Tough times for language comprehension models

It’s a bit of a miracle: brain activity related to language comprehension can be measured as extremely weak electrical signals on the scalp. If the sentence ‘On his sandwich, Jan was having…’ is finished off with the word ‘socks’, this leads to the reader having an N400, a negative peak in the signal that occurs 400 milliseconds after the reader registers the word ‘socks’. Most researchers think that this N400 is a measure of the ease with which a sentence is understood. Harm Brouwer, however, manages to prove this wrong in an article in Brain Research, which is almost certain to result in at least five language comprehension models being discarded.

In 2004 Brouwer’s academic supervisor, John Hoeks, and colleague Laurie Stowe investigated how sentences such as ‘The white teeth have brushed the child’ are processed. They expected that readers would have difficulty in understanding such sentences, which would then lead to an N400. This did not prove to be the case. Does this mean that language users glibly accept that teeth can brush children? Or are readers perhaps – temporarily – in the clutches of a ‘semantic illusion'?

New models

Research groups around the world have been coming up with new language processing models to explain this phenomenon ever since. These models, which strongly differ in their specifics, are often extremely complex. What they do have in common is that they have an extra ‘route’ built into them, where one meaning of a sentence can be created based purely on the meaning of its separate words: ‘teeth, child, brushing, yeah, that sort of fits…’. This is the only way the classic interpretation of the N400 can be justified as a measure of language comprehension, under the motto: ‘It does nearly fit, so there’s no N400’.

Brouwer manages to prove in his article that extra routes and increasing complexity are to no avail: the new models are incapable of explaining current research results. The reason for this is an extremely simple one: the N400 is not a measure for language comprehension.

Mental lexicon

But what’s the deal with the N400 when you have ‘socks’ on your sandwich? According to Brouwer the N400 is purely a measure of looking up the meaning of a word in an imaginary dictionary in the mind. And this mental lexicon is built in such a fashion that after the words ‘teeth’ and ‘child’, the meaning of ‘brushing’ can be retrieved more quickly. ‘Socks’, however, does not carry this advantage: the words ‘Jan’, ‘sandwich’ and ‘is having’ do not even begin to suggest that a word like ‘socks’ could be in the offing.

P600

Of course language comprehension does not end when the meaning of a word has been retrieved. The meaning of a word needs to be integrated with the meaning of the words preceding it into a coherent whole. Brouwer thinks that the building of such a coherent whole can be discerned in a slightly later shift in electrical brain activity: the P600, a positive peak that usually occurs some 200 milliseconds after the N400.

Researchers in the field will have to get used to this new interpretation, given that ever since the 1990s the P600 has been linked to making or reviewing the syntactic analysis of a sentence.

Groningen researchers consider the P600 to be much more than that. It reflects the way a language user imagines what the other person wants to say. This has clear consequences for future research: the P600, not the N400, appears to be the key to how language is understood by humans.

Curriculum Vitae

In 2010, Harm Brouwer graduated cum laude in Cognitive Neurosciences at the University of Groningen. For his degree, he specialized in computational modelling of human linguistic capability. Brouwer is currently an NWO PhD candidate at the University of Groningen, working on a neurocomputational model of language comprehension. John Hoeks graduated cum laude in Cognitive Psychology and Artificial Intelligence, obtained a PhD in psycholinguistics, and has been assistant professor with the department of Communication and Information Sciences since 2006.

Note for the press

For more information: Harm Brouwer MSc

A preprint of the article ‘Getting real about Semantic Illusions: Rethinking the functional role of the P600 in language comprehension’ is available at: http://hbrouwer.github.com/papers/Brouwer2012GettingReal.pdf