1. Encoding and Navigating Structure
We can easily diagram sentence structures on a blackboard. But what do those diagrams correspond to in (neuro-)cognitive terms? And what does it mean to “form dependencies” and “manipulate the structures”? This is the research question that our team has worried about the most in recent years.
Two independent developments converged to make this a very productive topic for us.
Selective fallibility and linguistic illusions: around 2000 we started to explore whether detailed grammatical constraints are respected in comprehension. In other words, does the parser construct all and only the representations that are sanctioned by the grammar. The original hope was that the answer would be “yes”, and for a few years the results pointed to that conclusion.
But things got a lot more interesting once we dug into cases of linguistic illusions, i.e., cases where people appear to accept incoherent sentences, e.g., More people have been to Russia than I have, or agreement with an incorrect controller, e.g., The key to the cabinets were on the table, or licensing of a negative polarity item by an inappropriate negation, e.g., The bills that no democratic senators voted for will ever become law. The contrast between the cases of success and failure begs the question of what is responsible for the “selective fallibility”.
Content-addressable memory: Research on memory processes provides strong support for content-addressable memory (CAM), i.e., items in memory are accessed based on retrieval cues rather than on a memory address. This is why it’s hard to remember what you ate for dinner 4 days ago. Psychologists Brian McElree and Peter Gordon highlighted evidence that this memory architecture is also used for human language. CAM-based access has distinctive timing profiles (long ≠ slow) and interference profiles (mis-retrieval of items similar to the target memory). Computational psycholinguists Shravan Vasishth and Rick Lewis captured these properties in an implemented and accessible model (ACT-R). The model makes clear predictions, which only partly fit our selective fallibility profile.
This set the stage for research in which a cycle of modeling and experimentation seeks to explain the selective fallibility profile. This generated clear hypotheses about how structures are encoded and navigated mentally. In our team, Matt Wagers, Ellen Lau, Brian Dillon, and Ming Xiang set us down this path.
Phase 1 – Dependency-wise differences: our first bet was that selective fallibility should be understood at the grain size of dependency types, e.g., subject-verb agreement is prone to illusions but subject-reflexive licensing is not (Dillon et al. 2013). We ran many studies in pursuit of this idea. And we found ways to capture both sides of the selective fallibility profile within a CAM architecture.
Phase 2 – Turning illusions on/off: recent findings by Dan Parker and Dave Kush challenge our earlier framing of the problem. Dan found that robust illusions could be turned off (NPIs: Parker & Phillips 2014a), and that illusions could be turned on in cases that we had thought immune (reflexives: Parker & Phillips 2014b). Meanwhile, Dave’s studies on c-command sensitive phenomena such as bound variable anaphora suggested that comprehenders can show greater structure sensitivity than we had thought possible in a CAM-based architecture.
Getting to the bottom of this will keep us busy for a while yet.
2. Linguistic Architecture
Are locality constraints in grammar (islands etc.) and language processing (longer = harder) ultimately the same? This possibility attracts many in both fields, but it is challenged by our findings from head-final languages such as Japanese and Bengali, whose properties distinguish structural and linear/temporal locality. (Aoshima et al. 2004; Omaki et al. 2014; Chacón et al. 2014)
A routine assumption in linguistics and psycholinguistics alike is that combinatorial semantic interpretation closely follows syntactic structure. Recent ERP findings have been taken as challenges to this orthodoxy, motivating independent semantic composition. Our own ERP studies in English, Chinese, and Spanish support the standard view. Perhaps. (Stroud & Phillips 2012; Chow & Phillips 2013)
Comprehension-production differences present a prima facie argument for a traditional linguistic architecture with a task- and time-independent grammar. The grammar is what links the comprehension and production systems. Shota Momma, Bob Slevc and I are trying to better understand this relation, focusing initially on look-ahead in production. (Momma et al. 2014ab)
If there’s no task-specific comprehension system, distinct from the grammar, then we should be able to explain language processing phenomena by embedding a grammar in a general cognitive architecture, with no mechanisms specific to language processing. We have recently begun to explore the feasibility of this. (Lewis & Phillips 2014; Parker 2014)
We suspect that predictive mechanisms may be important contributors to the speed and robustness of language understanding. When comprehenders can anticipate what is likely to be said, they can process the input more quickly and more successfully in noise.
In earlier work we argued that predictive mechanisms are responsible for the ELAN, an ERP response whose speed is almost too good to be true (Lau et al. 2006). We suggested that predictive mechanisms also underlie some instances of extreme grammatical fidelity in comprehension (Phillips et al. 2011). And we argued that the N400 ERP response is a relatively direct reflection of lexical predictions in comprehension (Lau et al. 2008).
But Wing Yee Chow’s recent findings on the N400’s temporary “blindness” to highly predictive information makes things rather more interesting. Her studies on role-reversed sentences show that some types of predictive information is accessed more quickly than others. This clears a path for explicit models of how predictions are generated. More to come!
4. Variation and Learning
Can differences in language processing abilities aid our understanding of learning outcomes? Perhaps — it depends what you mean.
Explaining failures: language processing abilities may account for some of children’s failings. Errors in the interpretation of anaphora such as the notorious Delay of Principle B Effect may reflect immature reanalysis and cognitive control systems, rather than failure of grammatical learning (Conroy et al. 2009).
Distorted input? In light of the well-documented cases of children’s parsing errors (e.g., Trueswell et al. 1999), there is a danger that children’s misparsing could lead to distortion of crucial distributional evidence for learning. Example: if a child only ever accesses a local interpretation of an ambiguous wh-dependency, might she incorrectly conclude that the target grammar only allows local dependencies? Akira Omaki has paid close attention to this concern (Omaki et al. 2014), and he’ll continue to do so. The jury remains out on how serious this concern is.
Explaining success? Is there any way that population differences in language processing abilities could explain why children are more successful language learners than adults? Good question. Check back soon.
Publications in Language Processing
(including PhD dissertations)
|de Dios Flores, Iria; Muller, Hanna; Phillips, Colin: The selectivity and interpretation of NPI illusions. In: Glossa Psycholinguistics, Forthcoming, (in revision). (Type: Journal Article | Abstract | Links | BibTeX)|
This study investigates the illusory licensing of Negative Polarity Items (NPIs), a subtype of linguistic illusion whose behavior is informative for understanding real-time interpretation. We present the results from six experiments, including online/offline judgments and interpretation tasks, in an attempt to identify the processing error that underlies this brief deviation from the mental grammar. The results indicate that illusions are robust for intrusive quantificational negation such as "no authors" but reduced or absent for other forms of intrusive negation such as "didn't". We additionally find that although NPI illusion sentences are often interpreted as expressing a globally negative proposition, these interpretive errors are a consequence of the illusion, not a likely cause of it. These findings pose problems for current accounts which do not predict the error profile nor the interpretation patterns that we observe, including accounts that appeal to the erroneous retrieval of a non c-commanding negative item in memory (Vasishth et al. 2008), a pragmatic rescuing operation (Xiang et al. 2009) or the parser’s failure to assign an appropriate scope for negative quantifiers (Orth et al. 2020a). Alternatively, we argue for an account of NPI illusions that focuses on how sentences are interpreted in relation to scalar alternatives, invoking a style of grammatical explanation in the spirit of Fauconnier (1975a, 1975b), which emphasizes the critical role of the relation between the NPI and the context in which it appears.
| Huang, Nick; Phillips, Colin: When missing NPs make double center-embedding sentences acceptable. In: Glossa, 2021, (revised submission, to appear in Glossa). (Type: Journal Article | Abstract | Links | BibTeX)|
A number of languages, such as English, exhibit a grammaticality illusion in ungrammatical double center-embedding sentences where a VP is missing. This article shows that the illusion generalizes to ungrammatical Mandarin Chinese double center-embedding sentences where the head NP of a relative clause is missing.
The Mandarin illusion raises interesting questions for existing accounts of center- embedding illusions. Mandarin missing NP sentences consist of three transitive verbs and only three NPs; the clear shortage of NPs should affect the thematic relations built for such sentences, with potential consequences for acceptability. We explore these issues with acceptability judgment experiments. We show that these illusory sentences receive rather distinct thematic interpretations compared to their better-studied missing VP counterparts, in ways not predicted by structural forgetting or interference accounts. A computational simulation further shows that the Mandarin illusion is problematic for accounts that attribute cross-linguistic variation in the illusion to differences in language experience.
To capture cross-linguistic variation, we build on existing interference accounts, in which the parser mis-attaches a verb or NP to the main clause instead of a relative clause. We supplement this approach with a repair process, in which the parser tracks thematic relations, repairing them where necessary so no verb or noun is “orphaned.” We suggest that the illusion of grammaticality arises when the parser can establish thematic relations between all verbs and nouns. This interference-and-repair approach provides a unified analysis of the missing VP and missing NP illusions, while accounting for the observed difference in thematic relations.
| Malko, Anton: The role of structural information in the resolution of long-distance dependencies. University of Maryland, 2018. (Type: PhD Thesis | Abstract | Links | BibTeX)|
The main question that this thesis addresses is: in what way does structural information enter into the processing of long-distance dependencies? Does it constrain the computations, and if so, to what degree? Available experimental evidence suggests that sometimes structurally illicit but otherwise suitable constituents are accessed during dependency resolution. Subject-verb agreement is a prime example (Wagers et al., 2009; Dillon et al., 2013), and similar effects were reported for negative polarity items (NPIs) licensing (Vasishth et al., 2008) and reflexive pronouns resolution (Parker and Phillips, 2017; Sloggett, 2017). Prima facie this evidence suggests that structural information fails to perfectly constrain real-time language processing to be in line with grammatical constraints. This conclusion would fall neatly in line with an assumption that human sentence processing relies on cue-based memory (e.g McElree et al., 2003; Lewis and Vasishth, 2005; Van Dyke and Johns, 2012; Wagers et al., 2009, a.m.o.), the key property of which is the fragility of memory search, which can return irrelevant results if they look similar enough to the relevant ones. The attractiveness of such an approach lies in its parsimony: there is independent evidence that general purpose working memory is cue-based (Jonides et al., 2008), so we do not need to postulate any language specific mechanisms. Additionally, the processing of multiple linguistic dependencies can be analyzed within the same theoretical framework. Cue-based approach has also been argued to be the best one in terms of its empirical coverage: some of the experimental evidence was assumed to only be explainable within it (the absence of ungrammaticality illusions in subject-verb agreement is the main example, to which we will return in more detail later). However, recently several other approaches have been suggested which would be able to account for these cases (Eberhard et al., 2005; Xiang et al., 2013; Sloggett, 2017; Hammerly et al., draft.april.2018). These approaches usually assume separate processing mechanisms for different linguistic dependencies, and thus lose the parsimonious at- tractiveness of cue-based memory models. They also take a different stance on the role of structural information in real-time language processing, assuming that structural cues do accurately guide the dependency resolution. A priori there is no reason why they could not turn out to be true. But given the theoretical attractiveness of cue-based models in which structural information does not categorically restrain processing, it is important to critically evaluate these recent claims. In this thesis, we focus on reflexive pronouns and on the novel pattern reported in Parker and Phillips (2017) and Sloggett (2017): the finding that reflexive pronouns are sensitive to the properties of structurally inaccessible antecedents in some specific conditions (interference effect). The two works report consistent findings, but the accounts they give take opposite perspectives on the role of structural information in reflexive resolution. Our aim in this thesis is to assess the reliability of these findings and to experimentally investigate cases which would hopefully provide clearer evidence on how the structure guides reflexives processing. To this aim, we conduct two direct replications of Parker and Phillips (2017) and four novel experiments further investigating the properties of the interference effect. None of the six experiments provided strong statistical support for the previous findings. After ruling out several possible confounds and analyzing numerical patterns (which go in the expected direction and are consistent with previous results), we conclude that interference effect is likely real, but may be less strong than the previous studies would lead to believe. These results can be used for setting more realistic expectations for future studies regarding the size of the effect and statistical power necessary to detect it. With respect to our main goal of distinguishing between cue-based and alternative accounts of the interference effect, we tentatively conclude that cue-based approaches are preferred; however, one has to assume that some structural features are able to categorically rule out illicit antecedents. Further highly powered studies are necessary to verify and confirm these conclusions.
| Wellwood, Alexis; Pancheva, Roumyana; Hacquard, Valentine; Phillips, Colin: The anatomy of a comparative illusion. In: Journal of Semantics, 2018, (in press). (Type: Journal Article | Abstract | Links | BibTeX)|
Comparative constructions like "More people have been to Russia than I have" are generally perceived as acceptable and meaningful by native speakers of English; yet, upon closer reflection, they are judged to be incoherent. This mismatch between initial perception and more considered judgment challenges the idea that we perceive sentences veridically, and interpret them fully; it is thus potentially revealing about the relationship between grammar and language processing. This paper presents the first detailed investigation of these so-called ‘comparative illusions’. We test four hypotheses about their source: a shallow syntactic parser, some type of repair by ellipsis, an incorrectly-resolved lexical ambiguity, or a persistent event comparison interpretation. Two formal acceptability studies show that speakers are most prone to the illusion when the comparative supports an event comparison reading. A verbatim recall task tests and finds evidence for such construals in speakers’ recollections of the sentences. We suggest that this reflects speakers’ entertaining an interpretation that is initially consistent with the sentence, but failing to notice when this interpretation becomes unavailable at the than-clause. In particular, semantic knowledge blinds people to an illicit operator-variable configuration in the syntax. Rather than illustrating processing in the absence of grammatical analysis, comparative illusions thus underscore the importance of syntactic and semantic rules in sentence processing.
| Chow, Wing Yee; Lau, Ellen; Wang, Suiping; Phillips, Colin: Wait a second! Delayed impact of argument roles on on-line verb prediction. In: Language, Cognition, and Neuroscience, 2018. (Type: Journal Article | Abstract | Links | BibTeX)|
Comprehenders can use rich contextual information to anticipate upcoming input on the fly, but recent findings suggest that salient information about argument roles may not impact verb prediction. We took advantage of the word order properties of Mandarin Chinese to examine the time course with which argument role information impacts verb prediction. We isolated the contribution of argument role information by manipulating the order of pre-verbal noun phrase arguments while holding lexical information constant, and we examined its effects on accessing the verb in long-term semantic memory by measuring the amplitude of the N400 component. Experiment 1 showed when the verb appeared immediately after its arguments, even strongly constraining argument role information failed to modulate the N400 response to the verb. An N400 effect emerged in Experiment 2 when the verb appeared at a greater delay. Experiment 3 corroborated the contrast between the first two experiments through a within-participants manipulation of the time interval between the arguments and the verb, by varying the position of an adverbial phrase. These results suggest time is a key factor governing how diverse contextual information contributes to predictions. Here argument role information is shown to impact verb prediction, but its effect is not immediate.
| Momma, Shota; Phillips, Colin: The relationship between parsing and generation. In: Annual Review of Linguistics, vol. 4, pp. 233-254, Forthcoming. (Type: Journal Article | Abstract | Links | BibTeX)|
Humans use their linguistic knowledge in at least two ways. On one hand, they use their linguistic knowledge to convey what they mean to others or to themselves. On the other hand, they use their linguistic knowledge to understand what others say or they themselves say. In either case, they must assemble the syntactic structures of sentences in a systematic fashion, in accordance with the grammar of their language. In this article, we advance the view that a single mechanism for building sentence structure may be sufficient for structure building in comprehension and production. We argue that differing behaviors reduce to differences in the available information in either task. This view has broad implications for the architecture of the human language system, and provides a useful framework for integrating largely independent research programs on comprehension and production by both constraining the models and uncovering new questions that can drive further research.
|Jeffrey Lidz Dave Kush, Colin Phillips: Looking forwards and backwards: the real-time processing of Strong and Weak Crossover. In: Glossa: A journal of general linguistics, vol. 2, no. 70, 2017, (29pp). (Type: Journal Article | Abstract | Links | BibTeX)|
We investigated the processing of pronouns in Strong and Weak Crossover constructions as a means of probing the extent to which the incremental parser can use syntactic information to guide antecedent retrieval. In Experiment 1 we show that the parser accesses a displaced wh-phrase as an antecedent for a pronoun when no grammatical constraints prohibit binding, but the parser ignores the same wh-phrase when it stands in a Strong Crossover relation to the pronoun. These results are consistent with two possibilities. First, the parser could apply Principle C at antecedent retrieval to exclude the wh-phrase on the basis of the c-command relation between its gap and the pronoun. Alternatively, retrieval might ignore any phrases that do not occupy an Argument position. Experiment 2 distinguished between these two possibilities by testing antecedent retrieval under Weak Crossover. In Weak Crossover binding of the pronoun is ruled out by the argument condition, but not Principle C. The results of Experiment 2 indicate that antecedent retrieval accesses matching wh-phrases in Weak Crossover configurations. On the basis of these findings we conclude that the parser can make rapid use of Principle C and c-command information to constrain retrieval. We discuss how our results support a view of antecedent retrieval that integrates inferences made over unseen syntactic structure into constraints on backward-looking processes like memory retrieval.
| Atkinson, Emily; Wagers, Matthew W.; Lidz, Jeffrey; Phillips, Colin; Omaki, Akira: Developing incrementality in filler-gap dependency processing. In: Cognition, 2017, (revised version to appear). (Type: Journal Article | Abstract | Links | BibTeX)|
Much work has demonstrated that children are able to use bottom-up linguistic cues to incrementally interpret sentences, but there is little understanding of the extent to which children’s comprehension mechanisms are guided by top-down linguistic information that can be learned from distributional regularities in the input. Using a visual world eye tracking experiment and a corpus analysis, the current study investigates whether 5- and 6-year-old children incrementally assign interpretations to temporarily ambiguous wh-questions like What was Emily eating the cake with __?. In the visual world eye-tracking experiment, 6-year-old children showed an adult-like bias to incrementally complete the filler-gap dependency at the earliest position in the sentence (i.e., verb), while no evidence for this bias was found in 5-year-olds. These results suggest that adult-like incrementality in filler-gap dependency processing begins to emerge around age 6. The corpus analysis of filler-gap dependency structures in adult corpora and child corpora demonstrate that the distributional regularities in either corpora are equally in favor of early, incremental completion of filler-gap dependencies, suggesting that the distributional information in the input is either not relevant to this incremental bias, or that 5-year-old children are somehow unable to recruit this information in real-time comprehension. Taken together, these findings shed light on the origin of incremental processing bias in filler-gap dependency processing, as well as on the role of language experience in the development of incremental sentence processing mechanisms.
|Colin Phillips Claudia Felser, Matthew Wagers: Editorial: Encoding and navigating linguistic representations in memory. In: Frontiers in Psychology, vol. 8, pp. 164, 2017. (Type: Journal Article | Links | BibTeX)|
| Parker, Dan; Phillips, Colin: Reflexive attraction in comprehension is selective. In: Journal of Memory and Language, vol. 94, pp. 272-290, 2017. (Type: Journal Article | Abstract | Links | BibTeX)|
Many studies have shown that attraction effects are consistently found for linguistic dependencies like subject-verb agreement, e.g., *The key to the cabinets are on the table. However, not all dependencies are equally susceptible to attraction. A parade case involves reflexive-antecedent dependencies, which rarely show attraction effects. The contrast between agreement and reflexives with respect to attraction has motivated various proposals regarding the memory architecture for the parser, including the use of qualitatively different access mechanisms or the selective use of morphological features as retrieval cues for different dependencies. In this paper, we show how to systematically induce attraction effects for reflexives in three eye-tracking experiments. Furthermore, we show based on computational simulations that it is possible to derive both the presence and absence of reflexive attraction from the same retrieval mechanism, based on the ACT-R architecture. We then propose an account of why agreement and reflexives are differentially susceptible to attraction, based on the predictability of the dependency.
| Lago, Sol; Sloggett, Shayne; Schlueter, Zoe; Chow, Wing Yee; Williams, Alexander; Lau, Ellen; Phillips, Colin: Coreference and antecedent representation across languages. In: Journal of Experimental Psychology: Learning, Memory, and Cognition, 2017. (Type: Journal Article | Abstract | Links | BibTeX)|
Previous studies have shown that speakers of languages such as German, Spanish and French reactivate the syntactic gender of the antecedent of a pronoun in order to license gender agreement. As syntactic gender information is assumed to be stored in the lexicon, this has motivated the claim that pronouns in these languages reactivate the lexical entry of their antecedent noun. In contrast, in languages without syntactic gender such as English, lexical retrieval might be unnecessary. Using eye-tracking while reading, we examined whether antecedent retrieval involves rapid semantic and phonological reactivation. We compared German and English. In German, we found early sensitivity to the semantic, but not to the phonological features of the pronoun’s antecedent. In English, readers did not immediately show either semantic or phonological effects specific to coreference. We propose that early semantic facilitation arises due to syntactic gender reactivation, and that antecedent retrieval may vary cross-linguistically depending on the type of information relevant to the grammar of each language.
| Parker, Dan; Phillips, Colin: Negative polarity illusions and the format of hierarchical encodings in memory. In: Cognition, vol. 157, pp. 321-339, 2016. (Type: Journal Article | Abstract | Links | BibTeX)|
Linguistic illusions have provided valuable insights into how we mentally navigate complex representations in memory during language comprehension. Two notable cases involve illusory licensing of agreement and negative polarity items (NPIs), where comprehenders fleetingly accept sentences with unlicensed agreement or an unlicensed NPI, but judge those same sentences as unacceptable after more reflection. Existing accounts have argued that illusions are a consequence of faulty memory access processes, and make the additional assumption that the encoding of the sentence remains fixed over time. This paper challenges the predictions made by these accounts, which assume that illusions should generalize to a broader set of structural environments and a wider range of syntactic and semantic phenomena. We show across seven reading-time and acceptability judgment experiments that NPI illusions can be reliably switched “on” and “off”, depending on the amount of time from when the potential licensor is processed until the NPI is encountered. But we also find that the same profile does not extend to agreement illusions. This contrast suggests that the mechanisms responsible for switching the NPI illusion on and off are not shared across all illusions. We argue that the contrast reflects changes over time in the encoding of the semantic/pragmatic representations that can license NPIs. Just as optical illusions have been informative about the visual system, selective linguistic illusions are informative not only about the nature of the access mechanisms, but also about the nature of the encoding mechanisms.
| Chacón, Dustin Alfonso; Imtiaz, Mashrur; Dasgupta, Shirsho; Murshed, Sikder Monoare; Dan, Mina; Phillips, Colin: Locality and word order in active dependency formation in Bangla. In: Frontiers in Psychology, vol. 7, pp. 1235, 2016. (Type: Journal Article | Abstract | Links | BibTeX)|
Research on filler-gap dependencies has revealed that there are constraints on both possible and grammatical gap sites, and that real-time sentence processing is sensitive to these constraints. Additionally, much of the evidence used to establish the real-time construction of filler-gap dependencies relies on indirect cues, such as measuring sensitivity to a disrupted resolution site. However, neither the mechanisms that select preferred gap sites nor the mechanisms used to detect whether these preferences are met are well understood. In this paper, we report on three experiments in Bangla, a language in which gaps may occur in either a preverbal embedded clause or a postverbal embedded clause. This word order variation allows us to manipulate whether the first gap linearly available is contained in the same clause as the filler, which allows us to dissociate structural locality from linear locality. In Experiment 1, an untimed ambiguity resolution task, we show that there is a global bias to resolve a filler-gap dependency with the first gap linearly available, regardless of structural hierarchy. In Experiments 2 and 3, which use the filled-gap paradigm, we find sensitivity to disruption only when the blocked gap site is both structurally and linearly local, i.e., the filler and the gap site are contained in the same clause. This suggests that the comprehender may not show sensitivity to the disruption of all preferred gap resolutions.
| Malko, Anton; Ehrenhofer, Lara; Phillips, Colin: Theories and frameworks in second language processing. In: Bilingualism: Language and Cognition, 2016, (in press; brief commentary on target article by Ian Cunnings: "Parsing and working memory in bilingual sentence processing"). (Type: Journal Article | Abstract | Links | BibTeX)|
Analyzing L2 sentence processing in terms of cue-based memory retrieval is promising. But this useful general framework has yet to become a specific theory of L1-L2 differences.
| Momma, Shota: Parsing, generation, and grammar. University of Maryland, 2016. (Type: PhD Thesis | Abstract | Links | BibTeX)|
Humans use their grammatical knowledge in more than one way. On one hand, they use it to understand what others say. On the other hand, they use it to say what they want to convey to others (or to themselves). In either case, they need to assemble the structure of sentences in a systematic fashion, in accordance with the grammar of their language. Despite the fact that the structures that comprehenders and speakers assemble are systematic in an identical fashion (i.e., obey the same grammatical constraints), the two ‘modes’ of assembling sentence structures might or might not be performed by the same cognitive mechanisms. Currently, the field of psycholinguistics implicitly adopts the position that they are supported by different cognitive mechanisms, as evident from the fact that most psycholinguistic models seek to explain either comprehension or production phenomena. The potential existence of two independent cognitive systems underlying linguistic performance doubles the problem of linking the theory of linguistic knowledge and the theory of linguistic performance, making the integration of linguistics and psycholinguistic harder. This thesis thus aims to unify the structure building system in comprehension, i.e., parser, and the structure building system in production, i.e., generator, into one, so that the linking theory between knowledge and performance can also be unified into one. I will discuss and unify both existing and new data pertaining to how structures are assembled in understanding and speaking, and attempt to show that the unification between parsing and generation is at least a plausible research enterprise. In Chapter 1, I will discuss the previous and current views on how parsing and generation are related to each other. I will outline the challenges for the current view that the parser and the generator are the same cognitive mechanism. This single system view is discussed and evaluated in the rest of the chapters. In Chapter 2, I will present new experimental evidence suggesting that the grain size of the pre-compiled structural units (henceforth simply structural units) is rather small, contrary to some models of sentence production. In particular, I will show that the internal structure of the verb phrase in a ditransitive sentence (e.g., The chef is donating the book to the monk) is not specified at the onset of speech, but is specified before the first internal argument (the book) needs to be uttered. I will also show that this timing of structural processes with respect to the verb phrase structure is earlier than the lexical processes of verb internal arguments. These two results in concert show that the size of structure building units in sentence production is rather small, contrary to some models of sentence production, yet structural processes still precede lexical processes. I argue that this view of generation resembles the widely accepted model of parsing that utilizes both top-down and bottom-up structure building procedures. In Chapter 3, I will present new experimental evidence suggesting that the structural representation strongly constrains the subsequent lexical processes. In particular, I will show that conceptually similar lexical items interfere with each other only when they share the same syntactic category in sentence production. The mechanism that I call syntactic gating, will be proposed, and this mechanism characterizes how the structural and lexical processes interact in generation. I will present two Event Related Potential (ERP) experiments that show that the lexical retrieval in (predictive) comprehension is also constrained by syntactic categories. I will argue that the syntactic gating mechanism is operative both in parsing and generation, and that the interaction between structural and lexical processes in both parsing and generation can be characterized in the same fashion. In Chapter 4, I will present a series of experiments examining the timing at which verbs’ lexical representations are planned in sentence production. It will be shown that verbs are planned before the articulation of their internal arguments, regardless of the target language (Japanese or English) and regardless of the sentence type (active object- initial sentence in Japanese, passive sentences in English, and unaccusative sentences in English). I will discuss how this result sheds light on the notion of incrementality in generation. In Chapter 5, I will synthesize the experimental findings presented in this thesis and in previous research to address the challenges to the single system view I outlined in Chapter 1. I will then conclude by presenting a preliminary single system model that can potentially capture both the key sentence comprehension and sentence production data without assuming distinct mechanisms for each.
| Chow, Wing Yee; Momma, Shota; Smith, Cybelle; Lau, Ellen; Phillips, Colin: Prediction as memory retrieval: timing and mechanisms. In: Language, Cognition, and Neuroscience, 2016, (in press). (Type: Journal Article | Abstract | Links | BibTeX)|
In our target article (Chow, Smith, Lau, & Phillips, 2015) we investigated the predictions that comprehenders initially make about an upcoming verb as they read. We studied the impacts of the lexical meaning and structural roles of preverbal arguments on comprehenders’ verb predictions by examining their effects on the N400 ERP response at the verb. We provided evidence that comprehenders’ initial verb predictions are sensitive to the arguments’ lexical meaning but not their structural roles. This is compatible with a number of other recent findings using electrophysiology and other time-sensitive measures of prediction. The commentaries offered alternative accounts of our findings (Kim, Oines, & Sikos, 2015; Kuperberg, 2016). In this article we synthesize findings from our target article with other studies that show that verb predictions are sensitive to the arguments’ structural roles if more time is available for prediction. Our first main contention is that prediction involves computations that may require differing amounts of time. Our second main contention is that prediction can be usefully framed as a memory retrieval problem, linking prediction to independently well understood memory mechanisms in language processing. We suggest that the delayed impact of argument roles on verb predictions may reflect a mismatch between the format of linguistic cues and target event memories. We clarify points of agreement and disagreement with the commentaries, and give reasons why memory access mechanisms are better able to explain the time course of prediction.
| Phillips, Colin; Ehrenhofer, Lara: The role of language processing in language acquisition. In: Linguistic Approaches to Bilingualism, vol. 5, no. 4, pp. 409-453, 2015, (target article with 17 commentaries & response). (Type: Journal Article | Abstract | Links | BibTeX)|
Language processing research is changing in two ways that should make it more relevant to the study of grammatical learning. First, grammatical phenomena are re-entering the psycholinguistic fray, and we have learned a lot in recent years about the real-time deployment of grammatical knowledge. Second, psycholinguistics is reaching more diverse populations, leading to much research on language processing in child and adult learners. We discuss three ways that language processing can be used to understand language acquisition. Level 1 approaches (“Processing in learners”) explore well-known phenomena from the adult psycholinguistic literature and document how they play out in learner populations (child learners, adult learners, bilinguals). Level 2 approaches (“Learning effects as processing effects”) use insights from adult psycholinguistics to understand the language proficiency of learners. We argue that a rich body of findings that have been attributed to the grammatical development of anaphora should instead be attributed to limitations in the learner’s language processing system. Level 3 approaches (“Explaining learning via processing”) use language processing to understand what it takes to successfully master the grammar of a language, and why different learner groups are more or less successful. We examine whether language processing may explain why some grammatical phenomena are mastered late in children but not in adult learners. We discuss the idea that children’s language learning prowess is directly caused by their processing limitations (“less is more”: Newport, 1990). We conclude that the idea is unlikely to be correct in its original form, but that a variant of the idea has some promise (“less is eventually more”). We lay out key research questions that need to be addressed in order to resolve the issues addressed in the paper.
| Phillips, Colin; Ehrenhofer, Lara: Learning obscure and obvious properties of languages. In: Linguistic Approaches to Bilingualism, vol. 5, no. 4, pp. 545-555, 2015, (response to 17 commentaries on target article "The role of language processing in language acquisition"). (Type: Journal Article | Links | BibTeX)|
| Parker, Dan; Lago, Sol; Phillips, Colin: Interference in the processing of adjunct control. In: Frontiers in Psychology, vol. 6, no. 1346, pp. 1-13, 2015. (Type: Journal Article | Abstract | Links | BibTeX)|
Recent research on the memory operations used in language comprehension has revealed a selective profile of interference effects during memory retrieval. Dependencies such as subject-verb agreement show strong facilitatory interference effects from structurally inappropriate but feature-matching distractors, leading to illusions of grammaticality (Dillon, Mishler, Sloggett, & Phillips, 2013; Gillespie & Pearlmutter, 2011; Pearlmutter, Garnsey, & Bock, 1999). In contrast, dependencies involving reflexive anaphors are generally immune to interference effects (Dillon et al., 2013; Sturt, 2003; Xiang, Dillon, & Phillips, 2009). This contrast has led to the proposal that all anaphors that are subject to structural constraints are immune to facilitatory interference. Here we use an animacy manipulation to examine whether adjunct control dependencies, which involve an interpreted anaphoric relation between a null subject and its licensor, are also immune to facilitatory interference effects. Our results show reliable facilitatory interference in the processing of adjunct control dependencies, which challenges the generalization that anaphoric dependencies as a class are immune to such effects. To account for the contrast between adjunct control and reflexive dependencies, we suggest that variability within anaphora could reflect either an inherent primacy of animacy cues in retrieval processes, or differential degrees of match between potential licensors and the retrieval probe.
| Chacón, Dustin Alfonso: Comparative psychosyntax. University of Maryland, 2015. (Type: PhD Thesis | Abstract | Links | BibTeX)|
Every difference between languages is a “choice point” for the syntactician, psycholinguist, and language learner. The syntactician must describe the differences in representations that the grammars of different languages can assign. The psycholinguist must describe how the comprehension mechanisms search the space of the representations permitted by a grammar to quickly and effortlessly understand sentences in real time. The language learner must determine which representations are permitted in her grammar on the basis of her primary linguistic evidence. These investigations are largely pursued independently, and on the basis of qualitatively different data. In this dissertation, I show that these investigations can be pursued in a way that is mutually informative. Specifically, I show how learnability concerns and sentence processing data can constrain the space of possible analyses of language differences.
In Chapter 2, I argue that “indirect learning”, or abstract, cross-contruction syntactic inference, is necessary in order to explain how the learner determines which complementizers can co-occur with subjects gaps in her target grammar. I show that adult speakers largely converge in the robustness of the that-trace effect, a constraint on complementation complementizers and subject gaps observed in languages like English, but unobserved in languages like Spanish or Italian. I show that realistic child-directed speech has very few long-distance subject extractions in English, Spanish, and Italian, implying that learners must be able to distinguish these different hypotheses on the basis of other data. This is more consistent with more conservative approaches to these phenomena (Rizzi, 1982), which do not rely on abstract complementizer agreement like later analyses (Rizzi, 2006; Rizzi & Shlonsky, 2007). In Chapter 3, I show that resumptive pronoun dependencies inside islands in English are constructed in a non-active fashion, which contrasts with recent findings in Hebrew (Keshev & Meltzer-Asscher, ms). I propose that an expedient explanation of these facts is to suppose that resumptive pronouns in English are ungrammatical repair devices (Sells, 1984), whereas resumptive pronouns in island contexts are grammatical in Hebrew. This implies that learners must infer which analysis is appropriate for their grammars on the basis of some evidence in linguistic environment. However, a corpus study reveals that resumptive pronouns in islands are exceedingly rare in both languages, implying that this difference must be indirectly learned. I argue that theories of resumptive dependencies which analyze resumptive pronouns as incidences of the same abstract construction (e.g., Hayon 1973; Chomsky 1977) license this indirect learning, as long as resumptive dependencies in English are treated as ungrammatical repair mechanisms. In Chapter 4, I compare active dependency formation processes in Japanese
and Bangla. These findings suggest that filler-gap dependencies are preferentially resolved with the first position available. In Japanese, this is the most deeply embedded clause, since embedded clauses always precede the embedding verb (Aoshima et al., 2004; Yoshida, 2006; Omaki et al., 2014). Bangla allows a within-language comparison of the relationship between active dependency formation processes and word order, since embedded clauses may precede or follow the embedding verb (Bayer, 1996). However, the results from three experiments in Bangla are mixed, suggesting a weaker preference for a linearly local resolution of filler-gap dependencies, unlike in Japanese. I propose a number of possible explanations for these facts, and discuss how differences in processing profiles may be accounted for in a variety of ways.
In Chapter 5, I conclude the dissertation.
| Chacón, Dustin Alfonso; Momma, Shota; Phillips, Colin: Linguistic representations and memory architectures: The devil is in the details. In: Behavioral and Brain Sciences, 2015, (in press, commentary on target article by Christiansen & Chater). (Type: Journal Article | Abstract | Links | BibTeX)|
Attempts to explain linguistic phenomena as consequences of memory constraints require detailed specification of linguistic representations and memory architectures alike. We discuss examples of supposed locality biases in language comprehension and production, and their link to memory constraints. Findings do not generally favor Christiansen and Chater’s approach. We discuss connections to debates that stretch back to the 19th century.
| Omaki, Akira; Lau, Ellen; White, Imogen Davidson; Dakan, Myles; Apple, Aaron; Phillips, Colin: Hyper-active gap filling. In: Frontiers in Psychology, vol. 6:384, 2015, (doi: 10.3389/fpsyg.2015.00384). (Type: Journal Article | Abstract | Links | BibTeX)|
Much work has demonstrated that speakers of verb-final languages are able to construct rich syntactic representations in advance of verb information. This may reflect general architectural properties of the language processor, or it may only reflect a language-specific adaptation to the demands of verb-finality. The present study addresses this issue by examining whether speakers of a verb-medial language (English) wait to consult verb transitivity information before constructing filler-gap dependencies, where internal arguments are fronted and hence precede the verb. This configuration makes it possible to investigate whether the parser actively makes representational commitments on the gap position before verb transitivity information becomes available. A key prediction of the view that rich pre-verbal structure-building is a general architectural property is that speakers of verb-medial languages should predictively construct dependencies in advance of verb transitivity information, and therefore that disruption should be observed when the verb has intransitive subcategorization frames that are incompatible with the predicted structure. In three reading experiments (self-paced and eye-tracking) that manipulated verb transitivity, we found evidence for reading disruption when the verb was intransitive, although no such reading difficulty was observed when the critical verb was embedded inside a syntactic island structure, which blocks filler-gap dependency completion. These results are consistent with the hypothesis that in English, as in verb-final languages, information from preverbal NPs is sufficient to trigger active dependency completion without having access to verb transitivity information.
