Against the single-domain constraint

Warglien et al. (2012) extends the theory of conceptual spaces (Gardenfors, 2000) to the semantics of verbs. The basic proposal is that the representation of an event contains at least an object (called the patient) and two vectors, namely a result vector representing a change in the properties of the object, and a force vector representing the cause of the change, with the structure of the event being determined by the mapping from force vector to result vector. In order to capture the lexicalization constraint proposed by Kiparsky (1997) and Hovav and Levin (2010), Warglien et al. (2012) add the so-called singledomain constraint, which says that the meaning of a verb is a convex region of vectors that depends only on a single domain. In the first part of this commentary we argue, both on theoretical and empirical grounds, against the single-domain constraint, by (i) pointing out that learnability considerations motivate at most a strong tendency for verbs to lexicalize either manner or result, and (ii) by arguing that some verbs expressing force dynamics (e.g. uberzeugen ‘to persuade’, zwingen ‘to force/compel’) and some verbs involving conventional consequences (buy, sell, inherit, bequeath) lexicalize both a manner and a result component. In the second part of the paper we make some critical remarks on the prospects of vector-based theories of event structure. Basically, our complaint is that the reduction of events to vectors makes a sufficiently rich analysis of verbal meanings impossible, as it ignores the temporal contour. In effect, what the authors do is to represent only the result, and ignore the manner of an event.