Abstract

It has been widely argued that the left inferior frontal gyrus (LIFG) is involved in the control of retrieval of information from long-term memory. Recent claims that the LIFG is involved in selecting among semantic alternatives have been challenged on the grounds that the manipulation of selection demands may have been confounded with controlled retrieval. The current study used an event-related functional magnetic resonance paradigm to re-examine the possibility that LIFG activation is involved in selection processes. In order to minimize potential confounding effects of controlled retrieval, we used an automatic retrieval task (picture naming) and held retrieval demands constant, while varying selection demands by way of competitor priming from earlier semantically related trials. We found significant activation in LIFG as a function of increased selection demands; activation centred on two peaks, one in anterior LIFG and a second more superior and posterior region. These data support the view that LIFG plays a role in selection among semantic information, even in the absence of controlled retrieval processes.

Introduction

The representation and processing of semantic information support a wealth of cognitive abilities, including language comprehension and production, object recognition, thinking and memory. For this reason, understanding the neural basis of the semantic system is a central goal of cognitive neuroscience. Lesion-deficit studies and neuroimaging of healthy brains have established that semantic processing recruits an extensive, largely left lateralized network of temporal, frontal and parietal regions (Vandenberghe et al., 1996; Mummery et al., 1998; Tyler et al., 2001). The question remains, however, as to the precise contributions of, and interactions among, the components of such a system.

In this paper we focus on the role of a region that has attracted considerable attention: the left inferior frontal gyrus (LIFG). Neuroimaging studies have frequently found the LIFG to be activated during semantic tasks (Petersen et al., 1989; Demonet et al., 1992; Demb et al., 1995; Martin et al., 1995; Thomspson-Schill et al., 1997; Gabrieli et al., 1998; Poldrack et al., 1999; Tyler et al., 2001). Recent studies have also shown that transcranial magnetic stimulation (TMS) of the anterior portion of the left inferior prefrontal cortex slows responses in a semantic decision task (Devlin et al., 2003; Kohler et al., 2003). However, in the neuropsychological literature, LIFG lesions have not typically been reported to result in semantic deficits, being traditionally associated with the morpho-syntactic and/or speech output impairments of Broca's aphasia (Goodglass, 1976; Zurif, 1995; Grodzinsky, 2000) although the lesion-deficit associations here are controversial (Dronkers et al., 2000). Semantic impairments are generally associated with temporal lobe damage (Hodges et al., 1992; Gainotti et al., 1995). It is widely argued, therefore, that the LIFG may support processes that operate over semantic representations rather than the representations themselves (Poldrack et al., 1999; Fletcher and Henson, 2001).

Two major hypotheses have recently emerged as to the nature of the semantic processes supported by the LIFG. The first is that LIFG is involved in selecting among competing semantic alternatives held in working memory; for example, in deciding which of many possible attributes of a word's meaning are relevant for a specific decision task (Thompson-Schill et al., 1997, 1998, 1999;Desmond et al., 1998; Gabrieli et al., 1998). The second is that LIFG guides the controlled retrieval of semantic information that cannot be generated on the basis of automatic stimulus–response associations; for example, retrieving the semantic information necessary to find a connection between two weakly related words (Buckner, 1996; Fiez, 1997; Bokde et al., 2001; Wagner et al., 2001; Gold and Buckner, 2002).

In the current study we examine the LIFG role in semantic selection when potentially confounding effects of controlled retrieval are minimised. Before discussing these issues further, we note that there is variation in the exact regions of interest across studies, with some researchers focusing specifically on the LIFG (e.g. Thompson-Schill et al., 1997, 1999), while others suggest that semantic processing is limited to only the more anterior portion of ventrolateral frontal cortex (Poldrack et al., 1999) and yet others refer more generally to left inferior prefrontal cortex (LIPC). For consistency, we use the term LIFG throughout this paper.

Semantic Selection and Retrieval

Drawing on theoretical accounts of prefrontal cortex function in non-semantic domains (e.g. Kimberg and Farah, 1993), Thomspon-Schill et al. (1997) suggested that the LIFG is involved in selecting among competing semantic alternatives, rather than retrieval of semantic information per se. In a similar vein, Gabrieli et al. (1998) suggested that the LIFG is activated to the extent that a greater amount of semantic information must be held in working memory in order to select among alternative answers. To test their hypothesis, Thompson-Schill et al. (1997, 1999) carried out a series of functional magnetic resonance imaging (fMRI) studies, manipulating selection demands in several tasks, while aiming to hold semantic retrieval constant, and reported a significant increase in LIFG activation in conditions that increased selection demands, as well as a decrease in activation when selection demands were reduced by repetition of task demands. They also found that patients with lesions including the LIFG were impaired specifically in the high selection conditions of a verb generation task (Thompson-Schill et al., 1998). Further evidence in support of left frontal involvement in selection processes comes from a stem completion study (Desmond et al., 1998), although activation centred on left middle rather than inferior frontal cortex, and from a recent study employing a proactive interference paradigm (Dolan and Fletcher, 1997; Fletcher et al., 2000).

An alternative view is that the LIFG supports the controlled retrieval of semantic information (Buckner, 1996; Fiez, 1997; Bokde et al., 2001; Wagner et al., 2001; Gold and Buckner, 2002; reviewed in Poldrack et al., 1999). On this view, it is claimed that the cases where LIFG activation is not observed — such as the low-selection conditions of Thompson-Schill et al.'s studies — involve automatic rather than controlled retrieval and it is this variable that accounts for the lack of LIFG recruitment. Controlled retrieval of semantic information is an example of a more general process of cognitive control (Norman and Shallice, 1986; Miller, 2000; Wagner et al., 2001; Miller et al., 2002). The claim is that the prefrontal cortex is a source of top-down signals, which guide goal-directed behaviours and processes that are not automatically triggered by strong associations to environmental stimuli. Controlled behaviours, including the strategic interrogation of conceptual knowledge to generate specific pieces of information for a given task, are attention-demanding and driven by internal goals and knowledge. Automatic behaviours, in contrast, are well-learned responses triggered by environmental stimuli (Miller et al., 2002).

Wagner et al. (2001) suggest that in Thompson-Schill et al.'s studies, controlled retrieval and selection were confounded such that increases in LIFG activation observed in ‘high selection’ conditions could be attributed to simultaneous increase in controlled retrieval demands. For example, in Thompson-Schill et al. (1997)'s classification task, the high selection condition involved judging whether a specific attribute was true or false of a pictured object (e.g. car–expensive), while the low selection condition involved judging whether a word was the appropriate basic level name for a pictured object. The logic of the task is that in the former case, participants must select the relevant attribute from the many possible semantic properties of car in order to make the correct response, whereas in the case of the picture-name match, only one name is appropriate, so no selection is required. However, the difference between the tasks could be reinterpreted as a manipulation of controlled versus automatic retrieval, i.e. the controlled retrieval of a specific, task-relevant piece of information in order to decide that a car is expensive in contrast to the automatic stimulus–response association of an object and its name. In Thompson-Schill et al. (1999)'s attribute generation study, participants generated either a colour or action associated with a stimulus word such as apple. When they were subsequently asked to produce a different attribute to the same concept (e.g. first the colour of the apple and then an action associated with an apple), LIFG activation increased, presumably as a function of the selection demands associated with competition from the previously generated, now irrelevant, information. Thompson-Schill et al. argue that if LIFG supported retrieval of semantic information rather than selection, its activity should be reduced when the same concept (e.g. apple) is repeated, regardless of whether the same or different attribute is generated, since retrieval of this concept has been primed. However, proponents of the retrieval hypothesis could, again, reinterpret the ‘different attribute’ trials as involving greater controlled retrieval demands, since access to semantic memory in this task could be viewed as guided search for specific pieces of information associated with a word, rather than as a passive spread of activation to a range of attributes from which one is then selected. Searching for an action associated with an apple may be more difficult when it is necessary to suppress the irrelevant search criterion from an earlier ‘colour’ trial [see also Badre and Wagner (2002) for other interpretations of this study].

In summary, several studies strongly suggest that the LIFG plays a role in selecting among semantic information. However, an alternative explanation in terms of controlled retrieval remains a possibility in any task where subjects are required to retrieve a specific piece of information that is not automatically activated by an existing stimulus-driven association; in each case, an increase in selection demands can be reinterpreted as the need to retrieve more information, more distantly related information or to change search criteria from trial to trial. We avoided this problem in our study by manipulating selection demands in the context of a task that exploits a strong pre-existing stimulus–response association, so minimizing the role of controlled retrieval and therefore any possibility that it could inadvertently covary with selection manipulations.

