Getting the verb classes right
Levin and Hovav (2005: 13) note that ‘there is more than one way of semantically characterizing most verbs, and it is not always a priori obvious which characterization is appropriate for argument realization’. How do we know, for example, that cut, hack, bash and hit contain a motion component, whereas pat and touch don’t? It’s not at all clear. The fact that the classifications are often indeterminate in this way might seem to undermine the very idea of semantically based verb classes, and hence undo the attempt to derive argument-structure alternations from semantics. As a result, a critic could make the following objection to Levin and Hovav: if several different descriptions of the meaning of the same verb can be validly given, isn’t it arbitrary to claim that just one drives argument alternation? What about all the incompatible semantic descriptions that also fit the verb?
This objection raises an interesting issue about the process of empirical research in a domain like linguistics. We normally think of theories being based on some sort of solid evidence: we start out with an unambiguous set of observed facts, and then try to come up with a theory that explains them. The theory matches the facts, which are therefore evidence for the truth of the theory. For example, the facts on which Copernicus based his theory of the solar system are ultimately observations made of the sky. Because the Copernican theory fits these observations, we have a reason to think the theory is true.
This is only part of the story, however. Often, the facts aren’t solid and unambiguous. It’s frequently the case that we’re not exactly sure what all the observational facts actually are: maybe our measurements aren’t always precise enough, maybe there’s disagreement about whether a phenomenon is relevant, maybe there are conflicting observations about the same phenomenon. In the case of astronomy, maybe there weren’t good enough instruments to check the details of the heliocentric theory directly. In this sort of situation, it’s often argued to be appropriate to let the theory tell you what the facts are. As long as we have enough clear cases, it’s OK to allow the unclear ones to be interpreted in whatever way is most favourable to the theory. Just as we adjust the theory to reflect the facts, so we can sometimes adjust our idea of what the facts are in order to fit the theory. Applied to verb classes, if we follow this approach we could accept that the semantic classification of some verbs isn’t obvious. But we wouldn’t see this as a threat to the broader theory. Instead, we could just say that considerations from the theory as a whole allow us to resolve the uncertainties of verb meaning. Independent of the theory, there are arguments both for and against including motion in the semantic representation of pat and touch. But since the theory only works if we say that this component is absent, that in itself is all the justification we need.
