Volume 8 Issue 1 (2010)
Note: Linguistic Discovery uses Unicode characters
to represent phonetic symbols. Please see Optimizing Display
for requirements to accurately reproduce this page.
A Multitude of Approaches to Make Semantic Maps
Comment on ‘Semantic Map Geometry: Two Approaches’ by
Joost Zwarts (2010)
Max Planck Institute for Evolutionary Anthropology,
The opposition between the matrix-driven and the
space-driven approach to establishing semantic maps, as proposed by Zwarts
(2010), is unnecessarily confronting. I agree completely that there are
different techniques for arriving at semantic maps, and I wholeheartedly
subscribe to Zwarts’ plea for an alliance between these. However, I would
like to sketch a different typology of approaches to semantic maps, based on the
kinds of evidence for the relations between the points in semantic
Before I get to this typology, I would like to stress that there are two
independent aspects of semantic maps: the ‘points’ in semantic space
(viz. the functions/meanings investigated) and the relations between these
points. Ultimately, I think that the reason for selecting particular points in
semantic space is independent of the approach for establishing the relations
between them. In practice, a particular investigation will often choose the
points in connection with the method used to establish the relations between
them (so it appears that points and method are related). However, in principle,
fixing the points is a separate (and difficult) issue. The typology of semantic
maps that I will propose only deals with establishing the relations, not with
selecting the points.
In section 2.2, Zwarts mentions three different examples in which the
conceptual space is defined independently of the cross-linguistic data (his
“space-driven approach”): the color term case, with reference to
Regier et al. (2007), the reciprocal example from Dalrymple et al. (1998), and
over example from Tyler and Evans (2001). However, I do not see much
coherence in these examples, except for the fact that they do
cross-linguistic evidence. Taking these three examples of making semantic maps,
and adding the cross-linguistic approach as exemplified by Haspelmath (2003), I
would like to propose the following classification.
Relations between functions/meanings can be based on linguistic evidence or on non-linguistic evidence, and they can be based on evidence from one language or from multiple languages. This
classification results in three different possibilities because the parameter “one vs. multiple languages” only makes sense when using linguistic evidence. The analysis of over by Tyler and Evans (2001) presents an example of using linguistic evidence from just one language, while Haspelmath (2003) uses linguistic evidence from multiple languages. By contrast, semantic maps can also be based on non-linguistic evidence. One possibility is exemplified by Dalrymple et al. (1998), who establish the relations purely on the basis of the logical structure of the definitions of the “points” in
semantic space. Another non-linguistic approach is exemplified by Regier et al. (2007), who base the relations between their color chips on the physical characteristics of the colors. I do not find Zwarts’ opposition between approaches “using multiple languages” and approaches “not using multiple languages” very helpful, because the more crucial point of these different approaches seems to be whether linguistic information is used or not.
Finally, I would like to make a short comment on Zwarts’
distinction between “graph-based” and “scale-based”
semantic maps. He discusses some weaknesses of both of these approaches, and I
completely agree that neither of these approaches is ideal. However, I would
want to add that the pictures produced by such methods are not the real semantic
map. As I argue in this issue (Cysouw 2010), both the graph-based and the
scale-based visualizations should not be interpreted as the empirically observed
structure of geometrical space. Visualizations are useful for human beings to
make sense of the underlying metric, which is the real result of the
investigation of semantic relations. Any visualization (necessarily) abstracts
away from many details in the underlying geometrical space. This implies, for
example, that calculations (e.g. correlations between different data sets, or
significance testing) should not be done on the basis of the visualization, but
on the basis of the underlying original figures.
Cysouw, Michael. 2010. Semantic maps as metrics on meaning.
Linguistic Discovery, this issue. doi:10.1349/ps1.1537-0852.a.346
Dalrymple, Mary, M. Kanazawa, Y. Kim, Sam A. Mchombo, and S.
Peters. 1998. Reciprocal expressions and the concept of reciprocity. Linguistics
and Philosophy 21/2.159-210.
Haspelmath, Martin. 2003. The geometry of grammatical meaning:
Semantic maps and cross-linguistic comparison. The new psychology of language:
Cognitive and functional approaches to language structure, ed. by Michael
Tomasello, vol. 2, 211-242. Mahwah, NJ: Erlbaum.
Regier, T., P. Kay and N. Khetarpal. 2007. Color naming reflects
optimal partitions of color space. Proceedings of the National Academy of
Tyler, Andrea and Vyvyan Evans. 2001. Reconsidering prepositional
polysemy networks: The case of over. Language 77/4.724-765. doi:10.1353/lan.2001.0250
Zwarts, Joost. 2010. Semantic map geometry: Two approaches.
Linguistic Discovery, this issue. doi:10.1349/ps1.1537-0852.a.357
Author's contact information:
Department of Linguistics
Max Planck Institute for Evolutionary Anthropology
Deutscher Platz 6