length of the phrases at S5 does not report a significant difference (t (29) = − 1.235, p = 0.227 > 0.05). The slight morphological difference is normally believed not to affect the analysis of late components of EEG signals (N400, P600, LAN). To further reveal the syntactic processing difference in spill-over effect, responses of the next segment (e.g., S6, the verb of the main sentence, “is”) were also recorded. The stimuli were coded by E-prime 3.0. SPSS25.0 was used for statistical analyses.

1. The news _i / that/ they/ heard _ti / from radios/ is/ delightful. (ORC).

   S1 S2 S3 S4 S5 S6 S7.

2. The news _i / that/ they/ heard _(ti) / some voices/ is/ delightful. (AC).

   S1 S2 S3 S4 S5 S6 S7.

To clearly reveal the difference of 8a-b in the deep structure, we illustrate the syntactic diagrams (see Fig. 1, 2). In ORC, the object of the verb (e.g., “heard” in 8a) leaves a trace in place (t_i) as it moves to the head “news”, and this displacement does not exist in AC. In AC, the clause, serving to explain or supplement, has a complete structure and meaning, and therefore there is an appositive relation between the filler and the clause.

[See PDF for image]

Fig. 1

The syntactic diagram of 8a

[See PDF for image]

Fig. 2

The syntactic diagram of 8b

Participants

According to the calculation of G*Power, the present study needs at least 12 participants for an adequate effect size (f = 0.4) and high statistical power (1−β = 0.8) (Cohen, 1992). Totally 26 college students from a university in Beijing volunteered for this experiment. The participants learned English at the age of 10 or 11, and they have learned English for 11–15 years (M = 13.10, SD = 2.02). Before the experiment, all participants were required to do the grammar test in Oxford Placement Test (Allan 2004), and their scores ranged from 30 to 41 (total score 50; M = 34.38, SD = 3.68), showing that their English proficiency is at the level of intermediate to advanced proficiency. All the participants are right-handed and have normal or corrected-to-normal vision. In addition, none of them have any brain or mental disorders. They all signed informed consent forms before the experiment and were remunerated for their participation.

Stimuli and procedure

There were 60 pairs of stimuli and 60 fillers (180 trials in total) in this experiment. Each pair contains an ORC and an AC (e.g., 8a, b). An English native speaker was invited to assess the naturalness and plausibility of sentences and made necessary revisions. Then acceptability and familiarity of amended stimuli were assessed among 42 Chinese EFL learners who did not participate in the online experiment by 1–5 scale questionnaires (1 = absolutely unfamiliar/ unacceptable, 5 = absolutely familiar/ acceptable). The statistical results reported the insignificant differences between ORCs and ACs in terms of acceptability (t (19) = 1.528, p = 0.143 > 0.05) and familiarity (t (21) = 1.347, p = 0.192 > 0.05). In addition, some simple questions (e.g., 9a-b), which are semantically corresponded to stimuli sentences were prepared to check participants’ comprehension.

1. They heard the news from TV.

2. They did not hear the voices.

The ERPs experiment was completed in a quiet room. During the experiment, participants were first presented a fixation “ + ”, as a reminder for the presence of stimuli materials. The sentences were presented segment by segment, and the duration for segments was controlled. Between each two segments, there was a blank screen. Before the formal experiment, we conducted three pre-tests to make sure whether the epoch time is reasonable and devices work normally. The data of pre-tests were not included in the final statistical analyses. Based on feedback from the pre-tests, the epoch time for each stimuli segment is set up to 800 ms and the blank screen lasts for 200 ms. All sentences were presented in a randomized order, and half of them are followed by a semantically related comprehension sentence for consistency judgement.

EEG recording & analysis

Data recording

The EEG signals were continuously collected and synchronously digitized at the sample rate of 1000 Hz by using MCScap-ACs64 (AF3, AF4, AF7, AF8, F1, F2, F3, F4, F5, F6, F7, F8, FZ, FP1, FP2, FPZ, FT7, FT8, FT9, FT10, FC1, FC2, FC3, FC4, FC5, FC6, FCZ, T7, T8, C1, C2, C3, C4, C5, C6, CZ, TP7, TP8, TP9, TP10, CP1, CP2, CP3, CP4, CP5, CP6, CPZ, P1, P2, P3, P4, P5, P6, P7, P8, PZ, PO3, PO4, PO7, PO8, POZ, O1, OZ, O2) and NVX-64 digital DC EEG amplifier, and finally transmitted to NeuroRec system for further process. The signals were filtered online applying a band-pass of 0.05–300 Hz and referenced online to the both earlobes. For electrodes, the impedance was kept below 20 kΩ. The extracted ERP epoch ranged from pre-stimulus 200 to 1000 ms post-stimulus.

