Fast of Simple Perceptual Discriminations Reduces Activation in Working Memory and in High-level Auditory Regions

Luba Daikhin and Merav Ahissar Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021

Abstract ■ Introducing simple regularities facilitates learning tion induced a larger activation in frontoparietal areas known to of both simple and complex tasks. This facilitation may reflect be part of the working memory network. However, only the an implicit change in the strategies used to solve the task when condition with a reference showed fast learning, which was successful predictions regarding incoming stimuli can be accompanied by a reduction of activity in two regions: the left formed. We studied the modifications in brain activity associa- intraparietal area, involved in stimulus retention, and the pos- ted with fast perceptual learning based on regularity detection. terior superior-temporal area, involved in representing auditory We administered a two-tone frequency discrimination task and regularities. We propose that this joint reduction reflects a re- measured brain activation (fMRI) under two conditions: with duction in the need for online storage of the compared tones. and without a repeated reference tone. Although participants We further suggest that this change reflects an implicit strategic could not explicitly tell the difference between these two con- shift “backwards” from reliance mainly on working memory net- ditions, the introduced regularity affected both performance works in the “No-Reference” condition to increased reliance on and the pattern of brain activation. The “No-Reference” condi- detected regularities stored in high-level auditory networks. ■

INTRODUCTION rarely been studied, although it is probably crucial to The dynamics of perceptual learning, particularly its ini- subsequent learning dynamics (e.g., Ortiz & Wright, tial stages, are not well understood. Previous studies have 2009; Hawkey, Amitay, & Moore, 2004; Karni, Jezzard, focused on the specificity of learning to trained stimuli, Adams, Turner, & Ungerleider, 1995). which was shown to be consistent with the specificity One of the key features of the training procedure, par- of the sensory areas (Spang, Grimsen, Herzog, & Fahle, ticularly at the early training stages, is the consistency of 2010; Van Wassenhove & Nagarajan, 2007; Amitay, Hawkey, stimuli across consecutive trials. Consistent training with & Moore, 2005; Seitz & Watanabe, 2005; Demany & similar stimuli leads to fast, condition-specific (Cohen, Semal, 2002; Ahissar & Hochstein, 1993, 1996; Levi & Daikhin, & Ahissar, 2013) learning (e.g., Otto, Herzog, Polat, 1996; Karni & Sagi, 1991). However, such speci- Fahle, & Zhaoping, 2006), whereas training with a broad ficity mainly characterizes later stages of learning, when range of stimuli, whose sequence is not predictable, leads some expertise had been obtained (Jeter, Dosher, Liu, & to slow learning (Parkosadze, Otto, Malaniya, Kezeli, & Lu, 2010; Ahissar & Hochstein, 1997; Karni & Sagi, Herzog, 2008) if any (e.g., Kuai, Zhang, Klein, Levi, & 1993). Ahissar and Hochstein (Ahissar, Nahum, Nelken, Yu, 2005; Adini, Wilkonsky, Haspel, Tsodyks, & Sagi, & Hochstein, 2009; Ahissar & Hochstein, 1997, 2004) 2004; Yu, Klein, & Levi, 2004). A very clear example of suggested that when finer resolution is required, per- this dissociation was recently reported in the auditory ceptual learning may progress backwards along the per- modality for training on frequency (pitch) discrimination ceptual hierarchy from crude generalizing representations between sequentially presented tones. Whereas dis- to more local ones. This theory, termed the Reverse crimination between tones whose frequency was ran- Hierarchy Theory, posits that perceptual learning is not domly chosen from a broad frequency range improved limited to a specific brain site and progresses from high- slowly (within hundreds of trials), substantial and fast to lower-level areas with practice. Nevertheless, it does improvement was achieved when the first tone in a pair not address the brain mechanisms underlying the very had a fixed frequency (Nahum, Daikhin, Lubin, Cohen, & early stages of learning, when the task and its broad Ahissar, 2010). This rapid improvement was attributed to stimulus characteristics need to be sorted out. This initial the ability to form effective predictions for the incoming stage is typically short and difficult to track and hence has stimuli when training with stimuli that obeyed a simple regularity (Ahissar et al., 2009; Ahissar & Hochstein, 2004). Here we inquired whether the impact of intro- The Hebrew University of Jerusalem ducing simple regularities that facilitate learning, perhaps

© 2015 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 27:7, pp. 1308–1321 doi:10.1162/jocn_a_00786 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 by facilitating the “reverse hierarchy” process, is accom- initial stages of learning the regularity and then decrease panied by a detectable concurrent change in the pattern its activity with repetitions of this regularity (Karni et al., of brain activation. 1995, 1998), as long as the reference containing condi- Although serial discrimination is considered a simple tion is not interrupted. perceptual task, it requires two types of management To test these hypotheses, we measured both processes, both of which involve frontoparietal networks. and the BOLD response when participants performed a First, as in any new task (or situation), its basic structure simple perceptual two-tone frequency discrimination

