Problems with cognitive semantics
Cognitivist analyses of meaning focus on metaphor and metonymy, encyclopaedic meaning description and semantic extension, and enable a much more detailed representation of semantic content than is possible in more formal or componential approaches. The radial network models and image-schema diagrams allow a rich description of meaning that seems to make contact with perceptual and cultural aspects of language – aspects which are easily left out in other types of analysis. In spite of these attractions, however, cognitivist analyses of meaning have been criticized for a number of reasons. We will consider three:
• the ambiguity of diagrammatic representations;
• the problem of determining the core meaning; and
• the indeterminate and speculative nature of the analyses.
Diagrams like those in Figures 7.3 and following have often been criticized as inherently ambiguous. If the diagrams are to correctly indicate the meanings of the prepositions, they must not be ambiguous with entirely different concepts. This is exactly the same requirement we would place on a verbal definition: if a definition of over is to be accepted, it must correctly distinguish the meaning of over from that of other, non-synonymous expressions. The diagrams in Figure 7.3 and following, however, fail to meet this very requirement. For instance, consider Figure 7.8. This is presented as an explanation of the meaning of over, but there would seem to be many other ways of taking it. For example, how do we know that it is not intended as a representation of the verb hover? Any theory which rep resented over and hover as having identical meanings should surely be rejected. In a similar spirit, Figure 7.10 seems just as suited as a representation of the words round or curve as it does of the reflexive sense of over. The problem here is that the highly abstract nature of the diagrams allows them to apply well beyond their intended range. Over does not have the same meaning as hover, round or curve, yet the representations can be interpreted as referring to these concepts. This is clearly an indication that there is more to the meanings than can be captured in a visual format. If image schemas are advanced as underlying the use of terms like over, the diagrammatic representations we give of them seem not to go nearly far enough in bringing out the full details of the meanings involved.
Another problem concerns the radial model of category structure dis cussed in 7.2.5. The main cognitivist model assumes a central meaning which serves as a basis for various metaphorical and metonymic extensions from it. The problem here is that it is often hard to pin down exactly what the nature of this basic core meaning is and, as a result, exactly when we have a metonymic or metaphorical extension from it. For example, consider the metonymies in (13a) and (13d) above. In (13a), for instance, what makes us so sure that we have a metonymic extension from the basic head ICM? How do we know that this ICM excludes the interpretation ‘thinking people’ that seems active in this context? After all, it is part of the broad knowledge we have of heads that they come attached to people. Similarly, in (13d), can we be certain that the interpretation ‘headache’ results from a metonymic extension? We know that people often suffer from pains in the head; why should this information be considered as an extension from the basic head ICM, and not part of it?
Lastly, cognitivist theories are often criticized for their arbitrary and speculative character. The model of categorization proposed in cognitive semantics is offered as a psychologically realistic model of conceptualization, but has not yet been subjected to significant psychological experimentation, although this is beginning (Boroditsky 2000; Boroditsky & Ramscar 2002; Matlock et al. 2005). The analyses have largely been based on linguistic evidence – a problematic state of affairs for a theory which wants to develop a psychologically realistic model. Kamp and Reyle point to the circularity of any attempt to explain meaning (‘content’) by way of the mental representations lying behind uses of language:
it won’t do to base whatever one has to say about mental representations of content solely on what can be learned from studying the linguistic expressions through which these contents are publicly expressed, and then to offer mental representation as explaining the content of the corresponding expressions of the public language. (Kamp and Reyle 1993: 10)
In other words, since the main evidence for the nature of the underlying representations (ICMs, image schemas, etc.) is language itself, we have no independent means of checking that the putative representations are in fact psychologically realistic! If we want to explain the meanings of words in a psychologically realistic way, we need to do more than simply develop a theory which fits the evidence of the words themselves: there is a crucial additional step, which is to look for non-linguistic evidence of conceptualizations, and to conduct and develop experimental ways of testing the models advanced in linguistics. Without this experimental, psychologically grounded work, much about the theory seems unmotivated, and there is little to stop theorists developing models in a psychological vacuum, with nothing other than quite general hunches about what a psychologically realistic model of the mind should look like. Lakoff’s analysis of over, for instance, has spawned a huge amount of discussion and alternative models (Vandeloise 1990, Dewell 1994, Kreitzer 1997, Tyler and Evans 2001), with, as yet, no clear way of discriminating between them. Clearly, the development of psychological or other means of testing cognitive semantics models would be welcome. In the meantime, many cognitivists would probably acknowledge that these concerns are challenges for the model. But they might also counter that the benefits of the approach – rich meaning description – are more than reason enough to pursue the programme in the hope of resolving any outstanding problems.
