Abstract

Selection between competing responses and stimulus-response association strength is thought to affect performance during verb generation. However, the specific contribution of these two processes remains unclear. Here we used fMRI to investigate the role of selection and association within frontal and BG circuits that are known to be involved in verb production. Subjects were asked to generate verbs from nouns in conditions requiring either high or low selection, but with constant association strength, and in conditions of weak or strong association strength, now with constant selection demands. Furthermore, we examined the role of selection and association during noun generation from noun stimuli. We found that the midpart of the left inferior frontal gyrus was more active in conditions requiring high compared with low selection, with matched association strength. The same left inferior frontal region activated irrespective of verb or noun generation. Results of ROI analyses showed effects of association strength only for verb generation and specifically in the anterior/ventral part of the left inferior frontal gyrus. Moreover, the BG were more active when weakly associated verbs had to be produced relative to weakly associated nouns. These results highlight a functional segregation within the left inferior frontal gyrus for verb generation. More generally, the findings suggest that both factors of selection between competing responses and association strength are important during single-word production with the latter factor becoming particularly critical when task-irrelevant stimuli interfere with the current task (here nouns during verb production), triggering additional activation of the BG.

INTRODUCTION

Cognitive control mechanisms are essential for the flexible, adaptive behavior that is required to meet changing environmental demands such as when a large pool of competing alternatives is afforded by a situation. In the domain of language production, substantial evidence indicates that the left inferior frontal region is critical for the performance of tasks that demand access and selection of a response from among other possible options. Thus, word generation tasks have often been used to investigate mechanisms of semantic retrieval and competition in this brain area (Martin & Cheng, 2006; Thompson-Schill, Kan, & Oliver, 2006; Persson et al., 2004; Thompson-Schill et al., 1998; Thompson-Schill, D'Esposito, Aguirre, & Farah, 1997; Petersen, Fox, Snyder, & Raichle, 1990; Petersen, Fox, Posner, Mintun, & Raichle, 1988). However, the cognitive control mechanisms involved in these kinds of tasks and the anatomical and functional organization of processing in the left inferior frontal regions are strongly debated. For instance, the speed and the accuracy of word generation may depend both on the number of possible responses (response selection) and on the strength of association between cues and responses (associative strength) (Martin & Cheng, 2006; Thompson-Schill & Botvinick, 2006). From a related perspective, previous behavioral and neuroimaging studies have explored the mechanisms and neural substrates involved in verb generation tasks and have emphasized either the role of response selection (Persson et al., 2004; Thompson-Schill et al., 1997, 1998) or the role of associative strength (Martin & Cheng, 2006).

According to the competitive view, the critical process for verb production is the selection among potential responses. Using fMRI, Thompson-Schill et al. (1997) investigated the production of verbs from noun cues in conditions of either low or high selection (i.e., items with few associated responses or with a clear dominant response vs. items with many appropriate associated responses without any clearly dominant response, respectively). They found slower RTs and increased activation of the left inferior frontal gyrus (LIFG) in the high-selection condition, indicating that LIFG is involved in selection from among alternative responses (for similar conclusions in the context of both fMRI and neuropsychological studies, see also Moss et al., 2005; Robinson, Shallice, & Cipolotti, 2005; Persson et al., 2004; Zhang, Feng, Fox, Gao, & Tan, 2004; Barch, Braver, Sabb, & Noll, 2000; Robinson, Blair, & Cipolotti, 1998; Thompson-Schill et al., 1998).

Other researchers have suggested that the critical factor for explaining verb production performance is the strength of association between the noun cue and the possible verb responses (Martin & Byrne, 2006; Martin & Cheng, 2006). According to this view, strong associative strength can drive automatic retrieval, irrespective of the number of verb competitors and without any behavioral cost (Badre & Wagner, 2002). By contrast, weakly associated responses would necessitate controlled retrieval, leading to slower performance. Martin and Cheng (2006) investigated response selection and association strength during verb production in a patient with LIFG damage and in two samples of young and older controls. To dissociate these two processes, they compared three critical conditions: (1) high selection and strong association (HS-SA), (2) high selection and weak association (HS-WA), and (3) low selection and strong association (LS-SA). Participants' responses were affected by the strength of association (longer RTs for weak than strong association conditions) and the patient was only impaired in the condition involving weak associations. The level of selection (high vs. low) did not influence performance either in the patient or in the two control groups. Accordingly, Martin and Cheng concluded that the LIFG is important for controlled semantic retrieval (for related positions, see Danker, Gunn, & Anderson, 2008; Wagner, Paré-Blagoev, Clark, & Poldrack, 2001).

Recently Badre, Poldrack, Paré-Blagoev, Insler, and Wagner (2005) attempted to reconcile the selection and the association strength accounts, providing fMRI evidence of some functional segregation within the LIFG. Subjects were presented with a noun cue, and they had to choose one stimulus from two to four possible targets. In different conditions, the association strength between the target and the cue was either strong or weak, or the selection requirement was high or low (e.g., two vs. four possible targets). The results revealed that the left mid-ventrolateral prefrontal cortex (VLPFC, inferior frontal gyrus pars triangularis and pars opercularis, BA 44/45) was activated in the high-selection versus low-selection conditions, whereas the more anterior section (including the inferior frontal gyrus pars orbitalis, BA 47) was activated during controlled retrieval (weak vs. strong association conditions). Nonetheless, it should be noted that none of the conditions presented in this study required the subjects to generate target words; thus, the results may not apply fully to word generation tasks.

Recent neuropsychological and neuroimaging studies have suggested that not only the frontal cortex but also the BG structures are important for the selection and retrieval of competing response alternatives (Castner et al., 2007, 2008; Longworth, Keenan, Barker, Marslen-Wilson, & Tyler, 2005; Crosson et al., 2003; Redgrave, Prescott, & Gurney, 1999). In a recent study on single-word generation, Crescentini, Mondolo, Biasutti, and Shallice (2008) administered the paradigm developed by Martin and Cheng (2006) to a group of Parkinson's disease (PD) patients and elderly normal controls. The results showed that when patients generated verbs from noun cues, both high-response selection demands and weak association strength decreased performance, with the latter having the most pronounced influence (consistent with the findings of Martin & Cheng, 2006, but now in patients with BG pathology). In addition, this study examined also the generation of nouns from noun cues. This revealed that weak association strength was the key factor determining slower RTs in both patients and elderly controls. Direct comparisons between verb and noun generation tasks revealed, in both subject groups, poorer performance with the former than the latter task, particularly on the weak association condition. In more detail, PD patients were particularly impaired in producing weakly associated verbs because of the many nouns that were produced instead (intrusion errors). This suggested that nonverb responses (nouns) interfere particularly with the production of weakly related verbs. This interpretation was in line with the proposal of Thompson-Schill and Botvinick (2006) who also claimed that nonverb competitors play an important role in verb generation, particularly in situations of weak association. Accordingly, these authors have claimed that response override is required during verb generation, namely, in some circumstances, a strong nonverb response needs to be overridden in favor of a weaker but task-relevant response (see Levy & Anderson, 2002; Botvinick, Braver, Barch, Carter, & Cohen, 2001).

The aim of the current study was to dissociate response selection and association strength during single-word production and to investigate the role of these two factors in the activation of frontal–striatal circuits (cf. Crescentini et al., 2008; Martin & Cheng, 2006). We used both a verb and a noun generation task to investigate possible commonalities and differences for these two types of stimuli. Furthermore, we recorded overt vocal responses during fMRI, which enabled us to monitor word generation performance on each trial.

We investigated the effect of selection by comparing high-selection versus low-selection conditions, with matched association strength (i.e., items without any clearly dominant response versus items with a clear dominant response; HS-SA > LS-SA), and the effect of association strength by comparing weak versus strong conditions, with equalized selection demands (i.e., items with many weakly associated responses versus items with many strongly associated responses; HS-WA > HS-SA). It should be noted that selection and association cannot be manipulated in fully factorial manner because there are effectively no cue words to act as triggers in a “low selection and weak association” condition (cf. also Martin & Cheng, 2006). Therefore, our analyses can only compare relevant conditions pairwise (e.g., high vs. low selection) but critically avoiding any confounding effect of the other factor (i.e., the level of association strength in this example).

We expected that high-selection demands would activate the mid-LIFG during verb generation (Badre et al., 2005; Persson et al., 2004; Barch et al., 2000; Thompson-Schill et al., 1997, 1998), and we assessed here for the first time the role of selection during noun generation. As far as the effect of association strength is concerned, we hypothesized that there could be a possible involvement of the more anterior/ventral section of the left VLPFC in the weak association condition following Badre et al. (2005), but now using active word generation tasks rather than merely target discrimination. Finally, following our previous results in PD patients (Crescentini et al., 2008), we predict increased activation in BG for the weak association condition, specifically during verb generation. This would support the hypothesis that the BG play a critical role in situations when noun competitors interfere with the production of weakly associated verbs.

METHODS

Participants

Fourteen right-handed healthy volunteers (6 men, 30.5 ± 4.5 years) participated in the study. All participants had no existing neurological or psychiatric illness. All subjects gave written informed consent, and the study was approved by the independent ethics committee of the Fondazione Santa Lucia (Scientific Institute for Research, Hospitalisation and Health Care).

Design

The design of the current fMRI experiment was based on the previous behavioral study of Martin and Cheng (2006). Accordingly, we tested for the effect of response selection comparing two conditions that differed for selection demands but were equated for association strength (LS-SA and HS-SA), and we tested for the effect of association comparing two conditions with different levels of association strength but with the same selection demands (HS-WA and HS-SA). In addition, our design included both verb and noun generation tasks, allowing us to investigate the effects of selection and association in two different types of generation tasks. A baseline “read” condition was also included. This permitted us to investigate brain activation common to the verb and the noun tasks (irrespective of selection and association conditions) but without any trivial confounds related to common low-level sensory-motor processes (e.g., processing of the sensory cue, overt vocalization, etc.).