Data cleaning

EEG data were further analyzed by the software WinEEG. In order to ensure the objectivity and reliability of EEG data, the signals of those participants who had the accuracy lower than 50% were eliminated. The selected data was first browsed to eliminate large drifts and delete bad channels if any, and then offline filtered with a band-pass of 0.053–30 Hz. Eye blinks, movements and power noise were successfully removed from the original EEG signal with the ICA method. When the epoched data were processed to reject artifacts, trials with the amplitude outside ± 100 μV were excluded with the baseline correction manipulated in reference to the 200 ms pre-stimulus voltage amplitude. Following this, waveforms were averaged across all trials per condition except for those that rejected more than 50% of trials (less than 30 trials per condition). WinEEG outputs the amplitude data for each participant in specified time windows. All the procedures together with the data could be accessed on the project website https://osf.io/czeyt/?view_only=06331fabff3642518a4a27a6ebb693c5.

Results

Among the 26 participants, 5 participants’ data were invalid due to high rejection rates for trials (more than 50%). Therefore, there were only 21 sets of valid data used in the final test (13 males and 8 females, aged between 23 to 32 years old with the average age of 25.1).

Accuracy

The accuracy rates can be seen from Table 1 below. Participants’ average accuracies for their end-of-trial responses to comprehension questions were 73.81% (range 69.23–92.31%) and 77.73% (range 64.71–94.12%) respectively for ORC and AC conditions. Paired samples t-tests revealed no significant difference in comprehension accuracy between the two types of stimulus sentences (t (20) = −1.603, p = 0.128 > 0.05).

Table 1. The average accuracy of sentence comprehension for ORCs and ACs

	ORCs	ACs	General
Accuracy (%)	73.81	77.73	75.77

ERP results

We examined the ERP effects at S5 (the critical phrases) and S6 (the subsequent word following critical phrases). The average amplitudes within time windows of 300–500 ms for N400 and LAN, and 500–700 ms for P600 were further statistically analyzed. For statistical analysis, 9 electrodes were grouped on the basis of Laterality (Left hemisphere: F3/C3/P3; middle-line: Fz/Cz/Pz; Right hemisphere: F4/C4/P4) and Site (Frontal: F3/Fz/F4; Central: C3/Cz/C4; Parietal: P3/Pz/P4).

The ERP results at S5 (indicator of clause type)

As shown in Fig. 3, both the ORCs and ACs had negative-going and positive-going deflections, accompanied with the peaks at around 400 and 600 ms, within the LAN, N400 and P600 time window respectively.

[See PDF for image]

Fig. 3

The grand average waveforms within 300–500 ms for ORCs and ACs at S5, respectively

N400

In Fig. 3, the visible inspection only showed a tiny difference in the amplitude between ORCs and ACs in the time window of 300–500 ms, which was assessed by the repeated measures ANOVAs with Sentence type (ORCs/ACs) * Laterality (Left/Middle-line/Right) * Site (Frontal/Central/Parietal). A possible effect was quantified by average amplitude for F3/z/4, C3/z/4 and P3/z/4, and the figures at these electrodes can be seen from Table 2.

Table 2. The average amplitude of the nine electrodes at S5 within 300-500 ms time window for ORCs and ACs

Interaction effect of ST *site		Post hoc of site
	F (2, 19) (Sig.)	Site	Mean amplitude (μV)		MDORC-AC (μV) (Sig.)
			ORC	AC
N400 at S5	3.733 (*0.043 < 0.05)	Frontal	1.562	− 0.019	1.581 (**0.009)
		Central	0.203	0.003	0.200 (0.592)
		Parietal	0.638	0.896	− 0.258 (0.543)

The ANOVAs reported that the interaction effect of Sentence type*Laterality showed no significance for the average amplitude within 300-500 ms time window (F (2, 19) = 1.039, p = .373 > 0.05, partial η² = 0.099). Sentence type interacted with Site (F (2, 19) = 3.733, p₂ = 0.043 < 0.05, partial η² = 0.282), showing significant Site effects (ACs minus ORCs: MD₁ = − 1.581 μV, p₁ = 0.009 < 0.01, MD₂ = − 0.200 μV, p₂ = 0.592 > 0.05 and MD₃ = 0.258 μV, p₃ = 0.543 > 0.05 at frontal, central and parietal sites, respectively). The results revealed main effects of Laterality (F (2, 19) = 9.670, p = 0.001 < 0.01, partial η2 = 0.504) and Site (F (2, 19) = 4.668, p = 0.022 < 0.05, partial η2 = 0.329). However, there were no significant main effects of Sentence type (F (1, 20) = 3.004, p = 0.098 > 0.05, partial η2 = 0.131). We tentatively speculate that the larger negativity within the 300–500 ms time window over the right frontal electrode is not the LAN effect since the distribution of this effect deviate from the typical LAN or AN. It is likely to be the N400 effect which coincides with Jessen and Felser’s (2019) finding, and this negativity’s distribution seem to be specific to L2 learners.

P600

The Fig. 4 demonstrates the average amplitude within the time window of 500–700 ms, the possible P600 was analyzed by measuring the average amplitude for the same nine electrodes with the three-factor repeated measures ANOVAs.

