A core tenet in linguistic theory is the principle of
compositionality, which holds that the meaning of a multi-word
utterance directly derives from the meanings of the individual words,
and the rules by which they are combined. Attempts at unifying formal
and computational models of lexical and compositional semantics have
proven challenging and often yield complex frameworks, in which word-
and utterance-level meanings are patched together to form a whole,
without fully integrating their semantic contributions. In this talk,
I revisit the principle of compositionality from the neurocognition of
language, which reveals that the human comprehension system harnesses
distinct models for lexical and compositional meaning, and that these
models are critically intertwined in a cyclic architecture for
language comprehension. I present the outlines of a neurocomputational
model that implements this notion of compositional integration, and
discuss the implications of this integrative approach towards modeling
the meaning of words and utterances.