Elsevier

Journal of Memory and Language

Volume 75, August 2014, Pages 117-139
Journal of Memory and Language

Coargumenthood and the processing of reflexives

https://doi.org/10.1016/j.jml.2014.05.006Get rights and content

Highlights

  • We examined memory retrieval for reflexives in different contexts.

  • ‘Standard’ reflexives preferred an antecedent in the same clause.

  • Reflexives in picture NPs also preferred the structurally closest antecedent.

  • Locality is a highly weighted retrieval cue for reflexives in different contexts.

Abstract

We report three eye-movement experiments and an antecedent choice task investigating the interpretation of reflexives in different syntactic contexts. This included contexts in which the reflexive and a local antecedent were coarguments of the same verbal predicate (John heard that the soldier had injured himself), and also so-called picture noun phrases, either with a possessor (John heard about the soldier’s picture of himself) or without (John heard that the soldier had a picture of himself). While results from the antecedent choice task indicated that comprehenders would choose a nonlocal antecedent (‘John’ above) for reflexives in either type of picture noun phrase, the eye-movement experiments suggested that participants preferred to initially interpret the reflexive in each context as referring to the local antecedent (‘the soldier’), as indexed by longer reading times when it mismatched in gender with the reflexive. We also observed a difference in the time-course of this effect. While it was observed during first-pass processing at the reflexive for coargument reflexives and those in picture noun phrases with a possessor, it was comparatively delayed for reflexives in possessorless picture noun phrases. These results suggest that locality constraints are more strongly weighted cues to retrieval than gender agreement for both coargument reflexives and those inside picture noun phrases. We interpret the observed time-course differences as indexing the relative ease of accessing the local antecedent in different syntactic contexts.

Introduction

Successful language comprehension involves forming dependencies between constituents that may span several words or sentences. A key case in point is anaphora resolution, where the ability to link reflexives and pronouns to their antecedents quickly and accurately is essential for successful sentence and discourse comprehension. Anaphor resolution thus provides an opportunity to examine the memory system that subserves language comprehension, as the interpretation of such forms relies on the retrieval of an item, the antecedent, from memory. A number of syntactic, semantic and discourse level factors are known to guide the retrieval of antecedents during anaphora resolution (see Nicol & Swinney, 2003 for review). In the current study, we focus on the role of syntactic constraints and gender agreement during the resolution of reflexives.

A growing number of studies have examined the time-course of anaphor resolution during the processing of reflexives (Badecker and Straub, 2002, Clackson et al., 2011, Cunnings and Felser, 2013, Dillon et al., 2013, Nicol and Swinney, 1989, Sturt, 2003, Xiang et al., 2009). These studies have sought to examine how syntactic constraints, specifically Principle A of the binding theory (Chomsky, 1981), and gender agreement influence antecedent retrieval during processing. However, to date, most existing research has focused on the processing of reflexives in a single type of syntactic environment, and only a few studies have examined reflexives in other constructions (e.g. Kaiser et al., 2009, Runner et al., 2003, Runner et al., 2006). This is unfortunate given that research in theoretical linguistics has long established that certain constructions pose problems for classic binding theory (e.g. Pollard and Sag, 1992, Reinhart and Reuland, 1993). The aim of the current study was to further examine how the syntactic environment in which a reflexive occurs influences the time-course of anaphor resolution during processing. We begin by discussing theoretical work on the characterisation of syntactic constraints on reflexives, before discussing previous research examining how binding constraints and gender agreement guide antecedent retrieval during comprehension.

In theoretical linguistics, constraints on the interpretation of reflexives have traditionally been accounted for by binding theory (Chomsky, 1981). Binding Principles A and B were formulated to account for interpretive preferences for reflexives and pronouns as in (1).

In (1), the reflexive himself can only be interpreted as referring to John, while the pronoun him must refer to some other, in this case unmentioned, antecedent. In standard binding theory, these preferences are accounted for by Principle A, which states that a reflexive must be bound to an antecedent within the local syntactic domain, in this case the same verbal predicate, and Principle B, which states that pronouns must be free within the local domain. We will refer to local antecedents of reflexives such as John in (1), which are predicted to be preferred according to standard binding theory as accessible antecedents, and other antecedents in a piece of discourse as binding theory inaccessible antecedents.

