Abstract
This study examines whether English native speakers and highly proficient non-native speakers make comparable use of plausibility information during online sentence processing. Two sentence types involving temporarily ambiguous structural configurations—subordinate-clause ambiguity sentences and sentences with adjacent/split verb-particle constructions (VPCs)—were tested in a self-paced reading task. In the subordinate-clause ambiguity sentences, the pattern of reading times indicated that both native and non-native speakers used plausibility to recover from initial structural misanalysis. Native speakers were also able to use this information during syntactic and semantic reanalysis in the sentences involving split VPCs. Non-native speakers, however, showed persistent processing difficulty for split VPC sentences, regardless of plausibility. These results are taken to indicate that both native speakers and non-native speakers use plausibility information to recover from misanalysis, even in sentences that require major syntactic revision. The only clear limit on non-native speakers’ ability to use this information related to lexico-syntactic/semantic processing difficulty, in that they appeared to be unable to use this information to recover from misanalysis associated with the structural properties of English VPCs.
Keywords
A core question in psycholinguistics is the extent to which language background influences comprehenders’ ability to use various sources of information during real-time language processing. This question is of particular interest in research related to second-language (L2) sentence comprehension, much of which has focused on (a) whether the first language (L1) affects L2 sentence processing and (b) whether structural computation in the L2 is characterised by the same depth and detail as in the L1. The present study takes up this issue by examining the processing of temporary structural ambiguities in native and non-native speakers of English. Specifically, this study investigates whether these comprehenders differ in their use of plausibility information when recovering from initial misanalysis in such sentences. Although a number of studies have indicated that both native and non-native readers use plausibility to inform structural processing (see, for example, Dussias & Piñar, 2010; Frenck-Mestre & Pynte, 1997; Hopp, 2015; Williams, 2006; Williams et al., 2001), there is evidence suggesting that non-native readers are limited in their ability to draw on this information during online reanalysis processes (Roberts & Felser, 2011). This study investigates whether this is indeed the case—and if so, the nature of these limits. This was done by focusing on the processing of two sentence types involving temporary structural ambiguity—subordinate-clause ambiguity sentences and sentences with verb-particle constructions.
The first of these sentence types involves the subordinate-clause ambiguity (sometimes referred to as object-subject ambiguity) in sentences like the following: (1) As the woman edited the magazine about fishing amused all the reporters.
In such sentences, studies have shown that both L1 and L2 comprehenders tend to initially interpret the main-clause subject (the magazine about fishing) as the object of the preceding subordinate-clause verb (edited), only to revise this analysis after they encounter the disambiguating verb (amused) in the main clause (L1 comprehension studies: Adams et al., 1998; Christianson et al., 2001; Clifton, 1993; Ferreira & Henderson, 1991, 1993; Frazier & Rayner, 1982; Mitchell, 1987; Pickering & Traxler, 1998; Slatterly et al., 2013; Staub, 2007; Stowe, 1989; Sturt et al., 1999; Tabor & Hutchins, 2004; Traxler, 2002; Van Dyke & Lewis, 2003; Van Gompel & Pickering, 2001; Warner & Glass, 1987; L2 comprehension studies: Frenck-Mestre & Pynte, 1997; Hopp, 2015; Jacob & Felser, 2016; Juffs, 1998, 2004; Juffs & Harrington, 1996).
Pickering and Traxler (1998) have also found that for English native speakers, this reanalysis process is influenced by the plausibility of the ambiguous noun phrase (NP) as the object of the preceding verb. This eye-tracking study investigated sentences like the following, in which syntactic ambiguity (i.e., the presence/absence of a disambiguating comma) and the local plausibility of the ambiguous NP were manipulated: (2a) As the woman sailed the magazine about fishing amused all the reporters. (2b) As the woman sailed, the magazine about fishing amused all the reporters. (2c) As the woman edited the magazine about fishing amused all the reporters. (2d) As the woman edited, the magazine about fishing amused all the reporters.
At and immediately after the head of the critical NP (the magazine), there was clear processing difficulty related to local plausibility. In particular, there were more first-pass regressions for locally implausible ambiguous sentences (2a) compared with their unambiguous controls (2b) as well as to locally plausible ambiguous sentences (2c). Locally implausible ambiguous sentences also had longer total reading times (RTs) than their unambiguous controls in these regions. At and after the main-clause verb (amused), there was also processing difficulty related to reanalysing the structure of ambiguous sentences—sometimes referred to as a garden-path effect—which was particularly strong when the ambiguous NP was a plausible object for the preceding verb. While there was only a weak garden-path effect for the implausible ambiguous sentences under the total time measure, this effect was observed across first-pass time, first-pass regressions, and total time for plausible ambiguous sentences. In short, this experiment found a plausibility effect at the temporarily ambiguous NP, followed by a reverse plausibility effect at and after the disambiguating verb. This pattern of results was taken to indicate that although readers misanalysed the ambiguous NP as the object of the subordinate-clause verb regardless of plausibility, the process of recovery and reanalysis was more difficult when this initial analysis led to a plausible interpretation. More specifically, it was argued that in the case of locally implausible ambiguous sentences, comprehenders are less committed to their initial structural analysis, allowing for easier revision of this analysis. (For comparable findings in L1 comprehension, see Clifton, 1993; Stowe, 1989.)
There is some evidence to suggest that L2 comprehenders, however, are not able to make comparable use of plausibility to recover from misanalysis in such sentences. This evidence comes from a self-paced reading study by Roberts and Felser (2011), in which sentences like the following were tested: (3a) While the band played the beer pleased all the customers. (3b) While the band played the song pleased all the customers.
Although non-native speakers had longer RTs at the ambiguous NP in locally implausible sentences (3a), there did not appear to be an especially large garden-path effect at or after the disambiguating main-clause verb (pleased) in their locally plausible counterparts (3b). That is, in terms introduced above, while these L2 comprehenders showed a plausibility effect at the ambiguous NP, they did not show a reverse plausibility effect at or after disambiguation. These results were taken to indicate that even though the non-native speakers were sensitive to plausibility during incremental sentence processing, they were nevertheless unable to use this information during structural reanalysis.