If we find LIFG activation association with increased selection in an automatic task, this does not, of course, rule out the possibility that left prefrontal cortex also plays a part in controlled retrieval. It is plausible that there is functional heterogeneity within the LIFG, such that different subregions may be associated with either controlled retrieval or selection. For example, Wagner et al. (2001) suggest that controlled retrieval is centred upon the anterior regions of left inferior frontal cortex, leaving the possibility for selection processes to be localized to the more posterior LIFG region. As mentioned above, the peak activations reported by Thompson-Schill et al. fall in a dorsal posterior region of the LIFG, an area suggested by Poldrack et al. (1999) to be commonly activated for phonological and lexical, as well as semantic processing — perhaps reflecting a domain-general selection and/or retrieval process (Gold and Buckner, 2002). The area associated most strongly with controlled retrieval in Wagner et al.'s study is more anterior — within the ventral anterior region (essentially BA 45/47) of the LIFG suggested by Poldrack et al. to be selectively associated with semantic processing. It is possible that this region subserves controlled retrieval (Fiez, 1997; Price et al., 1997; Wagner et al., 2001), although it is not clear that this would be a semantic-specific process, as suggested by Poldrack et al.'s analysis (see also Bokde et al. 2001) or a more general process of retrieval operating over different types of information (Gold and Buckner, 2002). If such regional fractionation of processes within the LIFG can be reliably observed, this would lend credence to the view that selection and controlled retrieval are distinct cognitive processes, which are recruited under different conditions.

The Current Study

We conducted an fMRI study, using a competitor priming task (Wheeldon and Monsell, 1994), which allowed us to manipulate semantic selection demands in a situation where retrieval of semantic information was (i) automatic rather than controlled and (ii) did not covary with high versus low selection demands. The participants' task was to provide the basic level name to a line drawing of a familiar, easily nameable object — so exploiting automatic retrieval only. Picture naming is an automatic task in the current context, in that there is a strong association between the stimulus — the pictured object — and the response — its name (Miller, 2002). [Note: We acknowledge that the automatic/controlled distinction is a controversial one, and that these terms are not always defined in the same way in different contexts. In the current study, we are using the term ‘automatic’ to refer to the stimulus-driven activation of the basic level name of a picture of a common object. Although one may decide whether to actually produce that name aloud, or may decide to output a different response to the picture (e.g. to label it as a living or non-living thing), we suggest that the activation of the name itself remains an automatic by-product of object recognition. In contrast, we do not claim that the competitor priming process is necessarily an automatic one. It is often argued in the priming literature that producing interference as well as facilitation of responses may be at least partially dependent on controlled or strategic processes, although a strategy-based account of the competitor priming effect is rejected by Wheeldon and Monsell (1994). Either way, the logic of our study rests on the automatic nature of the object naming task, and not of the competitor priming process.] Most cognitive theories of word production (including the more specific case of picture naming) include a well-defined process of selection, in which the correct abstract word representation (or lemma) must be activated by the semantic specification of the pictured object (e.g. Schriefers et al., 1990; Roelofs, 1992; Levelt, 1993; Wheeldon and Monsell, 1994; Levelt et al., 1999). Accounts differ as to whether lemma selection involves lateral inhibition among coactivated words, or whether a differential threshold must be reached in the absence of direct inhibition, but concur that there is a process of competition among coactivated candidates until one emerges as the ‘winner’ and goes on to activate its lexical form information and drive speech output. Consistent with this claim, imaging studies of picture naming typically report LIFG activation (e.g. van Turennout et al., 2000; Burgund et al., 2003), which may reflect this competition and selection process. Critically for our study, selection demands in picture naming can be manipulated by increasing the degree of competition among lemmas, for example by priming a related but incorrect picture name, using various word–picture interference or competitor priming paradigms.

We introduced three conditions in order to manipulate selection demands in the picture naming task:

Competition Condition

The picture to be named was preceded several trials earlier by a written definition of a highly related word (e.g. participants would produce the word whale in response to the definition the largest mammal that lives in the sea several trials before naming a picture of a shark). Several studies have shown that prior production of semantically related object names increases the degree to which they compete with the correct name for output (Brown, 1979; Vitkovitch and Humphreys, 1991; Vitkovitch et al., 1993, 2001; Wheeldon and Monsell, 1994; Levelt et al., 1999; Tree and Hirsh, 2003). Since the primes and targets are semantically similar (e.g. shark, whale) but not phonologically similar, competition for naming responses can be attributed to selection among coactivated lemma representations activated by similar conceptual representations, rather than form-based selection of the correct phonological specification for output (e.g. Vitkovitch et al., 1993; Wheeldon and Monsell, 1994).

Repetition Condition

The picture was preceded several trials earlier by a written definition of the same word (e.g. a definition of a shark before the picture of a shark). This condition should have the effect of reducing the demands on the selection of the appropriate candidate for output, since the correct name has just been produced, increasing its activation relative to possible competitors. Many earlier studies have demonstrated robust repetition priming (e.g. Durso and Johnson, 1979) and, more specifically that prior production of the same name in response to a written definition reliably facilitates picture naming (Wheeldon and Monsell, 1992; Monsell et al., 1992). [The hypothesized reduction of selection demands among competing semantic alternatives may be only one component of any such facilitation effect, since this condition also involves repeated activation of the same lexical and phonological representations. Repetition effects persist over a much longer time course than do competition priming effects, and so must be supported, at least in part, by different processes (Wheeldon and Monsell, 1994).]

Unrelated Control Condition

The picture to be named is preceded by unrelated definitions only (e.g. the picture of a shark is preceded by the definition, organ for pumping blood).

Our hypothesis is that LIFG activation will vary as a function of the selection demands of the picture naming task, as manipulated by the three priming conditions. Our main prediction is an increase in LIFG activation in the competition condition relative to the unrelated control condition. In addition we expect a reduction in LIFG activation in the repetition condition. It is important to note that our predictions are about increases or decreases in LIFG activation relative to the unrelated control condition, which may itself produce a degree of activation in this region relative to a low level baseline condition. In other words, there is a certain degree of selection required when participants name a picture, even in the absence of an explicit competitor prime; this may reflect competition among alternative names for the same object (e.g. sofa versus couch; bird versus robin) or unintended competition from other previous items in the test list. Previous studies have generally found that LIFG is activated in picture naming tasks relative to a fixation baseline (e.g. van Turennout et al., 2000; Burgund et al., 2003; although see Thompson-Schill et al., 1997).

Materials and Methods

Participants

Fifteen right-handed native English speakers between the ages of 18 and 35 were tested. Each gave informed consent and was paid for their participation.

fMRI Parameters

Scanning was carried out on a 3 T Bruker Medspec Avance S300 system at the Wolfson Brain Imaging Center, Cambridge, UK, using a gradient-echo EPI sequence (TR = 3600 ms; TA = 1100 ms; TE = 27.5 ms; flip angle = 90° FOV = 200 mm × 200 mm), 21 oblique slices, 4 mm thick (1 mm gap between slices, 64 × 64 in-plane resolution) with head coils, 100 kHz bandwidth and spin echo guided reconstruction). T1-weighted scans were acquired for anatomical localization.

In this experiment it was important to be able to record participants' overt naming responses to the pictures so that we could analyse the number of incorrect names generated in each condition and enter the error rate on each picture as a modulator into the analyses. Therefore we used a start–stop event-related imaging paradigm with a condensed TA (1.1 s) and extended TR (3.6 s). This allowed scanning acquisition phases to follow just after naming responses in the period during which the magnet gradient was not operating, so avoiding the head movement artefacts associated with speech production (Huang et al., 2001).

Materials

Participants produced a name in response to an alternating series of written definition sentences and line drawings presented on the screen in front of them. Probe pictures were preceded by definition primes, of one of three types (i) unrelated, (ii) competition, (iii) repetition. Primes and probes were interleaved such that there were always two items between a prime and its corresponding probe.

There were 60 prime–probe pairs in each of the three conditions. The 60 competitor priming pairs were predominantly those used by Wheeldon and Monsell (1994). Each pair consisted of a line drawing of an easily nameable object along with the written definition of a related competitor word. For example, a picture of a shark would be preceded by a definition of a whale (The largest creature that swims in the sea). These pairs were established in Wheeldon and Monsell's study to provide robust competitor priming effects as indexed by significant increases in naming latencies to the pictures when proceeded by the primes over a range of prime-target intervals and numbers of intervening items (from 2 to 30+ intervening items).