in terms of neural representations should be set (Miller task. On the basis of the observations of Nahum, Daikhin, Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021 & Cohen 2001). Many studies suggest that this task- et al. (2010), participants in the current study were setting is implemented by high-level networks, which administered the following two conditions. In one, the include extensive frontal and parietal regions. These same tone was consistently presented in the first interval networks are largely general-purpose and form the of each trial. This regularity is known to be detected “task-set” for various tasks (and were hence termed “the quickly and yields fast and substantial improvement multiple-demand” network; Duncan, 2010; Duncan & (Cohen et al., 2013; Nahum, Daikhin, et al., 2010). In Owen, 2000). Second, task performance requires the the second condition, the same task and similar stimuli retention of the relevant value of the first stimulus in each (though drawn from a broader frequency range) were trial during the interstimulus interval and a comparison of used, but there was no cross-trial tone repetition. In this this value with that of the second stimulus. This retain- condition, participants’ improvement has been reported and-compare process is a working memory operation to be very slow and does not reach the same level of (e.g., Romo, Brody, Hernández, & Lemus, 1999). Such performance even after many practice sessions. operations were also shown to activate frontoparietal Wepresentedblocksofthesetwoconditionsinaninter- regions, which were thus termed the working memory leaved manner (3 blocks of one condition followed by network (Fedorenko, Behr, & Kanwisher, 2011; Koelsch 3 blocks of the other condition). Because the stimuli were et al., 2009; Baldo & Dronkers, 2006; Rainer, Asaad, & similar and the task was the same, participants were unaware Miller, 1998). The exact role of this network in the retain- of the switch in conditions. We asked which brain areas and-compare operation is still being debated. Previous were sensitive to the difference between the two conditions, studies have suggested that these working memory areas and activity in which brain areas was modified as a function both manage and store the task-relevant stimuli. However, of the rapid improvement we anticipated in the condition very recent studies (reviewed in Sreenivasan, Curtis, & involving a simple, easily detected stimulus regularity. D’Esposito, 2014) posit that the stimuli are stored in pos- terior sensory areas, and the role of the working memory network primarily involves task-related management. METHODS “ ” Additional related term is attentional resources, whose Participants recruitment when a task is generally more demanding also activate partially overalapping posterior-parietal regions Nineteen participants (age = 29 ± 5 years; 10 women) (Magen, Emmanouil, McMains, Kastner, & Treisman, 2009). took part in the study. Each of them performed fre- The behavioral observation that a simple regularity in quency discrimination and another task (not reported perceptual discriminations leads to fast perceptual learn- here) in the magnet (except one, who was administered ing, which is specific to the trained regularity (Cohen only the frequency discrimination task) and had an addi- et al., 2013; Nahum, Daikhin, et al., 2010), implies that tional anatomical scan at the end of the session. Before the load on management processes decreases. This de- entering the scanner, participants practiced a short version crease is expected because utilizing the regularity, that of the behavioral protocol that they performed during is, the repeated reference, leads to increased reliance scanning. Participants signed an informed consent form on the internal representation (of the reference), which and were paid for their participation. partially replaces the need to actively retain the first stimulus in each trial. We therefore hypothesized that a Experimental Procedure condition with no regularity would place a heavier load on management processes and hence would induce In the frequency discrimination task participants were a higher activation in the frontoparietal network. We presented with pairs of tones and were asked to decide further hypothesized that frontoparietal activity would (and respond with a right/left button press) which tone quickly decrease when effective practice with the regular- was higher. The tones were 50 msec long, the ISI was ity containing condition led to the formation of a reliable ∼600 msec, and the trial duration (onset to onset) was prediction of the expected stimuli, and discrimination 2 sec. We measured frequency discrimination under two would increasingly rely on this stored regularity. More- conditions in a single fMRI run: (1) No-Reference condi- over, we reasoned that we may be able to track the for- tion (No-Ref; see schematic illustration in Figure 1A) with mation of this auditory prediction in a high-level auditory no cross-trial consistency. In this condition, the first tone area. This area is expected to show high activity at the was chosen from a frequency range of 800–1200 Hz and

Daikhin and Ahissar 1309 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021

Figure 1. Frequency discrimination—Experimental design. (A) A schematic illustration of five trials of the No-Ref condition (left), which contained no cross-trial stimulus repetitions, and of the Ref-1st condition (right), in which a 1000-Hz reference tone appeared first on every trial. (B) Frequency differences between the two tones in each trial on the 180 trials of each condition (No-Ref, blue; Ref-1st, green). Vertical dotted lines illustrate the division of the sequences into blocks of 12 trials as presented in the scanner. The condition was switched after a triad of blocks. (C) A schematic illustration of blocks composing the experimental procedure inside the fMRI scanner (No-Ref, blue; Ref-1st, green). Gray denotes periods of rest. Three blocks of one condition were followed by three blocks of the other condition.

1310 Journal of Cognitive Neuroscience Volume 27, Number 7 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 the second tone was chosen according to the frequency a cubic spline interpolation was applied. A temporal high differences shown in Figure 1B (see description below). pass with three cycles/points and linear trend removal (2) Reference-1st condition (Ref-1st; see schematic illus- were used for baseline correction of the signal. The tration in Figure 1A), which employed the same proce- functional images collected were coregistered with the dure, but the first tone in each pair was always 1000 Hz. anatomical images. Anatomical images were then trans- In both conditions, we administered sequences of tone formed into the Talairach space. pairs, with an initially large (20%) frequency difference, The statistical evaluation was based on a least-squares

which got gradually smaller. The specific characteristics estimation using the general linear model (GLM). The Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021 of these sequences (shown in Figure 1B) were based design matrix was generated using a hemodynamic re- on the average frequency differences (of naive per- sponse function. The time course of the BOLD signal formers in their first assessment) obtained in an adaptive obtained during the task was initially modeled using version of these conditions, which converged to 80% two predictors (one predictor per condition) as illus- correct (see Nahum, Daikhin, et al., 2010). The purpose trated by the pattern of coloring in Figure 1C. To inspect of using the sequences that both converged to the same within-condition changes, that is, the difference between level of performance (80% correct) was intended to con- the first and third blocks within a condition triad, the trol for difficulty differences between the conditions. time course of the BOLD signal was remodeled using Notethat,inanadaptiveprotocol(similaraccuracyof a separate predictor for each block within the triad performance), the Ref-1st condition converges to much (i.e., 3 predictors per condition × 2 conditions). lower thresholds (Figure 1B). Each condition consisted Multisubject random effects GLM and repeated-measures of 180 trials, presented in 15 blocks of 12 trials each ANOVAs of beta values with Condition (No-Ref, Ref-1st) (24 sec per block separated by 9 sec of rest). Each con- and Block (First, Third) as within-participant factors dition was presented in three consecutive blocks and wereappliedtothedata.Thedatawerez-transformed was then switched to the other condition (order counter- before entering the random effects analyses. The results balanced across participants). Thus, a single run contained were corrected for multiple comparisons using a cluster- five triads of blocks of each condition (see Figure 1C). size limitation. Applying a cluster-level statistical thresh- Participants were typically unaware of the condition old estimator, a minimal cluster size was determined at the switch. Participants were asked to keep their eyes closed chosen significance level (see figures) for each volume map. throughout the entire measurement. RT and accuracy of We first identified cortical areas that were positively and performance were collected while participants performed significantly activated by the frequency discrimination thetaskinsidethefMRIscanner.Thesewereanalyzed task: (all conditions) > rest, random effects GLM contrast. using repeated-measures ANOVAs with Condition (No- The obtained map served as a mask for testing our Ref, Ref-1st) and Block (First, Third) as within-participant hypotheses (see Figure 2A). Using voxel-wise repeated- factors. measures ANOVA of beta values, we examined which brain areas were sensitive to the differences between the conditions. To test sensitivity to the task conditions fMRI Scanning Procedure while controlling for the behavioral difference, an ANCOVA Scanning was performed in a 3T scanner (Magnetom on the beta values obtained from each of the condition- TimTrio System 3.0 T (Tim (102 × 32) TQ) Erlangen, sensitive regions (separately for each region) was run Germany). For each participant, functional (T2*-weighted) with condition as within-participant factor and behav- and high-resolution anatomical reference data sets (T1- ioral gain (ACC(Ref-1st) − ACC(No-Ref)) as a covariate. weighted) were acquired. Functional measurements were To assess within-condition learning-related changes, obtained with a single EPI sequence with an echo time of we remodeled the data using a different predictor for 30 msec and a repetition time of 3000 msec. Acquisition of each of the three blocks within a triad, obtaining six the slices was arranged uniformly within the repetition predictors—three for each condition. We then compared time interval. The matrix acquired was 80 × 80 with a the beta values obtained in the first and third blocks by field of view of 240 cm, resulting in an in-plane resolution applying voxel-wise repeated-measures ANOVA on the of 3 × 3 mm. The slice thickness was 3 mm. Anatomical areas within the mask (see Figure 2A), separately for scans were measured with a 3-D gradient-echo with a 1 × each condition. We examined which areas consistently 1×1mmresolution. changed their activity from the first to the third block of within-condition block triads. To assess learning-related modifications during the entire session, we remodeled fMRI Data Analysis the data using a different predictor for each block, obtain- Anatomical and functional data were analyzed using the ing 30 predictors—15 for each condition. We then com- Brain Voyager QX Software package (The Netherlands). pared the beta values obtained in the third block of the Functional data were corrected for motion using a tri- first triad to those obtained in the third block of the last linear estimation and interpolation. To correct for the triad by applying voxel-wise repeated-measures ANOVA temporal offset between the slices acquired in one scan, on the areas within the mask (Figure 2A).