Stimuli

The same stimuli were used as by Crescentini et al. (2008). There were three experimental conditions varying in selection demands and mean association strength. To choose stimuli for the verb production task, we selected 330 (di- and trisyllabic) nouns from the Veli Dictionary of Frequency for Italian Spoken Language. These stimuli were administered to 46 Italian subjects (range = 20–57 years old) who were asked to provide a related verb for each noun. In a similar way, for noun generation, we selected 250 (di- and trisyllabic) nouns from the Veli Dictionary of Frequency for Italian Spoken Language, and we administered each of these to a sample of 29 Italian subjects (range = 21–55 years old). They were required to provide another associated noun for each noun stimulus. Measures of selection demand and ratios of association strength were calculated for each stimulus. More specifically, as a measure of selection demand for each stimulus, we used the ratio of the response frequencies of the two most common responses to that stimulus (following Thompson-Schill et al., 1998). Moreover, the response frequency of a verb in the verb generation task or of a noun in the noun generation task was used as an index of its association strength to the stimulus noun (Crescentini et al., 2008; Martin & Cheng, 2006). The measure of stimulus-response association strength for each stimulus was given by the ratio of the mean association of the first two most common responses to the number of subjects who judged noun–verb or noun–noun stimuli (Crescentini et al., 2008). According to these data, we selected 32 stimuli for each of the three experimental conditions of the two tasks. Within each task, two experimental conditions were matched for the average stimulus-response associative strength but differed in selection demands, for instance, in verb generation, LS-SA—for example, the noun lattina (can) → bere (to drink) in 25/46 subjects who judged noun–verb stimuli and stappare (to broach) in 4/46 subjects; HS-SA—for example, the noun lamp elicited in 21 subjects accendere (to turn on) and in 17 subjects illuminare (to light on). The third experimental condition consisted of stimuli with high-selection demands and weak stimulus-response associative strength; HS-WA—e.g., the noun spada (sword) elicited in 8 subjects combattere (to fight) and in 8 subjects ferire (to wound). In both tasks, the HS-WA condition was matched to the HS-SA condition for selection demands, but the two conditions differed in the average associative strength.

The mean selection demands of the stimulus nouns in the four high-selection conditions of the two tasks varied between 1.3 and 1.6 with no significant differences between conditions. The conditions with low-selection demands did not differ for this measure (nouns = 11, verbs = 14). The two tasks were matched for the ratios of association strength as well; thus, the two HS-WA conditions had mean ratios of association strength of 0.13 and of 0.14 for verb and noun generation, respectively. In a similar way, the conditions with strong associations were matched across tasks (the mean ratios of association strength varied between 0.36 and 0.39 in these four conditions).

Moreover, we also presented all noun stimuli to 45 normal subjects (range = 24–55). We assessed free association asking subjects to report the first word (e.g., noun, verb, adjective) that came to mind for each noun stimulus. This analysis showed that 80% of the responses to the set of stimuli were nouns whereas only 6% were verbs (14% were adjectives). In other words, verb generation was associated with less task-relevant responses than noun generation in free association (see Crescentini et al., 2008). The stimuli in the three conditions of the verb generation task were matched for the strength of the first nonverb response given during free association. Moreover, in both tasks, the two conditions with strong associations (LS-SA and HS-SA) did not differ in the number of task-irrelevant responses that were produced in situations of free association (e.g., the number of nouns produced in the verb generation task), but this value was significantly lower than that in the conditions with weak association (HS-WA).

Procedures

Figure 1 reports a schematic representation of the experimental procedure.

Figure 1. 

Experimental design. The timing of the events in a trial is also reported. Read indicates the baseline condition in which noun stimuli have to be read. LS-SA = low selection and strong association; HS-WA = high selection and weak association; HS-SA = high selection and strong association. See main text for further details.

Figure 1. 

Experimental design. The timing of the events in a trial is also reported. Read indicates the baseline condition in which noun stimuli have to be read. LS-SA = low selection and strong association; HS-WA = high selection and weak association; HS-SA = high selection and strong association. See main text for further details.

Before fMRI scanning, the participants practiced the tasks during a 5-min training session. During practice, they were shown six blocks of stimuli, each consisting of four noun stimuli, each of which appeared on the computer screen at 4-sec intervals. There were two “read” blocks in which the stimuli had to be read by the subjects and two blocks for each of the two experimental tasks (noun and verb generation). The order of the blocks for the practice session was fixed: read–noun generation–verb generation, repeated twice. During practice, subjects were instructed on how to produce spoken responses while moving the face muscles as little as possible.

In the scanning session, the participants were asked to generate a verb (or a noun) related to a visually presented noun or to read the noun stimulus. For each noun, the participants responded aloud. The nouns were presented for 1 sec each; vocal responses were recorded during 3.5 sec starting from the presentation of the stimulus (intertrial interval = 4–5 sec). After the stimulus disappeared from the screen, a fixation cross appeared.

Noun stimuli were grouped in blocks of four. Thus, each block lasted approximately 20 sec. We blocked four trials of the same condition to minimize task-switching requirements and to maximize statistical power for the detection of differential effects between the different conditions (Friston, Zarahn, Josephs, Henson, & Dale, 1999). Moreover, task switches could well activate closely related areas to ones being investigated here (see meta-analyses of Derrfuss, Brass, Neumann, & von Cramon, 2005).

At the beginning of each block, an instruction was presented for 1500 msec, informing the subjects about the upcoming task. A group of eight blocks formed one presentation of an experimental task. Blocks within a group alternated according to a balanced Latin square design. In each group, six of the eight blocks were formed by stimuli of the three experimental conditions (two blocks per condition), whereas the remaining two blocks consisted of “read” stimuli (see Figure 1). In total, each group of eight blocks consisted of 32 stimuli.

The study was divided into two fMRI runs, each of these consisted of four groups of eight blocks (two groups per experimental task, see Figure 1). The order of repetition of the two tasks was counterbalanced across subjects; for half of the participants, the order was noun–verb–verb–noun for the first run and verb–noun–noun–verb for the second run. For the second half of the subjects, it was verb–noun–noun–verb for the first run and noun–verb–verb–noun for the second run. For the generate tasks (noun or verb), participants were not informed of the different conditions (LS-SA, HS-WA, and HS-SA). Before the beginning of the fifth block and at the end of the last block of each group of eight blocks, there were rest periods of 20 sec during which a fixation cross appeared on the screen. The total time of the fMRI session was approximately of 27 min (two fMRI runs of 13.5 min each).

Verbal Response Recording

To acquire overt verbal responses, we used a method similar to that of Barch et al. (2000). A plastic tube was fixed to the head coil through a plastic funnel and taped, outside the scanner room, to a microphone that was in turn attached to a computer. The plastic tube was used to allow placement of the microphone outside the bore of the scanner.

Each subject's response was recorded as a wav file. The noise reduction facility of Cool Edit Pro. 2.00 (Syntrillium Software Corporation, Phoenix, AZ) was used to isolate a subject's response from the scanner noise. The responses were transcribed and checked for accuracy by the first author. RTs were calculated from the onset of the stimulus to the onset of the subject's response. For all 14 subjects, almost all responses that were produced were recorded clearly. Most of the errors made by the subjects were missed responses; in a few cases, the subject did not pay attention to the instructions and generated nouns when verbs were required or generated verbs when noun generation was the task at hand. Nevertheless, subjects' behavior indicated that they generally carried out the tasks appropriately. Both behavioral and imaging data are referred to subjects' correct responses only.

fMRI Methods

Images were acquired using a 3-T MRI scanner (Siemens Medical Systems, Erlangen, Germany) equipped with a standard quadrature head coil and for EPI. Head movement was minimized by mild restraint and cushioning. Thirty-two slices of functional MR images were acquired using blood oxygenation level-dependent imaging (3 × 3 mm, 2.5 mm thick, repetition time = 2.08 sec, time echo = 30 msec), covering the entirety of the cortex. At the end of the scanning session, anatomical scans were also acquired for each subject, using a T1-weighted magnetization-prepared, rapid acquisition gradient-echo.

Experimental tasks were presented using Cogent 2000 (developed by the Cogent 2000 team at the FIL and the ICN, London). SPM5 (Wellcome Department of Cognitive Neurology; see Friston, 2004) was used for data preprocessing and statistical analyses. For all participants, we acquired 792 volumes (396 each fMRI run); the first 4 of these were discarded for each run. All images were then corrected for head movement. Slice-acquisition delays were corrected using the middle slice as reference. All images were normalized to the standard SPM5 EPI template and spatially smoothed using an 8-mm FWHM Gaussian filter. All images were high-pass filtered at the cutoff value of 128 sec.

All subsequent analyses of the functional images were performed using the general linear model implemented in SPM5. First, for each subject, the data were fitted at every voxel using a combination of effects of interest. The onset of each trial of the eight conditions (read, LS-SA, HS-WA, and HS-SA for noun generation and read, LS-SA, HS-WA, and HS-SA for verb generation) was convolved with the hemodynamic response function. Because the intertrial interval was relatively long and it is unlikely that activity remained sustained for an entire block, each trial with a correct response was modeled as a separate event (duration = 0). Error trials were modeled with a separate regressor and excluded from subsequent group-level analyses. Indeed, as covariates of no interest, both the parameters of the realignment (motion correction) and the onset of each error trial were also included in the design matrix. We then obtained six contrast images per subject subtracting the read condition form each of the six conditions of interest (LS-SA, HS-WA, and HS-SA noun, and verb generation). These six-contrast images per subject underwent an ANOVA for group-level random effects statistical inference. Correction for nonsphericity (Friston et al., 2002) was used to reduce for any nonindependent error terms in the repeated measures analysis. All reported activations survived a whole-brain pcorrected = .05 (cluster level, estimated at puncorrected threshold <.001). In addition, a one-sample t test was used to compare pairwise: generation versus fixation (pcorrected = .05, cluster level).

RESULTS

Behavioral Data

Reaction Time Data

The results of noun and verb generation tasks are shown in Figure 2. RTs data did not violate assumptions of normality (all Shapiro–Wilk p values >.2). A 2 × 3 repeated measures ANOVA showed the main effect of the task, F(1,13) = 7.12, p < .02, the main effect of condition, F(2,26) = 104.38, p < .001, and a significant task × condition interaction, F(2,26) = 34.85, p < .001. The main effect of task arose from verb generation, which was globally performed faster than noun generation (verb, M = 1585; noun, M = 1676). In view of the significant interaction, two repeated measure ANOVAs were carried out, one for each task. These analyses gave a significant main effect of condition both for noun generation, F(2,26) = 8.17, p < .01, and for verb generation, F(2,26) = 128.42, p < .001. Post hoc pairwise contrasts executed for the noun generation task showed significant differences (using Bonferroni corrections) between LS-SA versus HS-WA, F(1,13) = 10.00, puncorrected < .008 (LS-SA, M = 1639; HS-WA, M = 1746) and HS-SA versus HS-WA, F(1,13) = 11.94, puncorrected < .005 (HS-SA, M = 1642; HS-WA, M = 1746) but no difference between LS-SA versus HS-SA, F(1,13) = 0.01, p = .91 (LS-SA, M = 1639; HS-SA, M = 1642). The data show an effect of association strength but not of selection demands for RT in noun generation. Post hoc pairwise contrasts executed for the verb generation task showed a similar pattern of results: significant differences (using Bonferroni corrections) for comparisons between LS-SA versus HS-WA, F(1,13) = 121.54, puncorrected < .001 (LS-SA, M = 1415; HS-WA, M = 1862) and HS-SA versus HS-WA, F(1,13) = 196.79, puncorrected < .001 (HS-SA, M = 1479; HS-WA, M = 1862), but unlike noun generation, they also showed a difference between LS-SA versus HS-SA, F(1,13) = 11.95, puncorrected < .005 (LS-SA, M = 1415; HS-SA, M = 1479). Effects of association strength and of selection demands are present in the verb generation task.