A key prediction of binding theory as originally formulated is that interpretive preferences for reflexives and pronouns should be in complementary distribution. Whilst this prediction is true for cases such as (1), a number of counter-examples have long been noted (see e.g. Pollard and Sag, 1992, Reinhart and Reuland, 1993). One particularly well-discussed case is that of the so-called referential or picture noun phrase (henceforth PNP), as in (2).

While in (2a), the predictions of binding theory superficially may appear to hold, it is clear from example (2b) that reflexives inside PNPs can be bound to an antecedent outside of the local domain, in this case to an antecedent in a higher clause. Additionally, the complementary distribution of reflexives and pronouns appears to have broken down. In (2b) in particular, the acceptability of use of either a reflexive or pronoun to refer to John is not nearly as marked as in (1). The interpretation of reflexives in PNPs is also sensitive to non-syntactic factors, including pragmatic properties of the text such as ‘point of view’ or the ‘source’ of information (Kaiser et al., 2009, Kuno, 1987).

Considerations such as these led to a reformulation of the constraints on binding. Chomsky (1986) provided an account that maintained the core claims of structural binding theory, but in which the binding domain for reflexives can be enlarged if no binder is available within the most local domain. In this way, binding to the higher clause in (2b) is possible, as no antecedent is available in the reflexive’s most local domain. Others have proposed more radical departures from standard binding theory, and have in particular highlighted the importance of the coargument relationship (Pollard and Sag, 1992, Reinhart, 1983, Reinhart and Reuland, 1993, Reuland, 2001, Reuland, 2011). Although there are differences in the precise characterisation of these different theories, they share in common the importance of coargumenthood as an explanation for the contrast between (1), where the original predictions of binding theory seem to hold, and exceptional cases as in (2). Coargumenthood refers to a relationship between the core arguments of a predicate. In (1), the reflexive and antecedent are both core coarguments (subject and object) of the same verbal predicate. In such coargument contexts the predictions of binding theory are predicted to hold, such that the reflexive must be bound to one of its coarguments (and a pronoun must be free from it). Thus, complementarity is predicted to hold, and the reflexive must be bound, in this case, to John. In cases such as (2), the PNP itself is argued to form a nominal predicate. Here, the reflexive has no coarguments within the local domain (the nominal PNP predicate) that it can bind to and as such, under this account, binding constraints simply cannot apply. In this case, a breakdown in complementarity is expected. Pollard and Sag (1992) refer to such reflexives as being binding theory exempt anaphors, while others (e.g. Reuland, 2001, Reuland, 2011) restrict use of the term anaphor to coargument reflexives, and refer to other types of reflexives as logophors. In both cases, it is claimed that it is only coargument reflexives that are syntactically bound.

A related construction is that of the PNP with a possessor, or possessed picture noun phrase (henceforth PPNP), as in (3).

In contrast to debate regarding reflexives inside PNPs without a possessor, it has generally been assumed in the linguistics literature that PPNPs are restricted by binding constraints, such that a reflexive must bind to the possessor, while a pronoun must be free from it (Pollard and Sag, 1992, Reinhart and Reuland, 1993). If PNPs form nominal predicates, as argued by Pollard and Sag, an antecedent within this predicate (the possessor) must be bound to a reflexive in much the same way that reflexives have to be bound to an antecedent in the same verbal predicate in standard coargument contexts. In this way, the possessor of the PNP in (3) behaves just as the subject of the verb does in (1). However, in contrast to this account of binding in PPNPs, some have found that PPNP reflexives can indeed take antecedents other than the possessor (Keller and Asudeh, 2001, Runner et al., 2003, Runner et al., 2006). Based on evidence from a series of visual world paradigm studies (see below), Runner et al. claim that all reflexives inside PNPs, irrespective of whether or not they have a possessor, should be considered as being exempt from binding theory.

The formation of anaphoric dependencies during language processing clearly implicates the memory mechanisms that underlie language comprehension. A growing body of research has claimed that language comprehension is subserved by a content-addressable memory architecture (for review see Lewis et al., 2006, Van Dyke and Johns, 2012). Such theories assume that there is a severely limited focus of attention that is only capable of holding up to four items (Cowan, 2000), and potentially only one item (McElree, 2001), in the focus at one time. Retrieval is required for all other previously encountered items. Interpreting an anaphor will thus likely involve retrieval of the antecedent from memory, as it is unlikely to still be in the focus of attention when the anaphor is encountered.

A hallmark of content-addressable memory systems is that of direct access to items in memory without the need to search through irrelevant items. As items are accessed directly, this predicts that there should be no differences in retrieval speed as dependency length increases. In content-addressable architectures, retrieval occurs when an item in memory matches a set of retrieval cues. As retrieval cues are matched against all items in memory in parallel, all items that (partially) match the cues receive some amount of activation. This leads to the possibility of similarity-based interference, if a (partially) content-matching but ultimately incorrect item is retrieved. Key evidence that implicates such a memory architecture during language processing has been the demonstration of invariable access times to items in memory irrespective of dependency length (Martin and McElree, 2008, McElree, 2000, McElree et al., 2003), and of similarity-based interference from competitors (Van Dyke, 2007, Van Dyke and McElree, 2006, Van Dyke and McElree, 2011).

An important issue that such models must address relates to the question of what sources of information constitute cues to retrieval. Across different dependencies, a number of potential information sources could act as cues to memory retrieval during language processing. In the case of anaphora, a number of cues could potentially guide retrieval, including gender/number agreement, discourse prominence and syntactic constraints. That agreement might cue memory retrieval during language comprehension is most clearly demonstrated by ‘attraction’ effects in the processing of subject–verb agreement (Pearlmutter et al., 1999, Wagers et al., 2009). For example, Wagers et al. examined sentences such as (4).

Both (4a, b) are ungrammatical as the verb were mismatches in number agreement with the head noun the key. Wagers et al. found longer reading times shortly after the critical verb for ungrammatical in comparison to grammatical control sentences. However, the size of this effect was reliably attenuated when the local ‘attractor’ noun cells matched in number agreement with the verb. Wagers et al. concluded that this attraction effect strongly implicates a direct access content-addressable memory architecture with highly ranked agreement cues. When the correct subject head noun does not fully match with the retrieval cues of the verb (which cue for a number matching head noun), a grammatically illicit but number matching noun is sometimes retrieved.

In contrast to subject–verb agreement, reliable attraction effects have not been reported for reflexives in coargument contexts (Dillon et al., 2013, Sturt, 2003, Xiang et al., 2009). In two experiments, Sturt (2003) recorded participants’ eye-movements as they read sentences similar to (5).

Sturt manipulated gender agreement between the reflexive and both the binding-theory accessible antecedent the surgeon and the inaccessible antecedent (Jonathan/Jennifer). In his Experiment 1, the accessible antecedent was linearly closer to the reflexive than the inaccessible antecedent, as in (5a), while in his Experiment 2, the order was reversed as in (5b). Across both experiments, reading times were longer when the binding theory accessible antecedent the surgeon mismatched in stereotypical gender with the reflexive, during first-pass processing at the reflexive. Some effects of the nonlocal inaccessible antecedent were observed in Experiment 1, but they were restricted to second-pass processing measures, and were not in the direction predicted by attraction. At the sentence prefinal region (used syringe in (5)), in conditions when the local accessible antecedent matched the gender of the reflexive, reading times were longer when the nonlocal inaccessible antecedent mismatched the reflexive’s gender. The inaccessible antecedent did not reliably affect processing at any point in time in Experiment 2 however. As such, Sturt hypothesised that binding constraints act as an initial filter to anaphor resolution, which nevertheless may be defeasible during later stages of processing. In this way, binding constraints may guide the initial retrieval, but this initial retrieval can be subsequently reanalysed. Indeed, Sturt originally reasoned that one possible reason for a defeasible but not absolute filter would be to allow for the interpretation of reflexives that are not governed by binding constraints.

More recently, Dillon et al. (2013) directly compared attraction effects for both subject–verb agreement and reflexive binding. While they replicated the results of Wagers et al. (2009) for agreement, they failed to observe reliable attraction effects for reflexives in sentences similar to (5b), but which contained number agreement rather than gender agreement manipulations. They concluded that, at least for coargument reflexives, Principle A acts as a ‘hard constraint’ that solely guides antecedent retrieval during anaphor resolution.