Daikhin and Ahissar 1311 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 Figure 2. Functional anatomy of the two-tone frequency discrimination task. (A) A whole-brain activation map of the frequency discrimination, all-conditions > rest contrast, was obtained by applying random effects GLM (corrected by cluster size at a p < .05 threshold). Significant t values Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021 for positive fit between the BOLD signal and the modeled conditions are presented using an orange–yellow color scale. N = 19. Lateral (top) and medial (middle) views of the representative inflated cortex are shown. The bottom row shows two transverse anatomical slices that demonstrate the involvement of the left BG (right) and cerebellum (left). (B) Whole- brain activation map of the auditory-stimuli > rest contrast obtained from a subgroup of participants (random effects GLM; n = 5; corrected by cluster size at a p < .01 threshold) during the auditory localizer task. Significant t values for positive fit between the BOLD signal and the modeled condition are presented using an orange–yellow color scale.

To compare the areas that we found with those reported in of 18 stimuli each. The blocks were separated by 15 sec literature, we calculated the distance in anatomical (1 mm3) of rest. Participants were requested to listen to the stim- voxels between the peak voxels of the regions reported in uli with their eyes closed. They were not asked to per- the literature and the areas found in our study. form any task. Figure 2B shows the obtained auditory area resulting from the contrast: stimuli > rest (GLM ran- dom effects). The center of mass of the obtained area is sim- Comparison to the Primary Auditory Cortex ilar to the areas identified as primary auditory cortex in the Because we used simple auditory stimuli and a basic literature (x +1,y +1,z + 2 from the peak voxel reported auditory discrimination task, we were interested in the in Lockwood et al., 1999; x − 5, y +1,z +1fromthepeak impact of the experimental conditions on the primary voxel reported in Binder et al., 2000). This area was used as a auditory cortex, which shows automatic responses to control ROI to compare beta values and average time auditory stimuli. To specifically compare our results to courses with the regions that were obtained in each of our the dynamics of the signal there, we ran an auditory loca- experimental questions. Importantly, ROI-based repeated- lizer on a subgroup of participants (n = 5). During the measures ANOVAs of beta values in the auditory area yielded localizer period, participants were presented with audi- no significant effects for the frequency discrimination task. tory stimuli with rich and varying spectral content but with no clear semantic association. These included white RESULTS noise, broad-band noise, pink noise, pitch shifts, intensity modulations, and sound effects (fading in, fading out, The Pattern of Activation Induced by Two-Tone tremolo, stretching [paulstretch], amplitude modulation Frequency Discrimination [wahwah], inversion). The stimulus duration was 1 sec. Figure 2A shows the map of brain areas positively activated The stimuli were consecutively presented in seven blocks by the conditions of the frequency discrimination task. The

1312 Journal of Cognitive Neuroscience Volume 27, Number 7 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 map shows involvement of auditory areas in the superior in processing linguistic information and in auditory–motor temporal gyri and sulci of both hemispheres. It also shows interface (Friederici, Kotz, Scott, & Obleser, 2010; Obleser the somatosensory and motor areas in the precentral and & Kotz, 2010; Friederici, Makuuchi, & Bahlmann, 2009; postcentral gyri of the left hemisphere together with pre- Obleser, Wise, Alex Dresner, & Scott, 2007; Hickok & motor areas associated with participants’ motor responses Poeppel, 2000, 2004; Davis & Johnsrude, 2003). The right (with their right hand) and planning. Additionally, it shows middle temporal region was previously associated with activation of the inferior prefrontal regions and parietal auditory processing and working memory for pitch

areas, evident mainly in the left hemisphere. Extensive (Johnsrude, Penhune, & Zatorre, 2000; Zatorre & Samson, Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021 involvement of the cerebellum and BG is also shown. 1991). The involvement of the additional areas in the right Unexpectedly, we also found activation of the visual areas, hemisphere was not predicted by our working memory as is visible in the medial view of both hemispheres, hypothesis. although participants’ eyes were closed throughout the Figure 3B shows the beta values (averaged across par- assessments. Subsequent analyses were based on this map. ticipants and blocks) obtained from each of the condition- Figure 2B shows the whole-brain activation map ob- sensitive regions. Beta values from the auditory cortex tained from the subgroup of five participants who were are also presented for comparison. The plot shows that presented with an auditory localizer stimuli in the scanner. the condition effect stems from higher beta values in This localizer was composed of a sequence of auditory the No-Ref condition. In contrast, the auditory cortex stimuli with rich and varying spectral content. The marked shows no difference between the beta values of the two area was significantly more activated during the auditory conditions, F(1, 18) = 0, p = .99, in spite of being highly stimulus presentation compared to rest, when there was activated by the task. Figure 3C shows the time courses no auditory stimulation ( p < .01, corrected by cluster size). of the BOLD signal, indicating that the No-Ref condition This area served as a control ROI for comparing beta values induced a larger BOLD signal. Again, this condition- and average time courses with the regions obtained from specific increase in activity was not found in the auditory analyzing frequency discrimination activations. cortex. Although we aimed for attaining equal levels of difficulty (and hence of general attentional resources) in the two Sensitivity to Task Condition—With and conditions and therefore used larger frequency differences Without Stimulus Regularity in the No-Ref condition (based on Nahum, Daikhin, et al., To assess which areas were differentially activated by the 2010; see Methods), this condition was still slightly more two conditions (Condition effect), we applied a voxel-wise difficult. Specifically, participants were less accurate (95 ± repeated-measures ANOVA on the beta values obtained 1% vs. 89 ± 2% correct for Ref-1st vs. No-Ref; repeated- from the contrast: all-conditions > rest, shown in Figure 2A. measures ANOVA, main effect of Condition: F(1, 18) = The comparison between the two task conditions re- 19.33, p < .001) and somewhat slower (479 ± 20 msec vs. vealed several areas that were differentially activated by 530 ± 22 msec for Ref-1st vs. No-Ref, repeated-measures the two conditions. These were mainly high-level areas ANOVA, main effect of Condition: F(1, 18) = 19.2, p < in the left hemisphere, as shown in Figure 3A: lateral pre- .001), although they were not asked to be quick (but frontal (L-supPrefrontal; −46, −1, 33; L-infPrefrontal; there was a fixed time interval of 1.4 sec between trials). −52, 0, 23), premotor (L-Premotor; −26, −16, 53), pos- The difference in activity patterns between these condi- terior intraparietal (L-intraParietal; −32, −60, 47; L-intra- tions may thus be attributed to this small, yet significant, Parietal-2; −40, −47, 44), superior parietal (L-supParietal; difference in the required attentional resources rather than −19, −74, 46). As shown in Figure 3A, several small areas a difference in the ability to form effective predictions. in the right hemisphere also showed differential sensitivity To control for this alternative account, we compared to the task conditions: middle temporal (R-midTemporal; the beta values of the two conditions obtained for each 48, −30, 1), medial premotor (R-medial-Premotor; 4, 4, 48), condition-sensitive region (Figure 3A) by regressing out and medial occipital (R-medial-Occipital; 6, 70, −21). the behavioral difference. We ran an ANCOVA on the This condition-sensitive increase of activity in the left beta values obtained from each of the condition-sensitive prefrontal and parietal areas was in line with our pre- regions (separately for each region) with Condition as diction, as these regions are known to be part of the the within-participant factor and Behavioral gain (ACC working memory network involved in managing the (Ref-1st) − ACC(No-Ref)) as the covariate. In the parietal retention of sounds (Prefrontal: x, y +6,z +3from areas, the difference between the conditions remained the peak voxel reported in Zatorre, Perry, Beckett, significant even when the behavioral difference was Westbury, & Evans, 1998; x, y +4,z − 7; x, y − 2, z + controlled for (L-intraparietal: F(1, 17) = 5.3, p = .03; 7 from the peak voxels reported in Gaab, Gaser, Zaehle, L-intraparietal-2: F(1, 17) = 4.3, p = .05; L-sup-parietal: Jancke, & Schlaug, 2003; x, y +2,z +7;x, y − 4, z + F(1, 17) = 11.2, p = .004). Similarly, the left premotor 6 from the peak voxels reported in Koelsch et al., 2009; area and the right temporal region retained the signifi- parietal areas include the peak voxels reported in these stud- cant difference between conditions, F(1, 17) = 7, p = ies). The premotor region has been reported to be involved .02, and F(1, 17) = 5, p = .04, respectively. However,

