CLASP
The Centre for Linguistic Theory and Studies in Probability

Affordances for syntax and semantics

The ability to predict action sequences on the basis of perception and past experience is essential for autonomous agents exhibiting intelligent behaviour in unstructured environments. I present a view of syntactic processing as sensorimotor skills developed to grasp reliable actionperception regularities. Under this view, natural language elements like words and syntactic structures can be subsumed in models of affordances, properties of social settings [Heft, 1989] relative to (groups of) human agents who can explore or exploit them to gain access to these settings. I will present the formal architecture of a syntactic model (DS, Kempson et al. [2001]) that provides a way of capturing (joint) linguistic and physical actions by relying on the goaldirected, predictive nature of cognition.

Since linguistic phenomena also show that there is continuity between low-level “syntactic” and conceptual mechanisms, I will address various ways to unify DS with TTR (Cooper [2012], in prep) under a single formalism. In my view, TTR types, like syntactic DS types, can also be modelled as affordances, namely, real features of the sociocultural environment that are, nevertheless, defined relative to agents’ abilities and preferences. Standard TTR types are then abstract, higher level goal macrostates, referring to a (potentially) infinite set of low-level state graphs probabilistically accessible to agents in a particular context. From this point of view, judgements are not assignments of type labels to entities but, instead, the pursuance of affordances, action policies, in interactions.