The hypothesis that certain types of information might take priority over others during antecedent retrieval raises questions with regard to how different cues to retrieval are combined during processing. Some cue-based models (e.g. Lewis & Vasishth, 2007) assume that all sources of information, and thus all cues, are weighted equally. More recently however it has been claimed that at least for some dependencies, some cues might take priority over others. Van Dyke and McElree (2011), for example, claimed that certain cues may ‘gate’ access to other types of information. In their examination of constituent retrieval across relative clause boundaries they claimed that syntactic cues, in this case subjecthood, may gate access to semantic cues such as animacy. While attraction effects in subject–verb agreement suggest that agreement is a highly weighted cue in such cases, the results of Sturt (2003) and Dillon et al. (2013) suggest that for coargument reflexives, binding constraints may be more heavily weighted cues to retrieval than gender/number agreement.

Whether or not binding constraints uniquely cue memory access for coargument reflexives has however been the subject of considerable debate. In addition to the delayed effects observed by Sturt (2003) in his Experiment 1, in a self-paced reading experiment Badecker and Straub (2002) observed longer reading times when a nonlocal inaccessible antecedent matched in gender with the reflexive in comparison to when it did not, in sentences such as John/Jane thought that Bill owed himself another opportunity to solve the problem. Badecker and Straub interpreted this result as indicating competition between antecedents when multiple gender matching antecedents are available in the discourse. Cunnings and Felser (2013) also recently reported an eye-movement experiment which used materials similar to (5b), where the accessible antecedent was linearly more distant to the reflexive than the inaccessible one. However, in their experiment the inaccessible antecedent appeared as the subject of the relative clause, while in the materials tested by Sturt, as in (5b), it was the object. In comparison to a ‘double match’ condition in which both antecedents matched in gender with the reflexive (e.g. ‘the surgeon … Jonathan … himself…’), Cunnings & Felser observed longer reading times during first-pass processing at the reflexive in conditions when either antecedent mismatched in gender with the reflexive (e.g. ‘the surgeon … Jennifer … himself…’ and ‘the surgeon … Jonathan … herself…’). This effect however, was restricted to participants who scored comparatively lower on a standard reading span test (Daneman & Carpenter, 1980). These results indicate that individual differences may affect anaphor resolution, and suggest that for readers with a lower working memory span, a binding theory inaccessible antecedent may be retrieved if it intervenes between the reflexive and accessible antecedent, and is a discourse prominent subject.

In summary, the results of a number of studies suggest that binding constraints constitute highly weighted cues to antecedent retrieval during the processing of coargument reflexives. Whether such structural constraints uniquely cue retrieval has been debated. While effects of structurally inaccessible antecedents have been reported, they have not been observed consistently across experiments, and numerical trends are not always in the same direction. Badecker and Straub (2002), for example, observed longer reading times following multiple gender matching antecedents, whereas a different effect was observed by Sturt (2003) in his Experiment 1, and for lower working memory span readers by Cunnings and Felser (2013) in their Experiment 2. In contrast to these studies examining coargument reflexives, far fewer researchers have investigated the processing of reflexives inside PNPs.

We are aware of only one study to date that has examined the time-course of antecedent retrieval for reflexives inside PNPs without possessors. Kaiser et al. (2009) conducted a visual world paradigm experiment using sentences as in (6), in which the pragmatic properties of the discourse were manipulated. Participants in this task both viewed a visual scene of the participants described in the sentences while their eye-movements were monitored, and additionally had to provide an overt response as to who they thought the reflexive referred to.

In (6a), Peter is the ‘source’ of information and Andrew the ‘perceiver’, while in (6b) these roles are reversed. Kaiser et al. hypothesised that reflexives may prefer antecedents that are ‘sources’ of information (see Kuno, 1987). They observed that while participants generally preferred the subject as the antecedent of the reflexive over 80% of the time, this preference was stronger when the subject was a source of information. Participants’ eye-movements across a visual display also exhibited a subject advantage, which was stronger when the subject was the source of information, during an early time window starting 200 ms after the onset of the reflexive. Kaiser et al. concluded that the resolution of PNP reflexives is immediately sensitive to both structural (in this case subjecthood) and pragmatic (source vs. perceiver) properties of the discourse, a finding potentially compatible with the hypothesis that reflexives inside PNPs are exempt from binding constraints.