In addition to the 60 competitor prime trials, we generated an equal number of repetition and unrelated pairs, matched as closely as possible to the competitor prime pairs. Pictures were taken from the sets of line drawings compiled by Wheeldon and Monsell (1994) and Snodgrass and Vanderwart (1980). For the repetition condition, we generated definition sentences that participants could reliably name with the intended word (for example, the sentence A spiky desert plant was paired with a picture of a cactus). Unrelated definitions were generated that reliably produced a name that had no semantic relation to the picture (for example, the definition Organ for digesting food was paired with a picture of a padlock). Repetition and unrelated definitions were of a similar length to the competition definitions. As shown in Table 1, all pictures and sentences were closely matched across the three priming conditions on word/picture familiarity, word frequency, visual complexity of pictures and naming agreement for definitions. Where not already available from earlier norming studies (Snodgrass and Vanderwart, 1980), familiarity and visual complexity ratings for each picture were obtained from pre-tests in which 14 participants were asked to give ratings on a scale of 1–5. Naming agreement for the definitions was established on the basis of the number of correct names generated in the pilot behavioural experiment.

Table 1

Materials in each of the three priming conditions. Normative data come from the following sources: word frequency (Baayen et al., 1995); picture familiarity (Snodgrass and Vanderwort, 1980, and data collected at Centre for Speech and Language); word familiarity (Coltheart, 1981); visual complexity and definition naming agreement (data collected at Centre for Speech and Language).

 Priming condition
 
  
Variable
 
Competition
 
Repetition
 
Unrelated
 
Probe picture    
Word frequency of name 38.9 41.3 40.7 
Length of name (syllables) 1.55 1.65 1.5 
Picture familiarity 3.24 3.14 3.01 
Picture visual complexity 2.86 2.91 2.90 
Prime sentence    
Word frequency of name 49.6 41.3 44.8 
Word familiarity of name 542 539 523 
Definition naming agreement (%)
 
84
 
87.5
 
86
 
 Priming condition
 
  
Variable
 
Competition
 
Repetition
 
Unrelated
 
Probe picture    
Word frequency of name 38.9 41.3 40.7 
Length of name (syllables) 1.55 1.65 1.5 
Picture familiarity 3.24 3.14 3.01 
Picture visual complexity 2.86 2.91 2.90 
Prime sentence    
Word frequency of name 49.6 41.3 44.8 
Word familiarity of name 542 539 523 
Definition naming agreement (%)
 
84
 
87.5
 
86
 

Procedure

The scanning session for each participant was divided into four blocks, each containing 15 trials in each of the three conditions. Blocks were separated by a period of 8.8 s, during which participants saw a fixation cross on the screen. The conditions were randomly ordered within a block. In addition to the 45 test pictures and 45 test definitions in each block, one filler item was introduced at the beginning and one at the end to complete the interleaving. We also included a fixation point (+) continuously presented in the centre of the screen for the duration of seven scans at the start and end of each block (i.e. 14 fixation scans per block), although this was not modelled. The division of items into blocks was solely for the purposes of varying presentation order to avoid practice and/or fatigue consistently affecting specific items: half the participants saw blocks 1 and 2 followed by 3 and 4 and the other half saw 3 and 4 followed by 1 and 2. Thus the experiment had a fully between-items, within-subjects design. A practice block of 23 items preceded the main experimental sessions.

In order to avoid head movement artefacts associated with speech production, we used a ‘stop–start’ scanning technique in which a 1.1 s volume acquisition phase occurred shortly after participants named the definition or picture. In this way we were able to obtain overt naming responses and thus ensure that participants showed the behavioural effects of competition/repetition priming in terms of increased/reduced naming errors relative to the unrelated condition. Moreover, overt naming allowed us to determine whether participants produced the correct or incorrect name to each picture target, so that we could enter this information as a modulator into our analyses of the imaging data. The onset of each picture was jittered so that it occurred at any time between 0 and 800 s from the offset of the previous acquisition phase. The picture remained on the screen for 250 ms and participants had until the next scanner acquisition period to respond (between 1700 and 2500 ms, depending on the jitter). To encourage participants to respond quickly and so increase the requirement for rapid selection of the correct name, there was also a deadline tone at 600 ms after picture onset. Participants were instructed to try to ‘beat the beep’ and name the picture before they heard the tone. Written definition sentences were presented for 1600 ms, with a deadline tone at 2020 ms. The interval between the onset of a definition prime and its corresponding picture probe was between 11.05 and 11.85 s, depending on the degree of jitter for a given picture. These timings are summarized in Table 2 and illustrated Figure 1. Participants' naming responses were recorded through a sound delivery system onto a DAT recorder and later transcribed, so that error rates on picture trials in each condition could be computed.

Figure 1.

Illustration of timing of stimulus presentation and scan acquisition for a sample of trials. Scanning phases are timed not to overlap with participants' naming responses to avoid head movement artefacts.

Figure 1.

Illustration of timing of stimulus presentation and scan acquisition for a sample of trials. Scanning phases are timed not to overlap with participants' naming responses to avoid head movement artefacts.

Table 2

fMRI study: timing parameters for each picture and sentence


 
Time from onset of trial (ms)
 
Duration/timing of event (ms)
 
Scanning 1100 
Picture onset 1100–1900 (jittered) 250 
Deadline tone 1700–2600 (jittered) 600 after picture onset 
Scanning 3600 1100 
Sentence onset 4450 1600 
Deadline tone 6470 2020 after sentence onset 
Scanning 7200 1100 
Picture onset 8300–9100 250 
Etc.
 

 

 

 
Time from onset of trial (ms)
 
Duration/timing of event (ms)
 
Scanning 1100 
Picture onset 1100–1900 (jittered) 250 
Deadline tone 1700–2600 (jittered) 600 after picture onset 
Scanning 3600 1100 
Sentence onset 4450 1600 
Deadline tone 6470 2020 after sentence onset 
Scanning 7200 1100 
Picture onset 8300–9100 250 
Etc.
 

 

 

fMRI Analysis

Pre-processing (slice timing correction, image realignment into standard stereotactic space and smoothing) and statistical analysis of the data were performed using SPM99 software (Wellcome Institute of Cognitive Neurology, www.fil.ion.ucl.ac.uk), implemented in Matlab (Mathworks Inc. Sherborn, MA). Four time series were acquired for each subject (111 scans per series). The first three scans of each time series were discarded to allow for T1 equilibrium before the test trials started. Slice timing correction was followed by image realignment to account, respectively, for different slice acquisition times and head motion. The images were then spatially normalized to a standard EPI template based on the Montreal Neurological Institute (MNI) reference brain, using 7 × 8 × 7 non-linear basis functions. The spatially normalized images were smoothed with an isotropic 12 mm full-width half-maximal Gaussian kernel.

The data were analysed using the general linear model as implemented in Statistical Parametric Mapping (SPM99b, Wellcome Institute of Cognitive Neurology, www.fil.ion.ucl.ac.uk). The BOLD response for each event was modeled with the canonical haemodynamic response function (HRF). The time-series in each voxel were highpass-filtered to remove low-frequency noise and scaled to a grand mean of 100 over voxels and scans within each session. The LIFG was selected as an a priori region of interest (ROI). The ROI was introduced to the analyses in the form of mask, constructed from the Talairach and Tournoux (1988) atlas of the human brain, by outlining the regions on the atlas and transforming the outlines to a digital image. Contrast images from each participant were combined into a group random effects analysis. The use of a random effects analysis specifically models inter-subject variability to ensure that the group results reflect the response profile in all the individuals entered into the analysis. Error rates on picture trials were entered in the model as a parametric modulator (with linear expansion), since there may be specific patterns of neural activity associated with participants' awareness of producing an incorrect response (Gehring et al., 1993; Garavan et al., 2002). Except where indicated, results were thresholded at P < 0.001 uncorrected and only clusters that survived P < 0.05 corrected for multiple comparisons were considered significant. Since SPM coordinates are given in Montreal Neurological Institute (MNI) space, regions were identified by converting the coordinates to Talairach space with a non-linear transform (Brett et al., 2001).

Pilot Experiment

We carried out a pilot study to establish that our stimuli produced the predicted effects of competitor priming prior to running the experiment in the scanner. We used a lag of two or four intervening items, and prime–target inter-stimulus interval of 10.6 s — similar to the timings required in the fMRI experiment. The pilot study confirmed that there was a significant effect of priming condition [F1(2,16) = 39.7, P = 0.001; F2(2, 177) = 21.5, P < 0.001]. Planned comparisons revealed both an increase in error rate in the competition condition (22.6%) compared to the control condition [11.0%, t1(8) = 5.5, P = 0.001; t2(118) = 3.6, P < 0.001] and a decrease in naming errors in the repetition condition (5.8%) relative to the control condition (t1 = 3.2, P < 0.05; t2 = 2.3, P < 0.05). Since the inception of this study, Vitkovitch et al. (2001) have also reported a competition priming task with two intervening items under speeded naming conditions, which replicates the result of our pilot study.

