Lexical acquisition for computational grammars. A unified model

PhD ceremony: Mr. K.D. Cholakov, 14.30 uur, Academiegebouw, Broerstraat 5, Groningen

Dissertation: Lexical acquisition for computational grammars. A unified model

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

Faculty: Arts

Words are the building blocks in the implementation of many natural language processing systems. The lexical information in such systems is usually encoded in lexicons where the words are mapped to linguistic descriptions. However, lexicons will always be incomplete. Natural language is constantly evolving and new words emerge every day. It is impossible to list each word in a language in a lexicon. Kostadin Cholakov’s thesis describes a novel automated lexical acquisition model.

Cholakovs model learns the morphosyntactic properties of words which are not listed in lexicons employed by computational grammars of natural language. Two major aspects of the model set it apart from existing lexical acquisition techniques. First, it enables the acquisition of the full morphological paradigm of the unknown word. Second, different contexts of this word are considered during the acquisition process. This increases the amount and the diversity of the linguistic information available for the unknown word.

For each unknown word, a set of linguistic features is constructed automatically. Those features are used as an input to a statistical classifier which maps all forms in the paradigm of the unknown word to descriptions in the lexicon of the grammar. The lexical acquisition model is tested with computational grammars of Dutch and German. The results demonstrate its high-quality performance. Further, the model is applied to learn proper linguistic descriptions for words with wrong or incomplete entries in the lexicon of the grammar.

Finally, the work in this thesis goes beyond syntax. The lexical acquisition model is combined with vector-based semantic space techniques to acquire semantic properties of unknown words.