The cores of the verbssmoke 3 ,‘smoke (a cigar)’, andsmoke 5 ,‘smoke (a ham)’, can be seen as variants on the structure
(27a) as well.Ignoring thespecialized manner modifiers discussed in section 11.3, these are roughly‘cause smoke T to
go out of a cigar’and‘cause smoke 1 to go into a ham’respectively. (I leave the tree structures as an exercise for the
compulsive reader.)
This exercise shows how decomposition into functions brings out explicit similarities and differences among verb
meanings, how these cut across patterns of syntactic expression, and how relations among readings of polysemous
items can be made explicit. In particular,“thematic roles”such as Agent, Patient, Source, and Goal are determined in
structural terms inCS, so their partiallackofsystematicityinsyntax (section5.8) is a functionofthewayCSmaps into
words and syntactic structure.
There is no roo mhere to go into more detail, in particular the fascinating cross-linguistic differences in patterns of
incorporation,first brought to light by Talmy (1980); the way semantics affects alternations in the syntactic argument
structure of verbs (Pinker 1989; Levin 1993; Jackendoff 1990a); the maximal event structure associated with a clause
(Levin and Rappoport Hovav 1996; Rappaport Hovav and Levin 1998; Nikanne 1990), and the decomposition of
preposition meanings (Brugman 1981; Herskovits 1986; Vandeloise 1986; Landau and Jackendoff 1993; and many
papers in Bloo met al. 1996). The overall picture, however, is that it is of great interest to construct such
decompositions. Eachprimitivefunction, relativizedtosemanticfield, gives a conceptaccess tocharacteristicinference
patterns; and in the spatialfield, to perceptual patterns as well.
Fodor (1998) objects to such decomposition, on the grounds that we have