Evaluating and Combining Approaches to Selectional Preference Acquisition

Evaluating and Combining Approaches to Selectional Preference Acquisition Carsten Brockmann School of Informatics The University of Edinburgh 2 Buccleuch Place Edinburgh EH8 9LW, UK Carsten.Brockmann@ed.ac.uk MireIla Lapata Department of Computer Science University of Sheffield Regent Court, 211 Portobello Street Sheffield Si 4DP, UK mlap@dcs.shef.ac.uk Abstract Previous work on the induction of se- lectional preferences has been mainly carried out for English and has concen- trated almost exclusively on verbs and their direct objects. In this paper, we focus on class-based models of selec- tional preferences for German verbs and take into account not only direct ob- jects, but also subjects and prepositional complements. We evaluate model per- formance against human judgments and show that there is no single method that overall performs best. We explore a va- riety of parametrizations for our mod- els and demonstrate that model combi- nation enhances agreement with human ratings. 1 Introduction Selectional preferences or constraints are the se- mantic restrictions that a word imposes on the environment in which it occurs. A verb like eat typically takes animate entities as its subject and edible entities as its object. Selectional prefer- ences can most easily be observed in situations where they are violated. For example, in the sen- tence "The mountain eats sincerity." both sub- ject and object preferences for the verb eat are violated. The problem of quantifying the degree to which a given predicate (e.g., eat) semanti- cally fits its arguments has received a lot of atten- tion within computational linguistics. Several ap- proaches have been developed for the induction of selectional preferences, and almost all of them rely on the availability of large machine-readable cor- pora. Probably the most primitive corpus-based model of selectional preferences is co-occurrence frequency. Inspection in a corpus of the types of nouns eat admits as its objects will reveal that food, meal, meat, or lunch are frequent com- plements, whereas river, mountain, or moon are rather unlikely. The obvious disadvantage of the frequency-based approach is that no generaliza- tions emerge with respect to the observed pref- erences as it embodies no notion of semantic re- latedness or proximity Ideally, one would like to infer from the corpus that eat is semantically con- gruent with food-related objects and incongruent with natural objects. Another related limitation of the frequency-based account is that it cannot make any predictions for words that never occurred in the corpus. A zero co-occurrence count might be due to insufficient evidence or might reflect the fact that a given word combination is inherently implausible. For the above reasons, most approaches model the selectional preferences of predicates (e.g., verbs, nouns, adjectives) by combining ob- served frequencies with knowledge about the se- mantic classes of their arguments. The classes can be induced directly from the corpus (Pereira et al., 1993; Brown et al., 1992; Lapata et al., 2001) or taken from a manually crafted taxonomy (Resnik, 1993; Li and Abe, 1998; Clark and Weir, 2002; Ciaramita and Johnson, 2000; Abney and Light, 1999). In the latter case …

Content truncated to 500 words