Results

Behavioural Data

Errors were scored whenever a participant produced an incorrect name (including perseverations from the prime trials — e.g. calling the shark a whale) or was unable to produce a name within the time allowed. Error scores were entered into analyses of variance in which participants (F1) and items (F2) were treated as random variables. There was a significant effect of prime type [F1(2,28) = 59.8, P < 0.001; F2(2,168) = 20.4, P < 0.001]. Pairwise comparisons confirmed that compared to the unrelated condition (8.2% errors), there was a significant increase in errors when pictures were preceded by competitor primes (19.9%; t1 = 7.2, P < 0.001; t2 = 4.0, P < 0.001) and a decrease when pictures were preceded by repetition primes (4.2%; t1 = 2.7, P < 0.05; t2 = 2.34, P < 0.05). Thus, participants in the scanner were experiencing increased selection demands in producing the correct name of the picture in the competition condition. The majority of errors in the competition condition were caused by participants incorrectly producing the name of the semantically related competitor prime in response to the target picture, as can be seen from the error breakdown shown in Table 3. These results are consistent with those obtained in pilot studies outside the scanner.

Table 3

Behavioral data in the scanner: Overall error percentages in each condition and breakdown of errors by type in each priming condition

 Prime condition
 
  
Error type
 
Unrelated
 
Competition
 
Repetition
 
All errors as percentage of all trials 8.2 19.9 4.2 
Error breakdown: Percentage of total error score made up by each error type    
Omission 34 17.6 47.8 
Perseveration of prime name 57.4 n/a 
Perseveration of any other prior name 19.5 7.8 15.9 
Other incorrect name 46.2 17.5 37.8 
Significant hesitation 8.5 5.2 7.9 
Late response (cut off by onset of scanner noise)
 
1.2
 
0.6
 
0
 
 Prime condition
 
  
Error type
 
Unrelated
 
Competition
 
Repetition
 
All errors as percentage of all trials 8.2 19.9 4.2 
Error breakdown: Percentage of total error score made up by each error type    
Omission 34 17.6 47.8 
Perseveration of prime name 57.4 n/a 
Perseveration of any other prior name 19.5 7.8 15.9 
Other incorrect name 46.2 17.5 37.8 
Significant hesitation 8.5 5.2 7.9 
Late response (cut off by onset of scanner noise)
 
1.2
 
0.6
 
0
 

Activations

Our main question was whether we would see selective activation for the competition priming condition, relative to the unrelated control condition. As can be seen in Table 4, there was a small cluster of activation for the competition condition relative to the unrelated control condition, with two peaks in LIFG centring on BA 45 and 46. Second, we asked whether the competitor prime produced activation in the LIFG relative to the repetition condition. This was indeed the case with a larger cluster of significant activation relative to the repetition condition, peaking at a more inferior point in BA 47. The fact that the competitor condition produced a significant increase in LIFG activation relative to the repetition condition, as well as the control condition, suggests that this effect is due to the increased selection demands that are engendered by the competitor prime (but not by the repetition prime), rather than by other processes such as evocation of a memory of a related prior event — which would be common to both competition and repetition conditions compared to the unrelated control. The two regions of activation associated with the competitor priming condition are plotted in Figure 2a,b.

Figure 2.

(a) Plot showing LIFG activation for competition minus repetition contrast. The colour bars indicate strength of activation (beta values). The areas shown survived P < 0.05 correction for multiple comparisons at the cluster level and were thresholded at 0.001. Coordinates presented in MNI space. (b) Plot showing LIFG activation for competition minus control contrast. The colour bars indicate strength of activation (beta values). The areas shown survived P < 0.05 correction for multiple comparisons at the cluster level and were thresholded at 0.001. Coordinates presented in MNI space. (c) LIFG activation for competition minus repetition contrast at a lower threshold (P = 0.01). Coordinates presented in MNI space.

Figure 2.

(a) Plot showing LIFG activation for competition minus repetition contrast. The colour bars indicate strength of activation (beta values). The areas shown survived P < 0.05 correction for multiple comparisons at the cluster level and were thresholded at 0.001. Coordinates presented in MNI space. (b) Plot showing LIFG activation for competition minus control contrast. The colour bars indicate strength of activation (beta values). The areas shown survived P < 0.05 correction for multiple comparisons at the cluster level and were thresholded at 0.001. Coordinates presented in MNI space. (c) LIFG activation for competition minus repetition contrast at a lower threshold (P = 0.01). Coordinates presented in MNI space.

Table 4

Left inferior frontal gyrus activations

 Cluster level
 
 Voxel level
 
 Coordinates
 
  
Regions
 
Pcorrected
 
Extent
 
Pcorrected
 
Z
 
x
 
y
 
z
 
Competition minus repetition        
BA 47 0.014 34 0.101 3.24 −38 22 −8 
Competition minus control        
BA 45 0.039 14 0.085 3.25 −38 28 18 
BA 46   0.103 3.17 −46 34 20 
Additional cluster size/activation peaks for clusters significant at a lower threshold of P = 0.01        
Competition minus repetition        
BA 44 0.056 75 0.293 2.75 −54 24 
BA 44   0.329 2.69 −52 10 32 
BA 44   0.454 2.84 −40 32 
Competition minus control        
BA 46   0.072 3.31 −48 38 16 
BA 47   0.145 3.03 −36 26 −4 
Repetition minus control        
BA47
 
0.037
 
90
 
0.221
 
2.93
 
−30
 
14
 
−20
 
 Cluster level
 
 Voxel level
 
 Coordinates
 
  
Regions
 
Pcorrected
 
Extent
 
Pcorrected
 
Z
 
x
 
y
 
z
 
Competition minus repetition        
BA 47 0.014 34 0.101 3.24 −38 22 −8 
Competition minus control        
BA 45 0.039 14 0.085 3.25 −38 28 18 
BA 46   0.103 3.17 −46 34 20 
Additional cluster size/activation peaks for clusters significant at a lower threshold of P = 0.01        
Competition minus repetition        
BA 44 0.056 75 0.293 2.75 −54 24 
BA 44   0.329 2.69 −52 10 32 
BA 44   0.454 2.84 −40 32 
Competition minus control        
BA 46   0.072 3.31 −48 38 16 
BA 47   0.145 3.03 −36 26 −4 
Repetition minus control        
BA47
 
0.037
 
90
 
0.221
 
2.93
 
−30
 
14
 
−20
 

All clusters containing any voxels significant at P < 0.05 after statistical correction are reported. Cluster extents are presented at an uncorrected threshold of 0.001. Multiple peaks within an extent are shown on subsequent lines.

As a further comparison of competition, repetition and control conditions, we fitted percent signal change against peristimulus time (PST) at the activation peaks for competition versus control and competition versus repetition conditions (see Fig. 3). Also shown are plots of raw parameter estimates (plus standard error to indicate variability in response across individuals) in the three conditions. In both locations, a robust positive response is shown for competition relative to the other conditions. For the competition versus control peak (BA 45), a negative response is found for repetition and control conditions. At the competition versus repetition peak (BA 47), a positive response is shown for all three conditions, but for competition, the response peak is higher and occurs earlier than for repetition and control conditions. The parameter estimate plots are consistent with the random effects results, with a disproportionately larger and non-overlapping response in both locations for competition over control conditions.

Figure 3.

Fitted responses averaged across all subjects showing a simple plot of mean (fitted) response against peristimulus time (PST) at activation peaks for the competition versus control and competition versus repetition contrasts. Note: The fitted response is not corrected for confounds (global flow and high pass filtering). Individual (adjusted) data were not available in all conditions at these peaks. Also shown are plots of raw parameter estimates (plus standard error to indicate variability in response across individuals) in the three conditions.

Figure 3.

Fitted responses averaged across all subjects showing a simple plot of mean (fitted) response against peristimulus time (PST) at activation peaks for the competition versus control and competition versus repetition contrasts. Note: The fitted response is not corrected for confounds (global flow and high pass filtering). Individual (adjusted) data were not available in all conditions at these peaks. Also shown are plots of raw parameter estimates (plus standard error to indicate variability in response across individuals) in the three conditions.

