For the purposes of this thesis, Davis' emphasis on linking relations means that his representational architecture is not powerful enough. An issue which fails to be addressed adequately is that of complementary polysemy (Pustejovsky pustejovsky:95a), i.e. cases in which a single verb has multiple senses which are related to one another in some predictable way.
There are essentially two cases of polysemy relevant to this point. The first is that a verb can have a sense which derives from the interaction of its base sense with the semantics of a complement. This is the case for bake, for example, which can be interpreted as a change-of-state verb or as a creation verb with particular complements, as shown by dav10 (Pustejovsky pustejovsky:95a:122). This process is called co-composition.
John baked the potato. (change-of-state) John baked a cake. (creation)
The creation sense arises from the fact that a cake comes into
existence by baking, and so there is ``co-specification'' between the
verb and its complement. The difference between these two senses
cannot be captured in Davis' hierarchy as the semantic roles played by
the various semantic arguments are identical in each case, at least as
far as linking is concerned. There is no syntactic difference
displayed by these two forms. There is however a semantic difference
which is not reflected. This could easily be remedied by introducing
another subsort of und-rel, creation-rel, to correspond
to the entailment of creation of an entity.
Specific composition rules are then needed to control the
co-composition between the verb and complement to induce the shift
from ch-of-st-rel to creation-rel.
The second kind of complementary polysemy occurs when the primary semantic relationship between the semantic arguments is different in the different senses of a verb. In most cases of this kind of polysemy, the verb also appears in a distinct syntactic pattern for each sense. An example of this kind of polysemy is the causative alternation. Consider the sentences in dav14.
The ball rolled down the hill. John rolled the ball down the hill.
In dav14a we have an instance of a
move-displacement-rel, while in dav14b
we have an instance of a cause-move-rel. So each sense of
roll receives an entirely distinct relation under Davis'
architecture. Davis reflects correspondences between two senses with
statements about the relationship between two lexical semantic
structures; for the alternation in dav14 this would be something
like the statement in dav15. Note that Davis does not assume any
directionality in such relationships.
Davis states that it is not necessary to view such statements as lexical rules (ch. 3, p. 76), suggesting that verbs which exhibit alternations are systematically polysemous. He argues (ch. 7, pp. 3-6) that alternating verbs can be characterised as ``two senses sharing a lexeme'', for which the knowledge of relationships such as dav15 means that the presence of a lexical entry capturing either of its senses implies the existence of a lexical entry capturing the other sense. This perspective is directly in line with how I will suggest sense shifts should be viewed, but Davis does not explicitly address how he intends for this perspective to be formalised. In each of the following chapters of this thesis I will address this issue of the representation of polysemous verbs. The solutions will involve further use of the power of the multiple inheritance hierarchy and underspecification in lexical entries, following Bredenkamp et al bredenkamp:96 and marks_sadler:95.