To account for these findings, Roberts and Felser (2011) proposed that non-native speakers might be unable to use plausibility to facilitate reanalysis in sentences that require major structural revision. As noted above, reanalysis in subordinate-clause ambiguity sentences involves detaching the ambiguous NP from its position as the object of the subordinate-clause verb and reanalysing it as the subject of the main clause. Reanalysis of this type has been argued to be particularly difficult because it necessitates adjustments to both thematic and hierarchical structural relations during incremental sentence processing (see, for example, Pritchett, 1988, 1992; Sturt & Crocker, 1996; Sturt et al., 1999). In terms of thematic structure, the ambiguous NP must be removed from the thematic domain of subordinate-clause verb—roughly, a position in which it acts as an argument of this verb—so that it can be analysed in the thematic domain of the main-clause predicate (Pritchett, 1988, 1992). With regard to hierarchical structure, this NP must also be removed from the verb phrase (VP) headed by the subordinate-clause verb, and any dominance relationships involved in that initial structural representation (e.g., the dominance relationship between the VP node and the NP complement) must be deleted (Sturt & Crocker, 1996; Sturt et al., 1999). Roberts and Felser (2011; for detailed discussion, see pp. 324–325) argued that L2 comprehenders might have particular difficulty using plausibility to inform this kind of structural reorganisation. They further interpreted this proposal as consistent with models of L2 sentence comprehension that hold that structural processing is less automatic or reduced in non-native speakers compared with native speakers (see, for example, Clahsen & Felser, 2006).
It is important to point out, however, that this study also found evidence suggesting that this limitation can be overcome at higher levels of L2 proficiency and reading ability. Indeed, a post hoc analysis revealed a numerical trend suggesting that faster L2 readers—who also tended to be more proficient in English—showed a reverse plausibility effect after the disambiguating verb. This was taken to suggest that L2 comprehenders with sufficient proficiency/reading skill might be able to use plausibility information in the reanalysis of subordinate-clause ambiguity sentences, just like the L1 comprehenders in Pickering and Traxler (1998).
An eye-tracking study by Hopp (2015) revealed a pattern of results consistent with this follow-up analysis. Among the conditions that were tested in this experiment were sentences like following: (4a) When the girl was playing the boy made some funny noises. (4b) When the girl was playing the piano made some funny noises.
Although the L2 comprehenders in this study did not show processing difficulty at the ambiguous NP in locally implausible sentences (4a), these sentences were found to be easier to process (under second-pass and total time reading measures) than their locally plausible counterparts (4b) at and after the disambiguating verb (made). That is, while this study did not find a plausibility effect at the ambiguous NP, a clear reverse plausibility effect was observed at and after disambiguation. These results were thus inconsistent with the complete pattern of results for the L2 comprehenders in Roberts and Felser (2011), but they were in line with the trend obtained for the subgroup of more proficient learners/readers in that study.
With these somewhat conflicting findings in mind, the purpose of the present study was twofold. The first goal was to examine whether both L1 and L2 comprehenders of English are able to use plausibility information during reanalysis in subordinate-clause ambiguity sentences. Again, this is a sentence type in which recovery from an initial misanalysis requires an NP to be reassigned from a position as a subordinate-clause object to that of a main-clause subject. This reanalysis process therefore involves major adjustments to both the thematic and hierarchical structural relations relevant to this constituent, and as discussed above, there is conflicting evidence about whether L2 comprehenders can use plausibility to facilitate reanalysis of this type. The second, related goal was to test whether both L1 and L2 comprehenders are able to use plausibility during the processing of temporary structural ambiguity in a different sentence type—specifically, in sentences with verb-particle constructions. As will be explained in more detail below, such sentences also often involve structural reanalysis of an NP constituent. Crucially, however, these reanalysis procedures do not require changes to the thematic processing domain or the hierarchical structural relationships relevant to this constituent. This is important because, to the extent that L2 comprehenders are specifically limited in their ability to use plausibility information for structural reanalysis that involves such procedures, these limitations should not apply to the processing of this second sentence type.
Regarding the first of these goals, further investigation into the L2 processing of subordinate-clause ambiguity sentences is important for several reasons. First, as discussed above, the studies by Roberts and Felser (2011) and Hopp (2015) have revealed somewhat conflicting results with respect to whether L2 comprehenders can reliably use plausibility during reanalysis in these sentences. These findings can of course be reconciled, but only to the extent that the pattern of results for the L2 comprehenders in Hopp (2015) can be considered consistent with numerical trends for the more proficient L2 subgroup in Roberts and Felser (2011). It is also important to note that the interpretation of these studies is complicated by several discrepancies between their findings and those of previous L1 studies, and notably those of Pickering and Traxler (1998). As detailed above, the native speakers in Pickering and Traxler (1998) showed local plausibility effects at the ambiguous NP as well as reverse plausibility effects at and after the disambiguating verb. However, this pattern of results was not replicated in Roberts and Felser (2011) or in Hopp (2015)—for either their L2 comprehenders or for their L1 comparison groups. As reported above, the complete group of L2 comprehenders in Roberts and Felser (2011) showed only a plausibility effect at the ambiguous NP. And for its part, the native speaker comparison group did not show either a plausibility effect or a reverse plausibility effect for this sentence type. The only result along these lines was found in a post hoc analysis of slower L1 readers, who showed a (delayed) reverse plausibility effect after the disambiguating verb. A comparable pattern of results was found for both the L2 comprehenders and the L1 comparison group in Hopp (2015). As noted above, the L2 group did not show a plausibility effect at the ambiguous NP; rather, there was only a reverse plausibility effect at and after disambiguation. This was also the case for the L1 comparison group.
In both studies, these discrepancies were explained in part with reference to a key difference between their experimental sentences and those used in Pickering and Traxler (1998). For the sentences in this earlier L1 study, the head of the ambiguous NP (magazine) was separated from the disambiguating verb (amused) by intervening material (e.g., As the woman sailed the
To address the second goal of this study, the online comprehension of sentences involving verb-particle constructions (henceforth, VPCs) was also examined. VPCs are multi-word units that usually consist of a verb followed by an adverb- or preposition-like particle. Examples of these forms include chew out, run over, turn in, and call up, among many others. Despite the ubiquity of VPCs in English—there are over 3,000 VPCs (see, for example, McArthur & Atkins, 1974), and these forms occur quite frequently (Biber et al., 1999; Gardner & Davies, 2007)—they are notoriously difficult for L2 learners (for review, see (Blais & Gonnerman, 2013; Garnier & Schmitt, 2016). Indeed, a number of production and corpus studies have indicated that L2 English learners from various L1 backgrounds tend to avoid using VPCs (Dagut & Laufer, 1985; González, 2010; Hulstijn & Marchena, 1989; Liao & Fukuya, 2004; Matlock & Heredia, 2002; Siyanova & Schmitt, 2007). This avoidance appears to be influenced in part by learners’ L1 background and L2 English proficiency as well as by the semantic properties of VPCs. Specifically, advanced learners from L1s that also have VPCs, like Dutch and Swedish, tend to avoid these constructions less often than comparable learners from L1s that lack these forms (González, 2010; Hulstijn & Marchena, 1989; Laufer & Eliasson, 1993). Moreover, L2 learners—particularly at lower proficiency levels—have been shown to have a stronger dispreference for semantically opaque VPCs, or for figurative/idiomatic VPCs in which the meaning of the verb changes substantially in combination with the particle, as in the example chew out above (see, for example, Dagut & Laufer, 1985; González, 2010; Hulstijn & Marchena, 1989; Liao & Fukuya, 2004). These results have been taken to indicate that semantic transparency/opacity is a key factor influencing the L2 acquisition of VPCs (but see Garnier & Schmitt, 2016). (For findings indicating that L2 English learners are able to develop near-native-like explicit and implicit knowledge of VPC semantics at higher proficiency levels, see Blais & Gonnerman, 2013.)
