Online M-BIC Lecture: Frédéric Theunissen

The human cortical networks involved in speech and language processing have been identified in neurological studies and in controlled experiments that monitor brain activity while patients and subjects process speech sounds. A coarse functional brain map that describes multiple speech processing streams has emerged.  In the sound to meaning stream, speech sounds are known to be first decomposed by a modulation filter bank in primary and secondary auditory cortex in the dorsal superior temporal gyrus (dSTG).  dSTG projects to a phonemic area found in the medial superior temporal sulcus (mSTS) that shows selectivity for speech phonemes and words relative to matched non-speech sounds.  In the so-called ventral stream of the temporal lobe, the phoneme to meaning transformation begins: a dorsal to ventral projection and a posterior to anterior projection leads to lexical and combinatorial networks found in posterior middle temporal gyrus/inferior temporal sulcus (pMTG,pITS) and anterior MTG/ITS respectively. Although this working model is relatively well supported, it provides little insight on the actual computations that are performed at each processing stage and on the potential implementation of these computations within neural circuits. In this talk, I will show how brain imaging data obtained in fMRI experiments while subjects listen to natural narrated speech can be analyzed to decipher the intermediate steps in this processing stream.  I will first describe our efforts to describe the nature of the phonemic representation found in mSTS and how phonemic selective responses can be obtained from the spectro-temporal decomposition found in dSTG.   Second, I will describe our attempts to better identify the brain regions in the cortical speech network that could be involved in lexical retrieval. I will also show that it is essential to model BOLD responses obtained in fMRI to make correct inferences and how this modeling effort shows that the spatial resolution in human brain imaging studies are just as limiting as the well-known temporal limits of the sluggish BOLD signal.