Compositionality, context, and cognition: comment on "Embodied language, best fit analysis, and formal compositionality" by J. Feldman.

Feldman’s project [1] has the ambitious goal of giving a compositional account of the “full range of meanings” in human language, one which meets the high standards of formalization found in formal semantics but is built on a theory of “embodied cognition”, takes into account the neurobiological substrate of language, and makes use of representations of world knowledge and discourse and situational context as well as linguistic knowledge. His exposition is marred by some misconceptions about compositionality and about formal semantics. My goals in this comment are (i) to clarify what is meant by compositionality in formal semantics and how formal semanticists capture more aspects of meaning than Feldman suggests, and (ii) to discuss the striking differences between the methods and goals of formal semantics and those described by Feldman concerning modularity and the division or lack thereof between theories of syntax and semantics and psycholinguistic theories of language processing. What is compositionality? First of all, the principle of compositionality in semantics concerns the relation between form and meaning; it does not apply to forms or to meanings separately. Hence it is not straightforward to make sense of a notion of “compositionality of thought”, which Feldman wants to take as more basic; perhaps ‘compositionality’ is being confounded with ‘productivity’, the way that wholes are built up recursively from parts. The principle of compositionality is commonly stated as follows: