| Abstract |
In the task of automatic lexical acquisition, i.e. the induction of lexical information from texts, there have been no attempts to exploit theoretically-based models of the structure of the lexicon. Works like those of Bybee (1988) and Langacker (1987) propose a highly structured lexicon where words are related paradigmatically by phonological similarity and where lexical features are an emergent characteristic of the resulting structure. If so, a machine learning algorithm such as a Decision Tree (DT, Quinlan, 1945) should be able to learn the correlation between particular lexical features and the formal characteristics of words. In our experiment, the machine learner should be able to find a correlation between characters that form the words used for training it and the nominal feature /mass/. The ability of the trained learner to predict correctly whether nouns that it has not been shown in the training phase are mass nouns or not is proof that such a correlation exists and that it can be considered an emergent feature of the paradigmatic relations that relate words in the lexicon. The obtained results prove that a structured lexicon can provide information on lexical features. |
| BibTex |
@InProceedings{ELX08-010,
author = {Núria Bel, Sergio Espeja, Montserrat Marimon},
title = {The Structure of the Lexicon in the Task of Automatic Lexical Acquisition},
pages = {285-290},
booktitle = {Proceedings of the 13th EURALEX International Congress},
year = {2008},
month = {jul},
date = {15-19},
address = {Barcelona, Spain},
editor = {Elisenda Bernal, Janet DeCesaris},
publisher = {Institut Universitari de Linguistica Aplicada, Universitat Pompeu Fabra},
isbn = {978-84-96742-67-3},
} |
