next up previous
Next: Constraints on the Model Up: Modeling Language Acquisition Previous: MeaningConcepts, and Perception

Horizontal and Vertical Approaches to Language

Language is too large a domain to deal with in its entirety, and language scientists must slice up the problem space in some way or another. Normally the slices made are horizontal. A body of research covers some aspect of language, for example, syntax or the syntax of relative clauses, or some form of linguistic behavior, for example, syntactic parsing or the parsing of relative clauses. The goal is relatively thorough coverage of the behavior. Contact is often not made with other aspects of language or linguistic behavior; for example, research on syntax may not make reference to phonology or pragmatics and research on parsing may not make reference to production or acquisition. And contact is even less often made with non-linguistic aspects of cognition or with the external world.

Vertical approaches, on the other hand, make explicit contact across different aspects of language or linguistic behavior or across the boundary between linguistic and non-linguistic. While such approaches are ``tall,'' they are of necessity also ``thin;'' they can cover only very narrow aspects of language or linguistic behavior. Vertical approaches are associated in particular with cognitive linguistics, with anthropological linguistics, and with sociolinguistics. Figure 4 illustrates horizontal and vertical approaches to language.



Figure 4: Vertical and Horizontal Approaches to Language. Approaches may cover crucial interactions in toy worlds (vertical) or a single domain in a relatively broad fashion (horizontal).

While they have the advantage of broad coverage of a domain, horizontal approaches may miss crucial interactions between domains. A model such as the one we proposed in the last section, one in which language and non-linguistic perception exert mutual influence on one another, obviously requires a vertical methodology. Language, vision, and non-linguistic concepts must all be taken seriously. This is a tall order, enough to daunt even someone who believes in the sorts of interactions we are suggesting. Such an approach can succeed only if

  1. the range of linguistic phenomena covered is very narrow
  2. there is a body of established results or a coherent theory to guide the modeling in each of the relevant domains.

We have chosen to focus on the language of spatial relations and how it emerges in children. Besides the cross-linguistic study of how space is depicted linguistically and the acquisition of spatial language, our modeling will take us into the vision system, in particular, the visual representation of relations, and into the development of concepts of space in children. We believe that progress in each of these areas has reached the point where one may attempt to tie the various pieces together. Our eventual contribution will be an integrated picture of the development of spatial relations, linguistic and otherwise. In the following sections we summarize briefly some relevant facts from these four areas. First we discuss the language of spatial relations and how spatial relations are acquired in different languages. Then we talk about the perceptual end, about vision in particular, and about the acquisition of spatial concepts independent of language. Finally we look at the interactions between the two ends.

next up previous
Next: Constraints on the Model Up: Modeling Language Acquisition Previous: MeaningConcepts, and Perception

eliana colunga-leal
Mon Jun 23 04:27:19 EST 1997