On metonymy recognition for geographic IR

Metonymic location names refer to other, related entities and possess a meaning different from the literal, geographic sense. Metonymic names are to be treated differently to improve performance of geographic information retrieval (GIR). This paper presents a method for disambiguating location names in textual information to distinguish literal and metonymic senses, based on shallow features. The evaluation of this method is two-fold: First, we use a memory based learner to train a classifier and determine standard evaluation measures such as F-score and accuracy. Second, we perform retrieval experiments based on the GeoCLEF data (newspaper article corpus and queries) from 2005. We compare searching for location names in an index containing their literal and metonymic sense with searching in an index containing literal senses only. Evaluation results indicate that, using a large annotated corpus of location names, a classifier based on shallow features achieves adequate performance, and removing metonymic senses from a database index yields a higher performance for GIR.