Introduction

A significant part of our research effort is directed at the semantics component of the grammar equation. We are in particular interested in enabling artificial agents to autonomously conceptualize and interpret the kind of compositional semantics we want these agents to be able to express by means of the grammatical constructions under consideration in the language component of our research. The paradigms, goals and strategies of our research in general entails various requirements on the nature of such conceptual system. In particular the grounding and the separation of intension and extension are decisive design requirements.

fig 1. Integration of conceptual and language system

In line with the arguments of Winograd (1972), Davies (1972), Johnson-Laird (1977) and Hausser (2001), we postulate that such requirements necessitate some form of procedural semantics. We more specifically propose an embodied procedural semantics in which the intended meaning of an utterance is a structured composition of a number of computational primitives. Such a composition constitutes a program that the speaker wants the hearer to execute. The computational primitives employed in such programs are considered to be recruited from the general, non-language-specific, cognitive capabilities of the agents.

We choose to model the computational primitives as procedural constraints. These primitive constraints implement an omnidirectional relationship among a set of variables. By combining multiple primitives and having them hold over a shared set of variables, a compositional constraint system is composed. These composite constraint systems are what semantic programs are represented by.

Interpreting a semantic program is equivalent to solving a constraint satisfaction problem. Planning a semantic program, on the other hand, corresponds to solving what we call a constraint composition problem. A solver of such problems (the composer) takes an initial incomplete constraint system that has been set up in function of the communicative goal. For a referential communicative goal, for example, the initial state involves a variable to which the target referent is bound, while the context from which the target should be discriminated is provided by some arrangement of variables and context accessing primtives. The composer then adds constraints and variables until the system meets the requirements entailed by the linguistic domain. I.e. the system must be solvable and all decision variables must have a semantic domain, such as nominal categories, that can be lexicalized in an utterance.

Current research status

The core constraint satisfaction and composition engine has been largely completed. We are currently investigating the compositionality problem and how it can be solved bi-directionally in our systems. We are looking at how the system will abstract over successfully used compositions and make these available as composite constraints for re-use in incrementally more complex conceptualizations. We are also investigating which FCG-based mechanisms and rules are needed to enable a generic mapping between compositional semantic representations (in the form of constraint networks) and grammatically composed language utterances.