A Psychologically-Plausible Model for Spatial Language

EPSRC Grant GR/N38145

 

 

 

Summary     Objectives & Workplan       Background         Results        Publications

 

 

Summary

Computational models for spatial language have assumed that formalisms of geometrical space and abstract object modelling are adequate to account for the use and comprehension of spatial terms (Regier, 19967). However, there is growing evidence from psycholinguistic studies (see Coventry 1998 for a review) that talking about the spatial world involves extra-geometric variables (e.g. functions of objects, their shape and size) as well as geometric variables. The aim of this programme of work was to develop a psychologically plausible connectionist model for spatial prepositions which deals with both functional and geometric features.

The grant has resulted in a working neurocomputational model in which a visual input is presented to the model(e.g., a movie of a teapot pouring tea into/missing a cup), and the model returns a spatial description (e.g., teapot/over/cup)to describe the visual scene. The development of the model has been supported by new experimental results with human participants across a range of methodologies which show that the comprehension of spatial prepositions is influenced by a subtle interplay between "where" objects are in the scene,"what" they are and how they interact. Furthermore, given a shortened video showing intiial frames of a movie, the model and participants are able to predict where the liquid ends up, and this affects judgements of a range of prepositions for both model and participants. The experimental and computational work strongly supports the "functional geometric framework" to spatial language and spatial cognition.

People

Kenny Coventry (Principal investigator)

Angelo Cangelosi (Co-investigator)

Dan Joyce (now at UCL, London)

Lynn Richards (now at Peninsula Medical School, Plymouth & Exeter Universities)

Alison Bacon

Rohana Rajapakse

Related links

Spatial Language Group

Multi-agent Modelling of Language Evolution (EPSRC project)

Connectionist Modelling of Quantifiers (EPSRC project)

Instruction-based Learning for Mobile Robots (EPSRC project)

Laura Carlson at Notre Dame University

Simon Garrod at Glasgow University