Keller and Asudeh (2001) presented results from an acceptability judgement study using magnitude estimation examining PPNP reflexives. They found that participants did not rate sentences such as ‘Hanna found Peter’s picture of herself’ as being fully unacceptable. This finding suggests that, at least in an untimed task, reflexives in PPNPs do not necessarily have to be bound to the possessor, contrary to the predictions of both binding theory (Chomsky, 1981) and subsequently revised proposals (Pollard and Sag, 1992, Reinhart and Reuland, 1993). Runner et al., 2003, Runner et al., 2006 also examined PPNP reflexives in a series of visual world paradigm experiments. Participants in their studies acted out instructions as in (7) while their eye-movements were monitored.

Across their experiments, Runner et al. observed that participants chose an antecedent other than the possessor up to approximately 30% of the time. Additionally, looks to antecedents other than the possessor were observed in a time-window starting 300 ms after the onset of the reflexive. Here, they observed that participants were equally likely to fixate upon either the possessor or subject of the critical sentence, but not the antecedent in the lead-in sentences. Runner et al. concluded that antecedents other than the possessor are immediately considered during processing, and as such claimed that reflexives inside PPNPs should be treated as exempt from binding theory.

Kaiser et al. (2009) also examined PPNP reflexives in sentences such as (8) in a visual world paradigm experiment. As in their experiment on PNP reflexives without a possessor, they manipulated the source/perceiver properties of potential antecedents.

In contrast to their findings for PNP reflexives without a possessor, in this experiment Kaiser et al. observed no reliable effect of their pragmatic manipulation, and instead participants chose the possessor as the antecedent for the reflexive over 90% of the time. Participants’ eye-movements across a visual display were also focused on the possessor, with no reliable influence of the pragmatic manipulation at any point in time.

In summary, previous studies examining coargument reflexives have suggested that binding constraints constitute a highly weighted cue to antecedent retrieval. Kaiser et al. (2009) showed that the resolution of PNP reflexives is sensitive to pragmatic properties of the text, suggesting that the resolution of reflexives inside PNPs is not guided solely by structural cues. Research examining PPNP reflexives has shown that antecedents other than the possessor are sometimes considered immediately during processing (Runner et al., 2003, Runner et al., 2006), suggesting that such reflexives should be considered as being exempt from binding theory. However, this finding was not replicated in a subsequent study, in which there was no evidence that antecedents other than the possessor were being considered during early stages of processing (Kaiser et al., 2009).

Against this background, the aim of the current study was to further investigate how the syntactic context in which a reflexive appears influences the extent to which binding theory accessible and inaccessible antecedents are considered. In order to investigate both the time-course of anaphor resolution and the final interpretations given to reflexives in different contexts, we report three eye-movement experiments and an antecedent choice task. The eye-movement experiments were conducted to examine initial preferences and the relative time-course of antecedent retrieval in coargument contexts (Experiment 1), PNP contexts (Experiment 2) and PPNP contexts (Experiment 3). The antecedent choice task (Experiment 4) was designed to gauge the extent to which comprehenders are willing to ultimately consider local and nonlocal antecedents for coargument and (P)PNP reflexives. We are unaware of any previous studies that have compared coargument reflexives to reflexives in (P)PNPs with maximally similar materials. In contrast to Runner et al., 2003, Runner et al., 2006 and Kaiser et al. (2009), our eye-movement experiments adopted a reading paradigm rather than the visual world paradigm to investigate anaphora resolution. We are particularly interested in the early stages of anaphora resolution to investigate what cues are used to guide the earliest stages of antecedent retrieval in different structural contexts. To this end, we used the gender mismatch paradigm (Cunnings and Felser, 2013, Sturt, 2003, Xiang et al., 2009) to examine the extent to which binding theory accessible and inaccessible antecedents are preferentially retrieved during comprehension.