[See PDF for image]

Fig. 4

The grand average waveforms within 500–700 ms for ORCs and ACs at S5, respectively

The ANOVA indicated that neither the interaction of Sentence type*Laterality nor that of Sentence type*Site was significant (F₁ (2, 19) = 1,495, p₁ = 0.249 > 0.05, partial η² = 0.136; F₂ (2, 19) = 0.144, p₂ = 0.867 > 0.05, partial η² = 0.015). Besides, we could not find any significant main effect of Sentence type (F (1, 20) = 0.021, p = 0.887 > 0.05, partial η² = 0.001), Laterality (F (2, 19) = 2.388, p = 0.119 > 0.05, partial η² = 0.201) or Site (F (2, 19) = 1.193, p = 0.325 > 0.05, partial η² = 0.112).

The ERP results at S6 (predicate verb in the main clause)

At S6, all the data analyses were in conformity with the analyses in the above section. The average waveforms for two types of sentences, together with topographic map, are presented in Fig. 5.

[See PDF for image]

Fig. 5

The grand average waveforms within 300-500 ms for ORCs and ACs at S6, respectively

N400

Table 3 lists the average amplitude of the selected electrodes. The two-way ANOVA of average amplitudes claimed that there was no interaction of Sentence type*Laterality (F (2, 19) = 2.391, p = 0.119 > 0.05, partial η² = 0.201) and Sentence type*Site (F (2, 19) = 0.515, p = 0.605 > 0.05, partial η² = 0.051). The main effect of Sentence type (F (1, 20) = 4.965, p = 0.037 < 0.05, partial η² = 0.199) revealed that ACs elicited a more negative waveform than ORCs did (MD = − 0.793 μV). The further analyses reflected the N400 effect was more salient at parietal sites (ACs minus ORCs: MD = − 0.984 μV, p = 0.031 < 0.05) with no difference at the frontal sites (MD = − 0.632 μV, p = 0.095 > 0.05) but marginal significance at the central sites (MD = − 0.762 μV, p = 0.058 ≈ 0.05). Considering the typical scalp distribution of this component, this study reported that ACs incurred a larger N400 than ORCs, and the (L) AN effect is absent at this segment.

Table 3. The average amplitude of the nine electrodes at S6 within 300-500 ms time window for ORCs and ACs

Main effect of ST		Post hoc of site
	F (1, 20) (Sig.)	Site	Mean amplitude (μV)		MDORC-AC (μV) (Sig.)
			ORC	AC
N400 at S6	4.965 (*.037 < 0.05)	Frontal	− 0.618	− 1.25	0.632 (0.095)
		Central	− 0.331	− 1.094	0.763 (0.058)
		Parietal	0.833	− 0.151	0.984 (*0.031)

P600

At S6, within the time window of 500–700 ms, the visual inspection of the waveforms in Fig. 6 reveals that none of the comparisons shows significant differences, which was corroborated by the statistical comparisons, i.e., Sentence type (F (1, 20) = 0.021, p = 0.887 > 0.05, partial η² = 0.001), interaction effects (F_{Sentence type*Laterality} (2, 19) = 1.495, p = 0.249 > 0.05, partial η² = 0.136; F_{Sentence type*Site} (2, 19) = 0.144, p = 0.867 > 0.05, partial η² = 0.015).

[See PDF for image]

Fig. 6

The grand average waveforms within 500–700 ms for ORCs and ACs at S6, respectively

Discussion

The processing of filler-gap dependency structure is a complex procedure, which may involve many cognitive manipulations such as the identification of the filler, retention of the filler in the working memory, the identification or prediction of the gap, the reactivation of the retained filler to the semantic and syntactic integration of the filler and gap and etc. (Hesvik, 2007; Zhang, 2007). To test the above theoretical hypotheses, the present study adopted an ERP experiment to examine Chinese EFL learners’ processing of explicit and pseudo filler-gap dependency structures. Statistical analyses at the critical and latter segments showed that Chinese EFL learners have a significantly larger N400 for sentences with pseudo filler-gap dependencies (e.g., ACs) in comparison with sentences with filler-gap dependencies (e.g., ORCs). In ERP studies on the processing difficulty of syntactic structures, certain lexical factors are challenging to eliminate. For example, in research examining subject and object relative clauses, the contrasted word pairs typically consist of nouns and verbs. Researchers then infer the processing difficulty of the two types of relative clauses based on the N400 responses elicited by these lexical pairs (e.g., Zhang & Yang, 2010). To minimize the influence of lexical factors, we controlled for the morphological features of word pairs in the critical segment as much as possible, such as word length (no significant difference, t (29) = − 1.235, p = 0.227 > 0.05), difficulty and frequency (words selected from the National full-time senior high school English syllabus (Revised) (2004)). Furthermore, in a lexical-decision task using comparable materials, reading times for critical word pairs extracted from ORCs and ACs in non-context conditions showed no significant differences (Tang & Wu, 2025).

