The Idiom Processing Advantage is Explained By Surprisal

Abstract

It has been repeatedly found that idioms are processed faster than syntactically matched literal phrases, in both comprehension and production. This has led to debate about whether idioms are accessed as chunks or built compositionally, with different studies attempting to measure the effect of compositionality on processing, with differing conclusions. This paper looks at idiom processing through the lens of information update, in particular surprisal theory, which is a standard theory of sentence processing. Compositionality is just one aspect of a word's predictability; we argue that surprisal, as an expectation-based theory, provides a more general unifying framework for understanding the idiom processing advantage. In this paper, comprehension and production experiments on verb-object idioms reveal that the idiom processing advantage can be largely explained by the fact that idioms have lower surprisal than matched literal phrases. The results indicate that the idiom advantage manifests primarily on the noun in verb-object idioms.