CORDIS - EU research results

Neural and computational determinants of left cerebral dominance in speech and language

Final Report Summary - COMPUSLANG (Neural and computational determinants of left cerebral dominance in speech and language)

The goal of the COMPUSLANG is to characterize the neural operations taking place in the human brain during speech processing, and to understand the specific language disorders that arise when these operations are altered. Within the first half of the project we assessed the theoretical proposal that the coupling of slow (theta) and fast (gamma) cortical oscillations could serve to capture and process in a partly independent way (Giraud and Poeppel, Nature Neuroscience, 2012), speech cues occurring at different timescales in speech (phonemes, syllables, words etc.). Using a biophysically-inspired computational model we show that coupled theta/gamma oscillations can track syllables in speech and orchestrate the decoding of neural activity within syllabic chunks by higher-order brain regions (Fontolan et al., 2014; Hyafil et al., e-Life, 2015;). The output of this work has a direct technological application. The neurocomputational work using neural oscillations to segment speech was applied to automatic speech recognition and found to provide a novel method to syllabify continuous speech on-line, which state-of-the-art methods were hitherto unable to do (Hyafil and Cernal, Interspeech 2015).

Using MEG and combined EEG and fMRI we demonstrate that, in dyslexia, neural auditory processing at gamma-scale is accelerated (Lehongre et al., Neuron 2011). The consequence of this anomaly is that subjects with dyslexia segment speech into speech units that are shorter than standard phonemes, creating difficulties when they start to read and need to match their own mental representations of phonemes with universal ones (Giraud and Ramus, Current Opinion in Neurobiology, 2012). By contrast, we show that in autism, both the neural theta and gamma oscillatory activity is altered in auditory cortex. This double processing anomaly tightly predicts the severity of the language disorder in autism (Jochaut et al., Frontiers in Human Neurosciences). Using in combination computational neuroscience and human electrophysiology, we show that coupled auditory cortical oscillations are essential elements of speech processing by the human brain. These findings directly translate into new ideas for therapeutic interventions for improving speech reception in dyslexia and autism.

The COMPUSLANG project ultimately led to proposing a systematic way to relate the different forms of neural oscillation coupling to cognitive functions (Hyafil et al., Trends in Neurosciences 2015). It also led to new avenues for restoring a physiological functioning of the auditory cortex in subjects with autism and dyslexia, using transcranial alternative current stimulation.