Studies on the L1 comprehension of VPC sentences have also investigated the influence of semantic transparency/opacity and its interaction with structural properties of these sentences that might give rise to processing difficulty (Gonnerman, 2012; Gonnerman & Hayes, 2005). An interesting characteristic of VPCs is that many of these constructions allow the verb and its particle to be adjacent (e.g., The principal will chew out the class.) or split from other each other by intervening material, usually by an object NP (e.g., The principal will chew the class out.). (For more on the factors that influence particle placement in VPCs, see Bolinger, 1971; Chen, 1986; Fraser, 1976; Gries, 1999; Lohse et al., 2004). In a self-paced reading experiment, Gonnerman and Hayes (2005) examined the processing effects related to this particle placement (i.e., split vs. adjacent) during the L1 comprehension of VPC sentences that varied in terms of semantic transparency/opacity and that had object NPs of different lengths. Longer per-word RTs were observed for sentences with split particle placement. This processing cost was also found to be particularly large for sentences with semantically opaque VPCs as well as for those with longer object NPs.
In a similar study, Gonnerman (2012) examined the L1 comprehension of VPC sentences by manipulating particle placement, semantic transparency/opacity of the VPC, and syntactic complexity of the object NP. This study again found a clear processing cost for split particle sentences that was especially large for items with opaque VPCs. Of particular interest for the purposes of the present investigation, this study also explored a possible source of the interaction between these factors that relates to both structural ambiguity and plausibility. Consider, for instance, the following sentence: (5) The principal will chew the class of disruptive students out.
Gonnerman (2012) noted that in the incremental processing of such sentences, comprehenders might initially misanalyse the constituent beginning with the class as the object of the verb chew alone. In sentences with opaque VPCs, as in this example, this can often lead to local implausibility (i.e., The principal will chew the class . . .). Interestingly, a post hoc analysis suggested that readers were influenced by this temporary structural ambiguity and local implausibility. In particular, there was a numerical trend suggesting that the head noun of the object NP (class) was read more slowly in split versions of sentences with opaque VPCs when this NP was an implausible object for the verb alone—that is, a plausibility effect. There was also a trend suggesting attenuated processing costs—or a reverse plausibility effect—at the particle in these locally implausible, split-VPC sentences. Specifically, there was a numerical difference suggesting that split particles were read more slowly than adjacent particles in locally plausible sentences, but that there was essentially no difference between split and adjacent particles in locally implausible sentences as in the example above.
Although these post hoc results should be regarded with appropriate caution, they nevertheless suggest that, just as in the case of subordinate-clause ambiguity sentences, VPCs provide an opportunity to examine whether/how comprehenders use plausibility information in the reanalysis of temporarily ambiguous sentences. The present study therefore examined this issue directly, in both L1 and L2 sentence comprehension.
As alluded to above, this issue is particularly important in the case of L2 sentence comprehension in light of differences between the reanalysis operations involved in split VPC sentences and those that apply to subordinate-clause ambiguity sentences. On the surface, these sentence types are similar, in that both might involve recovery from the temporary misanalysis of an NP. However, this recovery process arguably requires larger adjustments to the developing structural representation in the case of subordinate-clause ambiguity sentences. As detailed above, in these sentences (While the band played the beer pleased the customers.), the ambiguous NP (the beer) tends to be initially analysed as the object of the subordinate-clause verb, only to be reanalysed as the main-clause subject. In this way, reanalysis involves adjusting hierarchical structural relations established during the incremental processing of this NP as well as assigning this constituent to a new thematic processing domain. However, the same procedures do not appear to apply to reanalysis in split VPC sentences. In such sentences (The principal will chew the class out.), if the relevant NP (the class) is misanalysed as the object of the verb (chew) alone, reanalysis would simply involve the construction of a more elaborated predicate—one that minimally includes a structural position for the particle and the linking of that particle with its verb counterpart. (For more on accounts of the syntactic analysis of split and adjacent VPCs, see, for example, Den Dikken, 1995; Haiden, 2006; Harley & Noyer, 1998; Johnson, 1991; Kayne, 1985; Koizumi, 1993.) Crucially, it is not necessary to remove the relevant NP from one predicate so that it can be assigned to another. This is important because a model in which non-native speakers are limited in their ability to use plausibility information to facilitate this kind of major structural reorganisation predicts that this limitation would not apply to the processing of split VPC sentences.
While these specific issues have not been addressed in research to date, several studies have shed light on the L2 comprehension of VPCs more generally. For instance, Matlock and Heredia (2002) examined global RTs on sentences involving the same set of words when they were used either as a VPC or as a verb + preposition (e.g., Paul went over the exam with his students. vs. Paul went over the bridge with his bicycle.) in English native speakers, early bilinguals, and late learners of English. The results showed that while the native speakers and early bilinguals read sentences with VPCs faster than those with corresponding verb + preposition structures, this was not the case for the late learners. These findings were taken to suggest that while native speakers and early bilinguals are able to automatically access the figurative meanings that are often involved in VPCs, late learners’ access to these meanings might be mediated by their literal interpretations. More recent studies have investigated item- and participant-level differences that might influence the L2 comprehension of VPC sentences. For example, in an L2 follow-up to Gonnerman and Hayes (2005), Herbay and colleagues (2018) found that French learners of English with more native-like knowledge of VPC semantics and with higher working memory capacities exhibited reading patterns comparable to those of the L1 comprehenders in Gonnerman and Hayes (2005). Specifically, they had processing difficulty for split particle sentences with opaque VPCs and with longer object NPs. Tiv and colleagues (2019) used eye-tracking to examine the comprehension of split and adjacent VPCs in similar French–English bilinguals as well as in L1 English readers. The results indicated essentially no first- or second-pass reading difficulty for split VPCs in the L1 English group. However, this group had especially fast first-pass RTs for adjacent VPCs when these constructions were opaque and highly frequent. The French-English bilinguals, on the other hand, showed a much clearer processing cost for split VPC sentences, with longer second-pass RTs for these items compared with their adjacent counterparts. This processing cost tended to be larger for lower proficiency readers and did not interact with item-level factors related to semantic transparency or frequency of the VPC. That is, this cost applied across split VPC items. This pattern of results was interpreted to indicate that while L1 readers might process VPCs—especially those that are semantically opaque and very common—as lexical units, L2 readers tend to process VPCs compositionally. The present study attempted to expand on these findings to examine how structural ambiguity and local plausibility might influence the processing of VPC sentences during both L1 and L2 comprehension.
