The separation of processing stages in a lexical interference fMRI-paradigm

Abstract

In picture–word interference paradigms, the picture naming process is influenced by an additional presentation of linguistic distractors. Naming response times (RTs) are speeded (facilitation) by associatively-related and phonologically-related words when compared to unrelated words, while they are slowed down by categorically-related words (inhibition), given that distractor onsets occur at appropriate stimulus onset asynchronies (SOAs). In the present study with healthy subjects, we for the first time integrated all four auditorily presented distractor types into a single paradigm at an SOA of − 200 ms, in order to directly compare behavioral and neural interference effects between them. The behavioral study corroborated results of previous studies and revealed that associatively-related distractors speeded RTs even more than phonologically-related distractors, thereby becoming equally fast as naming without distractors. Distractors were assumed to specifically enhance activation of brain areas corresponding to processing stages as determined in a cognitive model of word production (Indefrey, P., Levelt, W.J.M., 2004. The spatial and temporal signatures of word production components. Cognition 92, 101–144.). Functional magnetic resonance imaging (fMRI) at 3 T revealed activation of left superior temporal gyrus exclusively for phonologically-related distractors, and activation of left or right lingual gyrus exclusively for associatively-related and categorically-related distractors, respectively. Moreover, phonologically-related distractors elicited phonological–phonetic networks, and both semantic distractors evoked areas associated with mental imagery, semantics, and episodic memory retrieval and associations. While processes involved in distractor inhibition (e.g., conflict/competition monitoring) and high articulatory demands were observed for categorically-related distractors, priming of articulatory planning was revealed for associatively-related distractors. We conclude that activations of neural networks as obtained by the fMRI interference paradigm can be predicted from a cognitive model.

Introduction

In cognitive models of normal word production, picture naming has been segregated into several cognitive processing stages (cf. Levelt et al., 1999, Indefrey and Levelt, 2004; see also Dell et al., 1997, Chiliant et al., 2003). In the Levelt model, visual object recognition and selection of the appropriate semantic representation (lexical concept) are followed by selection of the lexical entry (lemma), retrieval and encoding of the phonological word form, as well as phonetic encoding and articulation of the picture name (cf. Fig. 1). In interference paradigms, participants are instructed to name pictures of simple objects while ignoring auditory or written distractor words. Processing of auditory distractors includes recognition of the word and selection of a lexical concept, probably guided by a visual image (imagery). Distractors differentially influence those production stages they interact or share with.

The time course of picture–word interference has been intensively investigated in a series of behavioral studies (e.g., Glaser and Düngelhoff, 1984, Schriefers et al., 1990, Starreveld and La Heij, 1995, Roelofs et al., 1996, Damian and Martin, 1999, Alario et al., 2000, Jescheniak and Schriefers, 2001, Damian and Bowers, 2003, Costa et al., 2005, Mahon et al., 2007). Both linguistic type and stimulus onset asynchrony (SOA) of the distractor, i.e. onset of the word in relation to onset of the picture, may affect naming latency. For example, Schriefers et al. (1990) investigated the impact of auditory distractors given before, simultaneously, and after picture presentation (SOA − 150, 0, + 150 ms). The authors reported that an unrelated word without any similarity to the target (e.g., roof when the naming target is dog) slowed down response times (RTs) compared to naming without distractors. Words belonging to the same semantic category as the target (e.g., cat when the target is dog) affected RTs only before picture presentation; they were slowed down compared to the unrelated condition. Distractors with phonological relation to the target (e.g., fog when the target is dog) speeded responses compared to the unrelated condition if given simultaneously or after the picture. However, we observed that SOAs which were optimal or at least sufficiently effective for each linguistic distractor type varied between studies (see below). The slowdown of RTs exerted by unrelated distractors can be explained by the dual demand of both distractor and picture processing and the need to functionally inhibit the distractor. Categorically-related distractors slow down responses because they additionally compete with the target name, while phonologically-related distractors speed responses because they facilitate it.