Daikhin and Ahissar 1313 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021

1314 Journal of Cognitive Neuroscience Volume 27, Number 7 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 the difference between the conditions decreased and ing voxel-wise repeated-measures ANOVAs on the areas became only marginally significant in the prefrontal areas within the mask (see Figure 2A), separately for each (L-sup-Prefrontal: F(1, 17) = 3.5, p = .08; L-inf-Prefrontal: condition. Figure 4B shows two regions that showed F(1, 17) = 3.2, p = .09). This reduction is in line with sensitivity to block in the Ref-1st condition. The No- previous reports of prefrontal sensitivity to task difficulty Ref condition is not shown because the comparison failed (Fuster, 2001; Grady et al., 1996). Areas in the right hemi- to reach significance for any of the activated areas (at the sphere were also sensitive to this control. The right medial- chosen significance level, p < .01, cluster-size corrected),

occipital area became only marginally condition-sensitive in line with the lack of significant behavioral improvement Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021 (R-medial-Occipital: F(1, 17) = 4.2, p = .06), and the right between consecutive blocks of this condition. medial-premotor area did not remain condition-sensitive The two regions that showed a main effect of Block (R-medial-Premotor: F(1, 17) = 1.7, p =.21). were located in the left hemisphere: in the intraparietal Taken together, the pattern of increased activity under area (L-intralParietal; −38, −45, 42) and in posterior the No-Ref condition indicates that this condition acti- superior temporal area (L-supTemporal; −49, −47, 13). vates working memory networks to a greater extent than The intraparietal area is associated with the storing of the Ref-1st condition, although the task was the same and information (Koelsch et al., 2009; Baldo & Dronkers, participants were unaware of the difference between the 2006), although it is probably not the site of storage itself conditions. The differences in brain activity, particularly (Sreenivasan et al., 2014; Magen et al., 2009). The pos- those related to the posterior parietal region, cannot be terior superior temporal region is associated with analysis attributed to a general difference in the overall atten- of temporal auditory structures at different levels of tional efforts required by these two conditions. complexity (Obleser & Kotz, 2010; Friederici et al., 2009; Davis & Johnsrude, 2003; Binder et al., 2000). Figure 4C and D shows a reduction in activity in these areas be- Fast Learning in the Regularity tween the first and third blocks. Beta values and time Containing Condition courses of activity for the auditory cortex are also pre- The behavioral advantage of the Ref-1st over the No-Ref sented. Here, in spite of high beta values and high condition was reflected in the different dynamics of per- BOLD signals, there was no significant effect of Block formance in the two conditions. Figure 4A shows the (ROI repeated-measures ANOVA, Block effect: F(1, 18) = average (cross-participant) accuracy in each block of each 1.1, p = .31), suggesting that this area is not part of the condition. As expected, in the Ref-1st condition (left fast “learning network” whose activity is modified across plot) performance improved quickly. Accuracy increased consecutive blocks of Ref-1st. between consecutive short blocks (12 trials each) of this The results described above only show the effects of condition (93 ± 1.7% in the first blocks of the block triad, within-triad learning. To assess the possible effects of vs. 97.6 ± 1% in the last blocks, t = −3.43, p = .003, in a learning during the entire session (across block triads), paired, two-tailed t test). However, this improvement was we compared the activity and the behavior in the third specific to this condition, that is, to the specific pattern of block of the first triad with that in the third block of stimuli, and was degraded whenever No-Ref blocks were the last triad. We remodeled the data using a different introduced. predictor for each block (see Methods) and applied a By contrast, performance in the No-Ref blocks (Figure 4A, voxel-wise repeated-measures ANOVA to the masked right plot) did not show significant improvement after voxels (Figure 2A) with Block (First third, Last third) and mild amounts of practice (88 ± 2 vs. 89.5 ± 2, t = Condition (No-Ref, Ref-1st) as within-participant factors. −1.18, p = .25, in a paired, two-tailed t test), in line with There was no significant difference in the measured brain previous findings of very slow improvement in this con- activity (no areas showed differential activity at the p <.05 dition (Nahum, Daikhin, et al., 2010). threshold, corrected by cluster size). Behavior did not To assess within-triad changes in brain activity, we improve during the session, and there was even a small remodeled the data using a different predictor for each of tendency for some accumulated fatigue (Ref-1st, 99.6% the three blocks within a triad, obtaining six predictors— vs. 95%: t = 1.46, p = .16; No-Ref, 97% vs. 90%: t = 2.1, three for each condition. We then compared the beta p = .05). There was no evidence of cross-triad learning in values obtained under the first and third blocks apply- any of the two conditions.