| Kush, Dave; Lidz, Jeffrey; Phillips, Colin: Relation-sensitive retrieval: evidence from bound variable pronouns. In: Journal of Memory and Language, vol. 82, pp. 18-40, 2015. (Type: Journal Article | Abstract | Links | BibTeX)|
Formal grammatical theories make extensive use of syntactic relations (e.g. c-command, Reinhart, 1983) in the description of constraints on antecedent-anaphor dependencies. Recent research has motivated a model of processing that exploits a cue-based retrieval mechanism in content-addressable memory (e.g. Lewis, Vasishth, & Van Dyke, 2006) in which item-to-item syntactic relations such as c-command are difficult to use as retrieval cues. As such, the c-command constraints of formal grammars are predicted to be poorly implemented by the retrieval mechanism. We tested whether memory access mechanisms are able to exploit relational information by investigating the processing of bound variable pronouns, a form of anaphoric dependency that imposes a c-command restriction on antecedent-pronoun relations. A quantificational NP (QP, e.g., no janitor) must c-command a pronoun in order to bind it. We contrasted the retrieval of QPs with the retrieval of referential NPs (e.g. the janitor), which can co-refer with a pronoun in the absence of c-command. In three off-line judgment studies and two eye-tracking studies, we show that referential NPs are easily accessed as antecedents, irrespective of whether they c-command the pronoun, but that quantificational NPs are accessed as antecedents only when they c-command the pronoun. These results are unexpected under theories that hold that retrieval exclusively uses a limited set of content features as retrieval cues. Our results suggest either that memory access mechanisms can make use of relational information as a guide for retrieval, or that the set of features that is used to encode syntactic relations in memory must be enriched.
| Lago, Sol; Shalom, Diego; Sigman, Mariano; Lau, Ellen; Phillips, Colin: Agreement processes in Spanish comprehension. In: Journal of Memory and Language, vol. 82, pp. 133-149, 2015. (Type: Journal Article | Abstract | Links | BibTeX)|
Previous studies have found that English speakers experience attraction effects when comprehending subject-verb agreement, showing eased processing of ungrammatical sentences that contain a syntactically unlicensed but number-matching attractor noun. In three self-paced reading experiments we examine whether attraction effects also occur in Spanish, a language where agreement morphology is richer and functionally more significant. We find that despite having a richer morphology, Spanish speakers show reliable attraction effects in comprehension, and that the magnitude and distributional profile of these effects are strikingly similar to those previously found in English. Further, we use distributional analyses to argue that cue-based memory retrieval is used as an error-driven mechanism in comprehension. We suggest the cross-linguistic similarities in agreement attraction result from speakers deploying repair or error-driven mechanisms uniformly across languages.
| Lewis, Shevaun; Phillips, Colin: Aligning grammatical theories and language processing models. In: Journal of Psycholinguistic Research, vol. 44, no. 1, pp. 27-46, 2015. (Type: Journal Article | Abstract | Links | BibTeX)|
We address two important questions about the relationship between theoretical linguistics and psycholinguistics. First, do grammatical theories and language processing models describe separate cognitive systems, or are they accounts of different aspects of the same system? We argue that most evidence is consistent with the one-system view. Second, how should we relate grammatical theories and language processing models to each other?
| Phillips, Colin; Parker, Dan: The psycholinguistics of ellipsis. In: Lingua, vol. 151, pp. 78-95, 2014, (published online Nov 27, 2013). (Type: Journal Article | Abstract | Links | BibTeX)|
This article reviews studies that have used experimental methods from psycholinguistics to address questions about the representa- tion of sentences involving ellipsis. Accounts of the structure of ellipsis can be classified based on three choice points in a decision tree. First: does the identity constraint between antecedents and ellipsis sites apply to syntactic or semantic representations? Second: does the ellipsis site contain a phonologically null copy of the structure of the antecedent, or does it contain a pronoun or pointer that lacks internal structure? Third: if there is unpronounced structure at the ellipsis site, does that structure participate in all syntactic processes, or does it behave as if it is genuinely absent at some levels of syntactic representation? Experimental studies on ellipsis have begun to address the first two of these questions, but they are unlikely to provide insights on the third question, since the theoretical contrasts do not clearly map onto timing predictions. Some of the findings that are emerging in studies on ellipsis resemble findings from earlier studies on other syntactic dependencies involving wh-movement or anaphora. Care should be taken to avoid drawing conclusions from experiments about ellipsis that are known to be unwarranted in experiments about these other dependencies.
| Kush, Dave; Phillips, Colin: Local anaphor licensing in an SOV language: implications for retrieval strategies. In: Frontiers in Psychology, vol. 5, no. 1252, 2014. (Type: Journal Article | Abstract | Links | BibTeX)|
Because morphological and syntactic constraints govern the distribution of potential antecedents for local anaphors, local antecedent retrieval might be expected to make equal use of both syntactic and morphological cues. However, previous research (e.g., Dillon et al., 2013) has shown that local antecedent retrieval is not susceptible to the same morphological interference effects observed during the resolution of morphologically-driven grammatical dependencies, such as subject-verb agreement checking (e.g., Pearlmutter, Garnsey and Bock, 1999). Although this lack of interference has been taken as evidence that syntactic cues are given priority over morphological cues in local antecedent retrieval, the absence of interference could also be the result of a confound in the materials used: the post-verbal position of local anaphors in prior studies may obscure morphological interference that would otherwise be visible if the critical anaphor were in a different position. We investigated the licensing of local anaphors (reciprocals) in Hindi, an SOV language, in order to determine whether pre-verbal anaphors are subject to morphological interference from feature-matching distractors in a way that post-verbal anaphors are not. Computational simulations using a version of the ACT-R parser (Lewis and Vasishth, 2005) predicted that a feature-matching distractor should facilitate the processing of an unlicensed reciprocal if morphological cues are used in antecedent retrieval. In a self-paced reading study we found no evidence that distractors eased processing of an unlicensed reciprocal. However, the presence of a distractor increased difficulty of processing following the reciprocal. We discuss the significance of these results for theories of cue selection in retrieval.
| Dillon, Brian; Chow, Wing Yee; Wagers, Matthew; Guo, Taomei; Liu, Fengqin; Phillips, Colin: The structure-sensitivity of memory access: evidence from Mandarin Chinese. In: Frontiers in Psychology, vol. 5, no. 1025, 2014, (doi: 10.3389/fpsyg.2014.01025). (Type: Journal Article | Abstract | Links | BibTeX)|
The present study examined the processing of the Mandarin Chinese long- distance reflexive ziji to evaluate the role that syntactic structure plays in the memory retrieval operations that support sentence comprehension. Using the multiple-response speed-accuracy tradeoff (MR-SAT) paradigm, we measured the speed with which comprehenders retrieve an antecedent for ziji. Our experimental materials contrasted sentences where ziji’s antecedent was in the local clause with sentences where ziji’s antecedent was in a distant clause. Time course results from MR-SAT suggest that ziji dependencies with syntactically distant antecedents are slower to process than syntactically local dependencies. To aid in interpreting the SAT data, we present a formal model of the antecedent retrieval process, and derive quantitative predictions about the time course of antecedent retrieval. The modeling results support the Local Search hypothesis: during syntactic retrieval, comprehenders initially limit memory search to the local syntactic domain. We argue that Local Search hypothesis has important implications for theories of locality effects in sentence comprehension. In particular, our results suggest that not all locality effects may be reduced to the effects of temporal decay and retrieval interference.
| Lago, Maria Sol: Memory and prediction in cross-linguistic sentence comprehension. University of Maryland, 2014. (Type: PhD Thesis | Abstract | Links | BibTeX)|
This dissertation explores the role of morphological and syntactic variation in sentence comprehension across languages. While most previous research has focused on how cross-linguistic differences affect the control structure of the language architecture (Lewis & Vasishth, 2005) here we adopt an explicit model of memory, content-addressable memory (Lewis & Vasishth, 2005; McElree, 2006) and examine how cross-linguistic variation affects the nature of the representations and processes that speakers deploy during comprehension. With this goal, we focus on two kinds of grammatical dependencies that involve an interaction between language and memory: subject-verb agreement and referential pronouns. In the first part of this dissertation, we use the self-paced reading method to examine how the processing of subject-verb agreement in Spanish, a language with a rich morphological system, differs from English. We show that differences in morphological richness across languages impact prediction processes while leaving retrieval processes fairly preserved. In the second part, we examine the processing of coreference in German, a language that, in contrast with English, encodes gender syntactically. We use eye-tracking to compare comprehension profiles during coreference and we find that only speakers of German show evidence of semantic reactivation of a pronoun’s antecedent. This suggests that retrieval of semantic information is dependent on syntactic gender, and demonstrates that German and English speakers retrieve qualitatively different antecedent representations from memory. Taken together, these results suggest that cross- linguistic variation in comprehension is more affected by the content than the functional importance of gender and number features across languages.
| Parker, Daniel: The cognitive basis for encoding and navigating linguistic structure. University of Maryland, 2014. (Type: PhD Thesis | Abstract | Links | BibTeX)|
This dissertation is concerned with the cognitive mechanisms that are used to encode and navigate linguistic structure. Successful language understanding requires mechanisms for efficiently encoding and navigating linguistic structure in memory. The timing and accuracy of linguistic dependency formation provides valuable insights into the cognitive basis of these mechanisms. Recent research on linguistic dependency formation has revealed a profile of selective fallibility: some linguistic dependencies are rapidly and accurately implemented, but others are not, giving rise to “linguistic illusions”. This profile is not expected under current models of grammar or language processing. The broad consensus, however, is that the profile of selective fallibility reflects dependency-based differences in memory access strategies, including the use of different retrieval mechanisms and the selective use of cues for different dependencies. In this dissertation, I argue that (i) the grain-size of variability is not dependency-type, and (ii) there is not a homogenous cause for linguistic illusions. Rather, I argue that the variability is a consequence of how the grammar interacts with general-purpose encoding and access mechanisms. To support this argument, I provide three types of evidence. First, I show how to “turn on” illusions for anaphor resolution, a phenomena that has resisted illusions in the past, reflecting a cue- combinatorics scheme that prioritizes structural information in memory retrieval. Second, I show how to “turn off” a robust illusion for negative polarity item (NPI) licensing, reflecting access to the internal computations during the encoding and interpretation of emerging semantic/pragmatic representations. Third, I provide computational simulations that derive both the presence and absence of the illusions from within the same memory architecture. These findings lead to a new conception of how we mentally encode and navigate structured linguistic representations.