The present study: design and predictions
This study examined the online processing of the two sentence types reviewed above—subordinate-clause ambiguity sentences and VPC sentences—using a self-paced reading task. The subordinate-clause ambiguity sentences were tested with materials adapted from Pickering and Traxler (1998), as in sentences 6a–d: (6a) Implausible/ambiguous While Emma drank the song about true love amused all the customers at the club. (6b) Implausible/unambiguous While Emma drank, the song about true love amused all the customers at the club. (6c) Plausible/ambiguous While Emma played the song about true love amused all the customers at the club. (6d) Plausible/unambiguous While Emma played, the song about true love amused all the customers at the club.
In sentences such as (6a) and (6c), there was an NP (the song . . .) that was ambiguous as to whether it was the object of the preceding subordinate-clause verb (drank, played) or the subject of the main clause. In sentences such as (6b) and (6d), however, a comma indicated that this NP was the main-clause subject. Also, while the ambiguous NP was an implausible object for the subordinate-clause verb in sentences like (6a) (drank the song . . .), it was a plausible object for the verb in sentences like (6c) (played the song . . .). It was predicted that if readers are sensitive to plausibility information and initially adopt the object analysis in ambiguous sentences, there should be processing difficulty for implausible/ambiguous sentences (6a) at and after the head noun of the ambiguous NP (song). If readers are then able to use plausibility information to revise their initial misanalysis, there should be a larger garden-path effect for plausible/ambiguous sentences (6c) than for implausible/ambiguous sentences (6a) at and after the disambiguating verb (amused).
The processing of VPC sentences was tested with items as in the following sample set: (7a) Opaque-implausible/split Mary ran the fox that crossed the road over when it jumped in front of her car. (7b) Opaque-implausible/adjacent Mary ran over the fox that crossed the road when it jumped in front of her car. (7c) Opaque-plausible/split Andy turned the key for his new office in and then left to catch the train. (7d) Opaque-plausible/adjacent Andy turned in the key for his new office and then left to catch the train. (7e) Transparent-plausible/split Albert called the family who visited his church up after he talked to his pastor. (7f) Transparent-plausible/adjacent Albert called up the family who visited his church after he talked to his pastor.
The split versions of these sentences were temporarily ambiguous in the sense that the NP immediately following the verb (the fox . . ., the key . . ., the family . . .) could be (mis)analysed as the object of the verb alone. This ambiguity was examined in three VPC sentence types: In opaque-implausible sentences, the VPC was semantically opaque, and the relevant NP in the split version was an implausible object for the verb on its own (ran the fox . . .). Opaque-plausible sentences also had relatively opaque VPCs, but in this case, the relevant NP in the split version was a plausible object for the verb on its own (turned the key . . .). Finally, transparent-plausible sentences contained semantically transparent VPCs, and again the relevant NP in the split version was a plausible object for the verb on its own (called the family . . .). (Note that sentences instantiating the fourth combination of these factors—transparent-implausible—are virtually impossible to create. This is because the meaning of transparent VPCs and their non-VPC counterparts is essentially the same. It is therefore difficult, if not impossible, to create sentences in which an NP is implausible for the verb on its own, but not for its split VPC counterpart.)
The predictions for these sentence types are comparable to those for the subordinate-clause ambiguity sentences. In this case, the critical word for local plausibility effects was the head of the object NP (fox, key, family). At and after this point, if readers initially analyse this NP as the object of the verb on its own in the split versions of these sentences, there should be particular processing difficulty for opaque-implausible/split sentences. If readers are then able to use plausibility information to revise their initial analysis of these sentences, there should be attenuated processing difficulty at and immediately after the disambiguating particle in opaque-implausible/split sentences. The two different types of plausible sentences—opaque and transparent—were included to evaluate the extent to which these processing costs are influenced by semantic, in addition to syntactic, reanalysis. It is important to note that in opaque-plausible/split sentences, semantic revision is also often necessary to select the correct meaning for the VPC and to interpret it in relation to its arguments. For instance, with reference to the example (7c) above, it is not clear that the VPC turned in should be interpreted as something similar to submitted until the particle in is encountered. However, this is not the case for transparent-plausible sentences, in which the VPC and its non-VPC counterpart have very similar meanings (e.g., call ≈ call up).
These sentence types were tested with both native and non-native speakers of English. The key prediction for these groups relates to the proposal that non-native speakers might be limited in their ability to use plausibility to facilitate major structural reanalysis. As discussed above, this refers to cases in which reanalysis involves detaching a misanalysed constituent from one predicate and assigning it to another. A process of this type applies to subordinate-clause ambiguity sentences, in which an NP that initially appears to be a subordinate-clause object turns out to be the main-clause subject. A similar process, however, does not appear to be relevant to the ambiguous NP in VPC sentences. In these sentences, an NP that initially appears to be the object of a verb on its own turns out to be the object of a VPC. Reanalysis in this case does not involve reassigning the relevant NP to another predicate; rather, it involves building a more complex predicate. If non-native speakers are limited in their ability to use plausibility during major structural reanalysis (of the type described above), this group should be unable to use this information effectively when recovering from misanalysis in subordinate-clause ambiguity sentences. However, they should not have similar difficulties using plausibility to recover from misanalysis in VPC sentences. In evaluating these predictions, it is also important to note that the present study tested Korean learners of English. This is important because Korean sentences that are analogous to English subordinate-clause ambiguity sentences do not involve similar ambiguity. Indeed, the canonical subject–object–verb word order in Korean as well as its overt case marking would not allow for a grammatical sentence in which an NP is temporarily ambiguous as either a subordinate-clause object or main-clause subject. It is also the case that Korean lacks VPCs. It is therefore unlikely that the L2 processing of either subordinate-clause ambiguity sentences or (split/adjacent) VPC sentences would be influenced by the direct transfer of L1-based processing procedures in this study.
Method
Participants
Two groups of participants were tested—72 English native speakers and 45 non-native speakers. The participants in the native-speaker group were undergraduate students at the University of Texas at Arlington who took part in the experiment for course credit. These participants indicated that English was their dominant language and that they had learned this language in early childhood (before the age of 5 years). Participants in the non-native group were proficient Korean learners of English who had completed or were studying in degree-granting undergraduate or graduate programmes in the U.S. at the time of the experiment. They were paid for their participation. The data from 13 of non-native participants were eliminated due to high error rates on comprehension questions during the reading experiment or on the VPC multiple-choice test (see below). For the participants who were included in the analyses, the mean age of onset for English learning was 10.65 (SD = 4.48), while the mean estimated age of English fluency was 19.06 (SD = 6.37). Although the average length of residence in the U.S. was 7.68 years (SD = 4.21), none of the participants reported that English had become their dominant language. On a 0–10 self-rated proficiency scale (0 = lowest, 10 = highest), these participants indicated that they were stronger in Korean than in English with respect to speaking (Korean: M = 9.77, SD = 0.56; English: M = 7.53, SD = 1.68), listening (Korean: M = 9.77, SD = 0.50; English: M = 7.78, SD = 1.30), and reading skills (Korean: M = 9.74, SD = 0.51; English: M = 7.87, SD = 1.25).