As noted in the Materials and Methods, the error rate associated with each picture was included in the analysis as a parametric modulator, since errors were not evenly distributed over conditions. A direct assessment of the effects of error across all conditions, modelled independently of other task elements, produced significant activation of LIFG (peak: −38 −32 −6), although this was anterior to that found in our competition versus repetition contrast. However, even with the effects of task error included in the model in this way, the competitor condition (relative to repetition and control) was associated with greater activation of LIFG.

In contrast, there were no significant activations associated with the repetition condition, when compared with either the unrelated control or the competition condition. This is not surprising, since the prediction for the repetition conditions was reduced activation in the LIFG relative to the unrelated control due to the correct naming response having already been produced on the prime trial. However, there was no statistical evidence for deactivation in the repetition condition either: the reverse contrasts (repetition minus control or repetition minus competition) revealed no significantly activated voxels or clusters.

In order to investigate the extents of activations further, we lowered the threshold to an uncorrected P = 0.01. At this very low threshold there was an additional cluster of activation in LIFG for the contrast of competition minus repetition (Fig. 2c). This region of superior and posterior LIFG (BA 44) overlaps in part with coordinates presented by Thompson Schill et al. (1997) which were related to high over low selection conditions (see below). Additional activation peaks were also found for the competition minus control condition in the more inferior area (BA47) which had shown up only in the competition–repetition contrast at the higher threshold.

In summary, we found two small but significant regions of activation in the LIFG associated with naming pictures in the competitor priming condition relative to the competition and repetition conditions. It may be that these are peaks within a single extent rather than two distinct and independent activations. The wider activations revealed when the threshold was lowered suggests that this is the case.

Comparisons with Earlier Studies

Our next question was whether our areas of activation are consistent with those associated with selection processes in earlier studies. Thompson-Schill et al. (1997) reported peak activations for the high selection conditions of three tasks, all of which clustered in a dorsal region of the LIFG. We have plotted these peaks in Figure 4a, to provide a comparison with the activations observed in the current study. It is clear that the Thompson-Schill peaks are located in a region considerably superior and posterior to either of our significant clusters. Nevertheless, when we plotted the percent signal change for each of the experimental conditions in our study at each of the peak coordinates identified by Thompson-Schill et al. (1997), we found greater activation in the competition priming condition than the repetition priming condition, consistent with the results of the competition minus repetition contrast at a reduced statistical threshold (Table 4, and Fig. 2c). This indicated that although our activations were centred on a more inferior region than in the earlier study, their extent overlaps to some degree.

Figure 4.

(a) Location of peak LIFG activations reported in Thompson Schill et al. (1997) superimposed on an axial brain slice. The plots show the results of competition minus control and repetition minus control contrasts from the present study at each location. Coordinates presented in MNI space. Approximate slice location indicated by red line on the three-dimensional rendered brain. (b) Plots show the results of competition minus control and repetition minus control contrasts from the present study at earlier reported activation peaks for words with weak versus strong semantic associations (Wagner et al., 2001) and semantic classification of words subsequently remembered (high confidence hits) versus words subsequently forgotten (misses; Wagner et al., 1998). Coordinates of axial brain slices. presented in MNI space. Approximate slice locations indicated by green lines on the three-dimensional rendered brain.

Figure 4.

(a) Location of peak LIFG activations reported in Thompson Schill et al. (1997) superimposed on an axial brain slice. The plots show the results of competition minus control and repetition minus control contrasts from the present study at each location. Coordinates presented in MNI space. Approximate slice location indicated by red line on the three-dimensional rendered brain. (b) Plots show the results of competition minus control and repetition minus control contrasts from the present study at earlier reported activation peaks for words with weak versus strong semantic associations (Wagner et al., 2001) and semantic classification of words subsequently remembered (high confidence hits) versus words subsequently forgotten (misses; Wagner et al., 1998). Coordinates of axial brain slices. presented in MNI space. Approximate slice locations indicated by green lines on the three-dimensional rendered brain.

We then carried out a similar analysis focusing on the coordinates for which activation was greatest in the weak > strong contrast in the Wagner et al. (2001) study. This contrast was intended to reflect increased demands on controlled retrieval while holding selection constant. The peak activation was in anterior LIFG and was suggested to be specifically associated with controlled retrieval of semantic information. We also include percent signal change for the conditions in our study at the peak coordinates presented in an earlier investigation of semantic retrieval found to be associated with the semantic processing of subsequently remembered relative to subsequently forgotten words (Wagner et al., 1998). Figure 4b shows the signal change in our competition and repetition conditions (relative to the unrelated control baseline) for the peaks in both Wagner et al.'s studies. Surprisingly, we find that there was greater activation in our competition than our repetition condition, at the anterior coordinate (−45 +27 −12) — the peak that Wagner et al. (2001) suggested was strongly associated with controlled retrieval of semantic information. Thus, we are finding significant activation for semantic selection without controlled retrieval in a very similar area to that previously suggested to be specifically associated with controlled retrieval without selection. In more superior peaks reported by Wagner et al. (1998) (associated with processing of words later remembered relative to words later forgotten), there was also more activation for competition relative to repetition conditions.

Discussion

In this fMRI study regions within the LIFG were activated as a function of increasing demands on the process of selecting among semantic alternatives. Unlike previous studies, there was no potential confounding of selection with controlled retrieval, since the task exploited the automatic retrieval of a name in response to a picture, and this did not vary across high and low selection conditions. Moreover, the LIFG activations in the competition condition cannot simply be assigned to an increase in some non-specific difference between conditions, such as task difficulty, since the effect was significant even with error rate independently modeled as a parametric modulator. Also, a non-specific difficulty factor is unlikely to be the basis for our results, since a number of studies suggest that tasks engaging semantic processing are associated with LIFG activation, even when the control tasks are subjectively more difficult and are associated with longer reaction times (e.g. Otten et al., 2001). Multiple regions of activation were associated with high selection (BA 47, 45, 46), consistent with selection processes extensively represented throughout LIFG rather than associated with a specific subregion. Indeed, at a lowered statistical threshold, additional recruitment of superior and posterior LIFG (BA 44) was observed during competition relative to repetition conditions.

Before discussing our interpretation of the competition and selection in the LIFG further, we first turn briefly to the issue of why we did not observe a reduction in LIFG activation in the repetition condition compared to the unrelated control condition. We predicted that the increased availability of the name of the target picture, brought about by activating that same name in response to a definition on a preceding prime trial, would make selection of the target lemma more efficient and so reduce LIFG activation. Several studies have found evidence of decreased neural activity — or repetition suppression — in regions including LIFG when stimuli and responses are repeated (Schachter and Buckner, 1998; Koutstaal et al., 2001; see Henson, 2003, for a review). However, the most robust effects are generally found when both stimulus and task are identical across trials (Thompson-Schill et al., 1999). Mixed results have been found when, for example, different exemplars of the same object are shown (Koutstaal et al., 2001; Vuilleumier et al., 2002). In our experiment, repetition trials were not identical, as the prime was a definition and the target a picture. The different modality of stimulus presentation may have reduced repetition suppression [although see Lebreton et al. (2001) for evidence of reduced but significant repetition suppression effects in left prefrontal regions for cross-format (word–picture) repetition presentations]. Moreover, there are mixed results as to whether repetition suppression is found in LIFG for a picture naming task, even when repeated trials are identical. While Chao et al. (2002) and van Turennout et al. (2000) report decreases in LIFG activation, Burgund et al. (2003) report transient increases in LIFG activation as a function of repeated trials in a picture naming task. Thus, although we did expect to see reduced LIFG activation in our repetition condition, it is possible that any such effect was reduced by the non-identical presentation format, or that any suppression in activation as a result of reduced competition in the selection phase was masked by possible priming related increases in activation that emerged from some other aspect of the task.

Returning to the main focus of the study, in line with many previous researchers, we have interpreted the interference effect of semantic primes on picture naming as emerging from competition within the process of selecting among coactivated lemmas in order to produce the target name (Brown, 1979; Schriefers et al., 1990; Vitkovitch and Humphreys, 1991; Roelofs, 1992; Vitkovitch et al., 1993, 2001; Levelt, 1993; Wheeldon and Monsell, 1994; Burke, 1999; Levelt et al., 1999). However, there is an alternative possible interpretation of the competitor priming paradigm that would locate the effect within the initial activation of semantic information on the basis of the picture, and so allow an explanation of both the behavioural effect and the increased LIFG activation in the fMRI study in terms of increased demands on controlled retrieval, rather than selection (Badre and Wagner, 2002). On this account the semantic similarity between the competitor prime and the target picture would lead to competition in the process of activating the correct semantic representation for the target; the increased accessibility of the semantic information associated with the competitor increases the probability that this set of semantic representation ‘wins’ the automatic retrieval stage. If this were the case, it would be plausible that to access the correct semantic and/or lexical representation for the target picture, a top-down control process is required to guide retrieval of this knowledge in the face of competition. On this account, the increased LIFG activation in the competitor condition would be attributable not, as we have claimed, to selection among coactivated semantic representations following automatic retrieval, but rather to the operation of a controlled retrieval process that is initiated in order to overcome competition within the automatic retrieval process itself.