Figure 2. 

Behavioral results. (Top) Response time (msec) performance in the noun (left) and verb (right) generation task. (Bottom) Accuracy performance (%) in both tasks (noun generation on the left). LS-SA = low selection and strong association; HS-WA = high selection and weak association; HS-SA = high selection and strong association. Bars indicate SD.

Figure 2. 

Behavioral results. (Top) Response time (msec) performance in the noun (left) and verb (right) generation task. (Bottom) Accuracy performance (%) in both tasks (noun generation on the left). LS-SA = low selection and strong association; HS-WA = high selection and weak association; HS-SA = high selection and strong association. Bars indicate SD.

In view of the interaction, we also performed post hoc pairwise contrasts (Bonferroni correction applied) between pairs of corresponding conditions of the two tasks. These contrasts showed that verbs were produced faster than nouns in the conditions with strong association: verb LS-SA versus noun LS-SA, F(1,13) = 21.47, puncorrected < .001 (verb LS-SA, M = 1415; noun LS-SA, M = 1639) and verb HS-SA versus noun HS-SA, F(1,13) = 34.02, puncorrected < .001 (verb HS-SA, M = 1479; noun HS-SA, M = 1642). However, verb generation was marginally slower than noun generation in the critical condition with weak association: verb HS-WA versus noun HS-WA, F(1,13) = 6.45, p < .25 (verb HS-WA, M = 1862; noun HS-WA, M = 1746).

Finally, two t tests showed that the read condition was performed faster than both the two generation tasks: read versus verb, t = −16.55 (13), p < .001 (read, M = 794 msec; verb, M = 1585 msec) and read versus noun, t = −25.87 (13), p < .001 (read, M = 794 msec; noun, M = 1676 msec).

Accuracy Data

Using Shapiro–Wilk tests of normality, four out of seven of the accuracy distributions (six experimental conditions plus the read condition) were not normal. An arcsine transformation was performed on the data. Following the transformation, only the accuracy distribution of the read condition was well below the threshold for normality.

A 2 × 3 repeated measures ANOVA on the proportion of correct responses showed a significant effect of condition, F(2,26) = 32.44, p < .001, and a significant Task × Condition interaction, F(2,26) = 6.20, p < .01, but the main effect of the task was not significant, F(1,13) = 0.16, p = .68.

In view of the significant interaction, two repeated measure ANOVAs were carried out, one for each task. These analyses gave a significant main effect of condition both for noun generation, F(1,13) = 14.46, p < .001, and for verb generation, F(1,13) = 29.80, p < .001. Post hoc pairwise contrasts executed for the noun generation task showed significant differences (using Bonferroni corrections) between LS-SA versus HS-WA, F(1,13) = 19.64, puncorrected < .001 (LS-SA, M = 0.94; HS-WA, M = 0.87) and HS-SA versus HS-WA, F(1,13) = 20.51, puncorrected < .001 (HS-SA, M = 0.95; HS-WA, M = 0.87) but no difference between LS-SA versus HS-SA, F(1,13) = 0.57, p = .46 (LS-SA, M = 0.94; HS-SA, M = 0.95). The data show an effect of association strength but not of selection demands for accuracy in noun generation. Post hoc pairwise contrasts executed for the verb generation task showed an analogous pattern of results: significant differences for comparisons between LS-SA versus HS-WA, F(1,13) = 36.18, puncorrected < .001 (LS-SA, M = 0.97; HS-WA, M = 0.82) and HS-SA versus HS-WA, F(1,13) = 31.61, puncorrected < .001 (HS-SA, M = 0.96; HS-WA, M = 0.82) but no difference between LS-SA versus HS-SA, F(1,13) = 1.60, p = .23 (LS-SA, M = 0.97; HS-SA, M = 0.96). An effect of association strength but not of selection demands was also present in the verb generation task.

In view of the interaction, we also performed post hoc pairwise contrasts (Bonferroni correction applied) between pairs of corresponding conditions of the two tasks. These contrasts showed a trend of verb generation performed better than noun generation for the LS-SA condition, verb LS-SA versus noun LS-SA, F(1,13) = 5.28, pcorrected = .039 (verb LS-SA, M = 0.97; noun LS-SA, M = 0.94) and no difference between tasks for both the HS-SA condition, verb HS-SA versus noun HS-SA, F(1,13) = 0.60, p = .45 (verb HS-SA, M = 0.96; noun HS-SA, M = 0.95) and the HS-WA condition, verb HS-WA versus noun HS-WA, F(1,13) = 2.11, p = .17 (verb HS-WA, M = 0.82; noun HS-WA, M = 0.87). Finally, Wilcoxon tests showed that the read condition was performed better than both the two generation tasks: read versus verb, Z = −3.18, p < .01 (read, M = 0.98; verb, M = 0.91); and read versus noun, Z = −3.19, p < .01 (read, M = 0.98; noun, M = 0.91).

The behavioral data show that the weak association conditions lead to longer RTs and are associated with poorer performance relative to the conditions with strong associations (see also Table 1). In addition, a selection demands effect is found in the response time data of the verb production task. Finally, verb generation is slower than noun production in condition of weak associations but faster in situations of strong association.

Table 1. 

Generation versus Fixation, Generation versus Read, and Verb versus Noun

Anatomical Localization
MNI Coordinates
pcorrected
Z
Voxels per Cluster
x
y
z
Regions More Active for Generation versus Fixation 
L middle occipital gyrus −28 −92 18 <.001 5.82 14,729 
R middle occipital gyrus 34 −94 <.001 4.94  
L cuneus −10 −80 <.001 5.53  
R cuneus 10 −80 <.001 4.75  
L cerebellum −40 −50 −24 <.001 5.27  
R cerebellum 36 −62 −32 <.001 5.04  
L fusiform gyrus −20 −92 −12 <.001 5.21  
R fusiform gyrus 32 −82 −16 <.001 4.54  
L lingual gyrus −12 −68 −4 <.001 4.58  
R lingual gyrus 18 −96 −6 <.001 4.85  
L inferior occipital gyrus −42 −84 −4 <.001 4.39  
L middle frontal gyrus −44 40 <.001 6.26 9516 
L inferior frontal gyrus −42 30 <.001 5.53  
L precentral gyrus −46 56 <.001 5.44  
L putamen −20 10 −2 <.001 5.24  
L insula −34 20 <.001 5.24  
L amygdala −22 −12 −10 <.001 4.97  
L parahippocampal gyrus −16 −14 −14 <.001 4.67  
L superior temporal gyrus −56 12 −8 <.001 4.50  
L thalamus −10 −14 <.001 3.73  
R thalamus 12 −16 <.001 3.39  
R substantia nigra −20 −14 <.001 4.44  
L superior frontal gyrus −6 16 54 <.001 6.20 2075 
L ACC −10 24 40 <.001 4.25  
R ACC 10 20 44 <.001 4.03  
R precentral gyrus 48 −10 34 <.001 5.23 1713 
R putamen 24 10 <.001 4.42 960 
R globus pallidus 20 −8 −4 <.001 4.16  
L superior parietal lobe −26 −60 54 <.001 5.53 623 
L precuneus −20 −64 50 <.001 4.63  
R superior temporal gyrus 52 −28 <.04 4.04 180 
 
Regions More Active for Generation versus Read 
L inferior frontal gyrus −40 30 <.001 >8 10,547 
L middle frontal gyrus −6 18 52 <.001 >8  
L ACC −8 24 40 <.001 7.3  
L superior frontal gyrus −20 30 58 <.001 3.29  
R lingual gyrus 10 −82 −30 <.001 >8 9534 
L lingual gyrus −12 −70 −10 <.001 5.83  
R fusiform gyrus 38 −64 −32 <.001 7.82  
L fusiform gyrus −38 −66 −28 <.001 5.23  
L cuneus −4 −100 14 <.001 4.98  
R cuneus 16 −98 <.001 4.33  
L inferior temporal gyrus −52 −52 −14 <.001 6.06  
R inferior occipital gyrus 30 −84 −6 <.001 3.57  
L thalamus −6 −6 <.001 6.13 3501 
L substantia nigra −6 −26 −18 <.001 5.45  
R substantia nigra −24 −16 <.001 4.93  
L putamen −16 −2 <.001 4.78  
R globus pallidus 10 <.001 4.92  
L parahippocampal gyrus −16 −14 26 <.001 4.34  
R nucleus caudate 18 −2 26 <.001 5.36 1044 
R parahippocampal gyrus 38 −36 −6 <.001 4.28  
R insula 30 −40 16 <.001 3.98  
L superior parietal lobule −28 −70 44 <.001 5.72 804 
L inferior parietal lobule −42 −42 38 <.001 3.46  
R inferior frontal gyrus 32 24 −6 <.003 5.22 545 
 
Regions More Active for Verbs versus Nouns 
L superior temporal gyrus −44 −20 <.007 4.01 423 
L middle temporal gyrus −46 −8 −6 <.007 3.36  
Anatomical Localization
MNI Coordinates
pcorrected
Z
Voxels per Cluster
x
y
z
Regions More Active for Generation versus Fixation 
L middle occipital gyrus −28 −92 18 <.001 5.82 14,729 
R middle occipital gyrus 34 −94 <.001 4.94  
L cuneus −10 −80 <.001 5.53  
R cuneus 10 −80 <.001 4.75  
L cerebellum −40 −50 −24 <.001 5.27  
R cerebellum 36 −62 −32 <.001 5.04  
L fusiform gyrus −20 −92 −12 <.001 5.21  
R fusiform gyrus 32 −82 −16 <.001 4.54  
L lingual gyrus −12 −68 −4 <.001 4.58  
R lingual gyrus 18 −96 −6 <.001 4.85  
L inferior occipital gyrus −42 −84 −4 <.001 4.39  
L middle frontal gyrus −44 40 <.001 6.26 9516 
L inferior frontal gyrus −42 30 <.001 5.53  
L precentral gyrus −46 56 <.001 5.44  
L putamen −20 10 −2 <.001 5.24  
L insula −34 20 <.001 5.24  
L amygdala −22 −12 −10 <.001 4.97  
L parahippocampal gyrus −16 −14 −14 <.001 4.67  
L superior temporal gyrus −56 12 −8 <.001 4.50  
L thalamus −10 −14 <.001 3.73  
R thalamus 12 −16 <.001 3.39  
R substantia nigra −20 −14 <.001 4.44  
L superior frontal gyrus −6 16 54 <.001 6.20 2075 
L ACC −10 24 40 <.001 4.25  
R ACC 10 20 44 <.001 4.03  
R precentral gyrus 48 −10 34 <.001 5.23 1713 
R putamen 24 10 <.001 4.42 960 
R globus pallidus 20 −8 −4 <.001 4.16  
L superior parietal lobe −26 −60 54 <.001 5.53 623 
L precuneus −20 −64 50 <.001 4.63  
R superior temporal gyrus 52 −28 <.04 4.04 180 
 