Materials and design
The subordinate-clause ambiguity items consisted of 40 sets of sentences, as in (6) above, in which plausibility (implausible, plausible) and temporary structural ambiguity (ambiguous, unambiguous) were manipulated. Plausibility was manipulated by pairing the second NP in each sentence with two different subordinate verbs—one for which it was an implausible object (While Emma drank the song. . .) and another for which it was a plausible object (While Emma played the song . . .). (See below for norming data related to this distinction.) Each subordinate verb appeared in two sets of items, once in plausible conditions and once in implausible conditions. Temporary structural ambiguity was manipulated by the presence/absence of a comma that indicated that the second NP was the subject of the main clause (ambiguous: While Emma played the song. . .; unambiguous: While Emma played, the song. . .). In each item, a three-word “buffer” region intervened between the head of the ambiguous NP and the disambiguating main-clause verb (While Emma played the song
The VPC items consisted of 72 sets of sentences, as in (7) above, in which VPC sentence type (opaque-implausible, opaque-plausible, transparent-plausible) and particle placement (split: Mary ran the fox. . . over. . . vs. adjacent: Mary ran over the fox. . .) were manipulated. The first of these factors related primarily to whether the NP immediately following the verb in the split versions of these sentences was implausible or plausible as the object of the verb on its own (implausible: Mary ran the fox . . . plausible: Andy turned the key . . ., Albert called the family . . .). (Again, see below for norming data related to this distinction.) This factor is also related to semantic transparency. All of the implausible items involved relatively opaque VPCs, in which the meaning of the construction depended substantially on both the verb and the particle (ran over). For the plausible sentences, however, one set was constructed from opaque VPCs (turned in), while the other was constructed from transparent VPCs (called up). The experiment included 24 (split/adjacent) items for each of these sentence types. In each sentence, a four-word “buffer” region followed the head of object NP (Mary ran the fox
The items also included 32 filler sentences. These filler items represented a range of structures and were comparable to the experimental items in terms of complexity and length (filler: M = 85.53 characters; experimental: M = 86.00 characters). Each filler item was followed by a yes/no comprehension question, with an equal number of correct “yes” and “no” answers (e.g., Jason complained about the waiter at the restaurant, and the chef gave him a free dessert. / Was Jason happy with the waiter?). Altogether, there were 144 items in the experiment, which were organised into four counterbalanced lists.
Plausibility norming
To examine whether the manipulations in the experimental items (40 subordinate-clause ambiguity sentences; 72 VPC sentences) produced the intended plausibility contrasts, 28 English native speakers (none of whom took part in the reading experiment) participated in a plausibility judgement task for course credit. This task was conducted online, using the web-based implementation of the DMDX software package (Forster & Forster, 2003; J. Witzel et al., 2013). In this task, participants judged the plausibility of the beginning portions of the experimental items on a 1–5 scale (1 = completely implausible, 5 = completely plausible). For the subordinate-clause ambiguity items, participants were presented with sentences consisting of each pairing of the verb and the immediately following NP, yielding an implausible and plausible version of each item (e.g., implausible: Emma drank the song./plausible: Emma sang the song.). For the VPC items, participants were presented with the beginning of each sentence under its split and adjacent versions (e.g., opaque-implausible: Mary ran the fox./Mary ran over the fox; opaque-plausible: Andy turned the key./Andy turned in the key; transparent-plausible: Albert called the family./Albert called up the family.). Two counterbalanced lists of items were constructed, with one version of each item appearing in each list. The items were presented in a different random order for each participant, and eight practice items were included at the beginning of the task.
For the subordinate-clause ambiguity sentences, the implausible versions were rated reliably lower than their plausible counterparts (implausible: M = 1.67, SEM = 0.06; plausible: M = 4.82, SEM = 0.06; F1(1, 26) = 740.46, p < .001; F2(1, 38) = 1071.44, p < .001). For the VPC items, the mean plausibility ratings for the split and adjacent versions of each sentence type are presented in Figure 1. An analysis of these ratings revealed a significant interaction of VPC sentence type and particle placement (F1(2, 52) = 110.19, p < .001; F2(2, 66) = 45.92, p < .001). Under pairwise comparisons, split sentences had significantly lower ratings than their adjacent counterparts only under the opaque-implausible condition (F1(1, 26) = 119.98, p < .001; F2(1, 22) = 98.74, p < .001). For the opaque-plausible items, there was a non-significant trend in the opposite direction, with the split versions rated slightly higher than the adjacent versions (F1(1, 26) = 6.35, p < .05; F2(1, 22) = 2.25, p = .15). For the transparent-plausible sentences, there was no reliable difference in the ratings for split and adjacent sentences (both F’s < 1). These findings clearly indicate that the plausibility manipulations in the experimental items produced the expected plausibility contrasts for both the subordinate-clause ambiguity sentences and the VPC sentences.
Mean plausibility ratings for the split and adjacent versions of each VPC sentence type.
Multiple-choice test
A multiple-choice test was created to assess the non-native speakers’ knowledge of the VPCs that were used in the experimental items. In this test, participants were asked to read sentences and choose the definition for the VPC that best fit the context (e.g., I
Procedure
The experiment used a word-by-word, self-paced, non-cumulative moving-window reading task (Just et al., 1982), which was run on desktop computers with DMDX software (Forster & Forster, 2003). Each trial began with a line of dashes on the computer screen in place of the words in the sentence. The first word was displayed when the participant pressed a button on a game controller. Each subsequent button press revealed the next word in the sentence and masked the previous word. The time between the presentation of each word and the subsequent button press was recorded. Each sentence was followed by a yes/no comprehension question, which the participant answered by pressing one of two buttons on the game controller. After answering this question, the participant received feedback (“correct”/“wrong”), and the next trial began automatically. Sentences and their accompanying questions were presented on a single line in the centre of the computer screen. Participants were instructed to read as quickly as possible, but carefully enough to provide correct answers to the comprehension questions. The task began with eight practice items. Experimental and filler items were then presented in a different random order for each participant in sets of 18, with a short break after each set.
After completing the reading task, participants in the non-native speaker group also took the VPC multiple-choice test, which was again administered using DMDX. Sentences that included the VPCs from the experimental items were presented one at a time on the computer screen, along with four possible definitions. For each item, the participant selected the definition of the VPC that best fit the context using the 1–4 keys on the keyboard. No feedback was provided during this test. The task began with three practice items. The rest of the items were then presented in a different random order for each participant.