This ‘retrieval’ account of the competitor priming paradigm is a logical possibility and similar accounts have been suggested previously to explain picture–word interference effects (Lupker and Katz, 1981; Rayner and Springer, 1986). However, it is called into question by considerable evidence that the semantic competition effect is reduced or disappears when a task is used that measures the activation of a semantic representation from a picture, without requiring a naming response. In this case, the effect of prior semantically related prime stimuli is generally facilitatory, not inhibitory (Kroll and Potter, 1984; Schriefers et al., 1990; Humphreys et al., 1995; cf. Bajo et al., 2003). For example, in data collected in our laboratory, we found a facilitatory effect of prior presentation of related trials in word–picture matching task. Participants' reaction times to decide whether a picture represented the same concept as a written word were facilitated as a function of the number of preceding trials from the same category (R2 = 0.394 P < 0. 001; Raposo et al., 2003). In fact, even when the target picture does have to be named, facilitatory semantic priming can outweigh competitor effects if the interval between prime and target is very short (Wheeldon and Monsell, 1994), if prime and target are associatively related rather than being semantically similar (Alario et al., 2000; Tree and Hirsh, 2003) and possibly, if the prime is not itself named (Damian and Bowers, 2003; Tree and Hirsh, 2003; although see Bajo et al., 2003). This transient facilitatory effect is thought to stem from spreading activation within the semantic system, which increases the availability of the semantic representation of the target concept to be named (Wheeldon and Monsell, 1994). Thus, we would question Badre and Wagner's assumption that ‘multiple representations [within semantic memory] can compete for processing through mutually inhibitory interactions’ (p. 207). In contrast, the interference or competitor effect of semantically related primes in picture naming tasks like the one used in our study seems to be located in the requirement to select a single lemma to drive the output of the correct name, rather than in any competition at the level of activating the correct semantic information for the picture Further direct evidence that semantic interference is located at the lemma selection level comes from a recent study in which prior presentation of a semantically related prime slowed performance in a task in which Spanish subjects made a grammatical gender decision in response to pictured objects (Bajo et al., 2003). Grammatical gender is claimed to be a syntactic feature linked to the lemma level of representation (Jescheniak and Levelt, 1994), and as such, competition effects in this task can be quite specifically tied to lemma selection rather than semantic activation.

For these reasons, we do not believe that the increased LIFG activation in the competition condition of our fMRI study can be attributed to the need for controlled retrieval processes such as top-down bias mechanisms to overcome competition from the prime at the level of activating representations within semantic memory. We turn now to an examination of potential selection processes, which remain rather imprecisely defined. In the current study we have assumed that selection demands vary as a function of the degree of competition among alternative candidates for a response. Thus, selection demands were argued to be high when one must name a picture of a shark on trials where an incorrect related response, such as whale, has been given an activation boost shortly beforehand. In other words, whale has become a pre-potent response that must be inhibited or ignored. Therefore, within the context of the current task, we have used the terms competition and selection interchangeably. However, the term selection has often been used in the literature to refer to more metalinguistic, effortful processes, in which participants must choose among, for example, different semantic attributes on which to base a response. It remains an open question as to whether these different interpretations of selection are referring to the same underlying process, or should be distinguished, and similarly, whether a distinction should be drawn between processes of competition on the one hand, and selection on the other.

The term competition is widely used in models of language comprehension and production to describe the emergence of a unique candidate as the ‘winner’ from a pool of coactivated representations. In normal circumstances, competition processes are rapid and automatic and not subject to conscious control. For example, when a spoken word starts to be heard, a cohort of candidate word representations is activated — e.g. on hearing, the sequence /ka…/, there is an increase in activation for the matching candidates cat, candle, cab, etc. Over time, these representations continue to accrue or lose activation as a function of their match or mismatch with the unfolding phonological sequence. Candidates compete with each other in proportion to their degree of activation, whether this be in terms of direct lateral inhibition between them, or the requirement for one to reach a sufficiently large advantage over other the others (e.g. Marslen-Wilson, 1990). Similarly, there is parallel activation of, and competition among, lexical representations (at lemma and/or lexeme level) in word production (Stemberger, 1985; Levelt et al., 1999).

In all the cases described above, a single candidate emerges automatically as the winner of the competition process, without involving controlled processing. In the broadest sense of the word, this candidate has been selected. However, we may want to reserve the term selection for the more specific case where it is necessary to carry out some additional processing relevant to current task goals. For example, in natural speech, the competition among cohort candidates is automatic and almost invariably the correct word is recognized either before or shortly after the end of the word, without any noticeable difficulty for words with large cohorts (e.g. cat) versus small cohorts (e.g. ocean). We could think of this as competition without selection. However, if a task were introduced in which the word were truncated after the initial two phonemes and people asked to guess which word they were hearing, as in the ‘gating paradigm’, then there would be a greater selection demand associated with generating a single response among all the possibilities for /ka…/ (Grosjean, 1980; Tyler and Wessels, 1983). In this case, there is competition and selection; task-related selection demands increase in line with the underlying competitor dynamics. This was also the situation in the speeded picture-naming task used in the current study. By demanding a speeded ‘beat the beep’ response, we forced people to select a naming response before the process of competition among possible candidates was fully resolved. Other things being equal, it will be relatively straightforward to select the correct name for production, since it will receive the majority of activation from the object recognition system; if it is clearly a picture of a shark, the lemma shark will activated more strongly than all others. However, in the competitor priming condition, the activation of a highly related lemma has been boosted, so that when the picture is presented, it reaches a level similar to that of the correct word — making it more difficult to reject in favour of the correct response; i.e. increasing selection demands. It is possible that with time, and continued activation from the pictured object, the competition between target and incorrect distracter would resolve such that the correct candidate emerges as the winner, but the task demands of speeded naming force participants to make a choice from within the unresolved competitor space. Within this framework we suggest that there is a task-specific process of selection that operates on the competition among candidates. Selection is defined relative to a particular task or goal. Note that it is possible to construct tasks that operate over the same competitor set (e.g. word candidates) but which do not require selection, at least within the same level of representation. For example, making a lexical decision simply requires a participant to decide whether there is a real word present or not. There is no need to discriminate among competing neighbours to select one of them among others; indeed, the more activation there is in the system as a whole, the more likely it is that there is a word present (Forster and Shen, 1996). Similarly, lexical decision is easier for words with many senses (e.g. twist as in twisted ankle versus twist in a story) rather than few senses (e.g. hotel) since there is no need to select just one sense from all those possible in order to make the decision (Rodd et al., 2002).

Although we suggest that selection in this sense is more than just competition, it is still a relatively rapid, automatic process. The tasks described above, such as picture naming and word identification, do not require that multiple possibilities be held in working memory, nor do they involve effortful, metalinguistic judgements — as do many tasks previously used in the investigation of LIFG function. We cannot determine at this stage whether these two kinds of selection are qualitatively different processes — one automatic and the other controlled — perhaps subserved by distinct subregions within the left prefrontal cortex, or whether they are endpoints of a continuum, in which a single underlying process may be subject to varying degrees of explicit control. The location of the LIFG activation we found to be specifically associated with an increase in automatic selection demands was in a more anterior inferior region of the LIFG than those reported by Thompson-Schill et al. (1997, 1999), which is consistent with the interpretation that we are tapping different kinds of selection process.

What we can conclude on the basis of the current results is that the selection process involved in naming a picture to deadline recruits the LIFG. It is unlikely in this case that LIFG activation can be attributed to confounding controlled retrieval demands with selection demands — since we used an automatic retrieval task, and there was no potential increase in controlled retrieval with the selection demands of the task. We have also argued against the potential reinterpretation of this effect as controlled retrieval operating to resolve inhibitory interactions among semantic representations (Badre and Wagner, 2002). Finally, we reiterate that the scope of our study was an investigation of the effect of increasing selection demands on LIFG activation when retrieval demands were held constant. Thus we do not attempt to make claims about whether or not any or all regions within the left prefrontal cortex are also involved in controlled retrieval (or indeed several other candidate processes). This question is not addressed by the design of the study.

