Reconstructing Maps from Text

摘要: Previous research has demonstrated that Distributional Semantic Models (DSMs) are capable of reconstructing maps from news corpora (Louwerse & Zwaan, 2009) and novels Benesh, 2012). The capacity for reproducing is surprising since DSMs notoriously lack perceptual grounding (De Vega et al., In this paper we investigate the statistical sources required in language to infer maps, resulting constraints placed on mechanisms semantic representation. Study 1 brings word co-occurrence under experimental control demonstrate direct necessary traditional successfully reproduce maps. 2 presents an instance-based DSM independent frequency city names.