Modelling Ordinary Natural Language

I have developed a strong interest for questions related to the computational modelling of unrestricted, ordinary natural language, and for the constraint-based approaches in particular. Natural language must be taken here in its

ordinary, common-sense notion (...) under which we can say that The Times in the UK, The New York Times in the USA, The Sydney Morning Herald in Australia, and other newspapers around the world, all publish in the same language --- though of course we would not deny that there may be local differences concerning which expressions are judged grammatical by the relevant editors. (Pullum and Scholtz, 2001)

Model-Theoretic Syntactic Parsing

MTS offers a radically different conception of what constitutes the syntactic description of an utterance than traditional Generative-Enumerative Syntax (GES). It was argued in several occasions in the literature that MTS is much better suited than GES for describing ordinary language---especially linguistic phenomena such as fragments, graded grammaticality, lexical openness, ... Yet, and despite that feature, MTS parsing has been poorly studied.

Different works showed that MTS parsing can be conceived as a configuration task, and implemented as a Constraint Satisfaction Problem (CSP). An interesting and promising aspect of configuration tasks is that they are found in very large-scale industrial problems, such as products configuration or software configuration. Solvers were successfully implemented in these areas, where a very high level of robustness is required---as it is the case when parsing ordinary language. Therefore an appealing question is to ask how, and to what extent advances in configuration problem solving may serve natural language parsing. Ultimately the aim would be to see the parsing activity just as any other large-scale configuration task, so as not have to worry about algorithmic aspects of constraint satisfaction, but rely on existing optimised generic solutions, and thus concentrate on problems of language modelling.

Applications for Language Technology

Model-theoretic, constraint-based parsing opens new perspectives for applications to grammar checking, language learning, and robust parsing more generally.

Please send comments or queries about this web site to jpprost@ics.mq.edu.au