Badre and Wagner (2002) have recently proposed that apparent selection effects should be reinterpreted along with controlled retrieval as reflecting the operation of a single top-down bias mechanism. In support of this claim, they cite a study in which they also found anterior LIFG activity associated with both selection and control demands — suggesting a unitary process rather than separate processes subserved by dissociable subregions within the left prefrontal cortex. This result is echoed in our own study, in which the activation cluster for a selection task is centred on the anterior inferior LIFG region identified by Wagner et al. (2001) to be most strongly associated with controlled retrieval of semantic information. Thus, very similar regions seem to be activated for processes that are apparently very different — both in terms of selection versus retrieval and in terms of their degree of automaticity. While we do not wish to claim that our results refute Badre and Wagner's elegant synthesis of the selection and retrieval approaches into a single mechanism, they nevertheless highlight just how broad a spectrum of apparently different processes must be spanned by any proposed unitary mechanism.

We are very grateful to Linda Wheeldon and Stephen Monsell for supplying us with their stimuli lists and pictures. P.F. is supported by the Wellcome Trust.

References

Alario FX, Segui J, Ferrand L (
2000
) Semantic and associative priming in picture naming.
Q J Exp Psychol
 
53A
:
741
–764.
Baayen RH, Piepenbrock R, Gulikers L (
1995
) The CELEX Lexical Database (CD-ROM). Linguistic Data Consortium, University of Pennsylvania, Philadelphia, PA.
Badre D, Wagner AD (
2002
) Semantic retrieval, mnemonic control, and prefrontal cortex.
Behav Cogn Neurosci Rev
 
1
:
206
–218.
Bajo MT, Puerta-Melguizo MC, Macizo P (
2003
) The locus of semantic interference in picture naming.
Psicologica
 
24
:
31
–55.
Bokde ALW, Tagamets MA, Friedman R, Horwitz, B (
2001
) Functional interactions of the inferior frontal cortex during the processing of words and word-like stimuli.
Neuron
 
30
:
609
–617.
Brett M, Christoff K, Cusak R, Lancaster J (
2001
) Using the Talairach atlas with the MNI template.
Neuroimage
 
13
:
S85
.
Brown AS (
1979
) Priming effects in semantic memory retrieval processes.
J Exp Psychol Hum Learn Mem
 
5
:
65
–77.
Buckner RL (
1996
) Beyond HERA: contributions of specific prefrontal brain areas to long-term memory retrieval.
Psychonom Bull Rev
 
3
:
149
–158.
Burgund ED, Lugar HM, Miezen FM, Petersen SE (
2003
) Sustained and transient activity during an object-naming task: a mixed blocked and event-related fMRI study.
Neuroimage
 
19
:
29
–41.
Burke DM (
1999
) Language production and aging. In: Constraints on language: aging, grammar and memory (Kemper S, Kliegl R, eds), pp. 3–27. Boston, MA: Kluwer.
Chao LL, Weisberg J, Martin A (
2002
) Experience-dependent modulation of category-related cortical activity.
Cereb Cortex
 
12
:
545
–551.
Coltheart M (
1981
) The MRC psycholinguistics database.
Q J Exp Psychol
 
33A
:
497
–505.
Damian MF, Bowers JS (
2003
) Locus of semantic interference in picture-word interference tasks.
Psychonom Bull Rev
 
10
:
111
–117.
Demb JB, Desmond JE, Wagner AD, Vaidya CJ, Glover GH, Gabrieli JDE (
1995
) Semantic encoding and retrieval in the left inferior prefrontal cortex: a functional MRI study of task difficulty and process specificity.
J Neurosci
 
15
:
5870
–5878.
Demonet JF, Chollet F, Ramsay S, Cardebat D, Nespoulous JL, Wise R, Rascol A, Frackowiak R (
1992
) The anatomy of semantic and phonological processing in normal subjects.
Brain
 
115
:
1753
–1768.
Desmond JE, Gabrieli JDE, Glover GH (
1998
) Dissociation of frontal and cerebellar activity in a cognitive task: evidence for a distinction between selection and search.
Neuroimage
 
7
:
368
–376.
Devlin JT, Matthews PM, Rushworth FS (
2003
) Semantic processing in the left inferior prefrontal cortex: a combined functional magnetic resonance imaging and transcranial magnetic stimulation study.
J Cogn Neurosci
 
15
:
71
–84.
Dolan RJ, Fletcher P (
1997
) Dissociating prefrontal and hippocampal function in episodic memory encoding.
Nature
 
388
:
582
–585.
Dronkers N, Redfern B, Knight R (
2000
) The neural architecture of language disorders. In: The new cognitive neurosciences (Gazzaniga MS, ed.), pp. 949–958. Cambridge MA: MIT Press.
Durso FT, Johnson, MK (
1979
) Facilitation in naming and categorising repeated pictures and words.
J Exp Psychol Hum Learn Mem
 
5
:
449
–459.
Fiez JA (
1997
) Phonology, semantics and the role of the left inferior prefrontal cortex.
Hum Brain Mapp
 
5
:
9
–83.
Fletcher, PC, Shallice T, Dolan RJ (
2000
) ‘Sculpting the response space’ — an account of left prefrontal activation at encoding.
Neuroimage
 
12
:
404
–417.
Fletcher PC, Henson RNA (
2001
) Frontal lobes and human memory: insights from neuroimaging.
Brain
 
124
:
849
–881.
Forster KI, Shen D (
1996
) No enemies in the neighborhood: absence of inhibitory neighborhood effects in lexical decision and semantic categorization.
J Exp Psychol Learn Mem Cogn
 
22
:
696
–713.
Gabrieli JDE, Poldrack RA, Desmond JE (
1998
) The role of left prefrontal cortex in language and memory.
Proc Natl Acad Sci USA
 
95
:
906
–913.
Gainotti G, Silveri MC, Daniele A, Giustolisi L (
1995
) Neuroanatomical correlates of category-specific impairments: a critical survey.
Memory
 
3
:
247
–264.
Garavan H, Ross TJ, Murphy K, Roche RAP, Stein EA (
2002
) Dissociable executive functions in the dynamic control of behavior: inhibition, error detection, and correction.
Neuroimage
 
17
:
1820
–1829.
Gehring WJ, Goss B, Coles MGH, Meyer DE, Donchin E (
1993
) A neural system for error detection and compensation.
Psychol Sci
 
4
:
385
–390.
Gold BT, Buckner RL, (
2002
) Common prefrontal regions coactivate with dissociable posterior regions during controlled semantic and phonological tasks.
Neuron
 
35
:
803
–812.
Goodglass H (
1976
) Agrammatism. In: Studies in neurolinguistics (Whitaker H, Whitaker HA, eds), Vol 2, pp. 237–260. New York: Academic Press.
Grodzinsky Y (
2000
) The neurology of syntax: language use without Broca's area.
Behav Brain Sci
 
23
:
47
–117.
Grosjean F (
1980
) Spoken word recognition processes and the gating paradigm.
Percept Psychophys
 
28
:
267
–283.
Henson RNA (
2003
) Neuroimaging studies of priming
progress in Neurobiology
 
70
:
53
–81.
Hodges JR, Patterson K, Oxbury S, Funnell E (
1992
) Semantic dementia: progressive fluent aphasia with temporal lobe atrophy.
Brain
 
115
:
1783
–1806.
Huang J, Carr TH, Cao Y (
2001
) Comparing cortical activations for silent and overt speech using event-related fMRI.
Hum Brain Mapp
 
15
:
39
–53.
Humphreys GW, Lloyd-Jones TJ, Fias W (
1995
) Semantic interference effects in naming using a postcue procedure: tapping the links between semantics and phonology with pictures and words.
J Exp Psychol Learn Mem Cogn
 
21
:
961
–980.
Jescheniak JD, Levelt WJM (
1994
) Word frequency effects in speech production: retrieval of syntactic information and of phonological form.
J Exp Psychol Learn Mem Cogn
 
20
:
824
–843.
Kimberg DG, Farah MJ (
1993
) A unified account of cognitive impairments following frontal lob damage: The role of working memory in complex organized behavior.
J Exp Psychol General
 
122
:
411
–428.
Kohler S, Paus T, Buckner RL, Milner B (
2003
) Effects of left inferior prefrontal stimulation on episodic memory formation: a two-stage fMRI–rTMS study.
J Cogn Neurosci
 
16
:
178
–188.
Koutstaal W, Wanger AD, Rotte M, Mard A, Buckner RL, Schacuter DL (
2001
) Perceptual specificity in visual object priming: Functional magnetic resonance imaging evidence for a laterality difference in fusiform cortex.
Neuropsychologia
 
