CCNCS Seminar Details
Searchlight Representational Similarity Analysis of Complex Morpholexical Processes
|Speaker:||Dr Su Li|
|Date/Time:||Wednesday 7 December 2011, 4.30pm|
Although the neural basis of speech comprehension is an important focus for neuroimaging research, dynamic neural models of morphosyntactic processing are notably absent. Here we explore how underlying properties of lexical constituents are computed in neural networks situated in bilateral fronto-temporal brain regions. A novel multivariate pattern analysis technique (MVPA) that reveals the fine grained structure of neural computation (with centimetre and millisecond precision) has been developed based on representational similarity analysis (RSA) of MEG/EEG data in source space using whole-brain searchlight techniques. This allows us to search brain volumes in time and space for neurocomputational signatures correlated to different theoretical models. A phonological model differentiates words according to their shared phonological markers (here the presence of the English inflectional rhyme pattern) but does not separate real inflection from pseudo-inflection (played vs trade). A second, morphosyntactic model classifies words according to their grammatical properties (regular past tense vs. non-inflected). RSA searchlight analyses show that each type of process differentially engages bilateral fronto-temporal networks, with morphosyntactic processing seen relatively late and primarily in left inferior frontal areas. The combination of MEG/EEG with RSA makes it possible to map out the complex neural systems underlying the real-time dynamics of human language comprehension.