The role of the left inferior frontal gyrus in implicit semantic competition and selection: An event-related fMRI study

Abstract

Recent research suggests that the left inferior frontal gyrus (LIFG) plays a role in selecting semantic information from among competing alternatives. A key question remains as to whether the LIFG is engaged by the selection of semantic information only or by increased semantic competition in and of itself, especially when such competition is implicit in nature. Ambiguous words presented in a lexical context provide a means of examining whether the LIFG is recruited under conditions when contextual cues constrain selection to only the meaning appropriate to the context (e.g., coin-mint-money) or under conditions of increased competition when contextual cues do not allow for the resolution to a particular meaning (e.g., candy-mint-money). In this event-related fMRI study, an implicit task was used in which subjects made lexical (i.e., word/nonword) decisions on the third stimulus of auditorily presented triplets in conditions where the lexical context either promoted resolution toward a particular ambiguous word meaning or enhanced the competition among ambiguous word meanings. LIFG activation was observed when the context allowed for the resolution of competition and hence the selection of one meaning (e.g., coin-mint-money) but failed to emerge when competition between the meanings of an ambiguous word was unresolved by the context (e.g., candy-mint-money). In the latter case, there was a pattern of reduced activation in frontal, temporal and parietal areas. These findings demonstrate that selection or resolution of competition as opposed to increased semantic competition alone engages the LIFG. Moreover, they extend previous work in showing that the LIFG is recruited even in cases where the selection of meaning takes place implicitly.

Introduction

Much current research has focused on the neural systems underlying the ability to attend to one stimulus in the face of multiple competing stimuli (for a recent review, see Kan and Thompson-Schill, 2004a). One prime example of this process occurs during language comprehension, where competition takes place at multiple levels of the grammar (e.g., phonetic, phonological, lexical-semantic and syntactic) and must be quickly resolved by the language processing system in order for the intended message to be understood. The goal of the present investigation is to examine the neural substrates of competition and selection during lexical-semantic processing.

Recent neuroimaging evidence suggests that the left inferior prefrontal cortex (LIPFC) may be critically involved in semantic processing. This evidence comes out of research showing that, in addition to posterior brain regions of the left hemisphere, frontal areas also contribute to certain aspects of processing word meaning (Devlin et al., 2003, Gabrieli et al., 1996, Gabrieli et al., 1998, Gitelman et al., 2005, Poldrack et al., 1999, Roskies et al., 2001). In fact, current theorizing about the LIPFC suggests that this region may play a key role in functions such as the controlled retrieval of semantic knowledge (Badre and Wagner, 2004, Badre et al., 2005, Gold and Buckner, 2002, Gold et al., 2005, Wagner et al., 2001), the manipulation of semantic knowledge in explicit tasks (Spitzer et al., 1996), and crucially the selection of semantic information among competing alternatives (Kan and Thompson-Schill, 2004a, Kan and Thompson-Schill, 2004b, Thompson-Schill et al., 1997, Thompson-Schill et al., 1999, Thompson-Schill et al., 1998, Thompson-Schill et al., 2002).

Of primary importance to the present work is the proposal put forth by Thompson-Schill and colleagues (Kan and Thompson-Schill, 2004a, Kan and Thompson-Schill, 2004b, Thompson-Schill et al., 1997, Thompson-Schill et al., 1998, Thompson-Schill et al., 1999, Thompson-Schill et al., 2002) that the LIPFC is engaged in the selection of meaning from competing alternatives. Evidence for this hypothesis was originally derived from Thompson-Schill et al. (1997), who, in an fMRI study, varied the demands placed on selection (i.e., high versus low) across three different semantic tasks. In the generation task, subjects covertly generated a verb in response to a visually presented noun which had either many associated verb responses (high selection) or few associated responses (low selection). In the classification task, subjects classified pictures of common objects according to a specific attribute of the object (high selection) or the basic level object name (low selection). Finally, in the comparison task, subjects made comparisons of a target word to several probe words and then decided which probe was most similar in terms of specific attributes (high selection) or global similarity (low selection). In all cases, in the high selection condition, retrieval of object-specific semantic knowledge made available both task-relevant semantic information and task-irrelevant semantic information, such as inappropriate associations or irrelevant features of the object. In contrast, in the low selection condition, all of the retrieved semantic information was task-relevant, either in generating the most dominant association, identifying the name of the object or making a global similarity judgment. Crucially, when comparing the high versus low selection conditions across all three tasks, Thompson-Schill et al. found increased activation in the LIPFC, specifically in the left inferior frontal gyrus (LIFG), for the high selection condition. This result was argued as evidence that this region plays a critical role in the selection of relevant semantic information in the presence of competing information.

While Thompson-Schill et al. (1997) associate LIFG activation with selection, their results are also consistent with an account where increased competition leads to LIFG activation. On the one hand, the high selection condition is also one of high competition, where many semantic properties are accessed and compete for selection. On the other hand, the low selection condition is one of low competition, where few semantic properties are retrieved and compete for selection. Thus, the LIFG activation observed by Thompson-Schill et al. may be due to either increased competition occurring prior to selection or resolution of the competition itself (i.e., selection).

