Rapid Cortical Plasticity Induced by Active Learning of Novel Words in Human Adults

Whether short-term learning of new words can induce rapid changes in cortical areas involved in distributed neural representation of the lexicon is a hotly debated topic. To answer this question, we examined magnetoencephalographic phase-locked responses elicited in the cerebral cortex by passive presentation of eight novel pseudowords before and immediately after an operant conditioning task. This procedure forced participants to perform an active search for unique meaning of four word-forms that referred to movements of their own body parts. While familiarization with novel word-forms led to bilateral repetition suppression of cortical responses to all eight pseudowords, these reduced responses became more selectively tuned towards newly learned action words in the left hemisphere. Our results suggest that stimulus repetition and active learning of semantic association have separable effects on cortical activity. They also evidence rapid plastic changes in cortical representations of meaningful auditory word-forms after active learning.

presentation of two pseudoword types before and after an associative learning procedure. Since 98 consistent association might potentially affect both phonological and semantic aspects of 99 lexicality, cortical changes, to be considered as truly "semantic", are assumed to occur in the 100 higher-tier cortical areas that underlie semantic processing of real words. Such changes are in fact 101 observed in these regions after consolidation (Davis & Gaskell, 2009).

102
To the best of our knowledge, there are only two EEG studies dedicated to associative 103 learning of novel auditory words. In one of them (Fargier et al., 2014), ERPs to passive 104 pseudoword presentation before and after learning were compared for pseudowords associated 105 during learning either with short movies of reaching-and-grasping movements or with abstract 106 visual images. As a result of learning, ERP started to differentiate both types of pseudowords 107 within 100-400 ms after stimulus onset. In line with fMRI findings, reliable learning-induced 108 changes in ERP occurred only on the second day after learning, supposedly after night-sleep 109 consolidation. Thus, this ERP study provided little to no confirmation for semantic "fast mapping" 110 in word learning. 111 The other available EEG study (François,  between syllables embedded into a continuous auditory stream. However, these ERP findings did 118 not provide evidence for the "fast mapping" hypothesis. Indeed, since the participants were 119 required to listen carefully to the auditory stream with the task of discovering new words, learning-120 related enhancement of N400 might have been elicited by an on-line attentional modulation, i.e., 121 attention biased toward auditory word-forms associated with pictures during the learning session. 122 Cortical plasticity induced by active learning of novel words 6 To prove semantic cortical plasticity, enhanced N400 should be observed during passive exposure 123 to the newly learned word-forms. 124 In summary, the current picture of "fast mapping" in word learning is obviously far from 125 complete. It is a controversial topic with a body of associated literature, yet the mere existence of  In the current MEG study, which employed an operant conditioning task, we sought 139 evidence for putative cortical "fast mapping" of two interactive but separate processes: formation 140 of a new acoustic word-form discrimination and semantic analysis of the newly-formed coherent 141 item. To pursue this goal, we engaged our participants in the pseudoword-action associative 142 learning task to let them actively find unique associations between four auditory pseudowords and 143 their own body part movements, whereas the other four auditory pseudowords were not supposed 144 to be associated with any motor action. To reveal the learning effect on word-form-related and 145 semantic-related cortical activity, we compared responses to passive presentations of the two 146 pseudoword types before and after learning sessions. We used the MEG neuroimaging technique,

151
During the experiment, participants were presented with eight pseudowords (Table 1; 152 Figure 1). The active task performed by participants was to learn specific associations between 153 action pseudowords (APW) and motor actions by their hands and feet, while refraining from any 154 responses to non-action pseudowords (NPW ; Table 1). MEG was recorded during 'Passive 155 block 1', which preceded word-form learning, and during 'Passive block 2', which followed 156 learning ( Figure 1B). participants during the active performance block was between 0 and 21 out of 320 trials.

Familiarization effects (sensor-level analysis)
196 Figure 2A shows the root mean square (RMS) waveforms, calculated across gradiometers 197 within left-and right-hemispheric regions of interest (ROIs), for passive presentation of APW and 198 NPW in "before learning" and "after learning" conditions. These data illustrate the time courses    sensitive effects were similar for APW and NPW ( Figure 2B). Considering the lack of an a priori 236 hypothesis regarding these early effects and their rather weak statistical reliability, we did not 237 proceed with their further analysis.