It is still debated at which processing stage the distractor words finally exert their effects. The facilitative phonological effect has generally been attributed to phonological retrieval or encoding (e.g., Schriefers et al., 1990, Starreveld and La Heij, 1995, Damian and Martin, 1999). The categorical effect has been located at semantically driven processing stages; it may either originate from an early decision which concept to lexicalize (conceptual selection account, e.g., Costa et al., 2005) or from later competition between lexical entries (lexical competition account; e.g., Roelofs et al., 1996, Starreveld and La Heij, 1995, Damian and Bowers, 2003). However, the fact that semantic information influences the time required to name a picture need not mean that the effect occurs at a semantic stage (Mahon et al., 2007). Contrarily, a decision mechanism which is sensitive to semantic information may operate at a post-lexical stage (response exclusion hypothesis).

The impact of associative distractors in the written modality has been studied by Alario et al. (2000). They examined semantic compared to unrelated distractors at early and late SOAs before picture presentation (SOA − 234 and − 114 ms) and revealed that associative-semantic distractors (e.g., bone when the target is dog) only elicited an early facilitative effect, and categorical-semantic distractors (e.g., door when the target is window) only exerted a later inhibitory effect. Thus, associative facilitation clearly preceded categorical inhibition. The authors argued that associative distractors exert their support early because they need ample time for activation spreading to the associated lexical target which is under selection. On the contrary, categorically-related distractors elicit their competing activation later on only, when their processing has not been completed and lexical selection of the target still takes place. These results are in accordance with recent RT results reported by Mahon et al. (2007). They compared the impact of written distractor types at SOA 0 ms and obtained categorical inhibition as well as phonological facilitation, but no effect for associative distractors. At the same SOA, semantic facilitation was reported for written distractors corresponding to related parts of the depicted object (e.g., bumper when the target is car; Costa et al., 2005). This suggests that activation spreading to words with part/whole relation may occur faster. The facilitatory associative effect may arise from conceptual rather than lexical activations (La Heij et al., 2006).

Thus, we can distinguish four linguistic distractor types: Unrelated words, associatively-related words including situational and part/whole relationships, categorically-related words, and phonologically-related words. A direct comparison of all four distractor types in a single paradigm has not been performed yet, probably due to the findings that optimal SOAs differ between them. But regarding the literature on distractors, there may be a certain point in time at which effective SOAs overlap, and at which therefore all distractor types might evoke their predicted interference effects, either inhibitory or facilitative. Fig. 2 presents a selection of studies reviewed and illustrates that it may in principle be possible to find both phonological facilitation (Phon) and categorical inhibition (Categ) between SOA − 200 and + 100 ms, while associative facilitation (Assoc) may be restricted to early SOAs. We concluded that an SOA − 200 ms was an optimal compromise between all four auditory distractor types. By applying the auditory picture–word interference paradigm with one single SOA in functional magnetic resonance imaging (fMRI), possible differences in brain activity between distractor types cannot be attributed to onset of distractor presentation, and neural networks for overt picture naming with varying distractor types can be compared in one and the same subject.

Fig. 1 illustrates stages of auditorily distracted picture naming according to the Levelt model, activation enhancements, and response times (RTs) as realized in the four experimental conditions. In a behavioral study outside the scanner, we intended to measure and compare RTs of picture naming with all distractor types as well as naming without distractors to test our materials and procedures. We expected that RTs of categorical distractors were slowest, of unrelated distractors faster, and of associative/phonological distractors fastest (RTs: C > U > A/P). The “default” RT of unrelated distractors has been shown to be slower than RTs for naming without distractors (cf. Schriefers et al., 1990). To our knowledge, neither RT comparisons of associative and phonological distractors, nor RT comparisons between specific distractor types and naming without distractors have been performed yet.

