Ferdinand de Haan, U of Arizona

On Representing Semantic Maps

One of the most intriguing ideas to come out of the typological linguistic research is the application of semantic maps for cross-linguistic research. The idea is similar to the semantic networks of AI, but there are some important differences. Semantic maps (AKA mental maps, semantic space, mental space) are used to describe, and to make predictions about, the patterns of multifunctionality of grammatical morphemes. Semantic maps consist of nodes and links between those nodes, thus making a network. The nodes and links are derived via cross-linguistic research and morphemes of individual languages are then mapped onto this semantic space (see Haspelmath 2003 for an introduction of semantic maps). This presentation will discuss the computational applications of semantic maps in frameworks like EMELD which rely on ontologies for cross-linguistic comparisons. A simple example (from Haspelmath 2003:236) is shown in (1), a semantic map of tense/aspect. There are three nodes and two links:

(1) habitual - progressive - future

The prediction this map makes is that there are languages that use the same morphemes for habitual and progressive (the Spanish Present), languages that use one morpheme for progressive and future (the English Progressive) and languages that use one morpheme for all three (the German Present), but that there are no languages that use one morpheme for habitual and future, but not progressive.

This presentation will focus on how to represent semantic maps computationally and what is needed on the linguistic side to make semantic maps work. I.e., what must be present in the database to construct semantic maps. Also, how can semantic maps be constructed from existing data? The linguistic illustration for this presentation will be the concept of irrealis. It is well-known that morphemes which has been labeled as irrealis in the literature differ wildly in their semantic content (cf. De Haan forthcoming), so much so that using the label "irrealis" for such morphemes is misleading at best and false at worst. In keeping with Best Practice principles it is perhaps better to use a semantic map approach in which morphemes are defined according to their position on the semantic map.

REFERENCES
De Haan, F. (forthc.) "Typological approaches to modality". To appear in W. Frawley (ed.) The expression of modality in natural language. Berlin: Mouton.

Haspelmath, M. (2003). "The Geometry of Grammatical Meaning: Semantic Maps and Cross-Linguistic Comparison." In M. Tomasello (ed.) The new psychology of language, Vol 2. Mahwah, NJ:Erlbaum, 211-42.