Regions More Active for Generation versus Read 
L inferior frontal gyrus −40 30 <.001 >8 10,547 
L middle frontal gyrus −6 18 52 <.001 >8  
L ACC −8 24 40 <.001 7.3  
L superior frontal gyrus −20 30 58 <.001 3.29  
R lingual gyrus 10 −82 −30 <.001 >8 9534 
L lingual gyrus −12 −70 −10 <.001 5.83  
R fusiform gyrus 38 −64 −32 <.001 7.82  
L fusiform gyrus −38 −66 −28 <.001 5.23  
L cuneus −4 −100 14 <.001 4.98  
R cuneus 16 −98 <.001 4.33  
L inferior temporal gyrus −52 −52 −14 <.001 6.06  
R inferior occipital gyrus 30 −84 −6 <.001 3.57  
L thalamus −6 −6 <.001 6.13 3501 
L substantia nigra −6 −26 −18 <.001 5.45  
R substantia nigra −24 −16 <.001 4.93  
L putamen −16 −2 <.001 4.78  
R globus pallidus 10 <.001 4.92  
L parahippocampal gyrus −16 −14 26 <.001 4.34  
R nucleus caudate 18 −2 26 <.001 5.36 1044 
R parahippocampal gyrus 38 −36 −6 <.001 4.28  
R insula 30 −40 16 <.001 3.98  
L superior parietal lobule −28 −70 44 <.001 5.72 804 
L inferior parietal lobule −42 −42 38 <.001 3.46  
R inferior frontal gyrus 32 24 −6 <.003 5.22 545 
 
Regions More Active for Verbs versus Nouns 
L superior temporal gyrus −44 −20 <.007 4.01 423 
L middle temporal gyrus −46 −8 −6 <.007 3.36  

Stereotactic MNI coordinates for significant clusters (random effects, cluster-level pcorrected < .05, estimated at puncorrected < .001) given in millimeter with effect sizes (z scores) and cluster extent. In the voxels per cluster column, cluster extent is reported in correspondence of the main peak. Subpeaks were selected dividing each cluster into Brodmann's areas and then selecting peaks within each area.

Neuroimaging Data

Generation versus Fixation, Generation versus Read, and Verb versus Noun

When the two generation tasks were combined and compared with the fixation condition, activation was found in seven clusters (see Table 1). There was a large posterior cluster with the peak activation located in the left middle occipital gyrus. A second large cluster involved the left middle frontal gyrus and extended to the LIFG, left precentral gyrus, left putamen, left insula, left amygdala, left parahippocampal gyrus, left superior temporal gyrus, left and right thalamus, and right substantia nigra. In addition, the superior frontal gyrus and bilaterally the ACC were activated as part of another cluster.

The contrast generation versus read resulted in the activation of six clusters (see Table 1). The largest cluster had its peak of activation in the LIFG. Activity in this cluster extended also to the left middle frontal gyrus, left superior frontal gyrus, and left ACC. Another large cluster involved posterior brain regions such as, bilaterally, the lingual gyrus, the fusiform gyrus, and the cuneus, whereas on the left the inferior temporal gyrus and on the right the inferior occipital gyrus. Moreover, subcortical areas such as the substantia nigra (bilaterally), the left thalamus and putamen, and the right globus pallidus and nucleus caudate were also more active for generation than read.

The two generation tasks were also contrasted directly against each other. No brain regions were more active for nouns than verbs. By contrast, the contrast verb > noun led to the activation of the left superior and middle temporal gyri (see Table 1 and the bottom part of Figure 3). The activation plots of the bottom part of Figure 3 report activation in these temporal regions for each condition of the tasks. The signal plots indicate that verb generation activates these regions in a similar way to the read condition.

Figure 3. 

Effects of selection demands, conjunction analysis, and verb versus noun contrast. The HS-SA versus LS-SA contrast is reported for noun and verb generation at the top of the figure (noun generation on the left). The results of the conjunction analysis (HS-SA/LS-SA, verb/noun) are reported in the middle of the figure. The results of the contrast verb versus noun are reported at the bottom of the figure. Brain activity in the LIFG is shown for each condition of the two tasks in the graph in the middle of the figure. Brain activity in the left superior and middle temporal gyri is also reported for the conditions of the two tasks in the graphs at the bottom of the figure. Plots depict activity in experimental conditions relative to the read condition (in arbitrary units [a.u.], ±90% confidence interval). LS-SA = low selection and strong association; HS-WA = high selection and weak association; HS-SA = high selection and strong association.

Figure 3. 

Effects of selection demands, conjunction analysis, and verb versus noun contrast. The HS-SA versus LS-SA contrast is reported for noun and verb generation at the top of the figure (noun generation on the left). The results of the conjunction analysis (HS-SA/LS-SA, verb/noun) are reported in the middle of the figure. The results of the contrast verb versus noun are reported at the bottom of the figure. Brain activity in the LIFG is shown for each condition of the two tasks in the graph in the middle of the figure. Brain activity in the left superior and middle temporal gyri is also reported for the conditions of the two tasks in the graphs at the bottom of the figure. Plots depict activity in experimental conditions relative to the read condition (in arbitrary units [a.u.], ±90% confidence interval). LS-SA = low selection and strong association; HS-WA = high selection and weak association; HS-SA = high selection and strong association.

High Selection versus Low Selection

The main aim of the fMRI analyses was to investigate whether selection demands play a role in verb (and/or noun) production once the potential confound of associative strength is removed. Thus, for each generation task (verb or noun), we compared high- versus low-selection conditions with matched levels of association strength (i.e., (HS-SA)–(LS-SA)). The upper left part of Figure 3 shows the brain areas that activated during the noun generation task (see also the first contrast of Table 2). The high- minus low-selection contrast resulted in the activation of several clusters. The peak activation was in the left postcentral gyrus. In the same cluster, the LIFG, the left precentral gyrus, the left middle and superior temporal gyri, and some subcortical areas (left putamen, nucleus caudate, and pallidum) were also activated. Another large cluster was located in the right hemisphere (precentral gyrus, superior and middle temporal gyri, putamen, insula, and claustrum). The upper right part of Figure 3 shows the regions that activated for the high- minus low-selection contrast during verb generation (see also the second contrast of Table 2). In the same way as for noun generation, this contrast revealed activation of the LIFG. As part of a different cluster, the left lingual gyrus and the left cerebellum also showed increased activation for this contrast.

Table 2. 

Direct Contrasts and Conjunction Analysis

Anatomical Localization
MNI Coordinates
pcorrected
Z
Voxels per Cluster
x
y
z
(1) Regions More Active for HS-SA Nouns versus LS-SA Nouns 
L postcentral gyrus −44 −20 32 <.001 5.53 13,163 
L inferior frontal gyrus −38 24 −6 <.001 4.50  
L precentral gyrus −58 −10 32 <.001 4.92  
L superior temporal gyrus −52 10 −2 <.001 5.49  
L middle temporal gyrus −56 −12 −12 <.001 4.69  
L cingulum −22 −36 −6 <.001 4.84  
L nucleus caudatus −32 −14 −10 <.001 4.91  
L pallidum −14 −4 <.001 4.37  
L putamen −22 14 −8 <.001 4.35  
R precentral gyrus 50 −4 20 <.001 5.23 9647 
R superior temporal gyrus 56 −32 14 <.001 4.85  
R middle temporal gyrus 56 −16 −4 <.001 4.22  
R putamen 24 12 <.001 4.79  
R claustrum 30 −2 12 <.001 4.45  
R insula 38 −14 <.001 4.31  
L ACC −6 20 22 <.001 3.98 1272 
R parieto-occipital sulcus 22 −62 18 <.002 4.49 590 
L precuneus −4 −66 32 <.002 3.54  
L parieto-occipital sulcus −22 −64 18 <.01 3.70 387 
 
(2) Regions More Active for HS-SA Verbs versus LS-SA Verbs 
L inferior frontal gyrus −42 32 <.001 4.56 901 
L lingual gyrus −14 −74 −12 <.01 4.18 390 
L cerebellum −10 −60 −22 <.01 3.41  
 
(3) Regions Showing an Interaction between Selection Demands and Task 
L superior temporal gyrus −46 −20 −8 <.009 4.23 397 
L inferior temporal gyrus −54 −10 −20 <.009 3.39  
R posterior cingulate cortex 22 −60 18 <.002 4.08 572 
R precuneus −62 32 <.002 3.56  
 
(4) Conjunction: Regions More Active for HS-SA Verbs U Nouns versus LS-SA Verbs U Nouns 
L inferior frontal gyrus −52 34 <.008 3.86 499 
L precentral gyrus −42 −10 42 <.008 3.25  
 
(5) Regions More Active for HS-WA Verbs versus HS-WA Nouns 
L putamen −18 10 −4 <.04 3.59 283 
R cerebellum 12 −38 −40 <.05 4.26 252 
Anatomical Localization
MNI Coordinates
pcorrected
Z
Voxels per Cluster
x
y
z
(1) Regions More Active for HS-SA Nouns versus LS-SA Nouns 
L postcentral gyrus −44 −20 32 <.001 5.53 13,163 
L inferior frontal gyrus −38 24 −6 <.001 4.50  
L precentral gyrus −58 −10 32 <.001 4.92  
L superior temporal gyrus −52 10 −2 <.001 5.49  
L middle temporal gyrus −56 −12 −12 <.001 4.69  
L cingulum −22 −36 −6 <.001 4.84  
L nucleus caudatus −32 −14 −10 <.001 4.91  
L pallidum −14 −4 <.001 4.37  
L putamen −22 14 −8 <.001 4.35  
R precentral gyrus 50 −4 20 <.001 5.23 9647 
R superior temporal gyrus 56 −32 14 <.001 4.85  
R middle temporal gyrus 56 −16 −4 <.001 4.22  
R putamen 24 12 <.001 4.79  
R claustrum 30 −2 12 <.001 4.45  
R insula 38 −14 <.001 4.31  
L ACC −6 20 22 <.001 3.98 1272 
R parieto-occipital sulcus 22 −62 18 <.002 4.49 590 
L precuneus −4 −66 32 <.002 3.54  
L parieto-occipital sulcus −22 −64 18 <.01 3.70 387 
 
(2) Regions More Active for HS-SA Verbs versus LS-SA Verbs 
L inferior frontal gyrus −42 32 <.001 4.56 901 
L lingual gyrus −14 −74 −12 <.01 4.18 390 
L cerebellum −10 −60 −22 <.01 3.41  
 
(3) Regions Showing an Interaction between Selection Demands and Task 
L superior temporal gyrus −46 −20 −8 <.009 4.23 397 
L inferior temporal gyrus −54 −10 −20 <.009 3.39  
R posterior cingulate cortex 22 −60 18 <.002 4.08 572 
R precuneus −62 32 <.002 3.56  
 
(4) Conjunction: Regions More Active for HS-SA Verbs U Nouns versus LS-SA Verbs U Nouns 
L inferior frontal gyrus −52 34 <.008 3.86 499 
L precentral gyrus −42 −10 42 <.008 3.25  
 
(5) Regions More Active for HS-WA Verbs versus HS-WA Nouns 
L putamen −18 10 −4 <.04 3.59 283 
R cerebellum 12 −38 −40 <.05 4.26 252 

Stereotactic MNI coordinates for significant clusters (random effects, cluster-level pcorrected < .05, estimated at puncorrected < .001) given in millimeter with effect sizes (z scores) and cluster extent. In the voxels per cluster column, cluster extent is reported in correspondence of the main peak. Subpeaks were selected dividing each cluster into Brodmann's areas and then selecting peaks within each area. LS-SA = low selection and strong association; HS-WA = high selection and weak association; HS-SA = high selection and strong association.