Results
The data from participants with accuracy rates of less than 75% on the complete set of comprehension questions (i.e., the questions for experimental and filler items) were eliminated from the analyses. Based on this criterion, the data from four native speakers and 11 non-native speakers were excluded. For the non-native speakers, the data from two participants with accuracy rates of less than 80% on the VPC multiple-choice test were also removed. Thus, the data from 68 native speakers and 32 non-native speakers were included in the analyses. The mean accuracy rate on comprehension questions was 89.26% (SD = 4.48) for the native speakers and 83.06% (SD = 3.23) for the non-native speakers. The mean accuracy rate on the VPC multiple-choice test for the non-native speaker group was 90.23% (SD = 5.77).
The statistical analyses for the subordinate-clause ambiguity sentences and the VPC sentences are presented separately below. The RT data were analysed only for trials on which the participant answered the comprehension question correctly. Prior to the analysis of these data, RTs that were two standard deviations above or below a participant’s mean for a given region were replaced with the value two standard deviations above or below that mean (4.91% of the data).
Subordinate-clause ambiguity sentences
The data were analysed with mixed-effects regression models using the lme4 package (Bates et al., 2015) in R (version 4.0.5; R Development Core Team, 2021). For comprehension question accuracy, these models were fitted using the glmer function; for the RT data, they were fitted with the lmer function. In these models, language group (native, non-native), local plausibility (plausible, implausible), and ambiguity (unambiguous, ambiguous) were contrast-coded fixed effects. The models included the maximal random effects structure whenever possible (Barr et al., 2013). In cases where the maximal model failed to converge, the random effects structure was simplified—by iteratively eliminating the random slope associated with the highest order interaction—until a converging model was achieved. (Appendix B in the online Supplementary Material provides the structure of each model reported below.) The p-values for the analysis of comprehension question accuracy were based on the z distribution; p-values for the RT analyses were estimated using the Satterthwaite approximation implemented in the lmerTest package (Kuznetsova et al., 2017). The RT analyses were conducted separately for five regions: a comma region, consisting of the subordinate clause verb and the immediately following article in the main-clause subject (drank(,)/played(,) the); a subject noun region, consisting of the head noun of the main-clause subject (song); a buffer region, consisting of the three words following the subject head noun (about true love); a main-clause verb region (amused); and a spillover region, consisting of the three words following the main-clause verb (all the customers). For multi-word regions, the RT corresponded to the mean RT over the component words. The raw RTs for each region were log transformed prior to analysis to meet the distributional assumptions of the statistical models. Estimated marginal means and standard errors for the log-transformed RTs were calculated based on the mixed-effects model reported for each region using the emmeans package (Lenth, 2020). This package was also used to conduct relevant pairwise comparisons. For these pairwise comparisons, the p-values were based on the z distribution—the default for data of this size.
Table 1 presents the mean accuracy rates on comprehension questions by language group and sentence condition. Table 2 presents the results of the mixed-effects regression analysis of these accuracy data. This analysis revealed a significant effect of group, indicating generally lower comprehension accuracy rates for the non-native speakers. There was also a reliable effect of ambiguity, indicating that both groups responded less accurately to comprehension questions following temporarily ambiguous items. Finally, there was a significant interaction of group and plausibility, reflecting the fact that the native speakers answered comprehension questions less accurately after plausible sentences.
Mean accuracy rates on comprehension questions by (language) group and sentence condition, subordinate-clause ambiguity sentences.
Mixed-effects regression estimates for the fixed effects of (language) group, plausibility, and ambiguity as well as their interactions on log odds of error for comprehension questions, subordinate-clause ambiguity sentences. Values in bold indicate significant (p < .05) and marginally significant (p < .10) effects.
SE: standard error.
Table 3 presents the estimated marginal means and standard errors for the log-transformed RTs in each analysis region by language group and sentence condition. Table 4 presents the results of the mixed-effects regression analyses of these RT data. Across all analysis regions, there was a robust effect of group, indicating longer RTs for non-native speakers. In the comma region, there was a clear clause-boundary effect. This was indicated by a reliable effect of ambiguity, with longer RTs for unambiguous sentences—that is, sentences with a comma—than for ambiguous sentences. At the subject noun, there were no reliable effects related to the sentence condition factors. However, there was a clear local plausibility effect for implausible/ambiguous sentences in the immediately following buffer region. In this region, there were reliable effects of plausibility and ambiguity as well as a significant interaction of these factors, indicating especially long RTs for implausible/ambiguous sentences in both native and non-native speakers. Indeed, in both participant groups, there were longer RTs for implausible/ambiguous sentences than for their unambiguous controls (native speakers: β = 0.05, SE = 0.02, z = 2.09, p = .036; non-native speakers: β = 0.14, SE = 0.04, z = 3.97, p < .001), but there were no reliable differences between plausible/ambiguous sentences and their unambiguous counterparts (native speakers: β = –0.02, SE = 0.02, z = –1.11, p = .267; non-native speakers: β = –0.03, SE = 0.03, z = –0.92, p = .358). In this region, there was also a marginally significant interaction of group and plausibility as well as a significant three-way interaction of group, plausibility, and ambiguity. These interactions indicate that the local plausibility effect in this region was larger for non-native speakers. A similar pattern of results was observed at the (disambiguating) main-clause verb. In this region, there was a marginally significant effect of plausibility as well as a reliable effect of ambiguity that was modulated by significant interaction of plausibility and ambiguity. Again, in both participant groups, there were longer RTs for implausible/ambiguous sentences than for their unambiguous controls (native speakers: β = 0.09, SE = 0.03, z = 3.24, p = .001; non-native speakers: β = 0.12, SE = 0.05, z = 2.71, p = .007), but there were no reliable differences between plausible/ambiguous sentences and their unambiguous counterparts (native speakers: β = 0.01, SE = 0.03, z = 0.27, p = .786; non-native speakers: β = 0.05, SE = 0.05, z = 0.98, p = .326). A different pattern of results was observed in the immediately following spillover region. In this region, there was marginally significant interaction of group and plausibility, reflecting a trend whereby native speakers had longer RTs for plausible sentences, while non-native speakers had longer RTs for implausible items. More importantly, this region revealed a robust effect of ambiguity that was modulated by a marginally significant interaction of plausibility and ambiguity. This pattern of results indicated a clear garden-path effect that was especially large for the plausible sentences. Indeed, while native speakers had longer RTs for ambiguous sentences under both the implausible and plausible sentence conditions (implausible: β = 0.04, SE = 0.02, z = 2.63, p = .009; plausible: β = 0.06, SE = 0.02, z = 3.05, p = .002), non-native speakers showed a reliable effect of ambiguity only for plausible sentences (implausible: β = 0.02, SE = 0.03, z = 0.96, p = .336; plausible: β = 0.09, SE = 0.03, z = 3.21, p = .001).
Estimated marginal means and standard errors (in parentheses) for the log-transformed RTs in each analysis region by (language) group and sentence condition, subordinate-clause ambiguity sentences. Millisecond equivalents for the estimated marginal means are provided in brackets.