Researchers also observed that functional words and content words may elicit distinct fluctuations in ERP components, with differences primarily emerging in early phases such as the N280 and eLAN (Friederici, Hahne, & Mecklinger, 1996). Additionally, word concreteness has been shown to modulate N400 amplitude, with concrete words (e.g., “apple”) eliciting a larger N400 than abstract words (e.g., “freedom”) (Barber et al., 2013; Kounios & Holcomb, 1994). In the present study, the critical words in ACs consist of determiners with concrete nouns (e.g., “some voices”), while those in ORCs involve prepositions with concrete nouns (e.g., “from radios”). Since both determiners and prepositions are functional words, it is less likely that word class contributed significantly to N400 fluctuations. Instead, the enhanced N400 amplitude observed in ACs during this segment is more likely to reflect sentence processing difficulty.

In the current study, the subsequent segment containing identical critical words (i.e., “is” in both ACs and ORCs) also elicited a significantly stronger N400 effect for ACs. This finding further supports the interpretation that the observed N400 differences reflect sentence-level processing distinctions rather than lexical-level variations. No significant deviation was found for the P600 responses between ACs and ORCs either at the critical segment nor at the following segment. In general, Chinese EFL learners have processing advantage for ORCs, which is in line with time-based studies (Zheng & Zhou, 2018; Tang & Wu, 2025).

N400 & the filler-gap dependency establishment

As stated in previous literature, the N400 indexes the retrieval difficulty of word meaning in the working memory. The N400 responses in the current study can reflect how semantic integration plays a role in (pseudo) filler-gap dependency processing. The larger N400 for ACs was found at the S5 (the critical segment, PPs under ORC condition vs. NPs under AC condition) and S6 (“is”, the verb of the main sentence). In the current study, if the readers predict a gap, they would find the PP in conformity with the expectation, and no semantic or syntactic difficulties would be evoked; in contrast, whereas there is a NP which is not in line with their expectations, a surprise effect may lead to dramatic EEG fluctuations. The appearance of the N400 effect in S5 provides evidence that the readers showed unexpectation, but it was more about the semantic integration difficulty.

According to the Active Filler Strategy (AFS; Frazier, 1987), the parser actively predicts potential gap positions during sentence processing. The filler, maintained in working memory, is reactivated upon gap identification to establish syntactic dependencies. Under this framework, the unexpected conditions (ACs) in our experimental design should theoretically elicit both semantic surprise (indexed by N400) and syntactic reanalysis (reflected in P600). However, our results revealed no significant P600 effects, suggesting that L2 learners may not consistently implement the AFS during parsing. This deviation from theoretical predictions may stem from typological differences between Chinese and English relative clause structures. In head-final languages like Chinese, the filler (e.g., “消息”, the news) appears postpositionally following a clause marker (“的”, de). Importantly, the relative clause precedes the head noun, requiring the parser to first identify a clause-internal gap before immediately encountering the head noun that fills it. In Chinese ACs (e.g., 10a), the clause initially resembles a canonical SVO structure until the appearance of the clause marker (“的”). This structural configuration suggests that native Chinese speakers may rely less on predictive parsing mechanisms or AFS-based strategies during comprehension.

1. 他们/从收音机/听到/的/消息/是/令人开心的。

They/ from radio/ heard/ de/ news/ is/ delightful.

Based on this, the online processing strategies of Chinese EFL learners may be influenced by L1 typological features. Without actively predicting the syntactic structure, they only showed difficulty in the semantic integration of noun phrases of ACs. These findings align with well-established evidence of L1 transfer of syntactic processing strategies across multiple domains, including: (1) relative clause attachment preferences (Papadopoulou & Clahsen, 2003; Tang & Wu, 2023, 2025), and (2) word order processing (Cuza, 2013).

Alternatively, the DAH claims that syntactic structure building is a lexically driven process triggered by the automatic mental reconstruction of the subcategorizer’s argument structure (Hestvik et al., 2007; Pickering, 2001; Pickering & Barry, 1991; Sag & Fodor, 1994). The lexical collocation is established via the argument structure satisfaction, that is, in the current study, the head noun is directly collocated with the verb in the ACs, and the appearance of another noun phrase led to the semantic surprise effect. It is therefore argued that Chinese EFL learners’ higher processing difficulty in ACs may reflect the lexical-driven surprise, which is in supportive of DAH.

The results are partly in accordance with previous studies for L1 speakers (Felser & Jessen, 2020; Friederici et al., 1993 & 2002; Jessen & Felser, 2019; Phillips et al., 2005). Phillips et al. (2005) found that L1 speakers showed a larger N400 at the position where the gap may potentially posit for the sentences with the filler-dependencies. In Jessen and Felser (2019)’s study, both the L1 speakers and L2 learners showed a larger N400 after the onset of the verb for implausible objects than for plausible objects, but the N400 for L2 learners is right frontally distributed, which coincides with the result of the current study.

The robust N400 observed at S6 further proves that Chinese EFL learners have difficulty in understanding ACs. We assume that the incomplete semantic processing at the end of the AC further gives rise to the difficulty in semantic integration of the main sentence. In sum, there were controversial empirical results and inconsistent theoretical hypothesis as to whether L2 learners, especially learners from the background lack of corresponding grammatical properties, can automatically use syntactic cues or strategies in L2 processing. The results suggest potential crosslinguistic influence while also aligning with predictions of the lexical-driven hypothesis.

