Overlapping but Language-Specific Mechanisms in Morphosyntactic Processing in Highly Competent L2 Acquired at School Entry: fMRI Evidence From an Alternating Language Switching Task

Many bilingual individuals acquire their second language when entering primary school; however, very few studies have investigated morphosyntax processing in this population. Combining a whole-brain and region of interest (ROI)-based approach, we studied event-related fMRI during morphosyntactic processing, specifically person-number phi-features, in Turkish (L1) and Persian (L2) by highly proficient bilinguals who learned Persian at school entry. In a design with alternating language switching and pseudorandomized grammaticality conditions, two left-lateralized syntax-specific ROIs and 11 bilateral ROIs involved in executive functions (EF) were analyzed for the intensity of activation relative to a resting baseline. Our findings indicate a strong overlap of neural networks for L1 and L2, suggesting structural similarities of neuroanatomical organization. In all ROIs morphosyntactic processing invoked stronger activation in L1 than in L2. This may be a consequence of symmetrical switch costs in the alternating design used here, where the need for suppressing the non-required language is stronger for the dominant L1 when it is non-required as compared to the non-dominant L2, leading to a stronger rebound for L1 than L2 when the language is required. Both L1 and L2 revealed significant activation in syntax-specific areas in left hemisphere clusters and increased activation in EF-specific areas in right-hemisphere than left-hemisphere clusters, confirming syntax-specific functions of the left hemisphere, whereas the right hemisphere appears to subserve control functions required for switching languages. While previous reports indicate a leftward bias in planum temporale activation during auditory and linguistic processing, the present study shows the activation of the right planum temporale indicating its involvement in auditory attention. More pronounced grammaticality effect in left pars opercularis for L1 and in left pSTG for L2 indicate differences in the processing of morphosyntactic information in these brain regions. Nevertheless, the activation of pars opercularis and pSTG emphasize the centrality of these regions in the processing of person-number phi-features. Taken together, the present results confirm that morphosyntactic processing in bilinguals relates to composite, syntax-sensitive and EF-sensitive mechanisms in which some nodes of the language network are differentially involved.

GPT-2’s activations predict the degree of semantic comprehension in the human brain

Language transformers, like GPT-2, have demonstrated remarkable abilities to process text, and now constitute the backbone of deep translation, summarization and dialogue algorithms. However, whether these models actually understand language is highly controversial. Here, we show that the representations of GPT-2 not only map onto the brain responses to spoken stories, but also predict the extent to which subjects understand the narratives. To this end, we analyze 101 subjects recorded with functional Magnetic Resonance Imaging while listening to 70 min of short stories. We then fit a linear model to predict brain activity from GPT-2 activations, and correlate this mapping with subjects’ comprehension scores as assessed for each story. The results show that GPT-2’s brain predictions significantly correlate with semantic comprehension. These effects are bilaterally distributed in the language network and peak with a correlation above 30% in the infero-frontal and medio-temporal gyri as well as in the superior frontal cortex, the planum temporale and the precuneus. Overall, this study provides an empirical framework to probe and dissect semantic comprehension in brains and deep learning algorithms.

Myeloarchitectonic Asymmetries of Language Regions in the Human Brain

One prominent theory in neuroscience and psychology assumes that cortical regions for language are left hemisphere lateralized in the human brain. In the current study, we used a novel technique, quantitative magnetic resonance imaging (qMRI), to examine interhemispheric asymmetries in language regions in terms of macromolecular tissue volume (MTV) and quantitative longitudinal relaxation time (T1) maps in the living human brain. These two measures are known to reflect cortical myeloarchitecture from the microstructural perspective. One hundred and fifteen adults (55 male, 60 female) were examined for their myeloarchitectonic asymmetries of language regions. We found that the cortical myeloarchitecture of inferior frontal areas including the pars opercularis, pars triangularis, and pars orbitalis is left lateralized, while that of the middle temporal gyrus, Heschl’s gyrus, and planum temporale is right lateralized. Moreover, the leftward lateralization of myelination structure is significantly correlated with language skills measured by phonemic and speech tone awareness. This study reveals for the first time a mixed pattern of myeloarchitectonic asymmetries, which calls for a general theory to accommodate the full complexity of principles underlying human hemispheric specialization.

Planum Temporale grey matter volume asymmetries in new-born monkeys (Papio anubis)

The Planum Temporale (PT) is one of the key hubs of the language network in the human brain. The gross asymmetry of this perisylvian region toward the left brain was considered as the most emblematic marker of hemispheric specialization of language processes in the brain. Interestingly, this neuroanatomical signature was documented also in newborn infants and preterms, suggesting the early brain’s readiness for language acquisition. Nevertheless, this latter interpretation was questioned by a recent report in nonhuman primates of a potential similar signature in newborn baboons Papio anubis based on PT surface measures. Whether this “tip of the iceberg” PT asymmetry is actually reflecting asymmetry of its underlying grey matter volume remain unclear but critical to investigate potential continuities of cortical specialization with human infants. Here we report a population-level leftward asymmetry of the Planum Temporale grey matter volume in in vivo 34 newborn baboons Papio anubis, which showed intra-individual positive correlation with PT surface’s asymmetry measures but also a more pronounced degree of leftward asymmetry at the population-level. This finding demonstrates that PT leftward structural asymmetry in this Old World monkey species is a robust phenomenon in early primate development, which clearly speaks for a continuity with early human brain specialization. Results also strengthen the hypothesis that early PT asymmetry might be not a human-specific marker for the pre-wired language-ready brain in infants.

Structural MRI Study of the Planum Temporale in Individuals With an At-Risk Mental State Using Labeled Cortical Distance Mapping

Background: Recent studies have demonstrated brain structural changes that predate or accompany the onset of frank psychosis, such as schizophrenia, among individuals with an at-risk mental state (ARMS). The planum temporale (PT) is a brain region involved in language processing. In schizophrenia patients, gray matter volume reduction and lack of normal asymmetry (left > right) of PT have repeatedly been reported. Some studies showed progressive gray matter reduction of PT in first-episode schizophrenia patients, and in ARMS subjects during their development of psychosis.Methods: MRI scans (1.5 T field strength) were obtained from 73 ARMS subjects and 74 gender- and age-matched healthy controls at three sites (University of Toyama, Toho University and Tohoku University). Participants with ARMS were clinically monitored for at least 2 years to confirm whether they subsequently developed frank psychosis. Cortical thickness, gray matter volume, and surface area of PT were estimated using FreeSurfer-initiated labeled cortical distance mapping (FSLCDM). PT measures were compared among healthy controls, ARMS subjects who later developed overt psychosis (ARMS-P), and those who did not (ARMS-NP). In each statistical model, age, sex, intracranial volume, and scanning sites were treated as nuisance covariates.Results: Of 73 ARMS subjects, 18 developed overt psychosis (12 schizophrenia and 6 other psychoses) within the follow-up period. There were no significant group differences of PT measures. In addition, significant asymmetries of PT volume and surface area (left > right) were found in all diagnostic groups. PT measures did not correlate with the neurocognitive performance of ARMS subjects.Discussion: Our results suggest that the previously-reported gray matter reduction and lack of normal anatomical asymmetry of PT in schizophrenia patients may not emerge during the prodromal stage of psychosis; taken together with previous longitudinal findings, such PT structural changes may occur just before or during the onset of psychosis.