In the present study, differential RTs and expected brain processing of distractor conditions are taken to be the result of specifically enhanced activations (see Fig. 1; cf. Mechelli et al., 2007, Roelofs, 2008, Abel et al., in preparation). However, activation suppression has also been reported (cf. De Zubicaray et al., 2002, Bles and Jansma, 2008). We made three major assumptions to explain expected RT results: Firstly, unrelated distractors are taken to comprise the whole naming process including processing and functional inhibition of the distractor, and they are therefore called unspecific distractors. The other three distractors interfere with specific aspects of these processes and are therefore called specific distractors. Secondly, when picture naming and processing stage(s) affected by the distractor coincide, dual activation may specifically enhance activation. Considering previous behavioral interference studies, we assume that phonological facilitation occurs during phonological retrieval and encoding (e.g., Schriefers et al., 1990), and associative-semantic facilitation during conceptual preparation (La Heij et al., 2006). Due to this boost of target activation, demands on inhibition of the concurrent distractor should be minor (but cf. De Zubicaray et al., 2002). Categorically-related distractors enhance activation at the lexical selection stage (lexical competition account; see also Alario et al., 2000). Impeding the semantic competitor places high demands on distractor inhibition (De Zubicaray et al., 2001). Both semantic distractor types are presumed to enhance activation for visual processing (picture recognition and mental imagery). Mental imagery assists semantic comprehension, which is not decisive for phonological distractors. Thirdly, indirect enhancement of activation may evolve from feed-forward processing, from stages with enhanced activation to neighboring stages — especially in both facilitatory conditions.

The neural activation patterns of picture naming in general are relatively well-known due to advances of brain imaging techniques (cf. reviews of Cabeza and Nyberg, 2000, Indefrey and Levelt, 2004, Vigneau et al., 2006; see also Hickok and Poeppel, 2004, Hickok and Poeppel, 2007, Martin, 2003). Nevertheless, further investigation is needed to carefully separate neural correlates of cognitive processing stages. According to Indefrey and Levelt, the left-lateralized word production network includes the mid segment of the middle temporal gyrus (MTG) for lexical selection, posterior superior temporal gyrus (STG) and MTG including Wernicke's area for word form retrieval, posterior inferior frontal gyrus (IFG) including Broca's area for word form encoding, and finally parts of the central nervous motor system for articulation. Left STG and right mid STG are associated with self-monitoring, which also involves perception of one's own speech. Moreover, the anterior cingulate has been related to self-monitoring and response inhibition (Christoffels et al., 2007, Fu et al., 2006) as well as conflict/competition monitoring (for overview, see Roelofs, 2008). According to Vigneau et al. (2006), pars triangularis of the IFG (BA 45) is involved in phonological processing, pars orbitalis (orbIFG, BA 47) in semantic processing. A phonological store area may be located in the gyrus supramarginalis (SMG) and amodal conceptual-semantic processing in angular (AG) and fusiform gyrus (FG). Visual recognition was located in visual, mental imagery in occipital/temporal, and episodic memory retrieval and associations in medial parieto–occipital areas (cf. Cabeza and Nyberg, 2000).

Only a few studies investigated the effect of picture–word interference with fMRI. De Zubicaray et al., 2001, De Zubicaray et al., 2002) examined only one specific distractor in each experiment. The task of Mechelli et al. (2007) included overt production of the prime before production of the target, and their semantic condition comprised associative as well as categorical distractors. In general, distractors applied were restricted to the written modality. A paradigm limited to auditory distractors, as applied in the present study, has the advantage to be simpler, because oral and written modalities are not intermingled. By means of repeated measures ANOVA, we set out to find networks specific to our distractor conditions (subtraction method). Moreover, we intended to find brain areas common to a distractor type as analyzed by conjunction analysis. Based on previous interference studies, specific predictions were made to also define regions of interest (ROIs). For phonological distractors, we expected differential activations at least in left (posterior) STG (De Zubicaray et al., 2002, Bles and Jansma, 2008). Previously reported STG activations of the facilitatory effect consisted of activation suppression (signal reduction) instead of enhancement (signal increase). However, if interference arises from dual activation of the phonological stage as concluded from the cognitive model, we should observe an increase of activation. For categorical distractors, we predicted a specific increase of activation at least in left mid MTG (Mechelli et al., 2007; see also Indefrey and Levelt, 2004) due to lexical selection. For associative distractors, the locus of activation may either comprise the left MTG region or other brain areas responsible for (conceptual-)semantic processing, i.e. left AG and orbIFG (Mechelli et al., 2007; see also Vigneau et al., 2006).

As the unrelated condition places relatively high demands on all processing stages, we expected no brain area to be common to the unrelated condition compared to all specific distractors as measured by conjunction analysis. For the same reason, subtracting the unrelated distractor from a specific distractor should be less successful in revealing brain activations than subtracting another specific distractor (see also Mechelli et al., 2007).