The Electrophysiology of Language Comprehension: A Neurocomputational Model
Dr. Harm Brouwer
The measurement of Event-Related brain Potentials (ERPs) has been very important to our understanding of the neural basis of language comprehension. Some time ago, however, ERP studies have stumbled upon a puzzle that came to be known as the “Semantic Illusion”-effect.
Hoeks et al. (2004), for instance, found that syntactically well-formed, but semantically anomalous sentences like ‘De speer heeft de atleten geworpen’ (lit: ‘The javelin has the athletes thrown’) did not produce an increase in N400 amplitude, which was unexpected given that the N400 component is generally taken to reflect difficulty in semantic integration. Instead, the critical verb thrown was found to produce an increase in P600 amplitude, which was equally surprising as the P600 component is generally taken to reflect syntactic processing. These findings have been taken to indicate that the processer temporarily assumes (is under the “illusion”) that the sentence is semantically felicitous.
Explaining the “Semantic Illusion”-effect
To account for this “Semantic Illusion”-effect, five complex neurocognitive processing models have been put forward. On the basis of a critical review of these models (Brouwer, Fitz, & Hoeks, 2012), we have concluded that none of these models succeeds in explaining all of the relevant findings in the literature. We argued that the reason for this failure is not architectural, but rather due to wrong interpretations of the processes underlying the N400 and the P600 component.
We proposed a reinterpretation of these components, in which the N400 is interpreted as reflecting lexical retrieval processes (cf. Kutas and Federmeier, 2000; Lau et al., 2008; van Berkum, 2009), and the P600 as semantic integration. These reinterpretations of the N400 and the P600 suggest that language processing proceeds in biphasic N400/P600—Retrieval-Integration—cycles. We aligned this Retrieval-Integration account with neuroanatomy, and derived a minimal functional-anatomic mapping centered around the left posterior Middle Temporal Gyrus (lpMTG; BA 21) as an epicenter or hub for retrieval (~ N400), and the left Inferior Frontal Gyrus (lIFG; BA 44/45/47) as an epicenter for MRC composition (~ P600) (Brouwer & Hoeks, 2013).
A Neurocomputational Model
We provided a proof of concept of our Retrieval-Integration account in terms of a neurocomputional model. This model, which is an anatomically-constrained Simple Recurrent neural Network (SRN; Elman, 1990), constructs a thematic role assignment representation for an input sentence on an incremental, word-by-word basis. During processing, correlates of N400 amplitude are measured in a layer representing the lpMTG, and those of P600 amplitude in a layer representing the lIFG. It is shown that our neurocomputational model produces ERP elicitation behavior similar to that of humans (cf. Hoeks et al., 2004), and explains the “Semantic Illusion”-effect.
For more information, please see: http://www.hbrouwer.eu/
|Laatst gewijzigd:||30 maart 2017 14:51|