Reconstructing Maps from Text
This addresses the problem of understanding semantic representation mechanisms in computational linguistics, but it is incremental as it builds on prior map reconstruction studies.
The paper investigates the statistical sources in language needed to infer maps using Distributional Semantic Models (DSMs), finding that direct co-occurrence is necessary for traditional DSMs, while an instance-based DSM can reconstruct maps independently of city name co-occurrence frequency.
Previous research has demonstrated that Distributional Semantic Models (DSMs) are capable of reconstructing maps from news corpora (Louwerse & Zwaan, 2009) and novels (Louwerse & Benesh, 2012). The capacity for reproducing maps is surprising since DSMs notoriously lack perceptual grounding (De Vega et al., 2012). In this paper we investigate the statistical sources required in language to infer maps, and resulting constraints placed on mechanisms of semantic representation. Study 1 brings word co-occurrence under experimental control to demonstrate that direct co-occurrence in language is necessary for traditional DSMs to successfully reproduce maps. Study 2 presents an instance-based DSM that is capable of reconstructing maps independent of the frequency of co-occurrence of city names.