In view of the relatively different patterns of activations (e.g., bilaterally in temporal cortex), which were found between the two tasks for the (HS-SA)–( LS-SA) contrast, we tested for the interaction between selection demands and task (i.e., HS-SA/LS-SA noun/verb). Two clusters showed a significant interaction (see the third contrast of Table 2), the posterior cingulate cortex and the precuneus formed the largest cluster, whereas the left superior and inferior temporal gyri formed a second. The interaction indicates that the decrease in the left temporal cortex activation (relative to the read condition) for the nouns relative to the verbs, which gave rise to the verb–noun main effect, is somewhat reduced for the HS-SA condition of noun generation. In fact, in this condition, the level of activation of the left temporal cortex was similar for noun and verb generation (e.g., see plots of Figure 3).

Moreover, because we expected a similar effect of “high versus low selection” for verb and noun generation, we formally assessed this prediction by performing a conjunction analysis (see Nichols, Brett, Andersson, Wager, & Poline, 2005) of (HS-SA)–(LS-SA) for both generation tasks. The middle part of Figure 3 shows that the only regions that were significantly more active for the HS-SA than LS-SA condition in both tasks were the LIFG (BA 45) and the left precentral gyrus (BA 6; see also the fourth contrast of Table 2).

Finally, we also investigated whether the effects of selection demands, which the conjunction analysis showed to be located in the LIFG, occur in an area (1) close and functionally similar to that which Badre et al. (2005) reported to be primarily involved in response selection (i.e., left mid-VLPFC, BA 45) (2) but distant and presumably functionally different to the area that these authors reported to be sensitive to manipulation of associative strength (i.e., controlled retrieval, left anterior/ventral VLPFC). To test these hypotheses, two sphere ROIs of radius 8 mm were extracted using the Marsbar software (www.sourceforge.net/projects/marsbar). The peak coordinates of the first ROI were selected referring to the area that was most sensitive to the selection component in the study of Badre et al. Thus, this ROI was located in the left mid-VLPFC (−54 21 12). In a similar way, the peak coordinates of the second ROI were selected referring to the area that, in Badre et al., was most sensitive to association strength. The second ROI was located in the anterior/ventral VLPFC (−45 27 −15). It should be noted that the cluster of activation found in the main conjunction analysis (HS-SA > LS-SA for both verb and noun generation tasks) was part of the ROI in the mid/posterior VLPFC (−54 18 8) but not of the ROI in the anterior VLPFC.

As for the whole-brain analyses, we compared high- versus low-selection conditions with matched levels of association strength (i.e., (HS-SA)–(LS-SA)) in both tasks and in both ROIs. As far as the verb generation task is concerned, the results confirmed that activation in the mid/posterior VLPFC was significantly higher for HS-SA than for LS-SA (one-tailed t test), t(13) = 3.23, p < .001, whereas activation in the anterior VLPFC did not differ between these two conditions (one-tailed t test), t(13) = 1.09, p = .14. In the noun generation task, the HS-SA > LS-SA contrast revealed more activation not only in the mid/posterior VLPFC (one-tailed t test), t(13) = 3.83, p < .001, but also in the anterior VLPFC, unlike verb generation (one-tailed t test), t(13) = 4.88, p < .001. These results show an effect of selection demands focused only in the mid/posterior VLPFC for verb generation but extending also to the more anterior section of the LIFG in case of noun generation.

Effect of Weak Association

Next, we tested for the effect of association strength that had been found to have a major effect on the behavioral data. Accordingly, we compared, in both tasks, the HS-WA condition with the HS-SA condition. Contrary to the initial prediction, this contrast did not reveal any significant activation in either task. However, interesting results were obtained when effects of association strength were investigated within the ROIs extracted in the mid/posterior VLPFC and anterior/ventral VLPFC. With regard to the verb generation task, the anterior/ventral VLPFC region was significantly more active for the HS-WA than the HS-SA condition (one-tailed t test), t(13) = 1.70, p < .05, whereas no significant differences between these two conditions were found in the mid/posterior VLPFC (one-tailed t test), t(13) = 1.31, p = .1. The test for an Area × Condition interaction (mid-VLPFC/anterior VLPFC by HS-WA/HS-SA) did not reach statistical significance, F(1,13) = 0.21, p = .60, possibly because a trend toward greater activation for the “weak” versus the “strong” association condition also occurs in the mid/posterior VLPFC.

As far as the noun generation task is concerned, neither the anterior/ventral VLPFC nor the mid/posterior VLPFC was more active for HS-WA than for HS-SA (one-tailed t tests), t(13) = −2.05, p = .98 and t(13) = −3.27, p = .99 for anterior/ventral and mid/posterior VLPFC ROIs, respectively. In fact, contrary to our expectations, the high negative t values indicate that for noun generation, both sections of the left VLPFC were more active in the HS-SA than the HS-WA condition.

Finally, an important prediction of our study concerned the HS-WA condition during the two different generation tasks. In particular, we asked whether verb and noun generation would differ specifically in the generation of weakly associated items. As reported in the Introduction, we expected to find more activity in BG for verbs than nouns in the weak association condition. For this reason, we compared HS-WA verb minus HS-WA noun. Consistent with the prediction, two clusters of activation were identified: one in the left putamen, extending also to the left pallidum, and one in the right cerebellum (see Figure 4 and the fifth contrast of Table 2). Moreover, in view of these results, we tested for the interaction between association strength and task (i.e., HS-WA/HS-SA, verb/noun). The left putamen showed a significant interaction (t = 3.82, pcorrected < .05, cluster level, estimated at puncorrected threshold < .001), whereas the right cerebellum did not. These results further suggest that the left BG are particularly activated in the weak association verb condition.

Figure 4. 

Brain areas revealed by the contrast HS-WA verb versus HS-WA noun. Activity in left putamen is reported in the graph for each condition of the two tasks. Plots depict activity in experimental conditions relative to the read condition (in arbitrary units [a.u.], ±90% confidence interval). LS-SA = low selection and strong association; HS-WA = high selection and weak association; HS-SA = high selection and strong association.

Figure 4. 

Brain areas revealed by the contrast HS-WA verb versus HS-WA noun. Activity in left putamen is reported in the graph for each condition of the two tasks. Plots depict activity in experimental conditions relative to the read condition (in arbitrary units [a.u.], ±90% confidence interval). LS-SA = low selection and strong association; HS-WA = high selection and weak association; HS-SA = high selection and strong association.

DISCUSSION

Summary of Main Findings

The purpose of this study was to provide evidence related to the issue of the competitive versus associative accounts of verb generation. We used fMRI to address this matter by extending the classical paradigm of verb production (see Martin & Cheng, 2006) to the generation of nouns (see Crescentini et al., 2008). Our fMRI results show that the selection between competing responses is a key factor in single-word production. The LIFG is more active in situations of high selection than low selection, with association strength matched across conditions. This was the case both for verbs and for nouns as formally shown by conjunction analyses. More specifically, the results of ROI analyses showed that, for verb generation, effects of selection demands are found in the mid/posterior LIFG (i.e., ROI in the mid/posterior VLPFC) but not in the anterior/ventral LIFG (i.e., ROI in the anterior VLPFC), whereas for noun generation, the ROI analyses showed effects of selection demands in both sections of the LIFG.

The subjects' behavior on both experimental tasks shows that association strength is also an important factor for both noun and verb production. Production of weakly stimulus-associated responses requires more time and is more prone to errors than that of strongly stimulus-associated responses. Nevertheless, for both tasks, the condition with weak associations did not lead to any significant activation relative to the corresponding condition with strong associations, once selection demands were carefully matched in the two conditions (i.e., the contrast (HS-WA)–(HS-SA)). Of importance, the ROI analyses, however, gave rise to effects of association strength in verb generation but not in noun generation. Thus, for verb generation, the anterior/ventral VLPFC was more active for HS-WA than for HS-SA, whereas the mid/posterior VLPFC was equally active in the two conditions.

An important prediction of this study concerned the comparison between the HS-WA condition of the two tasks. Consistent with our hypothesis, we found that weakly stimulus-related verbs activated the left BG (left putamen) more than did weakly stimulus-related nouns. Activity in the right cerebellum was also found in this contrast. An interaction between association strength and task was found in the left putamen but not in the right cerebellum. In a previous study (Crescentini et al., 2008), we had argued that the coming to mind of nouns interferes with verb production but that the complementary situation in noun production generally does not occur (see also free association norms reported in Stimuli section). From this perspective, the left putamen could be a key structure involved in coping with the interference that occurs selectively during verb generation.

Another unexpected but noteworthy result concerns the main effect of verbs versus nouns. This revealed greater activation of the left superior and middle temporal gyri for verbs than nouns. In the following sections, we provide a possible account for this and the earlier findings referring to the most relevant brain areas for single-word generation tasks.

