Authors:

GARAGNANI, M., SHTYROV, Y., Wennekers, T. & PULVERMULLER, F.

Reference:

Year of publication:

2010

CBU number:

7094

Abstract:

We applied magneto- and electro-encephalography (MEG/EEG) experimental methods in conjunction with neuroanatomically grounded computational modeling to investigate the hotly debated issue of how linguistic knowledge is represented in the human brain, and whether (and to what extent) access to the cortical representation of lexical items is automatic or relies upon attentional processes. Recent simulations obtained with a neurobiologically motivated neural-network model of the left perisylvian language cortex suggest that long-term synaptic plasticity mechanisms can lead to the emergence of cortical representations for words, consisting of strongly connected circuits that are both distributed and functionally discrete. Our model replicates, explains and reconciles different patterns of neurophysiological data (N400 and mismatch negativity, MMN), and makes testable predictions about the brain responses to words and pseudowords under different degrees of attention. In particular: (I) If ample attentional resources are made available to linguistic processes, simulated brain responses to pseudowords were larger than those to words (as typically observed in the N400 pattern), whereas if attentional resources are scarce, the opposite pattern (words > pseudowords, as reported in MMN experiments) emerged; (II) The model predicts that neurophysiological responses to familiar words should not be significantly modulated by the availability of attentional resources, whereas responses to unfamiliar, “unrepresented” items (pseudowords) should show strong attention dependence. To test these predictions, we carried out novel MEG and EEG experiments and recorded event-related fields and potentials (ERFs/ERPs) elicited by the same syllables completing acoustically matched words and pseudowords while subjects were asked to attend to the spoken input or to ignore it. Both methodologies independently confirmed the model predictions. In particular: (I) MEG results showed that when subjects attended the auditory stimuli, the magnetic brain response to pseudowords was larger than that to words (as in the N400 pattern), whereas when attention was directed away from the linguistic input, the opposite pattern emerged (words > pseudowords). Similarly, EEG results showed that, under non-attend conditions, word-elicited responses (peaking at ~120 ms) were larger than pseudoword ones; however, when attention was directed towards the sounds, such word-pseudoword difference disappeared. In addition, and importantly (II): in both experiments, responses to words were unchanged by attentional variation, whereas MMN responses to pseudowords were significantly modulated by attention. We explain these results by robustness of word memory networks, whose strong internal connections guarantee rapid full-scale activation irrespective of the attentional load; conversely, the processing of pseudowords (which involves the simultaneous activation of several competing lexical representations) is strongly affected by the availability of attentional resources, even at its earliest stages. Topography analysis and source reconstruction indicated that left-perisylvian cortices are involved in mediating such attention effects. Our findings (i) confirm earlier suggestions that initial stages of word (but not pseudoword) processing are unaffected by attentional demands and may thus be automatic, and (ii) provide evidence in support of the hypothesis that lexical items are represented in the brain as strongly connected, functionally discrete and distributed action-perception circuits.