39
:
184
–199.
Kroll JF, Potter MC (
1984
) Recognising words pictures and concepts: a comparison of lexical object and reality decisions.
J Verb Learn Verb Behav
 
23
:
39
–66.
Lebreton K, Desgranges B, Landeau B, Baron J-C, Eustache F (
2001
) Visual priming within and across symbolic format using a tachistoscopic picture identification task: a PET study.
J Cogn Neurosci
 
13
:
670
–686.
Levelt WJM (
1993
) Accessing words in speech production: stages processes and representations. In: Lexical access in speech production (Levelt WJM, ed.), pp. 1–22. Cambridge MA: Blackwell.
Levelt WJM, Roelofs A, Meyer AS (
1999
) A theory of lexical access in speech production.
Behavorial and Brain Sciences
 
22
:
1
–38.
Lupker SJ, Katz AN (
1981
) Input decision and response factors in picture–word interference.
Mem Cogn
 
7
:
269
–282.
Marslen-Wilson WD (
1990
) Activation competition and frequency in lexical access. In: Cognitive models of speech processing: psycholinguistic and computational perspectives (Altmann GTM, ed.). Cambridge MA: MIT Press.
Martin A, Haxby JV, Lalonde FM, Wiggs CL, Ungerleider LG (
1995
) Discrete cortical regions associated with knowledge of colour and knowledge of action.
Science
 
270
:
102
–105.
Miller EK (
2000
) The prefrontal cortex and cognitive control.
Natl Rev Neurosci
 
1
:
59
–65.
Miller EK, Freedman DJ, Wallis JD (
2002
) The prefrontal cortex: categories concepts and cognition.
Phil Trans Biol Sci
 
357
:
1123
–1136.
Monsell S, Matthews GH, Miller DC (
1992
) Repetition of lexicalisation across languages: A further test of the locus of priming.
Q J Exp Psychol
 
44A
:
763
–783.
Mummery CJ, Patterson K, Hodges JR, Price CJ (
1998
) Functional neuroanatomy of the semantic system: divisible by what?
J Cogn Neurosci.
 
10
:
766
–777.
Norman DA, Shallice T (
1986
) Attention to action: willed and automatic control of behaviour. In: Consciousness and self-regulation (Davidson RJ, Schwartz GE, Shapiro D, eds), pp. 1–18. New York: Plenum.
Otten LJ, Henson RN, Rugg M D (
2001
) Depth of processing effects on neural correlates of memory encoding: relationship between findings from across- and within-task comparisons.
Brain
 
124
:
399
–412.
Petersen SE, Fox PT, Posner MI, Mintun M, Raichle ME (
1989
) Positron emission tomographic studies of the processing of single words.
J Cogn Neurosci
 
1
:
153
–170.
Poldrack RA, Wagner AD, Prull MW, Desmond JE, Glover GH, Gabrieli JDE (
1999
) Functional specialization for semantic and phonological processing in the left inferior frontal cortex.
Neuroimage
 
10
:
15
–35.
Price CJ, Moore CJ, Humphreys GW, Wise RSJ (
1997
) Segregating semantic from phonological processes during reading.
J Cogn Neurosci
 
9
:
727
–733.
Raposo A, Moss HE, Rodd J, Bright P, Stamatakis E, Tyler LK (
2003
) The representation of conceptual knowledge in the brain. Paper presented to the Cognitive Neuroscience Society, New York.
Rayner K, Springer CJ (
1986
) Graphemic and semantic similarity effects in the picture–word interference task.
Br J Psychol
 
77
:
297
–322.
Rodd JM. Gaskell MG, Marslen-Wilson WD (
2002
) Making sense out of ambiguity: semantic competition in lexical access.
J Mem Lang
 
46
:
245
–266.
Roelofs A (
1992
) A spreading activation theory of lemma retrieval in speaking.
Cognition
 
42
:
107
–142.
Schachter DL, Buckner RL (
1993
) Priming and the brain.
Neuron
 
20
:
185
–195.
Schriefers H, Meyer AS, Levelt WJM (
1990
) Exploring the time course of lexical access in language production: picture–word interference studies.
J Mem Lang
 
29
:
86
–102.
Snodgrass JG, Vanderwart M (
1980
) A standardised set of 260 pictures: norms for name agreement image agreement familiarity and visual complexity.
J Exp Psychol Hum Learn Mem
 
6
:
174
–215.
Stemberger JP (
1985
) An interactive activation model of language production. In: Progress in the psychology of language (Ellis A, ed.), pp. 143–183. Hillsdale, NJ: Erlbaum.
Talairach J, Tournoux P (
1988
) A co-planar stereotactic atlas of the human brain. Stuttgart: Thieme Verlag.
Thompson-Schill SL, D'Esposito M, Aguirre GK, Farah MJ (
1997
) Role of left inferior prefrontal cortex in retrieval of semantic knowledge: a re-evaluation.
Proc Natl Acad Sci USA
 
94
:
14792
–14797.
Thompson-Schill SL, Swick D, Farah MJ, D'Esposito M, Kan IP, Knight RT (
1998
) Verb generation in patients with focal frontal lesions: a neuropsychological test of neuroimaging findings.
Proc Natl Acad Sci USA
 
95
:
15855
–15860.
Thompson-Schill SL, D'Esposito M, Kan IP (
1999
) Effects of repetition and competition on activity in left prefrontal cortex in word generation.
Neuron
 
23
:
513
–522.
Tree JJ, Hirsh KW (
2003
) Sometimes faster sometimes slower: associative and competitor priming in picture naming with young and elderly participants.
J Neuroling
 
16
:
489
–514.
Tyler LK, Wessels J (
1983
) Quantifying contextual contributions to word-recognition processes.
Percept Psychophys
 
34
:
409
–420.
Tyler LK, Russell R, Fadili J, Moss HE (
2001
) The neural representation of nouns and verbs: PET studies.
Brain
 
124
:
1619
–1634.
Vandenberghe R, Price CJ, Wise R, Josephs O, Frackowiak RS (
1996
) Functional anatomy of a common semantic system for words and pictures.
Nature
 
585
:
254
–256.
van Turennout M, Ellmore T, Martin A (
2000
) Long-lasting cortical plasticity in the object naming system. Nat Neurosci 3: 1329–1334.
Vitkovitch M, Rutter C, Read A (
2001
) Inhibitory effects during object name retrieval: the effect of interval between prime and target on picture naming responses.
Br J Psychol
 
92
:
483
–506.
Vitkovitch M, Humphreys GW (
1991
) Perseverant responding in speeded naming to pictures: it's in the links.
J Exp Psychol Learn Mem Cogn
 
17
:
664
–680.
Vitkovitch M, Humphreys GW, Lloyd-Jones TJ (
1993
) On naming a giraffe a zebra: Picture naming errors across different object categories.
J Exp Psychol Learn Mem Cogn
 
19
:
243
–259.
Vuilleumier P, Henson RNA, Driver J, Dolan RJ (
2002
) Multiple levels of visual object constancy revealed by event-related EMRI of repetition priming.
Nature Neuro Sci
 
5
:
491
–499.
Wagner AD (
2001
) Cognitive control and episodic memory: contributions from prefrontal cortex. In: Neuropsychology of memory, 3rd edn (Squire LR, Schachter DL, eds). New York: Guildford Press.
Wagner AD, Schacter DL, Rotte M, Koutstaal W, Maril A, Dale AM, Rosen BR, Buckner RL (
1998
) Building memories: remembering and forgetting of verbal experiences as predicted by brain activity.
Science
 
281
:
1188
–1191.
Wagner AD, Pare-Blgoev EJ, Clark J, Poldrack RA (
2001
) Recovering meaning: left prefrontal cortex guides controlled retrieval.
Neuron
 
31
:
329
–338.
Wheeldon L, Monsell S (
1992
) The locus of repetition priming of spoken word production.
Q J Exp Psychol
 
44A
:
723
–761.
Wheeldon L, Monsell S (
1994
) Inhibition of spoken word production by priming a semantic competitor.
J Mem Lang
 
3
:
332
–356.
Wise R, Chollet F, Hadar U, Friston K, Hoffner E, Frackowiak R (
1991
) Distribution of cortical neural networks involved in word comprehension and word retrieval.
Brain
 
114
:
1803
–1817.
Zurif EB (
1995
) Brain regions of relevance to syntactic processing. In: An invitation to cognitive science (Gleitman L, Liberman M, eds), Vol. 1. Cambridge, MA: MIT Press.

Author notes

1Department of Experimental Psychology, University of Cambridge, UK and 2Department of Psychiatry, University of Cambridge, UK