Left Inferior Frontal Gyrus

In this study, we replicated and extended the findings of Persson et al. (2004), Barch et al. (2000), and Thompson-Schill et al. (1997, 1998) of LIFG activation on the high- versus low-selection comparison. Thus, we provided additional evidence to support the hypothesis that LIFG is important for the selection of information among competing alternatives. Of importance, an ROI analysis has shown that the left mid/posterior VLPFC rather than the left anterior/ventral VLPFC is involved in this function. This analysis was performed by selecting the areas of the LIFG that Badre et al. (2005) had shown to be primarily involved in response selection (mid/posterior VLPFC, BA 45) and controlled retrieval (anterior/ventral VLPFC, BA 47). The results have shown that, for verb generation, only the more mid/posterior section of the left VLPFC was more active for HS-SA than for LS-SA. Nevertheless, effects of selection demands also extended to the anterior left VLPFC for noun generation. Globally, these results fit with the finding of our conjunction analysis of the degree of selection demands ((HS-SA)–(LS-SA)) carried out for both tasks. Thus, this analysis showed an activation cluster in a region that partially overlapped with the mid/posterior left VLPFC ROI but not with the anterior/ventral left VLPFC ROI.

Thus, our study sheds further light on the role of the mid/posterior LIFG. Within the domain of semantic knowledge, the selection between alternative nouns or verbs activates this area in a very similar way (for LIFG activation related to the production of both action nouns and verbs, see Siri et al., 2008; and for further evidence on activation of LIFG for both nouns and verbs, see also Berlingeri et al., 2008). Moreover, the fact that here the strong association conditions (HS-SA and LS-SA) were carefully matched for the strength and the number of task-irrelevant responses in both tasks fits with the proposal that the posterior LIFG mediates selection of verb and noun responses from among task-relevant alternatives.

As shown above, the results of the ROI analyses also showed that the anterior left VLPFC was more active for HS-WA than HS-SA. This effect of association strength was found only for verb generation. These results are in line with the behavioral effects of association strength, which were much stronger for verb than noun generation (i.e., about 400 vs. 100 msec, respectively; see Reaction time data section). For the verb generation task, the results of the ROI analyses are in line with the functional segregation that Badre et al. (2005) found in the left VLPFC. The anterior/ventral section of this brain region is held to be involved in controlled retrieval from semantic memory with the mid/posterior section being more involved in selection in situations of high competition. However, we did not obtain any significant Area × Condition interaction (mid-VLPFC/anterior VLPFC by HS-WA/HS-SA) and although statistically not significant, the mid/posterior VLPFC was also somewhat more active for the weak than the strong association condition; thus, from our data, it remains possible that the left mid/posterior VLPFC is also involved in controlled retrieval.

It is not clear, however, why the HS-SA condition of noun generation leads to more activation of both sections of the left VLPFC relative to HS-WA. One possibility is that two roughly equally strongly associated weak association nouns do not compete with each other as much as two strong association nouns do; in other words, noun responses in the former situation produce weaker inhibitory effects on each other than noun responses in the latter situation do. By contrast, for verbs, the situation is different because, in the weak association condition, verb responses face competition from interfering noun responses.

Accordingly, a possible reason why the anterior VLPFC is more active for HS-WA than for HS-SA only in verb generation is that in the HS-WA condition of this task, retrieval of task-relevant responses (i.e., verbs), requires the operation of a top–down control mechanisms. Such a mechanisms would need to specify the class of an appropriate response because, as suggested by free association norms (see Stimuli; see also Crescentini et al., 2008), inappropriate responses (i.e., nouns) are frequently spontaneously activated by the stimuli in the verb generation condition. Thompson-Schill and Botvinick (2006) have recently proposed a Bayesian framework for verb generation, which included a top–down biasing mechanism that adjusts the prior probabilities of response candidates so making the probability of verbs higher than that of nouns during verb generation. The anterior left VLPFC is a candidate area for the locus of this process.

Our study suggests that the activation of the mid/posterior LIFG may not necessarily give rise to behavioral effects such as, for example, an increase of response times for the high selection condition; in fact, the HS-SA and the LS-SA conditions of noun generation led to similar RTs in our study. A possible explanation for the relatively similar RTs in the two conditions with strong associations (compared with the situations with weak associations) is that in an interactive activation model with positive feedback (e.g., McClelland & Rumelhart, 1981), having two competing alternatives with strong input increases the degree of activation overall but, due to the mutual inhibition, affects the speed of selection much less than when two competing alternatives have weak input.

As an alternative theoretical account, within the ACT-R framework (Anderson et al., 2004), Crescentini and Del Missier (submitted) simulated the performance of three subject populations (i.e., PD patients, older controls, and young adults) on the noun and verb generation tasks. Briefly, the simulations replicate the pattern of similar RTs on the HS-SA and LS-SA conditions. An important feature of the ACT-R associative memory theory is that the probability of generating a given response node is a competitive process that depends on other potential response nodes and so is affected by the degree of selection. On the other hand, the computation of the RT for a given response node directly reflects its activation level that, in turn, is given by the strength of stimulus-response associative links.

As already discussed, the conjunction analysis also shows that the main area that is activated by the high-selection versus low-selection contrast in both tasks is BA 45. Robinson et al. (1998, 2005) reported two dynamic aphasic patients who had lesions to this brain area; they had a severe impairment in propositional language production characterized by an exceptionally reduced spontaneous speech in the context of well-preserved naming, articulation, prosody, and repetition skills. These patients had particular difficulties in the selection of a verbal response when others were in competition. The authors proposed that this brain area is important in the selection of a response between competing alternatives. Our study supports this interpretation of these neuropsychological findings.

A general issue concerns the way in which association strength and selection demands were operationalized in our study. It has been argued that such variables are likely to be correlated, that is, conditions differing on one measure also tend to differ on the other (see Snyder & Munakata, 2008; Thompson-Schill & Botvinick, 2006). Recently, Snyder and Munakata (2008) addressed this issue proposing new measures based on latent semantic analyses (LSAs) designed to unconfound these two effects. Using LSA association values, Snyder and Munakata also calculated a new measure of competition, namely, “entropy,” which reflects the competition between all alternative responses rather than just the two most common responses (as in our current study). Snyder and Munakata's LSA measures may help to eliminate the potential problem of using relative, proportion-based measures, as used in this and in previous studies (Martin & Cheng, 2006; Thompson-Schill et al., 1997). LSA-based measures of association strength and competition could be useful in future fMRI studies and could help to further clarify the functional segregation of controlled retrieval and selection (Badre et al., 2005).

Basal Ganglia

In our study, BG activity was particularly high in the HS-SA conditions and also in the HS-WA condition of verb generation. Moreover, the left putamen showed an interaction between association strength and task. Recent studies give the BG a function of inhibiting competing alternatives, and this function is also held to be used in language production (Castner et al., 2007, 2008; Crescentini et al., 2008; Longworth et al., 2005). Indeed, an involvement of BG structures in lexical-semantic processing has been proposed to explain findings both of neuropsychological studies on patients with subcortical lesions (see Copland, 2003; Copland, Chenery, & Murdoch, 2000) and of neuroimaging studies (Crosson et al., 2003; Rossel, Bullmore, Williams, & David, 2001).

For instance, Rossel et al. (2001) studied the brain correlates of automatic and controlled processing in a semantic priming environment. Automatic and controlled processing were investigated by using short and long prime-target delays. The authors found that the putamen was preferentially activated at long intervals, and this led them to suggest a role for this structure in controlled semantic processes. The authors suggested that the putamen is involved in the processes of response selection and inhibition. A similar conclusion was reached by Copland (2003). He carried out a semantic priming study in which PD patients and controls were presented with auditory prime-target pairs in four different conditions (subordinate meaning unrelated, bat-river; dominant meaning unrelated, foot-money; subordinate meaning related, bank-river; and dominant meaning related, bank-money) and at two conditions of prime-target ISI (short and long). Subjects were required to perform lexical decisions to targets (which also included nonwords). The results suggested that PD patients had intact automatic lexical processes, as at short ISI they showed priming in both the dominant and the subordinate conditions. However, at long ISI, PD patients, unlike normal controls, did not show selective semantic facilitation of the dominant meaning. Copland argued that this was due to an impairment of inhibitory mechanisms in PD.

Our results are in agreement with this proposed role of the BG in the processes of response selection during lexical retrieval. However, we need to consider whether it is appropriate to go a step further by assuming that the activation of the left putamen in the HS-WA condition selectively during verb generation arises from the need to suppress task-irrelevant responses (i.e., nouns) in verb production. This would be consistent with recent evidence for a cognitive role of the BG (Longworth et al., 2005) and also with some recent accounts of cognitive deficits of PD patients. For example, in a recent study, Castner et al. (2007) suggest that PD patients have deficits in some aspect of inhibitory control, such as the inhibition of prepotent responses, as well as having problems in selection from competing responses. Findings from a recent computational model (Crescentini & Del Missier, submitted) also support a role of the BG in the suppression of nonverb responses in the HS-WA condition of verb generation. This model was able to fit the findings previously obtained on PD patients on both the noun and the verb generation tasks. This occurred when a parameter controlling the probability of inhibiting task-irrelevant responses was reduced compared with the value that best simulated the findings of both young adults and normal older controls. When this parameter was changed in this fashion, the model was as slow and made as many intrusion errors in the HS-WA condition of verb generation as the PD patients (Crescentini et al., 2008). The change in this parameter had much less effect on the ability of the model to produce noun responses in the HS-WA condition of noun generation; this indicates that this condition is less subject to intrusion from task-irrelevant responses (i.e., verb responses) than the corresponding condition of verb generation. The finding that young adults are slower with verbs than nouns only in the HS-WA condition (see Reaction time data section) provides further support for the claim that this condition is more prone to interference from nonverb competitors.

Nonetheless, another possibility is that the BG contribute in accentuating task-relevant information (i.e., verbs in the verb generation task) rather than inhibiting task-irrelevant one (i.e., nouns in verb generation). Accordingly, it has recently been proposed that the resolution of conflicts between incompatible responses may also occur through a top–down accentuation of task-relevant responses rather than inhibition of task-irrelevant information (Egner & Hirsch, 2005). Even more recently, McNab and Klingberg (2008) have stressed the functional coupling between prefrontal cortex and BG and proposed that they operate in concert to filter irrelevant information. In particular, they claimed that the BG are important for allowing only relevant information to enter working memory. Furthermore, more than two decades ago Norman and Shallice (1986; see also Robbins & Sahakian, 1983) had proposed a mechanism through which task-relevant responses may be accessed. These authors suggested that the BG may operate to potentiate the activation level of schemas in contention scheduling, thus biasing their selection in a task-relevant manner. The plausibility of this proposal has been supported more recently in an fMRI study of language production (Crosson et al., 2003). These authors have suggested that the BG maintain a bias toward a lexical alternative chosen from among others in competition during controlled word selection.

It is important to note that the two possibilities briefly discussed above are not mutually exclusive; in fact, recent evidence has shown that response selection and response inhibition may involve similar mechanisms both relying on the operations of fronto-striatal circuits (Mostofsky & Simmonds, 2008).

There is also a third possibility for why the BG are more active for verbs than nouns in the HS-WA condition. This is related to the way that verb retrieval may depend on the efficiency of the representation of actions (e.g., Silveri & Misciagna, 2000). According to this view, the BG could be involved in verb processing because of their well-known function in motor control (Mink, 1996) and through its functional links with the frontal lobe. For example, Signorini and Volpato (2006) found that PD patients were impaired on an action fluency task but not on semantic and letter fluency tasks.

Left Temporal Lobe

We found that verb generation activated the left superior and the middle temporal gyri more than noun generation (see the third contrast of Table 1). The generation of nouns semantically related to noun stimuli has not been widely studied yet. An exception is the PET experiment of Warburton et al. (1996). These authors studied noun and verb retrieval in normal subjects. Noun and verb generation tasks were also contrasted with a rest control condition. As found in the present study, Warburton et al. showed that compared with rest, both tasks activated the same regions in the superior temporal gyrus (bilaterally), LIFG, ACC, and SMAs plus several subcortical areas. Moreover, the direct contrast of verb minus noun generation revealed activation of the left inferior temporal gyrus (as well as the left inferior parietal lobe, LIFG, and SMA). Warburton et al. suggested that the inferior temporal gyrus and the posterior part of the inferior parietal lobe are important in lexical processing and in particular in the access to semantic fields (for further evidence showing that activation of the left middle temporal gyrus is increased for verbs relative to nouns, see also Yokoyama et al., 2006).

One possibility related to that put forward by Warburton et al. (1996) is that semantic priming effects may be stronger for noun than verb generation. Recent neuroimaging studies have shown decreased activation in several brain areas in conditions of repetition priming, including regions of the temporal lobe (for reviews, see Henson, Shallice, Josephs, & Dolan, 2002; Schacter & Buckner, 1998). In an fMRI study of implicit semantic priming, Rissman, Eliassen, and Blumstein (2003) studied the brain correlates of a lexical decision task in the context of prime-target pairs. The targets could be semantically related or unrelated to the prime or they could be nonwords. The authors found decreased activation in the left superior temporal gyrus (as well as in the left precentral gyrus, left and right middle temporal gyri, and right caudate) when the related target condition was compared with the unrelated condition (for similar findings of neural priming in left middle temporal gyrus in a lexical decision task, see also Copland et al., 2003). The authors argued that this finding could be due to the enhanced neural efficiency occurring in the recognition of related targets.

A possible account of our findings could be that noun stimuli elicited many other nouns with related semantic features fairly automatically; this would have the effect of reducing the left temporal gyrus activation in situations of noun generation. By contrast, during verb production, subjects have to explicitly attend to the semantic relationship between the stimulus and the response (see Martin & Byrne, 2006; Thompson-Schill & Botvinick, 2006). In agreement with this, previous analyses of free association responses using the same stimuli as in the current fMRI experiment (see Stimuli section; Crescentini et al., 2008) showed that nouns are spontaneously produced much more often than verbs in response to noun cues.

However, there is a difficulty for an interpretation in terms of semantic priming of the increased activation in the left temporal lobe for verbs relative to nouns. In situations of strong association, the generation of verbs is actually faster than noun generation (see Figure 2). An alternative interpretation for this result can be postulated based on the declarative memory theory of ACT-R (Anderson et al., 2004) and, more specifically, on the explanation that this theory gives of the so-called “fan” effect. ACT-R explains the “fan” effect in a similar way to Martin and Byrne's (2006) explanation of their findings on verb generation (i.e., weak association condition performed worse than strong association condition). This is that the associative strength between the cue stimulus, and each associated response tends to reduce as the number of associates increases (Danker et al., 2008). Martin and Byrne have proposed that, in verb generation, the activation of a concept “action” contributes with the cue stimulus (i.e., noun) to the selection of a verb response. In other words, the strength of the spread of activation in their network depends on the strength of the links between both the cue and its associated responses and the concept “action” and its associated verb responses.

From a related perspective, our free association norms indicate that cues are more often associated with noun than verb responses. Thus, in a potential extension of Martin and Byrne's (2006) model, the existence of the “fan” effect would predict that the associative links between the “action” node and the verb responses are stronger than those between the “noun” node and the noun responses. Stronger associative links lead to more active responses in ACT-R (see also Left inferior frontal gyrus section) and this in turn may explain why verbs are produced faster than nouns, at least in conditions of strong association. From the prospective of the fMRI data, the potential difference in the degree of activation spreading between noun and verb generation may explain why the left temporal lobe is more active for verbs than for nouns. In similar fashion, Martin and Byrne have argued that, in situations of strong stimulus-response association strength, the automatic search within semantic memory depends on the left temporal cortex.

Finally, as reported above, the main effect of the verb–noun contrast suggests that globally noun generation leads to less activation of the left temporal lobe than does verb generation. However, the results of the interaction between selection demands and task indicate that the (HS-SA)–(LS-SA) contrast activates the superior and the inferior parts of the left temporal cortex more for noun generation than for verb generation. A possible explanation for this difference could be that in the HS-SA condition of noun generation, two or more strongly associated items are activated. In an interactive activation process, the existence of competitors would limit the potential speed in the achieving of the threshold that semantic priming would otherwise provide.

Conclusions

We investigated the roles of selection and association during verb and noun generation. For both types of task, we found that the midpart of the LIFG was more active in conditions requiring high compared with low selection, highlighting the central role of this region in mediating the selection between competing responses. This was also suggested by an ROI analysis. Association strength influenced behavioral performance but did not produce any specific brain activation on a whole-brain analysis. However, when an ROI was extracted in the anterior/ventral region of the VLPFC (BA 47), an effect of association strength was found for the verb generation task but not for the noun generation task. These results suggest that controlled retrieval is subserved by this brain region and that it is needed more for verb than noun generation. Moreover, we found that the BG and the cerebellum were more active when weakly associated verbs had to be produced relative to weakly associated nouns. An interaction between association strength and task was found in the former but not in the latter structures. This suggests that association strength becomes critical when task-irrelevant stimuli interfere with the current task (here nouns during verb production), triggering additional activation of the BG. We conclude that both selections between competing responses and associative strength are important factors for single-word production. However, association strength would only become critical in the verb generation task and particularly when task-irrelevant responses interfere with word production.

Acknowledgments

This research was partially supported by a grant from PRIN to Tim Shallice. The authors are also thankful to the members of the Neuroimaging Laboratory, Santa Lucia Foundation, Rome, for their helpfulness throughout the study.

Reprint requests should be sent to Cristiano Crescentini, SISSA-International School for Advanced Studies, Cognitive Neuroscience Sector, Via Beirut, 2-4, 34014, Trieste, Italy, or via e-mail: crescent@sissa.it or cristianocresce@hotmail.com.

