Mathematical linguistics

Models for natural language

I am interested in formal models for natural language processing and description. Millstream systems ([D1] [C9] [C10]), for example, are a generic mathematical model for the description of language that makes it possible to formalise and reason about the interaction and interdependency of different linguistic aspects, such as phonology, morphology, syntax, and semantics. Linguistic aspects are described by any number of autonomous modules. The central component of a Millstream system is the interface consisting of interface rules in the form of logical expressions that establish links between the outputs generated by the autonomous modules. A Millstream system can be considered to perform independent concurrent derivations of autonomous modules, enriched by an interface that establishes links between the outputs of the modules.

Mild context-sensitivity

My research also concerns the concept of 'mild context-sensitivity' ([A1]), specifically investigations of mild context-sensitivity in (limited) parallel rewriting systems. This is interesting from a mathematical point of view, but may also be useful for natural language processing since humans seem not to process language sequentially, but in a parallel, albeit restricted mode ([C4] [E2]).

Formal language and automata theory

My research interests include the theories of formal languages and automata. For example, descriptional complexity of formal grammars (including CD grammar systems, Lindenmayer systems and tree adjoining grammars) ([B2] [B3]), as well as computational capacity of automata (like input-reversal and input-revolving automata) ([B4] [C5]).

Human-computer interaction

Dialogue modelling

I am interested in adequate formal models of dialogue structures. A human-machine interface for spoken language has to deal with various dialogue phenomena, such as disagreements, misunderstandings, change of opinion, change of topic, and there is a need for a stable formal model of dialogues. Co-operating distributed grammar systems, for example, can be identified to model dialgoue structure as joint activity ([B1] [C2] [E1]).

Learning from demonstration

Learning from demonstration is a well investigated robot learning technique but ambiguity in learning from demonstration is an overlooked problem. One of my ongoing task concerns defining ambiguity in this context and investigating disambiguation strategies.