| Chow, Wing Yee; Lewis, Shevaun; Phillips, Colin: Immediate sensitivity to structural constraints in pronoun resolution. In: Frontiers in Psychology, vol. 5, no. 630, 2014. (Type: Journal Article | Abstract | Links | BibTeX)|
Real-time interpretation of pronouns is sometimes sensitive to the presence of grammatically-illicit antecedents and sometimes not. This occasional sensitivity has been taken as evidence that structural constraints do not immediately impact the initial antecedent retrieval for pronoun interpretation. We argue that it is important to separate effects that reflect the initial antecedent retrieval process from those that reflect later processes. We present results from five reading comprehension experiments. Both the current results and previous evidence support the hypothesis that agreement features and structural constraints immediately constrain the antecedent retrieval process for pronoun interpretation. Occasional sensitivity to grammatically-illicit antecedents may be due to repair processes triggered when the initial retrieval fails to return a grammatical antecedent.
| Omaki, Akira; White, Imogen Davidson; Goro, Takuya; Lidz, Jeffrey; Phillips, Colin: No fear of commitment: children's incremental interpretation in English and Japanese wh-questions. In: Language Learning and Development, vol. 10, pp. 206-233, 2014. (Type: Journal Article | Abstract | Links | BibTeX)|
Much work on child sentence processing has demonstrated that children are able to use various linguistic cues to incrementally resolve temporary syntactic ambiguities, but they fail to use syn- tactic or interpretability cues that arrive later in the sentence. The present study explores whether children incrementally resolve filler-gap dependencies, using Japanese and English ambiguous wh-questions of the form Where did Lizzie tell someone that she was gonna catch butterflies?, in which one could answer either the telling location (main clause interpretation) or the butterfly– catching location (embedded clause interpretation). Three story-based experiments demonstrate two novel findings on children’s incremental interpretation of filler-gap dependencies. First, we observe that English-speaking adults and children generally prefer the main clause interpretation, whereas Japanese adults and children both prefer the embedded clause interpretation. As the linear order of main clause and embedded clause predicates differs between English (main first, embedded second) and Japanese (embedded first, main second), the results indicate that adults and children actively asso- ciate the wh-phrase with the first predicate in the sentence. Second, Japanese children were unable to inhibit their embedded clause interpretation bias when the sentence was manipulated to syntactically block such analyses. The failure to inhibit the preferred interpretation suggests that the wh-phrase was incrementally associated with the embedded clause. On the other hand, when the sentence was manipulated to semantically block a plausible interpretation for the embedded clause wh-association, children were able to overcome their strong embedded clause interpretation bias and favored the main clause interpretation. These findings suggest that syntactic and interpretability cues may have distinct impacts on children’s sentence comprehension processes.
| Wagers, Matthew W; Phillips, Colin: Going the distance: memory and control processes in active dependency construction. In: Quarterly Journal of Experimental Psychology, vol. 167, no. 7, pp. 1274-1304, 2014. (Type: Journal Article | Abstract | Links | BibTeX)|
Filler–gap dependencies make strong demands on working memory in language comprehension because they cannot always be immediately resolved. In a series of three reading-time studies, we test the idea that these demands can be decomposed into active maintenance processes and retrieval events. Results indicate that the fact that a displaced phrase exists and the identity of its basic syntactic category both immediately impact comprehension at potential gap sites. In contrast, specific lexical details of the displaced phrase show an immediate effect only for short dependencies and a much later effect for longer dependencies. We argue that coarse-grained information about the filler is actively maintained and is used to make phrase structure parsing decisions, whereas finer grained information is more quickly released from active maintenance and consequently has to be retrieved at the gap site.
| Chow, Wing Yee: The temporal dimension of linguistic prediction. University of Maryland, 2013. (Type: PhD Thesis | Abstract | Links | BibTeX)|
This thesis explores how predictions about upcoming language inputs are computed during real-time language comprehension. Previous research has demonstrated humans’ ability to use rich contextual information to compute linguistic prediction during real-time language comprehension, and it has been widely assumed that contextual information can impact linguistic prediction as soon as it arises in the input. This thesis questions this key assumption and explores how linguistic predictions develop in real- time. I provide event-related potential (ERP) and reading eye-movement (EM) evidence from studies in Mandarin Chinese and English that even prominent and unambiguous information about preverbal arguments’ structural roles cannot immediately impact comprehenders’ verb prediction. I demonstrate that the N400, an ERP response that is modulated by a word’s predictability, becomes sensitive to argument role-reversals only when the time interval for prediction is widened. Further, I provide initial evidence that different sources of contextual information, namely, information about preverbal arguments’ lexical identity vs. their structural roles, may impact linguistic prediction on different time scales. I put forth a research framework that aims to characterize the mental computations underlying linguistic prediction along a temporal dimension.
| Kush, Dave: Respecting relations: memory access and antecedent retrieval in incremental sentence processing. University of Maryland, 2013. (Type: PhD Thesis | Abstract | Links | BibTeX)|
This dissertation uses the processing of anaphoric relations to probe how linguistic information is encoded in and retrieved from memory during real-time sentence com- prehension. More specifically, the dissertation attempts to resolve a tension between the demands of a linguistic processor implemented in a general-purpose cognitive architec- ture and the demands of abstract grammatical constraints that govern language use. The source of the tension is the role that abstract configurational relations (such as c-command, Reinhart 1983) play in constraining computations. Anaphoric dependencies are governed by formal grammatical constraints stated in terms of relations. For example, Binding Principle A (Chomsky 1981) requires that antecedents for local anaphors (like the English reciprocal each other) bear the c-command relation to those anaphors. In incremental sentence processing, antecedents of anaphors must be retrieved from memory. Recent research has motivated a model of processing that exploits a cue-based, associative retrieval process in content-addressable memory (e.g. Lewis, Vasishth & Van Dyke 2006) in which relations such as c-command are difficult to use as cues for retrieval. As such, the c-command constraints of formal grammars are predicted to be poorly implemented by the retrieval mechanism. I examine retrieval’s sensitivity to three constraints on anaphoric dependencies: Principle A (via Hindi local reciprocal licensing), the Scope Constraint on bound-variable pronoun licensing (often stated as a c-command constraint, though see Barker 2012), and Crossover constraints on pronominal binding (Postal 1971, Wasow 1972). The data suggest that retrieval exhibits fidelity to the constraints: structurally inaccessible NPs that match an anaphoric element in morphological features do not interfere with the retrieval of an antecedent in most cases considered. In spite of this alignment, I argue that retrieval’s apparent sensitivity to c-command constraints need not motivate a memory access procedure that makes direct reference to c-command relations. Instead, proxy features and general parsing operations conspire to mimic the extension of a system that respects c-command constraints. These strategies provide a robust approximation of grammatical performance while remaining within the confines of a independently- motivated general-purpose cognitive architecture.
| Lewis, Shevaun: Pragmatic enrichment in language processing and development. University of Maryland, 2013. (Type: PhD Thesis | Abstract | Links | BibTeX)|
The goal of language comprehension for humans is not just to decode the semantic content of sentences, but rather to grasp what speakers intend to communicate. To infer speaker meaning, listeners must at minimum assess whether and how the literal meaning of an utterance addresses a question under discussion in the conversation. In cases of implicature, where the speaker intends to communicate more than just the literal meaning, listeners must access additional relevant information in order to understand the intended contribution of the utterance. I argue that the primary challenge for inferring speaker meaning is in identifying and accessing this relevant contextual information. In this dissertation, I integrate evidence from several different types of implicature to argue that both adults and children are able to execute complex pragmatic inferences relatively efficiently, but encounter some difficulty finding what is relevant in context. I argue that the variability observed in processing costs associated with adults’ computation of scalar implicatures can be better understood by examining how the critical contextual information is presented in the discourse context. I show that children’s oft-cited hyper-literal interpretation style is limited to scalar quantifiers. Even 3-year-olds are adept at understanding indirect requests and “parenthetical” readings of belief reports. Their ability to infer speaker meanings is limited only by their relative inexperience in conversation and lack of world knowledge.
| Dillon, Brian; Mishler, Alan; Sloggett, Shayne; Phillips, Colin: Contrasting intrusion profiles for agreement and anaphora: experimental and modeling evidence. In: Journal of Memory and Language, vol. 69, pp. 85-103, 2013. (Type: Journal Article | Abstract | Links | BibTeX)|
We investigated the relationship between linguistic representation and memory access by comparing the processing of two linguistic dependencies that require comprehenders to check that the subject of the current clause has the correct morphological features: sub- ject–verb agreement and reflexive anaphors in English. In two eye-tracking experiments we examined the impact of structurally illicit noun phrases on the computation of reflexive and subject–verb agreement. Experiment 1 directly compared the two dependencies within participants. Results show a clear difference in the intrusion profile associated with each dependency: agreement resolution displays clear intrusion effects in comprehension (as found by Pearlmutter, Garnsey, & Bock, 1999; Wagers, Lau, & Phillips, 2009), but reflexi- ves show no such intrusion effect from illicit antecedents (Sturt, 2003; Xiang, Dillon, & Phillips, 2009). Experiment 2 replicated the lack of intrusion for reflexives, confirming the reliability of the pattern and examining a wider range of feature combinations. In addition, we present modeling evidence that suggests that the reflexive results are best captured by a memory retrieval mechanism that uses primarily syntactic information to guide retrievals for the anaphor’s antecedent, in contrast to the mixed morphological and syntactic cues used resolve subject–verb agreement dependencies. Despite the fact that agreement and reflexive dependencies are subject to a similar morphological agree- ment constraint, in online processing comprehenders appear to implement this constraint in distinct ways for the two dependencies.
