Michaela Socolof

A fundamental property of language is that words can be combined into larger phrases in systematic and predictable ways. This property, known as compositionality, allows language users to produce novel, complex utterances using set rules. Yet compositionality is not absolute—some structures, like idioms, involve meanings that combine in non-systematic ways. If we have a way of precisely quantifying compositionality, then we can investigate the relationships between compositionality, linguistic structure, and processing. This talk presents a novel, computationally-precise approach to formalizing compositionality using information theory, with computational experiments showing how the approach sheds light on irregularities and asymmetries in syntactic and morphological structure. The talk then addresses implications of non-systematic combining structures on sentence processing, focusing on idioms and relative clauses.