Graphomotor memory in Exner's area enhances word learning in the blind

Abstract

Handwriting is thought to impede vocabulary learning in sighted adults because the motor execution of writing interferes with efficient audiovisual processing during encoding. However, the motor memory of writing may facilitate adult word learning when visual sensory inputs are severely restricted. Using functional MRI, we show that late-blind participants, but not sighted participants, learned novel words by recruiting the left dorsal premotor cortex known as Exner's writing area and its functional coupling with the left hippocampus. During later recall, the phonological and semantic contents of these words are represented in the activation patterns of the left hippocampus as well as in those of left frontotemporal language areas. These findings suggest that motor codes of handwriting help blind participants maintain word-form representations during learning and retrieval. We propose that such reliance on the motor system reflects a broad architecture of the cerebral language network which encompasses the limb motor system as a hardwired component.

Fig. 1. Experimental paradigm.

A The behavioral paradigm consisted of several study–test cycles for foreign language (FL) word learning. Participants first heard and learned a list of 20 word pairs in a study period and then recalled the meanings of FL words in a test period. In the writing condition, participants studied each FL word while spelling it out with the right index finger, while in the no-writing condition, they studied the target word while keeping the right hand relaxed. The two conditions were performed in two separate sessions spaced at least 3 weeks from each other. During the test period immediately after the study period, participants were presented with FL words and asked to recall their meanings. This cycle of study and test was repeated outside the MRI scanner until participants achieved >50% accuracy in the test period. Participants then lay in the MRI scanner and received one additional cycle of study and test during fMRI scanning. Spoken responses were recorded and analyzed as “immediate recall.” One week after fMRI scanning, participants received the same recall test for each condition (“delayed recall”). B Three theoretical representational dissimilarity matrices (RDMs) used for multi-voxel pattern analysis. The motor RDM was created to characterize motor-level dissimilarity in writing movements between FL words. Pairwise comparisons of all possible word pairs yielded the motor RDM as a 40 × 40 symmetrical matrix of which the off-diagonal values represent the dissimilarity for each pair of words at the motor level. For the syllabic RDM, syllabic dissimilarity between a pair of words was calculated as the proportion of shared syllables between the two words, yielding a 40 × 40 matrix whose off-diagonal values represent pairwise dissimilarities between words at the syllable level. Likewise, the 40 × 40 semantic RDM was constructed by computing semantic distance between words according to the taxonomy provided by the WordNet (see “Methods”).

Fig. 2. Behavioral results during the test period.

During immediate recall in the MRI scanner, blind and sighted participants identified FL word meanings with high accuracy (>90%) irrespective of the learning conditions. On delayed recall, however, blind participants tended to score higher in the writing than in the no-writing condition, whereas sighted participants showed an opposite trend between the writing and no-writing conditions. The differential impact of writing was confirmed as a robust interaction between the effects of condition and group (see “Results”). In this and all subsequent figures, box plot elements are defined as follows: center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range.

Fig. 3. GLM results.

A Neural effects of learning condition and group during the study period. The main effect of condition calculated in the writing vs. no-writing contrast was observed in the right cerebellum (30, -50, −26, Z = 4.88) and the left sensorimotor cortex (−32, −18, 50, Z = 6.50), which extended to the known coordinates of Exner’s area (left). No brain region showed the effect of group, either in the whole-brain analysis or in the ROI analyses (p > 0.1 for all). Significant interaction between condition and group was found in Exner’s area ROI (−20, −4, 52, p = 0.013, right). B Neural effects of condition × group interaction during the test period. The main effect of group was observed in bilateral frontal and parietotemporal regions showing greater activation in blind relative to sighted participants (top). In ROI analysis, Exner’s area showed a greater effect of learning condition for blind than for sighted participants, creating significant condition × group interaction (−24, 0, 48, p = 0.003, bottom). Note that this interaction was reversed in direction from the one observed during the study period. C Psychophysiological interaction with Exner’s area. The magnitude of connection strength is represented as percent signal change relative to the baseline. A middle part of the left hippocampus (−36, −24, −12, Z = 3.78) was identified as showing increased functional connectivity with Exner’s area across the study and test periods.

Fig. 4. Correlations with the theoretical RDMs in Exner’s area and left hippocampus.

During the study period, multi-voxel activation patterns in Exner’s area showed a significant effect of learning condition for the motor RDM (p = 0.013, shaded in gray). The main effects and their interaction were non-significant for the three RDMs during the test period (p > 0.1 for all). On the other hand, the left hippocampus showed significant crossover interaction with the syllabic RDM (p = 0.023), suggesting that handwriting yielded a differential impact on activation patterns between the blind and sighted groups. During the test period, the same segment of the left hippocampus showed significant condition × group interaction for the semantic RDM (p = 0.026), again suggesting that semantic information was more weakly represented for the blind relative to the sighted participants.

Fig. 5. Correlations with the theoretical RDMs during the test period.

A Left hemisphere ROIs. The left pMTG overall showed significant correlation with the motor and semantic RDMs (p = 0.050 for both, shaded in gray). On the other hand, the left IFG showed significant condition × group interaction (p = 0.030, shaded in gray) for the syllabic RDM. In the left SMG, there was a weak trend of writing with the syllabic RDM (p = 0.115). No other ROI showed significant main effects or their interactions with any of the three RDMs (p > 0.2 for all). For each ROI, box plots of Z-transformed Spearman correlation were calculated at the voxels showing the maximum effect for each RDM. B Searchlight RSA. Significant condition × group interaction for the syllabic RDM was found in the left IFG (−52, 26, 10, Z = 4.11) and the left hippocampus that extended from the anterior to middle segments (−24, −14, −24, Z = 3.99 and −28, −24, −16, Z = 3.67, respectively).