While Emma drank(,)/played(,) the song about true love amused all the customers at the club.
Mixed-effects regression estimates for the fixed effects of (language) group, plausibility, and ambiguity as well as their interactions on the log-transformed RTs in each analysis region, subordinate-clause ambiguity sentences. Values in bold indicate significant (p < .05) and marginally significant (p < .10) effects.
SE: standard error.
While Emma drank()/played() the song about true love amused all the customers at the club.
VPC sentences
The analysis procedures for the VPC sentences followed those detailed above for the subordinate-clause ambiguity sentences. In the statistical models for these sentences, language group (native, non-native), sentence type (transparent-plausible, opaque-plausible, opaque-implausible), and particle position (adjacent, split) were fixed effects. Language group and particle position were both contrast coded, while sentence type was treatment coded, with transparent-plausible as the reference level. (Again, see Appendix B in the online Supplementary Material for the structure of each model reported below.) The RT analyses were conducted separately for four regions: an object noun region, consisting of the head noun of the VPC object; a buffer region, consisting of the four words following the object head noun; the VPC particle; and a spillover region, consisting of the three words following the VPC.
Table 5 presents the mean accuracy rates on comprehension questions by language group and sentence condition. Table 6 presents the results of the mixed-effects regression analysis of these accuracy data. As with the subordinate-clause ambiguity sentences, this analysis revealed a significant effect of group, indicating generally lower comprehension accuracy rates for the non-native speakers. There was also a reliable interaction of group and particle position in the comparison of opaque-plausible sentences with the transparent-plausible baseline. This effect was driven by native speakers’ particularly low accuracy rates on split versions of opaque-plausible sentences. Indeed, while native speakers had lower accuracy rates on opaque-plausible/split sentences than on their adjacent counterparts, this was not the case for the baseline transparent-plausible sentences.
Mean accuracy rates on comprehension questions by (language) group and sentence condition, VPC sentences.
Mixed-effects regression estimates for the fixed effects of (language) group, sentence type, and particle position as well as their interactions on log odds of error for comprehension questions, VPC sentences. Values in bold indicate significant (p < .05) and marginally significant (p < .10) effects.
SE: standard error.
Table 7 presents the estimated marginal means and standard errors for the log-transformed RTs in each analysis region by participant group and sentence condition. Table 8 presents the results of the mixed-effects regression analyses of these RT data. As with the subordinate-clause ambiguity sentences, there was a robust effect of group across all analysis regions, indicating longer RTs for non-native speakers. At the object noun, there were no reliable effects related to the sentence condition factors. However, there was a clear local plausibility effect for opaque-implausible/split sentences in the immediately following buffer region. In this region, both sentence type comparisons (transparent-plausible vs. opaque-plausible; transparent-plausible vs. opaque-implausible) were significant, indicating that both opaque-plausible and opaque-implausible sentences had longer RTs than the transparent-plausible baseline. More importantly, there was a reliable effect of particle position in the comparison of opaque-implausible sentences with the transparent-plausible baseline. This effect reflects the fact that while both native and non-native speakers had longer RTs on the split versions of opaque-implausible sentences than on their adjacent controls (native speakers: β = 0.05, SE = 0.02, z = 3.60, p < .001; non-native speakers: β = 0.09, SE = 0.02, z = 4.02, p < .001), no such difference was observed for either group in transparent-plausible sentences (native speakers: β = 0.01, SE = 0.02, z = 0.63, p = .531; non-native speakers: β = 0.03, SE = 0.02, z = 1.36, p = .174). At the particle, there was reliable effect of particle position in the comparison of opaque-implausible sentences with the transparent-plausible baseline and in the comparison of opaque-plausible sentences with these baseline sentences. In addition, there was a significant interaction of participant group and particle position. These effects reflect the fact that for native speakers, there were longer RTs for split particles than for adjacent particles, but only for opaque-implausible and opaque-plausible sentences (opaque-implausible: β = 0.08, SE = 0.02, z = 3.64, p < .001; opaque-plausible: β = 0.06, SE = 0.02, z = 2.94, p = .003; transparent-plausible: β = 0.02, SE = 0.02, z = 0.78, p = .435). For non-native speakers, however, there were no reliable indications of processing time differences for split particles in any of the sentence types. A different pattern of results was obtained in the spillover region. In this region, there was a reliable effect indicating that opaque-plausible sentences had longer RTs than the transparent-plausible baseline. More importantly, there was a significant effect of particle position that was qualified by a reliable interaction of group and particle position. In this case, this interaction was driven by a larger effect of particle position for non-native speakers than for native speakers. Indeed, although non-native speakers had significantly longer RTs in the split versions of all three sentence types (opaque-implausible: β = 0.06, SE = 0.02, z = 2.39, p = .017; opaque-plausible: β = 0.06, SE = 0.03, z = 2.59, p = .010; transparent-plausible: β = 0.07, SE = 0.02, z = 2.79, p = .005), native speakers only had reliably longer RTs for split versions of opaque-plausible sentences (opaque-implausible: β = 0.03, SE = 0.02, z = 1.58, p = .113; opaque-plausible: β = 0.05, SE = 0.02, z = 3.06, p = .002; transparent-plausible: β = 0.01, SE = 0.02, z = 0.43, p = .667).
Estimated marginal means and standard errors (in parentheses) for the log-transformed RTs in each analysis region by (language) group and sentence condition, VPC sentences. Millisecond equivalents for the estimated marginal means are provided in brackets.
Opaque-implausible/split: Mary ran the fox that crossed the road over when it jumped in front of her car.
Opaque-plausible/split: Andy turned the key for his new office in and then left to catch the train.
Transparent-plausible/split: Albert called the family who visited his church up after he talked to his pastor.
Mixed-effects regression estimates for the fixed effects of (language) group, sentence type, and particle position as well as their interactions on the log-transformed RTs in each analysis region, VPC sentences. Values in bold indicate significant (p < .05) and marginally significant (p < .10) effects.
SE: standard error.
Opaque-implausible/split: Mary ran the fox that crossed the road over when it jumped in front of her car.
Opaque-plausible/split: Andy turned the key for his new office in and then left to catch the train.
Transparent-plausible/split: Albert called the family who visited his church up after he talked to his pastor.
