Sentence Comprehension as [Statistical [Constraint Satisfaction] Learning]

MacDonald, M. C. .

Statistical learning (SL) research and work on sentence comprehension can seem like ships passing in the night, sharing the same waters but heading for different destinations. Studies of SL of language often investigate interpretation of meaningless words outside of context, frequently focusing on a single factor, such as transition probability between adjacent words. By contrast, sentence comprehension research addresses how people?s rich history with language enable them to weigh extensive probabilistic information to interpret the input. This work often focuses on the long distance dependencies that are pervasive in natural languages but are sometimes thought to be a challenge for SL. The fields also recapitulate the divide between child and adult psycholinguistics more broadly, where behavior in SL tasks is said to reflect ?learning,? while results in sentence processing studies reflect ?comprehension?. Despite these forces dividing the fields, there is nonetheless ample opportunity for rapprochement, beginning with the point that the probabilistic constraint use investigated in sentence processing research must have arisen via SL. I will discuss natural and artificial language studies that aim to integrate these fields and consider how to promote further collaboration.