Authors:

GARAGNANI, M. & PULVERMULLER, F.

Reference:

Year of publication:

2007

CBU number:

6549

Abstract:

Meaningful material and senseless unknown stimuli lead to different patterns of brain activation. Traditionally, the major response indexing “sense” has been the N400, a late event-related brain response elicited under conditions where subjects are attending to the input. The N400 is larger for senseless materials (e.g., meaningless pseudowords) than for matched meaningful words. More recently, early MEG and EEG differences have also been recorded – for example, in the Mismatch Negativity (MMN, latency 100-250ms). The MMN is elicited even when subjects are heavily distracted and is larger for words than for pseudowords, thus exhibiting the reverse pattern seen for the N400. The question of why, exactly, these brain indicators of lexico-semantic processes arise at different latencies and present reversed “polarities” is still unanswered. We implemented a neuroanatomically grounded neural-network model of the left-perisylvian language cortex and used it to simulate brain processes of language learning. We repeatedly confronted the network with activation patterns and allowed it to adapt by means of biologically-plausible Hebbian mechanisms: we observed the formation of input-specific neuronal assemblies, i.e., sets of strongly connected neurons distributed over a range of areas which responded only to known patterns (“words”). We then used the trained model to simulate activation of the language cortex when meaningful familiar word and senseless unknown pseudoword are presented in input. We found that variation of the (I) global level of excitation (or internal noise) and (II) amount of feedback inhibition of the network similarly modulated the simulated brain response to word and pseudowords. More precisely, low noise levels and weak feedback inhibition produced – on average – late activation differences, with a stronger response to pseudowords than to words (mirroring the N400 pattern). In contrast, high starting levels of excitation or strong feedback inhibition lead to early activation differences, with a stronger response to words than to pseudowords (exactly paralleling the MMN data). Our computational model offers a unifying explanatory account for previously seemingly contradictory neurophysiological observations. In addition, it points to the global level of excitation and feedback inhibition as possible brain correlates of arousal and attention, and predicts that it is precisely these cognitive processes which modulate latency and polarity of event-related brain responses. This prediction has been confirmed by recent experimental evidence (see Shtyrov, Kujala and Pulvermuller, poster submitted to CNS 2007).