I have shown that lexical specification of defaults combined with syntactic control over the persistence of these defaults into the pragmatic component facilitates interpretation of logical metonymies in a discourse context. A purely pragmatic explanation (e.g. Hobbs et al 1993), i.e. an explanation not relying on lexicosyntactic factors but only on world knowledge and contextual influences, of such data would fail to account for the incoherence of the discourses beg30a,b and cog3a,b,c(i) in contrast to the coherence of the discourses beg30a,c and cog3a,b,c(ii).
The semantic representations for begin the book and begin
on/with the book in such an approach would be identical -- they
would both correspond to the logical form 
in
Figure 5.7, as there would be no lexical specification of
default interpretations assumed. There would therefore be no basis for
distinguishing their behaviour with respect to the influence of
discourse, even if we wanted to assume that the ability of discourse
information to influence the interpretation of a particular
construction were specified pragmatically. We probably do not wish to
make such an assumption in any case since the primary determining
factor of this behaviour seems to be syntagmatic knowledge of a specific
construction.
Furthermore, the specification of default interpretations of logical metonymies in the pragmatic component would result in a great loss of generality, because information such as the relationship between possible default interpretations and the semantics of the noun in the NP complement (i.e. the fact that default interpretations correspond to the telic or agentive roles) could not be captured in any straightforward way, and also because there are several different types of metonymies (not only verb+NP but also adjective+noun) which display the same interpretation patterns. A pragmatic approach would be forced to specify the default interpretation of each individual logical metonymy in an ad hoc manner.
A purely pragmatic analysis would also fail to accommodate lexical idiosyncrasies and the conventionalisation I have argued for throughout this chapter. How would the oddity of the began reading interpretation for John began the flyer be explained in light of its clear availability for John began the book? Surely the reading activity with respect to magazines is no less common than with respect to books (that is, reading must be the most common activity performed on both flyers and books). There is therefore no pragmatic basis for making this distinction. In fact, none of the data in beg74, repeated here, could be accounted for on a pragmatic account since all of the events specified in the intended interpretations are commonly performed on the objects denoted by the NPs. [5.28]
*John began the film. (watching) *John began the door. (opening, walking through) *John began the nails. (hammering in)