Parametric effects of syntactic–semantic conflict in Broca’s area during sentence processing

Abstract

The hypothesized role of Broca’s area in sentence processing ranges from domain-general executive function to domain-specific computation that is specific to certain syntactic structures. We examined this issue by manipulating syntactic structure and conflict between syntactic and semantic cues in a sentence processing task. Functional neuroimaging revealed that activation within several Broca’s area regions of interest reflected the parametric variation in syntactic–semantic conflict. These results suggest that Broca’s area supports sentence processing by mediating between multiple incompatible constraints on sentence interpretation, consistent with this area’s well-known role in conflict resolution in other linguistic and non-linguistic tasks.

Introduction

The role of Broca’s area in sentence processing is a fundamental and controversial issue within the neurobiology of language. Many neuroimaging studies have found greater activation in this region within the left inferior frontal cortex (LIFC)¹ for some syntactic structures over others (Ben-Shachar et al., 2003, Caplan et al., 1998, Fiebach et al., 2005, Just et al., 1996, Stromswold et al., 1996). Such effects have been interpreted previously as reflecting the specialization of Broca’s area for syntactic computations (Ben-Shachar et al., 2003, Caplan et al., 1998). More recently, however, there has been growing interest in whether this region’s well-known role in executive function – specifically, the mediation of conflict between multiple incompatible representations – might explain the observed sentence processing effects. Here, we examine this issue by manipulating conflict between syntactic and semantic cues in a sentence-processing task.

A plethora of evidence implicates LIFC in non-syntactic tasks requiring executive function, specifically the modulation of competition between incompatible representations (Jonides and Nee, 2006, Jonides et al., 1998, Milham et al., 2001). For example, in the Stroop task, in which participants identify the font color of written color terms, increased activation in LIFC for incongruent trials – trials where font color and word meaning do not match – is taken to reflect the resolution of representational conflict (Milham et al., 2001). These findings have naturally led to the hypothesis that such conflict resolution mechanisms might also play an important role during sentence processing (Novick, Trueswell, & Thompson-Schill, 2005). Sentence comprehension is incremental and determined by multiple (sometimes competing) information sources. As such, it might involve the resolution of conflict between earlier and later interpretations, or between different sources of information such as semantics and syntax. Consistent with this proposal, recent studies have co-localized syntactic and non-syntactic conflict resolution in LIFC using neuropsychological case studies (Novick, Kan, Trueswell, & Thompson-Schill, 2009) and functional magnetic resonance imaging (fMRI) in healthy adults (January, Trueswell, & Thompson-Schill, 2009).

Under an executive function account, increased LIFC activation for non-canonical structures (e.g., passives or object-relatives) might be taken to reflect the resolution of conflict that is engendered by a preference for canonical structures within the language processing system. For example, incremental processing of an unfolding utterance (The boy was kick…) might favor the default, active interpretation (that the boy was kicking something) until the syntactic structure indicates otherwise (The boy was kicked by…). This conflict between an earlier interpretation and the most recent sentence information could lead to increased recruitment of Broca’s area (Novick et al., 2005). If this account is correct, then factors that influence the probability of a default Agent-Action-Object interpretation should also modulate LIFC activation. Consistent with this, Chen and colleagues have reported that an increase in Broca’s area activation for object relative over subject relative structures was due to a subset of object relatives, specifically those where the relativized object noun was animate (e.g., The golfer that the lightning struck survived the incident) (Chen, West, Waters, & Caplan, 2006). In other words, non-canonical object-relative sentences led to greater LIFC activation compared to canonical subject-relative sentences if and only if the subject noun was animate and could therefore presumably bias toward a default interpretation of the subject as the agent (instead of the patient as indicated by the syntactic structure). Furthermore, activation within Broca’s area is modulated by animacy even in canonical structures. Kuperberg and colleagues found increased Broca’s area activation for simple active structures containing semantic–thematic violations (At breakfast, the eggs would eat…) compared to those that did not contain such violations (At breakfast, the boys would eat…) (Kuperberg, Sitnikova, & Lakshmanan, 2008). Together, these studies demonstrate that activation within Broca’s area is not determined by syntactic structure alone but rather is dependent on the conflict between multiple cues to sentence interpretation.

The current study extends previous investigations of Broca’s area’s function during sentence processing in three ways. First, we manipulate syntactic structure and conflict between semantic and syntactic cues within the same experiment and ask which factor accounts better for the variation in Broca’s area activation. Second, we vary conflict between semantic and syntactic cues parametrically rather than dichotomously and ask whether activation in Broca’s area would also vary parametrically. Last but not least, unlike the above-mentioned studies, our manipulation of conflict between semantics and syntax relies not on animacy but on world knowledge about relations between specific entities and specific actions (e.g., cops being handcuffed versus doing the handcuffing). Many linguistic theories distinguish animacy from such semantic factors. Thus our manipulation and the concomitant pattern of neurophysiological responses may shed a different light on Broca’s area function.

Participants silently read active and passive sentences [a–d]. None of the sentences were anomalous. We manipulated conflict within passive sentences [b–d] (see Sections 2 and 3 for details).

• Active: The father was calming the cranky girl.

• No-conflict Passive: The celebrity was interviewed by a reporter.

• Neutral Passive: The patient was interviewed by the attractive man.

• Conflict Passive: The journalist was interviewed by the undergraduate.

Relative to neutral passives, conflict passives were hypothesized to induce greater competition between a syntactically guided interpretation and a semantically guided one. While journalists can quite possibly be interviewees (as indicated by the syntax), they are more likely to be interviewers (as indicated by world knowledge). In contrast, no-conflict passives were hypothesized to ease processing due to the congruence of syntactic and semantic cues. Celebrities are both possible and likely interviewees. If Broca’s area does not mediate the type of conflict manipulated here, we would expect activation to reflect syntactic structure alone such that all passives are equivalent and produce more activation than actives ((Conflict = Neutral = No-conflict passives) > Actives). Alternatively, if Broca’s area underlies the modulation of competition between incompatible representations, activation should vary according to hypothesized conflict (Conflict > Neutral > No-conflict passives).

We tested these contrasting hypotheses in regions of interest (ROIs) in Broca’s area, defined in three different ways; the effects in each case supported the conflict resolution hypothesis.