An ideal test case of these alternative interpretations is that of semantically ambiguous words (e.g., mint: candy/money), where single lexical items share multiple unrelated meanings. In the psycholinguistic literature, it has been shown that all meanings of an ambiguous word are initially activated and compete for selection, at least for a brief period of time (Onifer and Swinney, 1981, Swinney, 1979). Selection is eventually accomplished by using contextual cues, if available, to constrain the choice to only the meaning appropriate to the specific context (Simpson, 1994). In other words, only one meaning is integrated into the surrounding context and selected for processing. These findings have direct relevance to the proposal of Thompson-Schill et al., as either competition among multiple unrelated meanings of ambiguous words or integration and ultimate selection of one meaning may recruit left frontal regions, in particular the LIFG.

The results from recent fMRI studies on semantic ambiguity suggest that the LIFG responds to the resolution of competition and not simply to the presence of competition inherent in ambiguous words (Chan et al., 2004, Copland et al., 2003, Hoenig and Scheef, 2005, Mason and Just, 2007, Rodd et al., 2005, Zempleni et al., 2007). In particular, if LIFG activation were merely an index of semantic competition inherent in the stimulus, then ambiguous words outside of a context (i.e., where resolution or selection of a particular meaning cannot take place) should elicit activation in left frontal areas. Findings of two recent studies are inconsistent with this claim (Chan et al., 2004, Copland et al., 2003). In one study, Chan et al. (2004) found that ambiguous words (relative to unambiguous words) yielded increased activation in left mid-superior rather than left inferior frontal regions. Moreover, the only LIFG activation observed in this study was for unambiguous words. In another study, Copland et al. (2003) found decreased activation in the LIFG when subordinate (i.e., less frequent) meanings, presumably involving increased competition, were primed. Thus, semantic competition inherent in ambiguous words does not seem to in and of itself engage the LIFG. Rather, having to resolve this competition (i.e., by integrating and selecting one meaning of an ambiguous word) appears to recruit the LIFG.

There are a series of studies that are consistent with the view that the IFG is recruited in processing ambiguous words when meaning competition is resolved (Hoenig and Scheef, 2005, Mason and Just, 2007, Rodd et al., 2005, Zempleni et al., 2007). In one study, Rodd et al. (2005) investigated the role of sentence context in resolving semantic ambiguity by comparing neural responses to sentences containing ambiguity (e.g., There were dates and pears in the fruit bowl) and those without ambiguity (e.g., There was beer and cider on the kitchen shelf). In one experiment, subjects decided whether a probe word presented after each sentence was related to the meaning of the sentence, whereas in the second experiment, subjects simply listened to the sentences. In both cases, Rodd et al. found increased neural activation in bilateral IFG for ambiguous relative to unambiguous sentences. Taken together, these findings suggest that the LIFG is recruited when the meaning of an ambiguous word is integrated within a sentence context and hence when a particular meaning of an ambiguous word is selected.

The goal of the present study was to further investigate the neural substrates of semantic competition and selection using an event-related fMRI paradigm. In contrast to previous studies which have for the most part used explicit tasks, the current study used an implicit task with semantic ambiguity (e.g., mint: candy/money) to examine whether LIFG activation was modulated by competition and/or selection during semantic processing. A semantic priming paradigm was employed, in which subjects were required to make a lexical (i.e., word/nonword) decision on the third stimulus of auditorily presented triplets in conditions where the lexical context either promoted selection of and resolution toward a particular ambiguous word meaning or enhanced the competition among ambiguous word meanings. In the concordant condition (e.g., coin-mint-money), both flanking words were related to the same meaning of the middle ambiguous word. Although the middle stimulus was semantically ambiguous, the flanking words reduced competition between the ambiguous word meanings and allowed for one meaning to be integrated and ultimately selected. In the discordant condition (e.g., candy-mint-money), the flanking words were related to different meanings of the ambiguous word. Thus, the context in which the ambiguous word occurred did not allow for integration and ultimate selection of a particular meaning to take place. In the neutral condition (e.g., gown-mint-money), only the final word was related to a particular meaning of the ambiguous word, reducing competition and favoring selection of one meaning but to a lesser extent than the concordant condition. Finally, in the unrelated condition (e.g., cake-rat-money), which served as a baseline against which all other conditions were compared to determine the amount of priming, all three words were semantically unrelated. If the LIFG is engaged primarily when the context allows for the resolution toward and hence integration and selection of a particular meaning, then increased activation in this region should be observed in the concordant condition and to a lesser extent the neutral condition relative to the unrelated condition. In contrast, if the LIFG is engaged by competition, then increased activation in this region should be observed in the discordant condition relative to the unrelated condition.

Note that these predictions are different from what one might expect based on previous neuroimaging studies of semantic priming, where related conditions often show decreased neural activation relative to unrelated conditions (Matsumoto et al., 2005, Rissman et al., 2003, Wheatley et al., 2005, Wible et al., 2006). In contrast to previous studies which used unambiguous word pairs, the present study employed ambiguous word triplets. Processing of these triplets invokes higher-level meaning integration and selection processes (i.e., one meaning must be integrated across a lexical context and ultimately selected), which are not involved in processing unambiguous word pairs. It is therefore highly likely that neural activation patterns in the current study will differ from the patterns previously shown in semantic priming studies using word pairs.