| Chow, Wing Yee; Phillips, Colin: No semantic illusion in the 'semantic P600' phenomenon: ERP evidence from Mandarin Chinese. In: Brain Research, vol. 1506, pp. 76-93, 2013. (Type: Journal Article | Abstract | Links | BibTeX)|
Recent observations of unexpected ERP responses to grammatically well-formed role-reversed sentences (the ‘‘Semantic P600’’ phenomenon) have been taken to bear directly on questions about the architecture of the language processing system. This paper evaluates two central pieces of evidence for accounts that propose a syntax-independent semantic composition mechanism, namely the presence of P600 effects and the absence of N400 effects in role-reversed sentences. Experiment 1 examined the relative contribution of the presence of an animacy violation and the semantic relations between words (‘combinability’) to the ERP responses to role-reversed sentences. Experiment 2 examined the ERP responses to role-reversed sentences that are fully animacy-congruous. Results from the two experiments showed that animacy-violated sentences with no plausible non-surface interpretation elicited the same P600 effect as both types of role- reversed sentences; additionally, semantically anomalous target words elicited no N400 effects when they were strongly semantically related to the preceding words, regardless of the presence of animacy violations. Taken together, these findings suggest that the presence of P600s to role- reversed sentences can be attributed to the implausibility of the sentence meaning, and the absence of N400 effects is due to a combination of weak contextual constraints and strong lexical association. The presence of a plausible non-surface interpretation and animacy violations made no unique contribution to the ERP response profiles. Hence, existing ERP findings are compatible with the long-held assumption that online semantic composition is dependent on surface syntax and do not constitute evidence for a syntax-independent semantic composition mechanism.
| Phillips, Colin: Some arguments and non-arguments for reductionist accounts of syntactic phenomena. In: Language and Cognitive Processes, vol. 28, pp. 156-187, 2013. (Type: Journal Article | Abstract | Links | BibTeX)|
Many syntactic phenomena have received competing accounts, either in terms of formal grammatical mechanisms, or in terms of independently motivated properties of language processing mechanisms (‘‘reductionist’’ accounts). A variety of different types of argument have been put forward in efforts to distinguish these competing accounts. This article critically examines a number of arguments that have been offered as evidence in favour of formal or reductionist analyses, and concludes that some types of argument are more decisive than others. It argues that evidence from graded acceptability effects and from isomorphism between acceptability judgements and on-line compre- hension profiles are less decisive. In contrast, clearer conclusions can be drawn from cases of overgeneration, where there is a discrepancy between accept- ability judgements and the representations that are briefly constructed on-line, and from tests involving individual differences in cognitive capacity. Based on these arguments, the article concludes that a formal grammatical account is better supported in some domains, and that a reductionist account fares better in other domains. Phenomena discussed include island constraints, agreement attraction, constraints on anaphora, and comparatives.
| Phillips, Colin: Parser-grammar relations: We don’t understand everything twice. In: Sanz, Montserrat; Laka, Itziar; Tanenhaus, Michael (Ed.): Language down the garden path: the cognitive basis for linguistic structure, pp. 294–315, Oxford University Press, 2013. (Type: Incollection | Links | BibTeX)|
| Dillon, Brian; Nevins, Andrew; Austin, Alison C.; Phillips, Colin: Syntactic and semantic predictors of tense in Hindi: An ERP investigation. In: Language and Cognitive Processes, vol. 27, no. 3, pp. 313-344, 2012. (Type: Journal Article | Abstract | Links | BibTeX)|
Although there is broad agreement that error signals generated during an unexpected linguistic event are reflected in ERP components, there are least two distinct aspects of the process that the ERP signals may reflect. The first is the content of an error, which is the local discrepancy between an observed form and any expectations about upcoming forms, without any reference to why those expectations were held. The second aspect is the cause of an error, which is a context-aware analysis of why the error arose. The current study examines the processes involved in prediction of morphological marking on verbal forms in Hindi, a split ergative language. This is a case where an error with the same local characteristics (illicit morphology) can arise from very different cues: one syntactic in origin (ergative case marking), and the other semantic in origin (a past tense adverbial). Results suggest that the parser indeed tracks the cause in addition to the content of errors. Despite the fact that the critical manipulation of verb marking was identical across cue types, the nature of the cue led to distinct patterns of ERPs in response to anomalous verbal morphology. When verbal morphology was predicted based upon semantic cues, an incorrect future tense form elicited an early negativity in the 200-400 ms interval with a posterior distribution along with a marginally significant P600 effect. In contrast, when verbal morphology was predicted based upon morphosyntactic cues, an incorrect future tense form elicited a right-lateralized anterior negativity (RAN) during the 300-500 ms interval, as well as a P600 response with a broad distribution.
| Stroud, Clare; Phillips, Colin: Examining the evidence for an independent semantic analyzer: An ERP study in Spanish. In: Brain and Language, vol. 120, pp. 107-126, 2012. (Type: Journal Article | Abstract | Links | BibTeX)|
Recent ERP findings challenge the widespread assumption that syntactic and semantic processes are tightly coupled. Syntactically well-formed sentences that are semantically anomalous due to thematic mismatches elicit a P600, the component standardly associated with syntactic anomaly. This ‘thematic P600’ effect has been attributed to detection of semantically plausible thematic relations that conflict with the surface syntactic structure of the sentence, implying a processing architecture with an independent semantic analyzer. A key finding is that the P600 is selectively sensitive to the presence of plausible verb-argument relations, and that otherwise an N400 is elicited (The hearty meal was devouring ... vs. The dusty tabletop was devouring ...: Kim & Osterhout, 2005). The current study investigates in Spanish whether the evidence for an independent semantic analyzer is better explained by a standard architecture that rapidly integrates multiple sources of lexical, syntactic, and semantic information. The study manipulated the presence of plausible thematic relations, and varied the choice of auxiliary between passive-biased fue and active-progressive biased estaba. Results show a late positivity that appeared as soon as comprehenders detected an improbable combination of subject animacy, auxiliary bias, or verb voice morphology. This effect appeared at the lexical verb in the fue conditions and at the auxiliary in the estaba conditions. The late positivity elicited by surface thematic anomalies was the same, regardless of the presence of a plausible non-surface interpretation, and no N400 effects were elicited. These findings do not implicate an independent semantic analyzer, and are compatible with standard language processing architectures.
| Alcocer, Pedro; Phillips, Colin: Using relational syntactic constraints in content-addressable memory architectures for sentence parsing. 2012. (Type: Unpublished | Abstract | Links | BibTeX)|
How linguistic representations in memory are searched and retrieved during sentence processing is an active question in psycholinguistics. Much work has independently suggested that retrieval operations require constant-time computations, are susceptible to interference, and operate under the constraint of a severely limited focus of attention. These features suggest a model for sentence processing that uses a content-addressable memory (CAM) architecture that accesses items in parallel while relying on only a limited amount of privileged, fast storage. A challenge for a CAM architecture comes from the potentially unbounded configurational relation c-command (equivalent to logical scope) that plays a pervasive role in linguistic phenomena. CAM is well-suited to retrieval of elements based on inherent properties of memory chunks, but relational notions such as c-command involve the properties of pairs of nodes in a structure, rather than inherent properties of individual chunks. To explore this problem, in this paper we adopt an explicit CAM-based model of sentence processing in the context of the ACT-R computational architecture (Lewis & Vasishth, 2005, Cognitive Science, 29, 375-419). We discuss why c-command is a challenge for CAM and explore algorithms for exploiting or approximating c-command online, and discuss their consequences for the model. We identify computational problems that any attention-limited, CAM-based model would have to address.
| Dillon, Brian: Structured access in sentence comprehension. University of Maryland, 2011. (Type: PhD Thesis | Abstract | Links | BibTeX)|
This thesis is concerned with the nature of memory access during the construction of long-distance dependencies in online sentence comprehension. In recent years, an intense focus on the computational challenges posed by long-distance dependencies has proven to be illuminating with respect to the characteristics of the architecture of the human sentence processor, suggesting a tight link between general memory access procedures and sentence processing routines (Lewis & Vasishth 2005; Lewis, Vasishth, & Van Dyke 2006; Wagers, Lau & Phillips 2009). The present thesis builds upon this line of research, and its primary aim is to motivate and defend the hypothesis that the parser accesses linguistic memory in an essentially structured fashion for certain long-distance dependencies. In order to make this case, I focus on the processing of reflexive and agreement dependencies, and ask whether or not non- structural information such as morphological features are used to gate memory access during syntactic comprehension. Evidence from eight experiments in a range of methodologies in English and Chinese is brought to bear on this question, providing arguments from interference effects and time-course effects that primarily syntactic information is used to access linguistic memory in the construction of certain long- distance dependencies. The experimental evidence for structured access is compatible with a variety of architectural assumptions about the parser, and I present one implementation of this idea in a parser based on the ACT-R memory architecture. In the context of such a content-addressable model of memory, the claim of structured access is equivalent to the claim that only syntactic cues are used to query memory. I argue that structured access reflects an optimal parsing strategy in the context of a noisy, interference-prone cognitive architecture: abstract structural cues are favored over lexical feature cues for certain structural dependencies in order to minimize memory interference in online processing.