238
Our hypothesis was focused on repetition effects at later latencies that were related to the 239 time when different auditory pseudowords started to be discriminable from each other as coherent from -20 ms to 550 ms in the right ROI. As seen in Figure 2A, the time intervals of significant 246 suppression for APW and NPW calculated separately substantially overlapped. To identify cortical areas that contributed to neural repetition suppression resulting from 256 familiarization with the novel pseudowords, we analyzed the data in the source-space. As   Semantic learning effects (sensor-level analysis) 280 To unravel the putative effect of association learning, we analyzed the differences in the 281 neural responses between APW and NPW "after leaning" (APW2 -NPW2) versus "before 282 learning" (APW1 -NPW1) during passive blocks ( Figure 4). Both RMS signal timecourses 283 ( Figure 4A) and ERF topographical maps ( Figure 4B) demonstrated that whereas cortical activity 284 evoked by the two pseudoword types did not differ before learning, the strength of differential 285 neural responses to APW significantly increased after the learning procedure in the left ROI.  The responses showed greater selectivity for the APW for the protracted response time  difference in the neural responses before and after learning did result from learning, we 306 additionally checked for the significance of the APW-NPW difference separately for the two 307 conditions ("before learning" and "after learning") using the same TFCE permutation statistical 308 procedure. Neural responses following APW and NPW trials started to statistically discriminate 309 the two types of pseudowords only after learning (significant from 145 to 615 ms), without any 310 significant differences detected before learning ( Figure 4). 311 Remarkably, unlike familiarization-related changes in the neural response, the semantic-

331
Statistically thresholded cortical topography for the APW versus NPW differential neural 332 responses "before learning" (APW1 -NPW1, top row) and "after learning" (APW2 -NPW2, uncorrected) were considered further (see Table 2 for the list of the respective clusters). Figure 6 shows the cortical location of the clusters reconstructed at each of the three sequential time frames, 347 as well as their activation timecourses before and after learning. Initially, around 190 ms post-UP, 348 a learning-related selective response to APW emerged in cortical areas surrounding the Sylvian 349 fissure: aSTS, ventral premotor cortex, and the anterior part of intraparietal sulcus and insula. Once 350 it appeared, differential activation in these areas was mostly sustained until response termination.

351
After 250 ms, activation spread to more anterior brain regions, and by 330 ms post-UP it reached 352 the pole of the left temporal lobe and the triangular part of the left IFG extending to its orbital part.   at the bottom represent grand-averaged differential response strength for the cortical clusters 366 across time for "before learning" (APW1 -NPW1, gray lines) and "after learning" (APW2 -367 NPW2, green lines) conditions. Shaded areas on timecourses represent standard errors.

369
Whether short-term learning of new words can induce rapid changes in cortical areas 370 involved in distributed neural representation of the lexicon is a hotly debated topic in the literature.

371
To answer this question, we examined the MEG phase-locked responses elicited in the cerebral 372 cortex by passive presentation of eight novel pseudowords before and after an operant conditioning 373 task. The task forced the participants to perform an active search for word-form meaning, as four 374 unique word-forms acquired meaning that referred to movements of participants' own body parts 375 (in a way similar to real action words) and the other four word-forms remained "empty lexical phase-locked ERF components elicited by auditory word onset well before the UP (Figure 2), a 425 long-lasting and highly reliable attenuation of phase-locked activity occurred approximately 300 426 ms after stimulus onset, when the word-form began to be discriminable from each other. In fact, the onset of this reduction started even 100 ms earlier than UP, probably as a response to the 428 appearance of the third phoneme in the word-form, which, unlike the UP, was not sufficient to 429 distinguish all eight pseudowords, but rather allowed identification of the difference between 430 APW-NPW pairs (see Methods for details).

