|
Computational psycholinguists study human language processing by
comparing quantitative predictions from language models to data from
psycholinguistic experiments. If a model can account for human data, it
may form a valuable cognitive model of language processing. The type of
data most often used in this context are word reading time
measurements, but several groups have recently begun to investigate if
(and how) neural activation can be related to computational language
model predictions. Indeed, generative language models can help explain
brain activation during reading or listening to text. I will present
some recent results and discuss what they imply for theories of human
language comprehension. |