P600 & syntactic integration

P600 as stated in previous literature, is a component indexing for the syntactic integration in sentences with filler-gap dependency (Fiebach, Schlesewsky & Friederici, 2001; Hestvik et al., 2007; Kaan et al., 2000; Liu & Jiang, 2016; Yang et al., 2010). However, the results at the time window of 500–700 ms did not display a significant difference with sentence types at the critical position and later segment, although L2 learners showed a subtly larger positive wave with ORCs. It further confirms the argument that the surprise effect is more semantic-driven rather than syntactic-driven, that is, the absence of P600 showed the lack of deep structure processing by Chinese EFL learners in filling gaps. And it supports the DAH, which believes the argument structure satisfaction; while it does not confirm the hypothesis of AFS as argued with English native speakers.

Our results are inconsistent with some findings that sentences with filler-gap dependency trigger a greater P600 (Fiebach, Schlesewsky & Friederici, 2001; Kaan et al., 2000). We assume that it is because of the different processing patterns between the L1 and L2, even if they both showed surprise effect in ACs. Alternatively, the observed effects may be attributed not only to inherent differences in structural complexity but also to L2-specific processing features. The Shallow Structure Hypothesis (SSH for short, Clahsen & Felser, 2006a) posits that L2 learners develop less detailed syntactic representations than native speakers, leading to over-reliance on lexical-semantic, contextual, or surface-level cues during comprehension. This “shallow” processing style could explain why learners exhibited greater sensitivity to semantic mismatch (N400) in ACs—a structure requiring precise integration of semantics with discourse context—while showing attenuated engagement with syntactic reanalysis mechanisms (P600). Such patterns mirror findings by Sabourin and Haverkort (2003), who demonstrated L2 learners’ reduced sensitivity to syntactic violations compared to native speakers, and Clahsen and Felser’s (2006b) work highlighting L2 reliance on lexical-thematic rather than structural cues. Critically, the current data extend the SSH framework by revealing structural ambiguity arising from missing filler-gap dependencies (as in ACs). While native speakers efficiently resolve such ambiguities through predictive parsing and detailed syntactic representations, L2 learners appear constrained by their reliance on shallow representations, resulting in delayed semantic integration (N400) rather than compensatory syntactic reanalysis (P600).

In comparison with the 3-phase model for language comprehension as proposed by Friederici et al. (1993, 1996 & 2002), the L2 learners’ patterns in processing clauses with or without filler-gaps showed some differences. Rather than experiencing the word class recognition, lexical-semantic and verb-argument structure building, and syntactic reanalysis or repair, L2 learners seem to simplify the procedure to the first two phases, and mainly rely on the lexical-semantic information. The present findings revealed the underneath reasons which were unclear in Zheng and Zhou (2018)’s self-paced reading study even though they observed Chinese EFL learners’ higher processing difficulty in ACs. By the dissociation of semantic and syntactic processing procedures via neurophysiological evidence, we probed the unique L2 processing patterns and therefore enriched our understanding to the general L1 and L2 processing differences.

In general, the findings of the current study revealed that ACs elicited significantly larger N400 amplitudes compared to ORCs. This pattern suggests that sentences containing relativizers without filler-gap dependencies (e.g., ACs) impose greater demands on semantic processing and real-time integration than those with explicit filler-gap relationships (e.g., ORCs). Specifically, the pronounced N400 effect observed for ACs likely reflects heightened difficulty in mapping syntactic structures to semantic representations potentially because of crosslinguistic influence from typologically distinct L1, as learners struggled to reconcile the absence of an overt gap with the presence of a relativizer. The lack of significant differences in later components such as P600, however, indicates the lack of syntactic integration due to the different L2 processing strategy or general L2 deficiencies. This dissociation between mid-latency components (N400) and late components (P600) aligns with the hypothesis that L2 learners prioritize semantic and discourse-level cues over detailed syntactic computation during real-time processing. It enriched the literature in exploration of L2 learners’ processing strategies to the filler-gap dependency and sheds light on the general L2 processing theories, particularly on the processing of typologically distinct syntactic structures in the L2.

Limitation

While this study provides valuable insights into filler-gap dependency processing among Chinese EFL learners, several limitations should be acknowledged. First, although we carefully controlled for variables distinguishing noun phrases and prepositional phrases, potential confounding effects of lexical concreteness and word class (functional vs. content words) on N400 modulation cannot be entirely ruled out. Future research employing an ERP-based lexical decision paradigm could more definitively isolate lexical effects during sentence processing. Second, our participant sample was limited to intermediate-to-high proficiency learners. A more comprehensive investigation incorporating learners across all proficiency levels (beginning, intermediate, and advanced) would better elucidate the developmental trajectory of L2 processing strategies. Finally, comparative studies with L2 learners whose native languages employ head-initial clause structures would help determine whether the observed processing patterns reflect language-specific transfer effects or more universal aspects of L2 acquisition.

