Object identification leads to a conceptual broadening of object representations in lateral prefrontal cortex

Abstract

Recent experience identifying objects leads to later improvements in both speed and accuracy ("repetition priming"), along with simultaneous reductions of neural activity ("repetition suppression"). A popular interpretation of these joint behavioral and neural phenomena is that object representations become perceptually "sharper" with stimulus repetition, eliminating cells that are poorly stimulus-selective and responsive and reducing support for competing representations downstream. Here, we test this hypothesis in an fMRI-adaptation experiment using pictures of objects. Prior to fMRI, participants repeatedly named a set of object pictures. During fMRI, participants viewed adaptation sequences composed of rapidly repeated objects (3-6 repetitions over several seconds) that were either named previously or that were new for the fMRI session, followed by single "deviant" object pictures used to measure recovery from adaptation and that shared a relationship to the adapted picture (a different exemplar of the same object, a conceptual associate, or an unrelated picture). Effects of adaptation and recovery were found throughout visually responsive brain regions. Occipitotemporal cortical regions displayed repetition suppression to previously named relative to new adapters but failed to exhibit pronounced changes in neural tuning. In contrast, changes in the slope of the recovery curves were found in the left lateral prefrontal cortex: Greater residual adaptation was observed to exemplar stimuli and conceptual associates following previously named adapting stimuli, consistent with greater rather than reduced neural overlap among representations of conceptually related objects. Furthermore, this change in neural tuning was directly related to the proportion of conceptual errors made by participants in the naming sessions pre- and post-fMRI, establishing that the experience-dependent conceptual broadening of object representations seen in fMRI is also manifest in behavior. In a follow-up behavioral experiment, we further show that recent naming experience leads to greater semantic priming when using the previously named pictures as briefly presented primes. Taken together, our results fail to support perceptual sharpening as the primary mediator between repetition suppression and behavioral priming at durations typically used to study priming and instead highlight an experience-dependent broadening of conceptual representations. We suggest that alternative mechanisms, such as increases in neural synchronization, are more promising in explaining priming in the face of repetition suppression.

Trial registration: ClinicalTrials.gov NCT00001360.

Keywords: Priming; Repetition suppression; Semantic memory; Sharpening; Synchrony; Tuning curve.

Published by Elsevier Ltd.

Figures

Figure 1. Using fMRI adaptation to measure neural tuning preferences and alterations in tuning with experience

(A) The BOLD response within stimulus-responsive fMRI voxels is temporarily reduced as the same image (“anchor”) is repeated several times (3-6) at a rate of 1 per second. These temporary decreases, referred to as “adaptation”, are thought to be cell- and synapse-specific. (B) By this logic, tuning preferences can then be inferred in individual fMRI voxels by the level of the recovered response to a new stimulus that shares a relationship with the adapted stimulus (e.g. visual form, conceptual, etc.). A greater recovered response is thought to indicate less neural representational overlap (i.e. fewer cells in common), whereas a weaker recovered response indicates a higher degree of overlap. (C) Changes in tuning preferences can be inferred by a change in the pattern of recovery from adaptation. If adapting to the picture of a cow (as in A) initially yields a “conceptual” recovery curve (retaining adaptation to identity, exemplar, and semantically related images), sharpening would be reflected as a shift in the curve toward a more image-selective curve (retaining adaptation only to an identical image and fully recovered to the other conditions) (i.e. shifting from “high overlap” toward “low overlap”). Conceptual broadening would be reflected as a shift in the opposite direction (from low overlap toward high overlap).

Figure 2. Trial structure of adaptation sequences in Experiment 1

Anchors were repeated at a rate of 1 per second (stimulus duration=200 ms), anywhere from 3 to 6 repetitions, and were followed immediately by a single deviant picture used to assess recovery from adaptation. Phase-scrambled baseline images of the animal pictures used in the experiment and pictures of man-made objects were randomly interleaved between adaptation sequences. Participants were instructed to press a response button to the man-made object pictures but were told to attend to all pictures. There were no additional delays or gaps in the stimulus displays, such that a single image of one of these types (anchor, deviant, baseline, man-made object) was being presented each second (duration of 200 ms followed by fixation for 800 ms).

Figure 3. Experimental design for Experiment 1

Repeated anchor stimuli (animals) were either named in the pre-fMRI picture naming session (Old) or were new for the fMRI session (New). Old and New anchors were counterbalanced across participants, and lists were matched for item properties and conceptual category membership. The deviant stimuli were all new for the fMRI experiment and shared one of three relationships with the anchor pictures: a different exemplar picture of the same type of animal, a semantic associate, or an unrelated picture. Anchor type (Old, New) was fully crossed with deviant condition (Exemplar, Semantic, Unrelated) for a total of 6 trial types.

