Using weird verbs as term extractors

Jakob Halskov (jhalskov <at> dsn (point) dk)

Danish Language Council

Le 8 octobre 2008 à 11 h 30

Salle C-9019, Pavillon Lionel-Groulx

The Danish Language Council monitors usage trends in Danish, including the identification of neologisms (both domain specific and domain independent). To this end the council is developing a prototype application called the “Ordtrawler” which is being used for text retrieval and text analysis. The material analyzed by the Ordtrawler includes dictionary search logs, newspaper articles and scientific journals. Based on an Ordtrawler experiment with a POS tagged corpus of a Danish medical journal (some 1.5 million tokens) I will present preliminary findings on the utility of using “weird” verbs as term extractors. Statistical measures of “weirdness” (cf. Khurshid Ahmad: “Pragmatics of Specialist Terms: The Acquisition and Representation of Terminology”. EAMT Workshop, pp. 51-76, 1993) involve a comparison of term frequency in the analysis corpus versus a (general language) reference corpus. In Automatic Term Recognition applications, however, the measure has typically been used to extract domain specific nouns, but domain specific verbs also have a lot to offer as markers of terminological information in natural language text. The approach is inspired by Lotte Weilgaard’s studies of Danish verbs as knowledge probes (cf. “Danish Verbs as Knowledge Probes in Corpus-based Terminology work”, LSP and Professional Communication, vol. 2, no. 2, 2002) which have yet to be implemented and evaluated computationally. Preliminary empirical results indicate that log odds ratio is an efficient measure of “weirdness” (more so than log-likelihood) and that giving preference to term candidates which co-occur with “weird” verbs can boost the precision of the automatic term extraction.

Pour recevoir les annonces hebdomadaires par courriel, visitez

Liste des autres séminaires