Abstract. Computational approaches in spatial language understandingdistinguish and use dierent aspects of spatial and contextual information.These aspects comprise linguistic grammatical features, qualitativeformal representations, and situational context-aware data. Weapply formal models and machine learning techniques to map spatialsemantics in natural language to qualitative spatial representations. Inparticular, we investigate whether and how well linguistic features canbe classied, automatically extracted, and mapped to region-based qualitativerelations using corpus-based learning. We structure the problemof spatial language understanding into two parts: i) extracting parts oflinguistic utterances carrying spatial information, and ii) mapping theresults of the rst task to formal spatial calculi. In this paper we focuson the second step. The results show that region-based spatial relationscan be learned to a high degree and are distinguishable on the basis ofdierent linguistic features.
展开▼