REFERENCES

REFERENCES
Anderson
,
J. R.
,
Bothell
,
D.
,
Byrne
,
M. D.
,
Douglass
,
S.
,
Lebiere
,
C.
, &
Qin
,
Y.
(
2004
).
An integrated theory of the mind.
Psychological Review
,
111
,
1036
1060
.
Badre
,
D.
,
Poldrack
,
R. A.
,
Paré-Blagoev
,
E. J.
,
Insler
,
R. Z.
, &
Wagner
,
A. D.
(
2005
).
Dissociable controlled retrieval and generalized selection mechanisms in ventrolateral prefrontal cortex.
Neuron
,
47
,
907
918
.
Badre
,
D.
, &
Wagner
,
A. D.
(
2002
).
Semantic retrieval, mnemonic control, and prefrontal cortex.
Behavioral and Cognitive Neuroscience Reviews
,
1
,
206
218
.
Barch
,
D. M.
,
Braver
,
T. S.
,
Sabb
,
F. W.
, &
Noll
,
D. C.
(
2000
).
Anterior cingulate and the monitoring of response conflict: Evidence from an fMRI study of overt verb generation.
Journal of Cognitive Neuroscience
,
12
,
298
309
.
Berlingeri
,
M.
,
Crepaldi
,
D.
,
Roberti
,
R.
,
Scialfa
,
G.
,
Luzzatti
,
C.
, &
Paulesu
,
E.
(
2008
).
Nouns and verbs in the brain: Grammatical class and task specific effects as revealed by fMRI.
Cognitive Neuropsychology
,
25
,
528
558
.
Botvinick
,
M. M.
,
Braver
,
T. S.
,
Barch
,
D. M.
,
Carter
,
C. S.
, &
Cohen
,
J. D.
(
2001
).
Conflict monitoring and cognitive control.
Psychological Review
,
108
,
624
652
.
Castner
,
J. E.
,
Chenery
,
H. J.
,
Silburn
,
P. A.
,
Coyne
,
T. J.
,
Sinclair
,
F.
,
Smith
,
E. R.
,
et al
(
2008
).
Effects of subthalamic deep brain stimulation on noun/verb generation and selection from competing alternatives in Parkinson's disease.
Journal of Neurology, Neurosurgery and Psychiatry
,
79
,
700
705
.
Castner
,
J. E.
,
Copland
,
D. A.
,
Silburn
,
P. A.
,
Coyne
,
T. J.
,
Sinclair
,
F.
, &
Chenery
,
H. J.
(
2007
).
Lexical-semantic inhibitory mechanisms in Parkinson's disease as a function of subthalamic stimulation.
Neuropsychologia
,
45
,
3167
3177
.
Copland
,
D. A.
(
2003
).
The basal ganglia and semantic engagement: Potential insights from semantic priming in individuals with subcortical vascular lesions, Parkinson's disease, and cortical lesions.
Journal of the International Neuropsychological Society
,
9
,
1041
1052
.
Copland
,
D. A.
,
Chenery
,
H. J.
, &
Murdoch
,
B. E.
(
2000
).
Processing lexical ambiguities in word triplets: Evidence of lexical-semantic deficits following dominant nonthalamic subcortical lesions.
Neuropsychology
,
14
,
370
390
.
Copland
,
D. A.
,
de Zubicaray
,
G. I.
,
McMahon
,
K.
,
Wilson
,
S. J.
,
Eastburn
,
M.
, &
Chenery
,
H. J.
(
2003
).
Brain activity during automatic semantic priming revealed by event-related functional magnetic resonance imaging.
Neuroimage
,
20
,
302
310
.
Crescentini
,
C.
, &
Del Missier
,
F.
(
submitted
).
Executive control of retrieval in noun and verb production: A computational model
.
Crescentini
,
C.
,
Mondolo
,
F.
,
Biasutti
,
E.
, &
Shallice
,
T.
(
2008
).
Supervisory and routine processes in noun and verb generation in nondemented patients with Parkinson's disease.
Neuropsychologia
,
46
,
434
447
.
Crosson
,
B.
,
Benefield
,
H.
,
Cato
,
M. A.
,
Sadek
,
J. R.
,
Moore
,
A. B.
,
Wierenga
,
C. E.
,
et al
(
2003
).
Left and right basal ganglia and frontal activity during language generation: Contributions to lexical, semantic, and phonological processes.
Journal of the International Neuropsychological Society
,
9
,
1061
1077
.
Danker
,
J. F.
,
Gunn
,
P.
, &
Anderson
,
J. R.
(
2008
).
A rational account of memory predicts left prefrontal activation during controlled retrieval.
Cerebral Cortex
,
18
,
2674
2685
.
Derrfuss
,
J.
,
Brass
,
M.
,
Neumann
,
J.
, &
von Cramon
,
D. Y.
(
2005
).
Involvement of the inferior frontal junction in cognitive control: Meta-analyses of switching and Stroop studies.
Human Brain Mapping
,
25
,
22
34
.
Egner
,
T.
, &
Hirsch
,
J.
(
2005
).
Cognitive control mechanisms resolve conflict through cortical amplification of task-relevant information.
Nature Neuroscience
,
8
,
1784
1790
.
Friston
,
K. J.
(
2004
).
Experimental design and statistical parametric mapping.
In R. S. Frackowiak, K. J. Friston, C. D. Frith, R. J. Dolan, C. J. Price, S. Zeki, et al. (Eds.),
Human brain function
(2nd ed., pp.
599
632
).
Oxford
:
Academic Press
.
Friston
,
K. J.
,
Glaser
,
D. E.
,
Henson
,
R. N.
,
Kiebel
,
S.
,
Phillips
,
C.
, &
Ashburner
,
J.
(
2002
).
Classical and Bayesian inference in neuroimaging: Applications.
Neuroimage
,
16
,
484
512
.
Friston
,
K. J.
,
Zarahn
,
E.
,
Josephs
,
O.
,
Henson
,
R. N.
, &
Dale
,
A. M.
(
1999
).
Stochastic designs in event-related fMRI.
Neuroimage
,
10
,
607
619
.
Henson
,
R. N.
,
Shallice
,
T.
,
Josephs
,
O.
, &
Dolan
,
R. J.
(
2002
).
Functional magnetic resonance imaging of proactive interference during spoken cued recall.
Neuroimage
,
17
,
543
558
.
Levy
,
B. J.
, &
Anderson
,
M. C.
(
2002
).
Inhibitory processes and the control of memory retrieval.
Trends in Cognitive Sciences
,
6
,
299
305
.
Longworth
,
C. E.
,
Keenan
,
S. E.
,
Barker
,
R. A.
,
Marslen-Wilson
,
W. D.
, &
Tyler
,
L. K.
(
2005
).
The basal ganglia and rule-governed language use: Evidence from vascular and degenerative conditions.
Brain
,
128
,
584
596
.
Martin
,
R. C.
, &
Byrne
,
M. D.
(
2006
).
Why opening a door is as easy as eating an apple: A reply to Thompson-Schill and Botvinick.
Psychonomic Bulletin & Review
,
13
,
409
411
.
Martin
,
R. C.
, &
Cheng
,
Y.
(
2006
).
Selection demands versus association strength in the verb generation task.
Psychonomic Bulletin & Review
,
13
,
396
401
.
McClelland
,
J. L.
, &
Rumelhart
,
D. E.
(
1981
).
An interactive activation model of context effects in letter perception: Part 1. An account of basic findings.
Psychological Review
,
88
,
375
407
.
McNab
,
F.
, &
Klingberg
,
T.
(
2008
).
Prefrontal cortex and basal ganglia control access to working memory.
Nature Neuroscience
,
11
,
103
107
.
Mink
,
J. W.
(
1996
).
The basal ganglia: Focused selection and inhibition of competing motor programs.
Progress in Neurobiology
,
50
,
381
425
.
Moss
,
H. E.
,
Abdallah
,
S.
,
Fletcher
,
P. C.
,
Bright
,
P.
,
Pilgrim
,
L. K.
,
Acres
,
K.
,
et al
(
2005
).
Selecting among competing alternatives: Selection and retrieval in the left inferior frontal gyrus.
Cerebral Cortex
,
15
,
1723
1735
.
Mostofsky
,
S. H.
, &
Simmonds
,
D. J.
(
2008
).
Response inhibition and response selection: Two sides of the same coin.
Journal of Cognitive Neuroscience
,
20
,
1
11
.
Nichols
,
T. E.
,
Brett
,
M.
,
Andersson
,
J.
,
Wager
,
T.
, &
Poline
,
J. B.
(
2005
).
Valid conjunction inference with the minimum statistic.
Neuroimage
,
25
,
653
660
.
Norman
,
D. A.
, &
Shallice
,
T.
(
1986
).
Attention to action: Willed and automatic control of behaviour.
In R. J. Davidson, G. E. Schwartz, & D. Shapiro (Eds.),
Consciousness and self-regulation: Advances in research
(
Vol. 4
, pp.
1
18
).
New York
:
Plenum Press
.
Persson
,
J.
,
Sylvester
,
C. C.
,
Nelson
,
J. K.
,
Welsh
,
K. M.
,
Jonides
,
J.
, &
Reuter-Lorenz
,
P. A.
(
2004
).
Selection requirements during verb generation: Differential recruitment in older and younger adults.
Neuroimage
,
23
,
1382
1390
.
Petersen
,
S. E.
,
Fox
,
P. T.
,
Posner
,
M. I.
,
Mintun
,
M.
, &
Raichle
,
M. E.
(
1988
).
Positron emission tomographic studies of the cortical anatomy of single-word processing.
Nature
,
331
,
585
589
.
Petersen
,
S. E.
,
Fox
,
P. T.
,
Snyder
,
A. Z.
, &
Raichle
,
M. E.
(
1990
).
Activation of extrastriate and frontal cortical areas by visual words and word-like stimuli.
Science
,
249
,
1041
1044
.
Redgrave
,
P.
,
Prescott
,
T. J.
, &
Gurney
,
K.
(
1999
).
The basal ganglia: A vertebrate solution to the selection problem?
Neuroscience
,
89
,
1009
1023
.
Rissman
,
J.
,
Eliassen
,
J. C.
, &
Blumstein
,
S. E.
(
2003
).
An event-related FMRI investigation of implicit semantic priming.
Journal of Cognitive Neuroscience
,
15
,
1160
1175
.
Robbins
,
T. W.
, &
Sahakian
,
B.
(
1983
).
Behavioral effects of psychomotor drugs: Clinical and neuropsychological implications.
In I. Creese (Ed.),
Stimulants: Neurochemical, behavioral and clinical perspectives.
New York
:
Raven Press
.
Robinson
,
G.
,
Blair
,
J.
, &
Cipolotti
,
L.
(
1998
).
Dynamic aphasia: An inability to select between competing verbal responses?
Brain
,
121
,
77
89
.
Robinson
,
G.
,
Shallice
,
T.
, &
Cipolotti
,
L.
(
2005
).
A failure of high level verbal response selection in progressive dynamic aphasia.
Cognitive Neuropsychology
,
22
,
661
694
.
Rossel
,
S. L.
,
Bullmore
,
E. T.
,
Williams
,
S. C. R.
, &
David
,
A. S.
(
2001
).
Brain activation during automatic and controlled processing of semantic relations: A priming experiment using lexical-decision.
Neuropsychologia
,
39
,
1167
1176
.
Schacter
,
D. L.
, &
Buckner
,
R. L.
(
1998
).
Priming and the brain.
Neuron
,
20
,
185
195
.
Signorini
,
M.
, &
Volpato
,
C.
(
2006
).
Action fluency in Parkinson's disease: A follow-up study.
Movement Disorder
,
21
,
467
472
.
Silveri
,
M. C.
, &
Misciagna
,
S.
(
2000
).
Language, memory, and the cerebellum.
Journal of Neurolinguistics
,
13
,
129
143
.
Siri
,
S.
,
Tettamanti
,
M.
,
Cappa
,
S. F.
,
DellaRosa
,
P.
,
Saccuman
,
C.
,
Scifo
,
P.
,
et al
(
2008
).
The neural substrate of naming events: Effects of processing demands but not of grammatical class.
Cerebral Cortex
,
18
,
171
177
.
Snyder
,
H. R.
, &
Munakata
,
Y.
(
2008
).
So many options, so little time: The roles of association and competition in underdetermined responding.
Psychonomic Bulletin & Review
,
15
,
1083
1088
.
Thompson-Schill
,
S. L.
, &
Botvinick
,
M. M.
(
2006
).
Resolving conflict: A response to Martin and Cheng.
Psychonomic Bulletin & Review
,
13
,
402
408
.
Thompson-Schill
,
S. L.
,
D'Esposito
,
M.
,
Aguirre
,
G. K.
, &
Farah
,
M. J.
(
1997
).
Role of left inferior prefrontal cortex in retrieval of semantic knowledge: A reevaluation.
Proceedings of the National Academy of Sciences, U.S.A.
,
94
,
14792
14797
.
Thompson-Schill
,
S. L.
,
Kan
,
I. P.
, &
Oliver
,
R. T.
(
2006
).
Neuroimaging of semantic memory.
In R. Cabeza & A. Kingstone (Eds.),
Handbook of functional neuroimaging
(2nd ed., pp.
149
190
).
Cambridge, MA
:
MIT Press
.
Thompson-Schill
,
S. L.
,
Swick
,
D.
,
Farah
,
M. J.
,
D'Esposito
,
M.
,
Kan
,
I. P.
, &
Knight
,
R. T.
(
1998
).
Verb generation in patients with focal frontal lesions: A neuropsychological test of neuroimaging findings.
Proceedings of the National Academy of Sciences, U.S.A.
,
95
,
15855
15860
.
Wagner
,
A. D.
,
Paré-Blagoev
,
E. J.
,
Clark
,
J.
, &
Poldrack
,
R. A.
(
2001
).
Recovering meaning: Left prefrontal cortex guides controlled semantic retrieval.
Neuron
,
31
,
329
338
.
Warburton
,
E.
,
Wise
,
R. J.
,
Price
,
C.
,
Weiller
,
C.
,
Hadar
,
U.
,
Ramsay
,
S.
,
et al
(
1996
).
Noun and verb retrieval by normal subjects. Studies with PET.
Brain
,
119
,
159
179
.
Yokoyama
,
S.
,
Miyamoto
,
T.
,
Riera
,
J.
,
Kim
,
J.
,
Akitsuki
,
Y.
,
Iwata
,
K.
,
et al
(
2006
).
Cortical mechanisms involved in the processing of verbs: An fMRI study.
Journal of Cognitive Neuroscience
,
18
,
1304
1313
.
Zhang
,
J. X.
,
Feng
,
C.
,
Fox
,
P. T.
,
Gao
,
J.
, &
Tan
,
L. H.
(
2004
).
Is the left inferior frontal gyrus a general mechanism for selection?
Neuroimage
,
23
,
595
603
.