Discussion
This study investigated whether English native speakers and non-native speakers make comparable use of plausibility during structural reanalysis in subordinate-clause ambiguity sentences and sentences with adjacent/split VPCs. For the subordinate-clause ambiguity sentences, both groups showed clear processing difficulty at and immediately after the comma in unambiguous control items. These results are in line with previous self-paced reading studies of this sentence type (Sturt et al., 1999; Traxler, 2002) and likely reflect processing costs related to establishing a clause boundary at this point. Although this effect was not of particular interest in this study, it is worth noting because it indicates that this comma induced the anticipated structural processing, and thus that sentences with this punctuation provided appropriate controls for their temporarily ambiguous counterparts. Both groups also showed local plausibility effects in these sentences, with longer RTs when the temporarily ambiguous NP was an implausible object for the subordinate-clause verb. This indicates that both native and non-native speakers initially analysed the ambiguous NP as the object of the subordinate-clause verb and that they experienced processing difficulty when this (mis)analysis led to an implausible interpretation. Interestingly, these plausibility effects were not obtained on the head of this NP—the word that provides the first indication of implausibility in these sentences. Rather, these effects were initially observed on words intervening between this head and the disambiguating main-clause verb (i.e., the buffer region in the analyses above). This points to the importance of including such words in experiments of this type, particularly those that use self-paced reading, in which processing time differences are regularly observed “downstream” of the predicted word/region (N. Witzel et al., 2012). Finally, and most importantly, both groups showed garden-path effects after the disambiguating verb in temporarily ambiguous sentences, which were attenuated for sentences in the implausible/ambiguous condition. This reverse plausibility effect was especially clear for the non-native speakers. For this group, there was a reliable garden-path effect in the spillover region only for sentences in which the temporarily ambiguous NP was a plausible object for the subordinate-clause verb. These results therefore indicate that English native and non-native speakers are able to pick up on local implausibility and are able to use this information to facilitate structural reanalysis in subordinate-clause ambiguity sentences.
A different pattern of results was found for VPC sentences. For this sentence type, both native and non-native speakers again showed local plausibility effects. Specifically, they had inflated RTs for split VPC sentences in which the ambiguous NP was an implausible object for the verb on its own (i.e., in opaque-implausible/split sentences). This indicates that both native and non-native speakers initially analysed the ambiguous NP as the object of the verb on its own and that they experienced processing difficulty when this (mis)analysis led to an implausible interpretation. Just as in the subordinate-clause ambiguity sentences, this plausibility effect was not observed on the head noun of the relevant NP, but rather in the immediately following region. The results at and after the disambiguating split particle, however, were different for native and non-native speakers. At the split particle, native speakers showed processing difficulty for sentences with opaque VPCs—or for sentences in which this particle required syntactic and semantic reanalysis. These costs then were attenuated in locally implausible sentences. Specifically, in the spillover region immediately following the split particle, there was processing difficulty for opaque-plausible sentences, but not for opaque-implausible sentences. Non-native speakers, however, showed processing difficulty for split VPC sentences across all three VPC sentence types in this spillover region. This indicates that while both native and non-native speakers were able index implausibility in split VPC sentences, only native speakers were able to use this information to facilitate structural reanalysis in these sentences.
So what do these findings suggest about possible differences between native and non-native speakers in terms of their ability to use plausibility to inform structural reanalysis? As discussed above, Roberts and Felser (2011) proposed that non-native speakers might be limited in their ability to use plausibility during reanalysis of the type involved in subordinate-clause ambiguity sentences—in particular, during reanalysis processes that require major adjustments to both the thematic and hierarchical structural relations relevant to temporarily ambiguous constituents. It was further argued that this proposal is consistent with the more general idea that structural processing is less automatic or reduced in non-native speakers compared with native speakers (see, for example, Clahsen & Felser, 2006). However, the findings for non-native speakers in the present study are not consistent with this proposed limitation. First, the non-native speakers in this study appeared to use plausibility to facilitate reanalysis in subordinate-clause ambiguity sentences. That is, reanalysis processes that should have been resistant to the influence of plausibility were in fact susceptible to its influence. The results for VPC sentences are also important to consider with respect to this proposed limitation. As detailed above, recovering from misanalysis in split VPC sentences does not necessitate major structural reorganisation of the type involved in subordinate-clause ambiguity sentences. Under the proposal above, non-native speakers should therefore have been able to use plausibility to facilitate reanalysis in split VPC sentences. But this was not the case. That is, reanalysis processes that should have been susceptible to the influence of plausibility were in fact resistant to its influence.
A relevant question then is whether the findings from this study suggest a different limitation. One possibility is that non-native speakers are limited in their ability to use plausibility information during L2 reanalysis processes (and in L2 structural processing more generally) in cases when they are unable to efficiently execute lower-level language processing, and lexical processing in particular. Such a proposal would be consistent with the lexical bottleneck hypothesis (see, for example, Hopp, 2018), which holds that non-native speakers’ inability to demonstrate native-like structural processing often stems from lexical processing difficulties in the L2. Specifically, this model posits that more effortful lexical processing in the L2 can overload computational capacities, leading to delays or disruptions in L2 structural processing. Moreover, under this hypothesis, diffuse lexical-level activation from L2 words can result in less efficient access to structurally relevant information encoded in L2 lexical representations. A model along these lines can be applied to the results for the VPC sentences in the present study. Again, it is important to reiterate that the native-speaker group had reanalysis-related processing difficulty only for split sentences with relatively opaque VPCs. Sentences with transparent VPCs did not trigger comparable processing difficulty at or after the split particle. These differences are likely due to the nature of the revision processes involved in these sentences. Although the particle required both syntactic and semantic reanalysis in split sentences with opaque VPCs, this was not the case for those with transparent VPCs. For these sentences, the particle did not necessitate semantic revision and thus could be parsed into the VPC predicate without discernible processing costs. This was not true, however, for the non-native speakers. This group experienced processing difficulty after the split particle regardless of the nature of the VPC. Interestingly, these findings are similar to those of Tiv and colleagues (2019), who also found a processing cost for split VPC sentences that was uninfluenced by item-level characteristics related to VPC frequency and semantic transparency. Taken together, these findings suggest that non-native speakers have general processing difficulty related to interpreting the components of VPC constructions, especially when these components are separated by intervening material. It could be the case that non-native speakers are unable to use plausibility information effectively during L2 structural reanalysis in the context of lexical-level processing difficulty of this type.
This proposed limitation should of course be investigated in future research. One general research strategy would be to manipulate lexical processing demands (e.g., by using words from high and low-frequency bands) and local plausibility during the comprehension of sentences that require structural reanalysis. The clear prediction is that non-native speakers should be able to use plausibility to inform structural processing when lexical processing demands are low, but not when these demands are high. With respect to the sentences of particular interest in the present study, it might also be interesting to investigate the processing of split VPC sentences with learners from different L1 backgrounds. As indicated above, learners from L1s that also have VPCs—like Dutch and Swedish—have been shown to have avoid these forms less often in their L2 English production compared with learners from L1s that lack these forms (González, 2010; Hulstijn & Marchena, 1989; Laufer & Eliasson, 1993). To the extent that these findings suggest that English VPCs are easier to process for learners whose L1s also have these forms, such learners might be able to use plausibility information to inform structural reanalysis in split VPC sentences.
Footnotes
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Funding for participant payment was provided by the University of Texas at Arlington Linguistics & TESOL Alumni Endowment.