Conclusion

This study explored how Chinese EFL learners process the filler-gap dependency in ACs and ORCs via ERP experiment. The results showed that L2 learners have a processing advantage for ORCs, which supports that L2 Learners resort to the Direct Association Hypothesis rather than Active Filler Strategy. Furthermore, the processing difficulty is mainly caused by the reanalysis of meaning instead of the syntactic prediction. To the best of our knowledge, no previous research explored L2 learners’ filler-gap dependency processing by the comparison of sentences with explicit and pseudo the filler-gap dependency via EEG methods. It thus enriched the literature in the domain of clauses processing and shed lights on L2 processing theories.

Acknowledgements

We acknowledge the Cognitive Neurolinguistics Research Center (CNLRC) for providing access to the experiment devices and analytic software. We sincerely appreciate the anonymous reviewers for their insightful suggestions.

Author contributions

MT: Writing—review & editing, Writing—original draft, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. JW: Writing—original draft, Methodology, Formal analysis, Data curation.

Funding

This study was supported by Science Foundation of China University of Petroleum, Beijing (No.2462023YXZZ006).

Data availability

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/supplementary material.

Declarations

Conflict of interests

The authors declare no conflict of interests.

Ethics approval

This study was approved by the Academic Ethics Committee of the author’s affiliated institution, China University of Petroleum, Beijing. The participants provided their written informed consent to participate in this study.

Consent for publication

All participants signed an informed consent form agreeing to the use of their data for publication. All authors unanimously agree to the publication.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

Aldwayan, SN. The syntax of relative and appositive clauses. Journal of King Saud University-Languages and Translation; 2012; 24, 2 pp. 95-100. [DOI: https://dx.doi.org/10.1016/j.jksult.2012.05.004]

Allan, D. Oxford placement test; 2004; UK, Oxford University Press:

Barber, HA; Otten, LJ; Kousta, ST; Vigliocco, G. Concreteness in word processing: ERP and behavioral effects in a lexical decision task. Brain and Language; 2013; 125, 1 pp. 47-53. [DOI: https://dx.doi.org/10.1016/j.bandl.2013.01.005]

Brouwer, H; Crocker, MW; Venhuizen, NJ; Hoeks, JCJ. A neurocomputational model of the N400 and the P600 in language processing. Cognitive Science; 2016; 41, S6 pp. 1318-1352. [DOI: https://dx.doi.org/10.1111/cogs.12461]

China Ministry of Education. National full-time senior high school English syllabus (revised); 2004; Beijing, People's Education Press:

Clahsen, H; Felser, C. Continuity and shallow structures in language processing. Applied Psycholinguistics; 2006; 27, 1 pp. 107-126. [DOI: https://dx.doi.org/10.1017/S0142716406060206]

Clahsen, H; Felser, C. Grammatical processing in language learners. Applied Psycholinguistics; 2006; 27, 1 pp. 3-42. [DOI: https://dx.doi.org/10.1017/S0142716406060024]

Cohen, J. A power primer. Psychological Bulletin; 1992; 112, 1 pp. 155-159. [DOI: https://dx.doi.org/10.1037/0033-2909.112.1.155]

Cuza, A. Crosslinguistic influence at the syntax proper: Interrogative subject-verb inversion in heritage Spanish. International Journal of Bilingualism; 2013; 17, 1 pp. 71-96. [DOI: https://dx.doi.org/10.1177/1367006911432619]

Dallas, A; Dede, G; Nicol, J. An event-related potential (ERP) investigation of filler-gap processing in native and second language speakers. Language Learning; 2013; 63, 4 pp. 766-799. [DOI: https://dx.doi.org/10.1111/lang.12026]

Delogu, F; Brouwer, H; Crocker, MW. Event-related potentials index lexical retrieval (N400) and integration (P600) during language comprehension. Brain and Cognition; 2019; 135, 103569 pp. 1-14. [DOI: https://dx.doi.org/10.1016/j.bandc.2019.05.007]

DeLong, KA; Urbach, TP; Kutas, M. Probabilistic word pre-activation during language comprehension inferred from electrical brain activity. Nature Neuroscience; 2005; 8, 8 pp. 1117-1121. [DOI: https://dx.doi.org/10.1038/nn1504]

Deng, L. Appositive clauses in EST and their translation. Bulletin of Science and Technology; 2013; 29, 5 pp. 208-212. [DOI: https://dx.doi.org/10.13774/j.cnki.kjtb.2013.05.003]

Fan, X; Wei, M. Appositive clause or attributive clause. Journal of Shanghai Maritime University; 2002; 23, 2 pp. 84-87.