431
The above considerations suggest that our findings of the strong suppression of neural 432 responses to novel acoustic word-forms, which started to be very familiar through the experimental 433 procedure, most probably reflect a mechanism of familiarization memory. This mechanism is one 434 of the components of the recognition memory system that is responsible for judging the prior  sharpening of neural representations did occur for APW, we would expect that after operant 470 conditioning, the cortical brain responses to APW would relatively increase compared with NPW.

471
This finding is exactly what we observed while contrasting APW-NPW differences before and 472 after learning ( Figure 6). 473 Indeed, the only factor that affected the auditory perception of APW and NPW stimuli 474 during the second passive presentation was their unique relatedness to a specific motor action in 475 the prior active blocks. Acoustical features across APW-NPW pairs were well counter-balanced 476 across the eight pseudowords (see Methods), and neural responses to pseudowords of both types 477 did not differ before learning (Figures 5 and 6). Additionally, our findings cannot be explained by 478 differences in selective attention to or in familiarization with APW-NPW pairs during learning.

479
The learning procedure itself did not introduce any bias toward APW word-forms, as it required 480 the subject to attentively discriminate between both stimulus types, which were repeated the same . However, in our case, enhanced neural response to APW spanned 140 ms after the first 509 meaningful phoneme and onwards, which clearly occurred later than the MMN wave. We 510 speculate that rather than reflecting a rapidly detected phonological difference in the fourth 511 phonemes between APW and NPW, the differential response to APW points to enhanced activity 512 of neuronal circuitry that mediates sensitivity for the temporal sequence of the phonemes that from the higher order areas. From this view, we assume that while the early differential activity in 562 the perisylvian areas appears to be stimulus-driven, the later activity there presumably depends on 563 top-down signaling from higher-order speech cortical areas involved in semantic retrieval.

564
In summary, we would argue that according to criteria proposed by Davis & Gaskell 565 (2009), and briefly reviewed here in the introduction, our data evidence that cortical 566 representations of both phonology and semantics of previously unfamiliar words may be formed 567 following 1-2 hours of active associative learning. This conclusion raises the question as to why a 568 rapid cortical activity modulation by a newly learned word would be found in our MEG study, 569 while the blood-oxygen-level-dependent (BOLD) response of cortical areas consistently remain 570 largely unaffected during the hours after associative learning (Davis & Gaskell, 2009 Another explanation, which is not necessarily mutually exclusive with the first one, focuses 582 on the difference in the associative learning procedure between our study and the previous fMRI Handedness Inventory (Oldfield, 1971). The study was conducted following the ethical principles During the experiments, participants were comfortably seated in the MEG apparatus that 647 was placed in an electromagnetically and acoustically shielded room (see below). Pseudowords 648 were presented binaurally via plastic ear tubes in an interleaved quasi-random order, at 90 dB SPL.

649
The experiment was implemented using the Presentation 14.4 software (Neurobehavioral systems,

651
The experiment consisted of four consecutive blocks with a fixed order across participants:

654
Two identical passive listening blocks were administered before and after the two active  After MEG data acquisition, participants underwent MRI scanning with a 1.5T Philips

713
Intera system for further reconstruction of the cortical surface.  and 182 ± 21 for APW and NPW stimuli, respectively, before learning, and 181 ± 21 and 182 ± 732 20 for the same stimuli after learning.

733
The baseline correction was computed using the interval from the -210 ms to 0 ms before 734 the stimulus onset (i.e., -610 --410 ms relative to the UP).

736
Analyses were performed in two steps. First, in search for the general familiarization effect 737 for the novel word-forms, the phase-locked cortical responses to APW and NPW were compared 738 between "before learning" and "after learning" conditions.

739
Secondly, we aimed to identify a putative effect of pseudoword associative learning on 740 neural activity elicited by pseudowords that acquired referential meaning. To this end, we analyzed 741 Cortical plasticity induced by active learning of novel words 32 the APW-NPW difference in phase-locked responses before and after learning. We expected that 742 while cortical responses to APW and NPW would not differ before learning, the differential 743 response to APW would emerge after learning as a result of fine-tuning of cortical representations 744 toward the respective auditory word-forms.