Figure 3. Sensitivity to task condition. (A) Voxel-wise repeated-measures ANOVA of the beta values within (all conditions) > rest contrast map—main effect of Condition (corrected by cluster size at a p < .05 threshold). (B) Beta values obtained from each of the condition-sensitive regions (averaged across participants and blocks) for the No-Ref (blue) and Ref-1st (green) conditions. Beta values from the primary auditory ROI are also presented for comparison. Error bars indicate cross-participant standard error. (C) Time courses of the BOLD signal obtained from each of these regions (averaged across participants and blocks) for the No-Ref (blue) and Ref-1st (green) conditions. Time courses of the BOLD signal from the primary auditory ROI are also presented for comparison. Talairach coordinates indicate the center of mass for each region. Error bars denote cross-participant standard error. Pink backgrounds denote the duration of a block.

Daikhin and Ahissar 1315 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021

Figure 4. Sensitivity to block. (A) Average performance accuracy for each of the blocks in the Ref-1st (left) and No-Ref (right) conditions. First blocks of each triad are marked by dashed bars, second blocks are marked by empty bars, and third blocks are marked by filled bars. Participants only showed fast, condition-specific improvement in the Ref-1st condition. (B) Voxel-wise repeated-measures ANOVA of the beta values within (all conditions) > rest contrast map; a main effect of Block (corrected by cluster size at a p < .01 threshold). Beta values are shown only for the Ref-1st condition because, consistent with the lack of behavioral improvement, no region showed a significant block effect in the first versus third block comparison under the No-Ref condition. (C) Beta values obtained from each of the block-sensitive regions in the Ref-1st condition for the first (dashed bars) and third (filled bars) blocks. Beta values from the auditory ROI are also presented for comparison. Error bars indicate cross- participant standard error. (D) Time courses of the BOLD signal obtained from each of the block-sensitive regions for the first (dashed) and the third (solid) blocks. Error bars indicate cross-participant standard error. Time courses of the BOLD signal from the auditory ROI are also presented for comparison. Talairach coordinates indicate the center of mass for each region. Pink backgrounds denote the duration of a block.

1316 Journal of Cognitive Neuroscience Volume 27, Number 7 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 DISCUSSION extreme case of roving, in which stimuli were randomly chosen from a flat distribution. Interestingly, this varia- We studied the dynamics of brain activation during the bility was sufficient to block the fast learning of the performance of a two-tone frequency discrimination task simple Ref-1st condition to the extent that performance in two conditions: with (Ref-1st) and without (No-Ref), did not improve between the first and last triad of this an easily detected regularity in the stimulation pattern condition. of consecutive trials. We conducted three ANOVAs that However, as expected, there was fast within-triad

tested (1) which areas were activated differentially under improvement in the Ref-1st condition. This improvement Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021 these two similar behavioral conditions and (2) which was specific to the simple predictable stimulation pattern areas modified their activity across consecutive blocks of this condition and was interrupted (performance was of the same condition (two separate ANOVAs for the degraded) by intervening No-Ref blocks. This interference two conditions, respectively). In addition, we tested was expected, because No-Ref blocks violate the expected potential impact of a behavioral difference between the pattern of stimulation that underlies the fast improve- two conditions on the activity in the condition-sensitive ment. The cross-block (first to third) behavioral improve- regions (ANCOVA results) as well as possible cross-triad ment was accompanied by modifications in two specific learning. The findings showed that participants were regions, namely, the left intraparietal area and the left typically unaware of the existence of the two different posterior superior temporal area. conditions. This is not surprising given the common behav- We interpret these results in the framework of the ioral task and trial structure and the similar range of stim- Reverse Hierarchy Theory, which suggests that successful uli (in the No-Ref condition 800–1200 Hz; in the Ref-1st detection of task-informative lower-level representations condition vs. 950–1050 Hz, except for the broader few enables a gradual reliance on lower-level populations first trials). (e.g., Ahissar et al., 2009). In other words, we propose We hypothesized that the condition effect would that the temporal region retains the detected auditory reveal different levels of activation within the working regularity whereas the intraparietal region controls this memory network, because the No-Ref condition, which retention. Auditory regularity was successfully detected contained no regularities, placed a heavier load on work- only in the Ref-1st condition. In this condition, the pop- ing memory processes. This is because online manage- ulation that best decodes the average frequency is a reli- ment of comparison and retention of stimuli was more able predictor in each trial, because this is the frequency demanding in the No-Ref condition (Cohen et al., 2013; of the first tone of every pair. In the No-Ref condition, the Nahum, Daikhin, et al., 2010). As hypothesized, the frequency within a trial could not be reliably predicted. condition-sensitive regions were mainly located in the Thus, when this simple regularity was detected, the left frontoparietal and premotor areas, which have been “managing effort” required from the intraparietal region associated with the working memory network for sound may have decreased. This claim of a division of labor is (Koelsch et al., 2009; Gaab et al., 2003; Zatorre et al., based on recent imaging studies that suggest that the 1998; premotor area–auditory–motor interface: Hickok high-fidelity representations of stimuli in working memory & Poeppel, 2000, 2004). Furthermore, increased reliance are kept in perceptual areas, whereas intraparietal regions on successful stimulus-specific predictions in the Ref-1st only “manage” retention efforts (reviewed in Sreenivasan condition perhaps also reduced the activity related to et al., 2014). Indeed, a study that aimed to assess this task-setting, because performance becomes more auto- question directly concluded that intraparietal regions are matic (Miller & Cohen, 2001). The lack of cross-triad not the site of storage itself but of the attentional re- learning in either of the two conditions suggests that this sources required for keeping online storage (Magen reduction cannot be explained as manifesting a general et al., 2009). decrease in difficulty and hence in the need to allocate An alternative account to the pattern of reduction of general attentional resources. activation relates to a general reduction in task difficulty, In contrast to the Ref-1st condition in which fast im- which perhaps was greater in the Ref-1st condition, provement was observed across consecutive short blocks, which was learned faster. This interpretation is unlikely. no such improvement was found in the No-Ref condition. Comparing the two conditions while controlling for the This was expected from previous studies using this dis- behavioral difference (ANCOVA results) did not eliminate crimination task (Nahum, Daikhin, et al., 2010) and other the condition effect in the posterior parietal region. It discrimination tasks when many stimuli were used with did, however, reduce the significance of the frontal no repeated pattern. In these studies, even with a more region in the condition effect, suggesting that for this limited range of stimuli, when several repeated references region, we cannot rule out a contribution of the small were used in a randomly chosen sequence (“roving con- difference in the overall difficulty of the two conditions. ditions”; Clarke, Grzeczkowski, Mast, Gauthier, & Herzog, Note, however, that participants were completely un- 2014; Herzog, Aberg, Frémaux, Gerstner, & Sprekeler, aware of the switch between conditions or a change in 2012; Parkosadze et al., 2008), improvement was either the effort they were required to allocate at different absent or small and very slow. Our No-Ref condition is an stages of the session. This reported introspection is in