Felser, C; Clahsen, H; Thomas, FM. Storage and integration in the processing of filler-gap dependencies: An ERP study of topicalization and wh-movement in German. Brain & Language; 2003; 87, 3 pp. 345-354. [DOI: https://dx.doi.org/10.1016/S0093-934X(03)00135-4]

Felser, C; Jessen, A. Torrens, V. Brain responses elicited by implausible fillers and filled object gaps in German. Typical and Impaired Processing in Morphosyntax; 2020; Amsterdam, John Benjamins Publishing Company: pp. 75-90. [DOI: https://dx.doi.org/10.1075/lald.64.04fel]

Fiebach, CJ; Schlesewsky, M; Friederici, AD. Syntactic working memory and the establishment of filler-gap dependencies: Insights from ERPs and fMRI. Journal of Psycholinguistic Research; 2001; 30, 3 pp. 321-338. [DOI: https://dx.doi.org/10.1023/A:1010447102554]

Frazier, L. Syntactic processing: Evidence from Dutch. Natural Language & Linguistic Theory; 1987; 5, 4 pp. 519-559. [DOI: https://dx.doi.org/10.1007/BF00138988]

Friederici, AD. Towards a neural basis of auditory sentence processing. Trends in Cognitive Sciences; 2002; 6, 2 pp. 78-84. [DOI: https://dx.doi.org/10.1016/S1364-6613(00)01839-8]

Friederici, AD; Hahne, A; Mecklinger, A. Temporal structure of syntactic parsing: Early and late event-related brain potential effects. Journal of Experimental Psychology: Learning, Memory, and Cognition; 1996; 22, 5 pp. 1219-1248. [DOI: https://dx.doi.org/10.1037/0278-7393.22.5.1219]

Friederici, AD; Pfeifer, E; Hahne, A. Event-related brain potentials during natural speech processing: Effects of semantic, morphological and syntactic violations. Brain Research; 1993; 1, 3 pp. 183-192. [DOI: https://dx.doi.org/10.1016/0926-6410(93)90026-2]

Gorrell, P. Evaluating the direct association hypothesis: A reply to Pickering and Barry. Language and Cognitive Processes; 1993; 8, 2 pp. 129-146. [DOI: https://dx.doi.org/10.1080/01690969308406951]

Gouvea, AC; Phillips, C; Kazanina, N; Poeppel, D. The linguistic processes underlying the P600. Language, Cognition and Neuroscience; 2010; 25, 2 pp. 149-188. [DOI: https://dx.doi.org/10.1080/01690960902965951]

Hagoort, P; Brown, CM; Groothusen, J. The syntactic positive shift (sps) as an ERP measure of syntactic processing. Language and Cognitive Processes; 1993; 8, 4 pp. 439-483. [DOI: https://dx.doi.org/10.1080/01690969308407585]

Hagoort, P; Brown, CM; Osterhout, L. Brown, CM; Hagoort, P. The neurocognition of syntactic processing. Neurocognition of Language; 1999; UK, Oxford University Press: pp. 273-361.

Hestvik, A; Maxfield, N; Schwartz, RG; Shafer, V. Brain responses to filled gaps. Brain and Language; 2007; 100, 3 pp. 301-316. [DOI: https://dx.doi.org/10.1016/j.bandl.2006.07.007]

Hsiao, F; Gibson, E. Processing relative clauses in Chinese. Cognition; 2004; 90, 1 pp. 3-27. [DOI: https://dx.doi.org/10.1016/S0010-0277(03)00124-0]

Jacobs, RA; Rosenbaum, PS. English transformational grammar; 1968; NY, Blaisdell Publishing Company:

Jessen, A; Felser, C. Reanalysing object gaps during non-native sentence processing: Evidence from ERPs. Second Language Research; 2019; 35, 2 pp. 285-300. [DOI: https://dx.doi.org/10.1177/0267658317753030]

Kaan, E; Harris, A; Gibson, E; Holcomb, P. The P600 as an index of syntactic integration difficulty. Language and Cognitive Processes; 2000; 15, 2 pp. 159-201. [DOI: https://dx.doi.org/10.1080/016909600386084]

Kluender, R; Kutas, M. Bridging the gap: Evidence from erps on the processing of unbounded dependencies. Journal of Cognitive Neuroscience; 1993; 5, 2 pp. 196-214. [DOI: https://dx.doi.org/10.1162/jocn.1993.5.2.196]

Kounios, J; Holcomb, PJ. Concreteness effects in semantic processing: ERP evidence supporting dual-coding theory. Journal of Experimental Psychology: Learning, Memory, and Cognition; 1994; 20, 4 pp. 804-823. [DOI: https://dx.doi.org/10.1037/0278-7393.20.4.804]

Kutas, M; Federmeier, KD. Electrophysiology reveals semantic memory use in language comprehension. Trends in Cognitive Sciences; 2000; 4, 12 pp. 463-470. [DOI: https://dx.doi.org/10.1016/S1364-6613(00)01560-6]

Kutas, M; Federmeier, KD. Thirty years and counting: Finding meaning in the N400 component. Annual Review of Psychology; 2011; 62, pp. 621-647. [DOI: https://dx.doi.org/10.1146/annurev.psych.093008.131123]

Lau, E; Phillips, C; Poeppel, D. A cortical network for semantics: (De) constructing the N400. Nature Review Neuroscience; 2008; 9, pp. 920-933. [DOI: https://dx.doi.org/10.1038/nrn2532]

