PhD ceremony Mr. D.J.A. de Kok: Reversible stochastic attribute-value grammars

PhD ceremony: Mr. D.J.A. de Kok, 16.15 uur, Academiegebouw, Broerstraat 5, Groningen

Dissertation: Reversible stochastic attribute-value grammars

Promotor(s): prof. G.J.M. van Noord, prof. J. Nerbonne

Faculty: Arts

A frequently discussed topic in linguistics is the question of whether a human has two independent `modules’ for language comprehension and production. In computational linguistics, comprehension (parsing) and production (generation) have been treated separately in recent history. The main thesis of De Kok’s work is that parsing and generation can be performed by computers using only one component without any degradation in performance compared to separate components for parsing and generation. The underlying theory is that many preferences should be shared between production and comprehension, for a parser to `understand’ the output of a generator.

To support this thesis, a generator was developer for Dutch. This generator was then integrated with an existing parser for Dutch. This work shows that there is indeed no significant difference between the performance of the integrated module and separate components for parsing and generation.

More insight into this combined model is gained through feature selection. Feature selection is a technique that isolates the most important properties that characterize a comprehendible and fluent sentence. This dissertation shows that comprehensibility and fluency can be determined using a small number of linguistically informed properties.