Daikhin and Ahissar 1317 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 line with the lack of a general improvement or a general However, these two perspectives can be reconciled. change in activity during the session. Raviv et al.’s model does not take into account the reli- ability of the prior, which differs considerably between the two conditions: in the No-Ref condition, its reliability The Benefit of the Regularity—Integration is low, whereas under the Ref-1st condition, its reliability of Interpretations is high. The weight assigned to the prior should depend The cognitive literature attributes a unique role to the on its reliability. Studies have shown that people are

detection of regularities in sounds. For example, the sensitive to the reliability with which recent data indi- Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021 MMN ERP component (Näätänen, 1992) is considered cate the current state of the data (Nassar et al., 2012). an automatic response of the auditory cortex to a viola- Moreover, this reliability is also reflected in pupil diameter, tion of regularities (e.g., Näätänen, Paavilainen, Rinne, & implying that it is tightly linked with activity in attentional Alho, 2007; Picton, Alain, Otten, Ritter, & Achim, 2000). systems. This finding is consistent with the idea that the Our own interpretation of the fast improvement in Ref- activity in the intraparietal regions is sensitive to the reli- 1st in fact stresses its easily detected regularity (within ability of the prior. It decreases when the estimated reliabil- fewer than 10 trials), as described in a series of previous ity is increased, because the required attentional resources studies (Cohen et al., 2013; Oganian & Ahissar, 2012; can consequently be reduced. Thus, the decrease in intra- Nahum, Daikhin, et al., 2010). parietal activity may reflect the gradual switch of reliance For example, Nahum, Daikhin, et al. (2010) interpreted from the externally presented stimulus to the temporally this improvement as stemming from a shift from an initial stored prior, as its estimated reliability is increased. working memory-based comparison of the stimuli pre- This interpretation suggests that learning the reliable sented in the two intervals of the trial to a comparison prior should be reflected in a decrease of activity in the with an internal representation of the constant reference. posterior superior temporal region. A decrease of activity This interpretation was also based on monkey studies is a marker of the initial, fast, but condition-specific stage (reviewed in Romo & de Lafuente, 2012) that found that, of learning (e.g., Karni et al., 1995, 1998). The mecha- in the No-Ref condition, well-trained monkeys compare nism underlying this “habituation-like” pattern is not well stimuli and activate working memory areas (premotor, understood and may reflect a match between the suc- prefrontal, parietal), thus producing “delayed activity.” cessfully detected prior and the incoming first tone, However, when trained on the Ref-1st condition, mon- which would lead to a stimulus-specific suppression, keys do not compare stimuli online and do not produce whereas the failure of such a match (mismatch) in the delayed activity in higher level areas (Romo & Salinas, No-Ref condition would not yield suppression. 2003; Brody, Hernández, Zainos, Lemus, & Romo, 2002; The notion that this area is involved in auditory regu- Romo et al., 1999; Hernández, Salinas, García, & Romo, larity detection and, perhaps, in the storage of detected 1997). Rather, they compare the second stimulus to the regularities is consistent with previous studies that asso- previously trained fixed reference stimulus maintained in ciated this area with the analysis of temporal structures at their long-term memory, although no neural signature different levels of complexity (peak voxel is located at: was found for the storage of this trained reference x +2,y +3,z + 4 from the peak voxel reported in Davis stimulus. & Johnsrude, 2003; x, y +5,z from the peak voxel However, we also found that participants keep track reported in Friederici et al., 2009; x, y +4,z + 5 from and are heavily affected by the statistics of the experi- the peak voxel reported in Obleser & Kotz, 2010). ment even when it contains no regularities, as in the case By extension, with additional, cross-day learning, the of No-Ref. Raviv, Ahissar, and Loewenstein (2012) sug- response to this prior should gradually increase, as was gested a simple model (inspired by Bayesian rules), reported by Karni et al., following several weeks of prac- accounting for these effects. The model proposes that tice on a simple finger sequencing task (Karni et al., 1995, rather than comparing the two stimuli within a trial, lis- 1998). It may evolve into an area of expertise that gradu- teners compare the second tone to a combined repre- ally stores more elaborate priors. sentation of the frequency of the first tone (which is noisy because of the working memory noise added No Indication of Regularity Learning in the during the retention interval) and the prior. The prior Auditory Cortex in this case is simply the average frequency of the first tone on previous trials. According to Raviv et al.’s model, As expected, our frequency discrimination paradigm acti- the same mechanism could have been automatically vated the primary auditory cortex. However, its response implemented in both the No-Ref and Ref-1st conditions. did not differ between conditions or blocks. Importantly, Indeed, the same simple model also accounts for partic- although the Ref-1st condition contained a smaller range ipants’ behavior when a reference is introduced (Raviv, of tones than the No-Ref condition, which could have Lieder, Loewenstein, & Ahissar, 2014), suggesting that, induced greater in areas with narrow fre- in spite of its substantial behavioral advantage, Ref-1st quency tuning curves, no such reduction was found. This may not be a qualitatively different condition. observation is congruent with observations both at the