Liu, T; Jiang, H. Neural mechanisms of syntactic movement: An ERPs study of Chinese passive sentences. Linguistic Sciences; 2016; 15, 6 pp. 612-624. [DOI: https://dx.doi.org/10.7509/j.linsci.201610.031506]

Osterhout, L; Holcomb, PJ. Event-related brain potentials elicited by syntactic anomaly. Journal of Memory and Language; 1992; 31, 6 pp. 785-806. [DOI: https://dx.doi.org/10.1016/0749-596X(92)90039-Z]

Osterhout, L; Holcomb, PJ. Event-related potentials and syntactic anomaly: Evidence of anomaly detectiion during the perception of continuous speech. Language and Cognitive Processes; 1993; 8, 4 pp. 413-438. [DOI: https://dx.doi.org/10.1080/01690969308407584]

Papadoppoulou, D; Clahsen, H. Parsing strategies in L1 and L2 sentence processing: A study of relative clause attachment in Greek. Studies in Second Language Acquisition; 2003; 25, 4 pp. 501-528. [DOI: https://dx.doi.org/10.1017/S0272263103000214]

Phillips, C; Kazanina, N; Abada, SH. ERP effects of the processing of syntactic long-distance dependencies. Cognitive Brain Research; 2005; 22, 3 pp. 407-428. [DOI: https://dx.doi.org/10.1016/j.cogbrainres.2004.09.012]

Phillips, C; Wagers, M. Gaskell, G. Relating structure and time in linguistics and psycholinguistics. Oxford Handbook of Psycholinguistics; 2007; UK, Oxford University Press: pp. 739-757.

Pickering, MJ. No evidence for traces in sentence comprehension. Behavioral and Brain Sciences; 2001; 23, 1 pp. 47-48. [DOI: https://dx.doi.org/10.1017/S0140525X00502398]

Pickering, M; Barry, G. Sentence processing without empty categories. Language and Cognitive Processes; 1991; 6, 3 pp. 229-259. [DOI: https://dx.doi.org/10.1080/01690969108406944]

Radford, A. Transformational syntax: A student’s guide to Chomsky’s extended standard theory; 1981; Cambridge, Cambridge University Press:

Sabourin, L; Haverkort, M. Neural substrates of representation and processing of a second language. Language Acquisition and Language Disorders; 2003; 30, pp. 175-196. [DOI: https://dx.doi.org/10.1075/lald.30.09sab]

Sag, I. A., & Fodor, J. D. (1994). Extraction without traces, In R. Aranovich, W. Byrne, S. Preuss, & M. Senturia. (Eds.). Proceedings of the 13th Annual Meeting of the West Coast Conference on Formal Linguistics (pp. 365–384). CSLI, Stanford, CA.

Tang, M; Wu, J. Chinese ESL learner’s processing strategy of English reflexive pronoun ambiguity. Journal of Beijing International Studies University; 2023; 45, 06 pp. 143-156. [DOI: https://dx.doi.org/10.12002/j.bisu.493]

Tang, M; Wu, J. Language learners’ processing strategies of filler-gap dependencies: A comparison of English object relative clauses and appositive clauses. Journal of Cultural Cognitive Science; 2025; [DOI: https://dx.doi.org/10.1007/s41809-025-00181-x]

Ueno, M; Kluender, R. On the processing of Japanese wh-questions: An ERP study. Brain Research; 2009; 1290, pp. 63-90. [DOI: https://dx.doi.org/10.1016/j.brainres.2009.05.084]

Wang, H; Ma, L; Wang, Y; Troyer, M; Li, Q. An ERP study in English relative clause processing by Chinese-English bilinguals. Chinese Journal of Applied Linguistics; 2015; 38, 1 pp. 3-35. [DOI: https://dx.doi.org/10.1515/cjal-2015-0001]

Williams, JN; Mobius, P; Kim, C. Native and non-native processing of English wh- questions: Parsing strategies and plausibility constraints. Applied Psycholinguistics; 2002; 22, 4 pp. 509-540. [DOI: https://dx.doi.org/10.1017/S0142716401004027]

Yang, CL; Perfetti, CA; Liu, Y. Sentence integration processes: An ERP study of Chinese sentence comprehension with relative clauses. Brain and Language; 2010; 112, 2 pp. 85-100. [DOI: https://dx.doi.org/10.1016/j.bandl.2009.10.005]

Zhang, Q; Yang, Y. The processing advantage of Chinese object relative clauses: Evidence from neuroelectrophysiological studies. Linguistic Science; 2010; 4, pp. 337-353.

Zhang, Y; Jiang, X; Huang, Y. Verbal working memory, sentence comprehension, and the processing of filler-gap dependencies. Advances in Psychological Science; 2007; 15, 1 pp. 22-28. [DOI: https://dx.doi.org/10.1016/j.conbuildmat.2005.08.001]

Zheng, W; Zhou, T. A study of object relative clause and appositive clause processing by Chinese EFL learners. Foreign Language Education; 2018; 39, 5 pp. 54-60. [DOI: https://dx.doi.org/10.16362/j.cnki.cn61-1023/h.2018.05.011]