| Phillips, Colin; Wagers, Matthew W; Lau, Ellen F: Grammatical Illusions and Selective Fallibility in Real-Time Language Comprehension. In: Runner, Jeffrey (Ed.): Experiments at the Interfaces. Syntax & Semantics, vol. 37, pp. 147–180, Emerald Group Publishing Limited, 2011. (Type: Incollection | Abstract | Links | BibTeX)|
Grammatical constraints impose diverse requirements on the relations between words and phrases in a sentence. Research on the online implementation of grammatical constraints reveals a strikingly uneven profile. The parser shows impressive accuracy in the application of some rather complex constraints, but makes many errors in the implementation of some relatively simple constraints. Just as the study of optical illusions has played an important role in the study of visual perception, the parser’s highly selective vulnerability to interference and ‘‘grammatical illusions’’ provides a valuable tool for understanding how speakers encode and navigate complex linguistic representations in real time.
| Omaki, Akira: Commitment and flexibility in the developing parser. University of Maryland, 2010. (Type: PhD Thesis | Abstract | Links | BibTeX)|
This dissertation investigates adults and children’s sentence processing mechanisms, with a special focus on how multiple levels of linguistic representation are incrementally computed in real time, and how this process affects the parser’s ability to later revise its early commitments. Using cross-methodological and cross-linguistic investigations of long-distance dependency processing, this dissertation demonstrates how paying explicit attention to the procedures by which linguistic representations are computed is vital to understanding both adults’ real time linguistic computation and children’s reanalysis mechanisms. The first part of the dissertation uses time course evidence from self-paced reading and eye tracking studies (reading and visual world) to show that long-distance dependency processing can be decomposed into a sequence of syntactic and interpretive processes. First, the reading experiments provide evidence that suggests that filler-gap dependencies are constructed before verb information is accessed. Second, visual world experiments show that, in the absence of information that would allow hearers to predict verb content in advance, interpretive processes in filler-gap dependency computation take around 600ms. These results argue for a predictive model of sentence interpretation in which syntactic representations are computed in advance of interpretive processes. The second part of the dissertation capitalizes on this procedural account of filler- gap dependency processing, and reports cross-linguistic studies on children’s long- distance dependency processing. Interpretation data from English and Japanese demonstrate that children actively associate a fronted wh-phrase with the first VP in the sentence, and successfully retract such active syntactic commitments when the lack of felicitous interpretation is signaled by verb information, but not when it is signaled by syntactic information. A comparison of the process of anaphor reconstruction in adults and children further suggests that verb-based thematic information is an effective revision cue for children. Finally, distributional analyses of wh-dependencies in child-directed speech are conducted to investigate how parsing constraints impact language acquisition. It is shown that the actual properties of the child parser can skew the input distribution, such that the effective distribution differs drastically from the input distribution seen from a researcher’s perspective. This suggests that properties of developing perceptual mechanisms deserve more attention in language acquisition research.
| Gouvea, Ana C; Phillips, Colin; Kazanina, Nina; Poeppel, David: The linguistic processes underlying the P600. In: Language and Cognitive Processes, vol. 25, no. 2, pp. 149–188, 2010. (Type: Journal Article | Abstract | Links | BibTeX)|
The P600 is an event-related brain potential (ERP) typically associated with the processing of grammatical anomalies or incongruities. A similar response has also been observed in fully acceptable long-distance wh-dependencies. Such findings raise the question of whether these ERP responses reflect common underlying processes, and what might be the specific mechanisms that are shared between successful processing of well-formed sentences and the detection and repair of syntactic anomalies. The current study presents a comparison of the ERP responses elicited by syntactic violations, garden path sentences, and long-distance wh-dependencies, using maximally similar materials in a within-subjects design. Results showed that a P600 component was elicited by syntactic violations and garden path sentences, but was less robustly elicited in the long-distance wh-dependency condition. Differences in the scalp topography, onset and duration of the P600 effects are characterised in terms of the syntactic operations involved in building complex syntactic structures, with particular attention to retrieval processes, which control the latency of the P600, and structure building processes, which control its duration and amplitude.
| Kazanina, Nina; Phillips, Colin: Differential effects of constraints in the processing of Russian cataphora. In: The Quarterly Journal of Experimental Psychology, vol. 63, no. 2, pp. 371–400, 2010. (Type: Journal Article | Abstract | Links | BibTeX)|
Anaphoric relations between pronouns and their antecedents are subject to a number of different linguistic constraints, which exclude the possibility of coreference in specific syntactic or discourse con- texts. Constraints on anaphora may, in principle, impact online sentence processing in a couple of different ways. They may act as constraints on the generation of interpretations, preventing illicit ana- phoric relations from ever being considered. Alternatively, they may act as later filters on interpret- ations, rejecting candidate interpretations after initial consideration. A number of previous studies have sought to determine which of these mechanisms accurately describes the online impact of con- straints on anaphora. The current studies present evidence that there is no uniform answer to this question, and that the two mechanisms are both used, for different constraints. Evidence for this is drawn from studies on the processing of two constraints on backwards anaphora or cataphora in Russian that apply in superficially similar contexts but that differ in a number of respects. One self-paced reading study and two judgement studies are reported. The self-paced reading study manipulates the gender congruency between a pronoun and a following name in three pairs of con- ditions. In conditions where the pronoun–name configuration violates no constraints on anaphora a gender mismatch effect was observed following the name, as in previous studies, suggesting that comprehenders actively search for an antecedent following a cataphoric pronoun. In conditions where the pronoun–name configuration violates Principle C of the classical binding theory no effect of the gender manipulation was observed, suggesting that comprehenders do not even consider the possibility of interpretations that violate this constraint. In conditions where the pronoun–name configuration violates a Russian-specific constraint on cataphora a gender match effect was observed following the name, the reverse of the finding in the no-constraint conditions, suggesting that the constraint applies as a filter on candidate interpretations.
| Lau, Ellen: The predictive nature of language comprehension. University of Maryland, 2009. (Type: PhD Thesis | Abstract | Links | BibTeX)|
This dissertation explores the hypothesis that predictive processing—the access and construction of internal representations in advance of the external input that supports them—plays a central role in language comprehension. Linguistic input is frequently noisy, variable, and rapid, but it is also subject to numerous constraints. Predictive processing could be a particularly useful approach in language comprehension, as predictions based on the constraints imposed by the prior context could allow computation to be speeded and noisy input to be disambiguated. Decades of previous research have demonstrated that the broader sentence context has an effect on how new input is processed, but less progress has been made in determining the mechanisms underlying such contextual effects. This dissertation is aimed at advancing this second goal, by using both behavioral and neurophysiological methods to motivate predictive or top-down interpretations of contextual effects and to test particular hypotheses about the nature of the predictive mechanisms in question. The first part of the dissertation focuses on the lexical-semantic predictions made possible by word and sentence contexts. MEG and fMRI experiments, in conjunction with a meta-analysis of the previous neuroimaging literature, support the claim that an ERP effect classically observed in response to contextual manipulations—the N400 effect—reflects facilitation in processing due to lexical- semantic predictions, and that these predictions are realized at least in part through top-down changes in activity in left posterior middle temporal cortex, the cortical region thought to represent lexical-semantic information in long-term memory,. The second part of the dissertation focuses on syntactic predictions. ERP and reaction time data suggest that the syntactic requirements of the prior context impacts processing of the current input very early, and that predicting the syntactic position in which the requirements can be fulfilled may allow the processor to avoid a retrieval mechanism that is prone to similarity-based interference errors. In sum, the results described here are consistent with the hypothesis that a significant amount of language comprehension takes place in advance of the external input, and suggest future avenues of investigation towards understanding the mechanisms that make this possible.
| Wagers, Matthew W; Lau, Ellen F; Phillips, Colin: Agreement attraction in comprehension: Representations and processes. In: Journal of Memory and Language, vol. 61, no. 2, pp. 206–237, 2009. (Type: Journal Article | Abstract | Links | BibTeX)|
Much work has demonstrated so-called attraction errors in the production of subject–verb agreement (e.g., ‘The key to the cabinets are on the table’, [Bock, J. K., & Miller, C. A. (1991). Broken agreement. Cognitive Psychology, 23, 45–93]), in which a verb erroneously agrees with an intervening noun. Six self-paced reading experiments examined the online mech- anisms underlying the analogous attraction effects that have been shown in comprehen- sion; namely reduced disruption for subject–verb agreement violations when these ‘attractor’ nouns intervene. One class of theories suggests that these effects are rooted in faulty representation of the number of the subject, while another class of theories suggests instead that such effects arise in the process of re-accessing subject number at the verb. Two main findings provide evidence against the first class of theories. First, attraction also occurs in relative clause configurations in which the attractor noun does not intervene between subject and verb and is not in a direct structural relationship with the subject head (e.g., ‘The drivers who the runner wave to each morning’). Second, we observe a ‘grammatical asymmetry’: attraction effects are limited to ungrammatical sentences, which would be unexpected if the representation of subject number were inherently prone to error. We argue that agreement attraction in comprehension instead reflects a cue-based retrieval mechanism that is subject to retrieval errors. The grammatical asymmetry can be accounted for under one implementation that we propose, or if the mechanism is only called upon when the predicted agreement features fail to be instantiated on the verb.