1318 Journal of Cognitive Neuroscience Volume 27, Number 7 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 level of single neurons (Kajikawa, de La Mothe, Blumell, by stimulus variability. & Psychophysics, 67, – & Hackett, 2005; Recanzone, Guard, & Phan, 2000; Ehret 691 698. Baldo, J. V., & Dronkers, N. F. (2006). The role of inferior & Schreiner, 1997; Howard et al., 1996) and with ERP parietal and inferior frontal cortex in working memory. (Daikhin & Ahissar, 2012) reports of broad adaptation , 20, 529–538. tuning. Nevertheless, we cannot exclude the possibility Binder, J. R., Frost, J. A., Hammeke, T. A., Bellgoan, P. S. F., that a different experimental design specifically aimed Springer, J. A., Kaufman, J. N., et al. (2000). temporal at measuring the working memory-based retention and lobe activation by speech and nonspeech sounds. Cerebral Cortex, 10, 512–528. comparison processes would have revealed working Brechmann, A., Gaschler-Markefski, B., Sohr, M., Yoneda, K., Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021 memory-related activity in the auditory cortex (e.g. Kaulisch, T., & Scheich, H. (2007). Working memory Brechmann et al., 2007; Zatorre & Samson, 1991). specific activity in auditory cortex: Potential correlates of In conclusion, we found that the degree of regularity sequential processing and maintenance. Cerebral Cortex, – affects the pattern of brain activity even in simple dis- 17, 2544 2552. Brody, C. D., Hernández, A., Zainos, A., Lemus, L., & Romo, R. crimination tasks. The frontoparietal network involved (2002). Analysing neuronal correlates of the comparison of in working memory is activated to a greater extent when two sequentially presented sensory stimuli. Philosophical no regularity is introduced. When a simple regularity is Transactions of the Royal Society of London, Series B, introduced, an effective prior is formed, leading to re- Biological Sciences, 357, 1843–1850. duced activity in a region that controls retention (left Clarke, A. M., Grzeczkowski, L., Mast, F. W., Gauthier, I., & Herzog, M. H. (2014). Deleterious effects of roving on intraparietal) and in a region that stores this effective learned tasks. Vision Research, 99, 88–92. prior (posterior superior-temporal region). We posit that Cohen, Y., Daikhin, L., & Ahissar, M. (2013). Perceptual learning this orchestrated modification in brain activity reflects a is specific to the trained structure of information. Journal of quick and implicit shift “backwards” when a reliable prior Cognitive Neuroscience, 25, 2047–2060. is detected. In other words, task performance relies more Daikhin, L., & Ahissar, M. (2012). Responses to deviants are modulated by subthreshold variability of the standard. on posterior networks that store effective priors than on Psychophysiology, 49, 31–42. laborious online computations. Davis, M. H., & Johnsrude, I. S. (2003). Hierarchical processing in spoken language comprehension. The Journal of Neuroscience, 23, 3423–3431. Demany, L., & Semal, C. (2002). Learning to perceive pitch Acknowledgments differences. The Journal of the Acoustical Society of America, 111, 1377–1388. This work was supported by ISF grant 616/11 and the HUJI and Duncan, J. (2010). The multiple-demand (MD) system of the EPFL Brain Collaboration. In addition, we thank Avi Mendelson brain: Mental programs for intelligent behaviour. and Tanya Orlov for their constructive comments and help with Trends in Cognitive Sciences, 14, 172–179. data analysis. Duncan, J., & Owen, A. M. (2000). Common regions of the Reprint requests should be sent to Merav Ahissar, Department human frontal lobe recruited by diverse cognitive demands. of and the Edmond and Lily Safra Center for Brain Trends in Neurosciences, 23, 475–483. Sciences, Hebrew University of Jerusalem, Israel 91905, or via Ehret, G., & Schreiner, C. E. (1997). Frequency resolution and e-mail: [email protected]. spectral integration (critical band analysis) in single units of the cat primary auditory cortex. Journal of Comparative . A, Sensory, Neural, and Behavioral Physiology, 181, 635–650. REFERENCES Fedorenko, E., Behr, M. K., & Kanwisher, N. (2011). Functional specificity for high-level linguistic processing in the human Adini, Y., Wilkonsky, A., Haspel, R., Tsodyks, M., & Sagi, D. brain. Proceedings of the National Academy of Sciences, (2004). Perceptual learning in contrast discrimination: U.S.A., 108, 16428–16433. The effect of contrast uncertainty. Journal of Vision, 4, Friederici, A. D., Kotz, S. A., Scott, S. K., & Obleser, J. (2010). 993–1005. Disentangling syntax and intelligibility in auditory language Ahissar, M., & Hochstein, S. (1993). Attentional control of early comprehension. Human Brain Mapping, 31, 448–457. perceptual learning. Proceedings of the National Academy of Friederici, A. D., Makuuchi, M., & Bahlmann, J. (2009). The role Sciences, U.S.A., 90, 5718–5722. of the posterior superior temporal cortex in sentence Ahissar, M., & Hochstein, S. (1996). Learning pop-out comprehension. NeuroReport, 20, 563–568. detection: Specificities to stimulus characteristics. Vision Fuster, J. M. (2001). The prefrontal cortex—An update: Time is Research, 36, 3487–3500. of the essence. Neuron, 30, 319–333. Ahissar, M., & Hochstein, S. (1997). Task difficulty and the Gaab, N., Gaser, C., Zaehle, T., Jancke, L., & Schlaug, G. (2003). specificity of perceptual learning. Nature, 387, 401–406. Functional anatomy of pitch memory—An fMRI study with Ahissar, M., & Hochstein, S. (2004). The reverse hierarchy sparse temporal sampling. Neuroimage, 19, 1417–1426. theory of visual perceptual learning. Trends in Cognitive Grady, C. L., Horwitz, B., Pietrini, P., Mentis, M. J., Ungerleider, Sciences, 8, 457–464. L. G., Rapoport, S. I., et al. (1996). Effect of task difficulty on Ahissar, M., Nahum, M., Nelken, I., & Hochstein, S. (2009). cerebral blood flow during perceptual matching of faces. Reverse hierarchies and sensory learning. Philosophical Human Brain Mapping, 4, 227–239. Transactions of the Royal Society of London, Series B, Hawkey, D. J. C., Amitay, S., & Moore, D. R. (2004). Early and Biological Sciences, 364, 285–299. rapid perceptual learning. Nature Neuroscience, 7, 1055–1056. Amitay, S., Hawkey, D. J. C., & Moore, D. R. (2005). Hernández, A., Salinas, E., García, R., & Romo, R. (1997). Auditory frequency discrimination learning is affected Discrimination in the of flutter: New psychophysical