Figure 4. Repetition priming effects in picture naming

(A) Mean response times to correctly named pictures in the pre-fMRI naming session decreased as a function of repetition number (1–5), while naming accuracy simultaneously increased (see text for description). (B) Repetition priming was assessed and found to be present in the post-fMRI picture naming session. The response times (and accuracies) to New and Old pictures in the post-fMRI session were comparable to those observed to the 1st and 5th repetitions, respectively, in the pre-fMRI session, suggesting that fatigue over the course of the three testing sessions was not an issue. Errors bars in both A and B depict the standard error of the mean (SE).

Figure 5. Regions showing significant effects of adaptation and recovery from adaptation

Voxels exhibiting both adaptation (First Anchor > Last Anchor) and recovery from adaptation (Last Anchor P<.05, and voxels showing both effects were retained and used to constrain later analyses of repetition suppression and changes in neural tuning.

Figure 6. Repetition suppression for previously named anchors in bilateral occipitotemporal cortex

(A) Repetition suppression was found bilaterally in occipitotemporal cortex by comparing New and Old anchor stimuli in the adaptation sequences, combining First, Middle, and Last conditions in a weighted contrast (New > Old). The effect in right occipitotemporal cortex survived multiple comparisons correction (P<.05), while the effect on the left showed a statistical trend (P<.1). Beta weights are shown separately for the New and Old anchors in units of percent signal change, averaged across participants and all voxels in the two ROIs (combined). (B) Shape of the recovery curves following New and Old anchors is shown separately for the two occipitotemporal ROIs using the beta weights that have been rescaled between 0 and 1 in each voxel for each participant, displayed to start from the same value in the identical condition to facilitate comparison of curve shape. Slope estimates calculated over the Exemplar through the Unrelated conditions (deviant levels 1-3) are displayed for each curve as thick dashed lines. Line plots have been used rather than bar plots for trends across the deviant conditions, although it is important to keep in mind that the deviant conditions along the x-axes are on an ordinal rather than a cardinal numerical scale. Little or no change in tuning was observed, with image-selective tuning (adaptation remaining only for an identical image) observed for both New and Old anchors in the ROI-averaged data.

Figure 7. Conceptual broadening of object representations in the left lateral prefrontal cortex

The slope of the recovery curve (calculated for the rescaled beta weights using the Exemplar, Semantic, and Unrelated conditions) was more positive following Old than following New anchors for a region in the left lateral prefrontal cortex, starting in the inferior frontal junction and extending anteriorly along the inferior frontal sulcus. Shown in the right panels is the conjunction of significant slope differences with overall effects of adaptation and recovery (all effects corrected to P<.05 prior to conjunction). Rescaled beta weights are shown to the left for the left prefrontal ROI (conventions as Figure 6), with slope estimates displayed as thick dashed lines.

Figure 8. Incidence of picture naming error types in the pre- and post-fMRI sessions

(Top) Error types in the pre- and post-fMRI picture naming sessions, shown by session (1st repetition in the pre-fMRI session, 2nd and 3rd repetitions combined, 4th and 5th repetitions combined, and post-fMRI session), and normalized by the total number of trials in each condition to yield the average error rate. Error types included omissions (O), visual/semantic errors (V/S), visual/semantic perseverations (V/S+P), and Other (see text for definitions/examples). (Bottom) Same data shown normalized instead by the total number of errors rather than the total number of trials.

Figure 9. Conceptual broadening in left lateral prefrontal cortex predicts the proportion of conceptual naming errors

Scatterplot across participants (N=18) of the relationship between the overall proportion of conceptual naming errors in all picture naming sessions (pooling V/S and V/S+P types) and the change in the slopes of the recovery curves in the left lateral prefrontal ROI (slope Old - slope New). A greater proportion of conceptual naming errors was significantly associated with a more positive slope change (i.e. a shift to more conceptual tuning).

Figure 10. Experimental design for Experiment 2

(A) Prime type of Old (previously named) versus New was crossed with prime-probe relatedness (Related semantically versus Unrelated). Each condition occurred with equal frequency during the course of the experiment. (B) Timing of events for an individual trial in the semantic priming session involved fixation (500 ms), followed by a briefly presented prime picture (Old or New) for 100 ms with backward masking (50 ms), a blank screen for 200 ms, and finally the probe picture to be named (200 ms). Probe pictures were either semantically related or unrelated to the prime picture.

Figure 11. Semantic priming increases for previously named primes

A scatterplot across participants (N=16) shows the positive relationship between the magnitude of repetition priming (in units of effect size) observed in the post-semantic-priming session and the change in semantic priming, or SP (Unrelated-Related probes), following previously named primes (Old) versus primes that were new for the semantic priming session (New) (SP Old - SP New). A box with dashed outline highlights the participants (N=9) with repetition priming magnitudes that were significant at P<.05.

Figure 12. Semantic priming for participants with significant repetition priming in the post-semantic-priming session

Response times and semantic priming effects are shown for the participants who exhibited significant repetition priming in the post-semantic-priming session (N=9, highlighted by the dashed box in Figure 11). Mean response times to the four conditions are shown in the left panel, and semantic priming effects (Unrelated-Related probe response times) across participants are shown separately for New versus Old primes in the right panel. Errors bars depict the standard error of the mean (SE) for each condition.