Section snippets

Experiment 1

To examine the time-course of anaphora resolution, we monitored participants’ eye-movements as they read a series of texts as shown in (9) below. The aim of Experiment 1 was to replicate Sturt (2003) and provide a ‘yardstick’ of the time-course of antecedent retrieval during anaphora resolution for coargument reflexives, to contrast with reflexives inside PNPs and PPNPs in Experiments 2 and 3 respectively. Gender congruence between a reflexive and two antecedents in the discourse was

Experiment 2

In Experiment 2, we examined the time-course of reference resolution for reflexives inside PNPs. The items used were identical to those in Experiment 1, except that the reflexive now appeared inside a picture noun phrase. The items from Experiment 1 were adapted as in (10).

(10a)Local antecedent match, nonlocal antecedent match
Jonathan was walking through the military barracks. He heard that the soldier had a picture of himself in the middle of the mess hall. The food being served for dinner did

Experiment 3

In Experiment 3, we adapted the materials from Experiments 1 and 2 to investigate the processing of reflexives in PPNP contexts. The items were adapted as in (12).

(12a)Local antecedent match, nonlocal antecedent match
Jonathan was walking through the military barracks. He heard about the soldier’s picture of himself in the middle of the mess hall. The food being served for dinner did not look very appetising.
(12b)Local antecedent match, nonlocal antecedent mismatch
Jennifer was walking through

Experiment 4

While Experiments 1–3 investigated initial retrieval preferences, the aim of Experiment 4 was to investigate the final interpretations that comprehenders are willing to give to reflexives in coargument, PNP and PPNP contexts. A subset of six conditions from the eye-movement experiments was tested as in (13). The nonlocal antecedent always matched in gender with the reflexive, while we manipulated gender congruence between the reflexive and local antecedent across the three construction types

General discussion

The primary aim of this study was to investigate the time-course of anaphora resolution for reflexives in different structural contexts. We were particularly interested in examining the extent to which antecedents hypothesised to be accessible and inaccessible according to binding theory are retrieved during processing for coargument reflexives in comparison to those inside picture noun phrases. The results of Experiment 4 indicated that comprehenders are willing to interpret a reflexive inside

Conclusion

The results of three eye-movement experiments showed that readers initially preferred to resolve a reflexive anaphor as referring to a structurally local antecedent. These results are compatible with claims that different cues to memory retrieval during language processing may not be equally weighted (Dillon et al., 2013, Van Dyke and McElree, 2011), and that locality constraints are more heavily weighted retrieval cues than gender for reflexives across different syntactic contexts. We also

Acknowledgements

We thank the editor and three anonymous reviewers for comments on earlier drafts of this paper. The research reported here was funded by a British Academy Postdoctoral Fellowship (pf100026).

References (47)

  • J. Van Dyke et al.

    Retrieval interference in sentence processing

    Journal of Memory and Language

    (2006)
  • J. Van Dyke et al.

    Cue-dependent interference in comprehension

    Journal of Memory and Language

    (2011)
  • M. Wagers et al.

    Attraction in comprehension: Representation and processes

    Journal of Memory and Language

    (2009)
  • M. Xiang et al.

    Illusory licensing effects across dependency types: ERP evidence

    Brain & Language

    (2009)
  • H. Baayen

    Analyzing linguistic data. A practical introduction to statistics using R

    (2008)
  • H. Baayen et al.

    Mixed-effects modeling with crossed random effects for subjects and items

    Journal of Memory and Language

    (2008)
  • W. Badecker et al.

    The processing role of structural constraints on the interpretation of pronouns and anaphors

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (2002)
  • N. Chomsky

    Lectures on government and binding

    (1981)
  • N. Chomsky

    Knowledge of language. Its nature, origin, and use

    (1986)
  • N. Cowan

    The magic number 4 in short-term memory: A reconsideration of mental storage capacity

    Behavioral and Brain Sciences

    (2000)
  • I. Cunnings et al.

    The role of working memory in the processing of reflexives

    Language and Cognitive Processes

    (2013)
  • D. Dahan et al.

    Looking at the rope when looking for the snake: Conceptually mediated eye movements during spoken-word recognition

    Psychological Bulletin & Review

    (2005)
  • M. Daneman et al.

    Individual differences in working memory and reading

    Journal of Verbal Learning and Verbal Behavior

    (1980)
  • Cited by (54)

    • Number attraction in verb and anaphor production

      2022, Journal of Memory and Language
    • Online reflexive resolution and interference

      2024, Language, Cognition and Neuroscience
    View all citing articles on Scopus
    View full text