745
At each step, we analyzed MEG data both at the sensor-and the source-level in order to 746 pinpoint the anticipated effects both in terms of their timing and involved cortical regions. and NPW1 before learning and APW2 and NPW2 after learning). The RMS signal was baseline-764 corrected using the interval from the -210 ms to 0 ms before the stimulus onset (-610 to -410 ms 765 relative to the UP). A low-pass 6 th -order Butterworth filter with a cutoff frequency 100 Hz was 766 applied in order to smooth the RMS signals before statistical analyses; this procedure was done in 767 order to reduce the signal-to-noise ratio. surrogate data, which were generated from real data by swapping the two conditions for the entire 792 time window in random subsets of participants. The significance level was set at p < 0.05 793 (corrected). Then, we repeated the same analysis separately for action pseudowords (APW2 versus 794 APW1) and non-action pseudowords (NPW2 versus NPW1).
For illustrative purposes, the differential (after learning minus before learning) topographic 796 maps for ERFs elicited by APW and NPW stimuli separately were plotted in 100 ms steps (data 797 averaged across 35 ms for each plot).

798
Additionally, although not the main purpose of the current study, we analyzed early 799 transient familiarization effects bound to the stimulus onset that, although prominent, did not 800 survive the TFCE correction procedure. For this purpose, all trials were pooled, we averaged over 801 timepoints within the 35-ms intervals centered on M100 and M200 peaks, and applied a paired 802 two-tailed t-test (for each hemisphere separately).

804
Cortical sources that exhibited the familiarization effect were reconstructed for time 805 windows during which the effect was significant at the sensor level in the left and right ROIs (see 806 above). The source-space data for APW and NPW types were collapsed and averaged over these 807 time intervals. We compared "before learning" and "after learning" conditions using vertex-wise 808 t-test with FDR correction performed for two hemispheres (20,484 vertices).

809
Next, for the sake of comparison with the previous passive word learning studies, two splitting the two time windows should be considered as exploratory, so the "before" and "after" 818 learning conditions were compared using vertex-wise t-test with FDR correction performed for 819 two hemispheres (20,484 vertices). 821 We sought to identify the semantic learning effect by analyzing the contrast between 822 cortical responses to APW and NPW types before and after learning.

823
First, the statistical analyses were performed for the contrast "APW1-NPW1" versus 824 "APW2-NPW2", where APW1 and APW2 stand for ERF time course to passive presentation of 825 APWs "before learning" and "after learning", respectively, while NPW1 and NPW2 designate the 826 responses to NPWs under the same two experimental conditions. The paired t-test with the TFCE 827 permutation statistical procedure (see above for details) was applied at each time point of the entire 828 RMS waveform (from -410 ms to 1000 ms relative to UP) to determine the response intervals that 829 demonstrated a significant difference between conditions. To ensure that the APW-NPW 830 difference in the neural responses before and after learning did result from learning, we 831 additionally checked for the significance of the APW-NPW difference for each of the two 832 conditions separately ("before learning" and "after learning") using the same TFCE permutation 833 statistical procedure.

834
To visualize the direction and dynamics of the effect, we plotted ERF topographic maps 835 for the (APW -NPW) difference before and after learning at 100 ms steps; at each step we 836 integrated the ERF signed values across 35 ms.

837
Semantic learning effects (source-level analysis) 838 To reveal cortical regions, activation of which contributed to the "semantic learning" 839 effect, the cortical sources of the effect were reconstructed within the time interval that was 840 identified at the sensor level. Since for the sensor-level data the effect already survived correction 841 for multiple comparisons, for source-space analysis we applied the uncorrected significance 842 threshold of p < 0.05 (see Gross et al., 2013). To this end, we used a vertex-wise paired two-tailed 843 t-test in order to contrast cortical activity averaged across the whole time interval for APW1 versus 844 NPW1 ("before learning") and for APW2 versus NPW2 ("after learning"). Further, in order to 845 explore the temporal dynamics of the semantic learning effect, we used a vertex-wise paired two-846 tailed t-test in order to contrast cortical activity for "APW1-NPW1" versus "APW2-NPW2" 847 differences between conditions. This was done at the time points corresponding to the lowest p-