Daikhin and Ahissar 1319 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 measurements in monkeys. The Journal of Neuroscience, Näätänen, R., Paavilainen, P., Rinne, T., & Alho, K. (2007). The 17, 6391–6400. mismatch negativity (MMN) in basic research of central Herzog, M. H., Aberg, K. C., Frémaux, N., Gerstner, W., & auditory processing: A review. Clinical Neurophysiology, Sprekeler, H. (2012). Perceptual learning, roving and the 118, 2544–2590. unsupervised bias. Vision Research, 61, 95–99. Nahum, M., Daikhin, L., Lubin, Y., Cohen, Y., & Ahissar, M. Hickok, G., & Poeppel, D. (2000). Towards a functional (2010). From comparison to classification: A cortical tool for neuroanatomy of speech perception. Trends in Cognitive boosting perception. The Journal of Neuroscience, 30, Sciences, 4, 131–138. 1128–1136. Hickok, G., & Poeppel, D. (2004). Dorsal and ventral streams: Nahum, M., Nelken, I., & Ahissar, M. (2010). Stimulus A framework for understanding aspects of the functional uncertainty and perceptual learning: Similar principles Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021 anatomy of language. , 92, 67–99. govern auditory and visual learning. Vision Research, 50, Howard, M. A., Volkov, I. O., Abbas, P. J., Damasio, H., 391–401. Ollendieck, M. C., & Granner, M. A. (1996). A chronic Nassar, M., Rumsey, K. M., Wilson, R. C., Parikh, K., Heasly, B., & microelectrode investigation of the tonotopic Gold, J. I. (2012). Rational regulation of learning dynamics by organization of human auditory cortex. Brain Research, pupil–linked arousal systems. Nature Neuroscience, 15, 724, 260–264. 1040–1046. Jeter, P. E., Dosher, B. A., Liu, S.-H., & Lu, Z.-L. (2010). Obleser, J., & Kotz, S. A. (2010). Expectancy constraints in Specificity of perceptual learning increases with increased degraded speech modulate the language comprehension training. Vision Research, 50, 1928–1940. network. Cerebral Cortex, 20, 633–640. Johnsrude, I. S., Penhune, V. B., & Zatorre, R. J. (2000). Obleser, J., Wise, R. J. S., Alex Dresner, M., & Scott, S. K. (2007). Functional specificity in the right human auditory cortex for Functional integration across brain regions improves speech perceiving pitch direction. Brain: A Journal of Neurology, perception under adverse listening conditions. The Journal 123, 155–163. of Neuroscience, 27, 2283–2289. Kajikawa, Y., de La Mothe, L., Blumell, S., & Hackett, T. A. Oganian, Y., & Ahissar, M. (2012). Poor anchoring limits (2005). A comparison of neuron response properties in dyslexics’ perceptual, memory, and skills. areas A1 and CM of the marmoset monkey auditory cortex: Neuropsychologia, 50, 1895–1905. Tones and broadband noise. Journal of Neurophysiology, Ortiz, J. A., & Wright, B. A. (2009). Contributions of procedure 93, 22–34. and stimulus learning to early, rapid perceptual improvements. Karni, A., Jezzard, P., Adams, M., Turner, R., & Ungerleider, L. Journal of : Human Perception (1995). Functional MRI evidence for adult motor cortex and Performance, 35, 188–194. plasticity during motor skill learning. Nature, 377, Otto, T. U., Herzog, M. H., Fahle, M., & Zhaoping, L. (2006). 155–158. Perceptual learning with spatial uncertainties. Vision Karni, A., Meyer, G., Rey-Hipolito, C., Jezzard, P., Adams, M. M., Research, 46, 3223–3233. Turner, R., et al. (1998). The acquisition of skilled motor Parkosadze, K., Otto, T. U., Malaniya, M., Kezeli, A., & performance: Fast and slow experience-driven changes in Herzog, M. H. (2008). Perceptual learning of bisection primary motor cortex. Proceedings of the National Academy stimuli under roving: Slow and largely specific. Journal of of Sciences, U.S.A., 95, 861–868. Vision, 8, 1–8. Karni, A., & Sagi, D. (1991). Where practice makes perfect in Picton, T. W., Alain, C., Otten, L., Ritter, W., & Achim, A. (2000). texture discrimination: Evidence for primary Mismatch negativity: Different water in the same river. plasticity. Proceedings of the National Academy of Sciences, Audiology and Neuro-Otology, 5, 111–139. U.S.A., 88, 4966–4970. Rainer, G., Asaad, W. F., & Miller, E. K. (1998). Selective Karni, A., & Sagi, D. (1993). The time course of learning a visual representation of relevant information by neurons in the skill. Nature, 365, 250–252. primate prefrontal cortex. Nature, 393, 577–579. Koelsch, S., Schulze, K., Sammler, D., Fritz, T., Müller, K., & Raviv, O., Ahissar, M., & Loewenstein, Y. (2012). How recent Gruber, O. (2009). Functional architecture of verbal and tonal history affects perception: The normative approach and its working memory: An fMRI study. Human Brain Mapping, heuristic approximation. PLoS Computational Biology, 8, 30, 859–873. e1002731. Kuai, S.-G., Zhang, J.-Y., Klein, S. A., Levi, D. M., & Yu, C. Raviv, O., Lieder, I., Loewenstein, Y., & Ahissar, M. (2014). (2005). The essential role of stimulus temporal patterning Contradictory behavioral biases result from the influence of in enabling perceptual learning. Nature Neuroscience, 8, past stimuli on perception. PLOS Computational Biology, 1497–1499. 10, e1003948. Levi, D. M., & Polat, U. (1996). Neural plasticity in adults Recanzone,G.H.,Guard,D.C.,&Phan,M.L.(2000). with . Proceedings of the National Academy of Frequency and intensity response properties of Sciences, U.S.A., 93, 6830–6834. single neurons in the auditory cortex of the behaving Lockwood, A. H., Salvi, R. J., Coad, M. L., Arnold, S. A., Wack, macaque monkey. Journal of Neurophysiology, 83, D. S., Murphy, B. W., et al. (1999). The functional anatomy 2315–2331. of the normal human auditory system: Responses to 0.5 Romo, R., Brody, C. D., Hernández, A., & Lemus, L. (1999). and 4.0 kHz tones at varied intensities. Cerebral Cortex, Neuronal correlates of parametric working memory in the 9, 65–76. prefrontal cortex. Nature, 399, 470–473. Magen, H., Emmanouil, T.-A., McMains, S. A., Kastner, S., & Romo, R., & de Lafuente, V. (2012). Conversion of Treisman, A. (2009). Attentional demands predict short-term sensory signals into perceptual decisions. Progress in memory load response in posterior parietal cortex. Neurobiology, 103, 41–75. Neuropsychologia, 47, 1790–1798. Romo, R., & Salinas, E. (2003). Flutter discrimination: Neural Miller, E. K., & Cohen, J. D. (2001). An integrative theory of codes, perception, memory and decision making. Nature prefrontal cortex function. Annual Review of Neuroscience, Reviews Neuroscience, 4, 203–218. 24, 167–202. Seitz, A., & Watanabe, T. (2005). A unified model for Näätänen, R. (1992). and brain function. Hillsdale, perceptual learning. Trends in Cognitive Sciences, 9, NJ: Erlbaum. 329–334.

1320 Journal of Cognitive Neuroscience Volume 27, Number 7 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021 Spang, K., Grimsen, C., Herzog, M. H., & Fahle, M. (2010). Yu, C., Klein, S. A., & Levi, D. M. (2004). Perceptual learning in Orientation specificity of learning vernier discriminations. contrast discrimination and the (minimal) role of context. Vision Research, 50, 479–485. Journal of Vision, 4, 169–182. Sreenivasan, K. K., Curtis, C. E., & D’Esposito, M. (2014). Zatorre, R. J., Perry, D. W., Beckett, C. A., Westbury, C. F., & Revisiting the role of persistent neural activity during Evans, A. C. (1998). Functional anatomy of musical working memory. Trends in Cognitive Sciences, 18, processing in listeners with absolute pitch and relative pitch. 82–89. Proceedings of the National Academy of Sciences, U.S.A., Van Wassenhove, V., & Nagarajan, S. S. (2007). 95, 3172–3177. Auditory cortical plasticity in learning to discriminate Zatorre, R. J., & Samson, S. (1991). Role of the right temporal modulation rate. The Journal of Neuroscience, 27, neocortex in retention of pitch in auditory short-term Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/27/7/1308/1783003/jocn_a_00786.pdf by MIT Libraries user on 17 May 2021 2663–2672. memory. Brain: A Journal of Neurology, 114, 2403–2417.

Daikhin and Ahissar 1321 Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/jocn_a_00786 by guest on 28 September 2021