Computational Linguistics Colloquium

Tuesday, 25 June, 14:15, Dekanatssitzungssaal Building 10
NOTE UNUSUAL DAY, TIME and PLACE

Metonymy is the figure of speech where "we are using one entity to refer to another that is related to it", according to a well-known definition by Lakoff and Johnson (1980). Typical examples are "The ham sandwich is waiting for his check" or "I read Shakespeare". Metonymy is pervasive in natural language and its importance has been shown for anaphora resolution, machine translation and question answering.

Although several automatic metonymy resolution procedures have been developed in the past, they share the following drawbacks: they are scarcely or not at all evaluated, the evaluation often uses toy texts or is performed using only the intuition of the developer as comparison standard, the methods use extensive hand-modelled lexical or world knowledge and are not robust.

I, therefore, propose metonymy resolution procedures that are firmly based in corpus investigations and human annotation standards and use statistical methods comparable to methods used in word sense disambigation.

In this talk I will present: firstly, the, to my knowledge, first metonymy annotation scheme which has been rigorously evaluated for reliability on natural language texts and an overview of the problems associated with metonymy annotation; secondly, a corpus annotated with this annotation scheme; thirdly, a novel view of metonymy resolution as a classification task using the categories of this annotation scheme as well as the experimental results of some statistical classification methods (decision lists, naive Bayes) using various feature types for metonymy recognition.

If you would like to meet with the speaker, please contact Katrin Erk.