The Mesoamerican languages are renowned for shape discriminations. The Mayan language Tzeltal is no exception, and the theoretical implications are here explored for one area of the vocabulary that plays an important role in locative descriptions. Tzeltal body-part terms are" metaphorically" mapped onto parts of inanimate objects strictly according to a complex volumetric analysis of shape. This raises a number of fundamental issues: a. In what sense is this a" metaphorical" process? b. What is the relation of the volumetric analysis reflected in the language to the volumetric analysis involved in visual object recognition? The answers given are the following: a'. The mapping of body-part terms to shapes is not done by any form of creative analogy, but by a precise geometrical algorithm. Contrary to assumptions in the Mesoamericanist and cognitive linguistics literature this then has few of the properties of metaphor. b'. There is a coincidence between the kind of volumetric analysis involved in visual object recognition and that involved in the application of these terms.

The conclusions from (br) may be far-reaching. According to modularity arguments, linguistic processes should have no access to strictly visual processes. Although the present facts are not decisive, together with other observations they favor models where there is shared linguistic and visual access to the underlying processes of volumetric shape analysis. The paper suggests that cross-linguistic data might play an important role in general speculations about the relation between different kinds of mental representation, and that the Mesoamerican languages might have a special pertinence to the relation between visual and linguistic representations.