Pallier, C. , Devauchelle, A. D. & Dehaene, S.
INSERM-CEA Cognitive Neuroimaging Unit, Gif-sur-Yvette, France
How the constituent structure of sentences is represented in the brain remains largely unknown. The current experiment was based on the premise that a brain region that encodes linguistic trees should show an activation that increases monotonically with the size of phrases that can build from the input.
Participants were scanned with fMRI while reading sequences of 12 words. The sequences differed by the maximal size of constituents that could be parsed. At the lowest level, sequences consited of 12 unrelated words. At the next levels, six unrelated constituents of two words were presented, then four constituents of three words, etc, all the way up to a well-formed right-branching sentence with 12 words. A second group of participants read similar sequences except that the content-words had been replaced by pseudowords, yielding jabberwocky-like constituents. This allowed us to control for semantic or transition probability confounds.
The results indicated that when constituents became larger, activation increased in left hemispheric regions located in the temporo-parietal junction (TPJ), posterior and anterior superior temporal sulcus (pSTS and aSTS), temporal pole (TP), inferior frontal gyrus (IFG, pars orbitalis and triangularis) and putamen. In the delexicalized “jabberwocky” version of the experiment, the pSTS, left IFG and the putamen still varied monotonically with constituent size, while the aSTS, TP and TPJ responses were abolished. There was thus a clear dissociation between regions which respond whenever syntactic constituents are present (putamen/pSTS/IFG) and regions which require the presence of additional lexico-semantic information (TP/aSTS/TPJ). The latter regions either integrate incoming words into semantic constituents or rely on word-based syntactic information (e.g. subcategorization frames) to build constituents.
The current experiment was performed on monolingual subjects but could be adapted to bilinguals to address the issue of the overlap of regions involved in L1 and L2 processing.