134 Yves Peirsman, Kris Heylen and Dirk Geeraerts
- Conclusions
The main goal of this article was to illustrate the use of word space models
in theoretically-oriented linguistics. These word space models, which cap-
ture the semantic relationship between two words in terms of their distribu-
tion in a corpus, have now long proved their use in the field of computa-
tional linguistics. We argued, however, that they can equally be applied in
the more theoretical disciplines, and function as a tool for quantitative in-
vestigations in fields like variational linguistics or Critical Discourse Anal-
ysis.
Word space models come in many flavors. Those that describe a word in
terms of its syntactic relations are best suited to find paradigmatic similari-
ty, as between synonyms. This is because they uncover words in the corpus
that tend to have the same functions or roles as the target word. Those
models that describe a word in terms of the articles in which it appears, are
best geared towards the modeling of syntagmatic, topical relations. In this
way, they can be used to automatically define lexical fields and to measure
the distance of between two such fields.
By way of a case study, we have shown how these word space models
can be put to practice in the field of usage-based theoretical (socio-) lin-
guistics. In particular, we studied the use of the religion names islam and
christendom in Dutch newspaper articles before and after the attacks of
September 11, 2001. We were able to pin down some important changes.
Both the document-based and the syntax-based model showed that islam
has become distributionally more similar to words related to terrorism and
politics. Christendom, by contrast, was still mainly characterized by cultur-
al and more positive dimensions. Many of these results would be difficult
to find with a manual analysis. A sociolinguistic study of word fields in the
several newspapers showed significant differences in their treatment of
islam. For Algemeen Dagblad in particular we found a high relatedness
value between Islam and terrorism – due to the fact that these topics often
occur together in its articles. Of the five newspapers, NRC Handelsblad and
Trouw appear to focus most on the cultural and religious background of the
news events.
Evidently, this article has introduced the word space models in the most
concise way. For those readers interested in the technical details of the
models, there is a wealth of literature in computational linguistics and cog-
nitive science. Moreover, our case study has only given a brief sketch of
the many possibilities that the models offer. Their automatic approach