The functions which Jackendoff (1983, 1990) assumes are summarised in Figure 2.4. A direct comparison of these functions with the semantic relations proposed by Davis can be found in Figure 2.5. It is obvious from this comparison that Jackendoff's representation has wider coverage than that discussed by Davis. The functions which have no correlates in Davis' representation have particularly to do with the internal structure of objects -- orient is for verbs like point in The sign points toward New York (Jackendoff 1990:44) for which an object is oriented along a path, ext is for verbs like extend and stative go as in The road goes from New York to San Francisco which describes the extension of an object along a path, and conf is for verbs like stand, and sit (Sally stood for hours on end, Jackendoff 1990:91) which express the internal spatial configuration of their argument.
The verbs which are represented using these functions impose very specific entailments on their arguments, ones which do not correlate well with the proto-roles and associated entailments Davis proposes, as shown in davis-pr. Those proto-roles are appropriate only for events rather than states, with the exception of the fig and grnd features in certain relations (e.g. loc-rel). However, the entailments of fig/grnd as presented by Davis do not quite reflect the same entailments as the functions orient, extend and conf. In the case of orient, the oriented entity is not located at specific location or contained within a path but has an orientation along that path. Similarly for extend, the extending entity covers the full extent of a path rather than being contained within the path. The function conf seems to convey a kind of fig-grnd-rel in which the single argument fills both the fig and grnd roles. I propose to introduce three subsorts of fig-grnd-rel to capture these specific entailments.
There are certain other discrepancies of coverage between Jackendoff's functions and Davis' relations. The function STAY, for example, does not have a clear correlate in Davis' sort hierarchy. However, Jackendoff (1990:44) suggests that this function might be unnecessary, since instances of this function can correspond to standard relations with the additional requirement that the relation holds over a certain period of time. So the sentence Bill stayed in the kitchen can simply be represented by a subsort of loc-rel which adds the entailment of stasis. This function will therefore not be added to Davis' hierarchy as an independent relation.
The function INCH can easily be accommodated in Davis' framework. As indicated in Figure 2.5, this function is similar to ch-of-st-rel in that a particular state results from the event. However, the subject of a sentence instantiating that relation is entailed to be an undergoer. The inchoative interpretation of sentences like those in dav17, which expresses an event of a particular state occurring (i.e.\ dav17a could be paraphrased as Bill came to be standing on the table), seems to be more accurately represented by a subsort of act-rel in that the subjects of the sentences act, effecting the change of state.
Bill stood on the table. Snow covered the hills.
In the event interpretation of these sentences, Bill and
snow not only participate in the resulting state, but they seem
to be causally involved with achieving the result. Contrast this with
the examples in dav18 which only have a stative interpretation:
the event interpretation is impossible because the subjects of these
sentences cannot be construed as actors.
The vase stood on the table. The blanket covered the table.
I therefore will introduce a subsort of act-rel, act-inch-rel, corresponding to Jackendoff's INCH function which is defined for an actor and a stative effect. The semantics of sentence dav17a can be represented as shown in dav19.
Jackendoff also extends the use of his functions to semantic fields other than the spatial domain, where some concept is abstracted yet conceptually related to spatial motion. He provides the following examples (Jackendoff 1990:25), arguing that the sentences in dav20, in the spatial semantic field, are similar to those in dav21, in the possessional semantic field.
The bird went to the tree. The bird is in the tree. Harry kept the bird in the cage.
The inheritance went to Philip. The money is Philip's. Susan kept the money.
There is a basic sense of the verbs which seems to be conveyed in each of the parallel sentences: the (a) sentences convey a change (of location in dav20 and of possession in dav21), expressing a GO function, the (b) sentences express the result state of the (a) sentences, a BE function, and the (c) sentences ``denote the causation of a state that endures over a period of time'' (Jackendoff 1990:26), treated as a STAY function in the original Jackendoff analysis. These abstract concepts are made precise in the different semantic fields, and associated with field-specific inferences, but certain generalisations follow from the abstract concepts.
To a certain extent Davis acknowledges these relationships, for example by the fact that his poss-rel is a subsort of loc-rel: so possession is a more specific kind of location. However, there is no relation in Davis' hierarchy at the level of generality of Jackendoff's GO function. As is clear from Figure 2.5, the spatial and possessional variants of Jackendoff's GO function correspond to entirely distinct relations in Davis' hierarchy. This treatment causes Davis to miss certain generalisations. He proposes, for example, a representation for the semantics of the verbs give and send as a cause-poss-rel, as in dav22 (based on Green 1974 and Pinker 1989).
Jackendoff, in contrast, treats these verbs as conveying a caused
GO 
function. The Davis/Green/Pinker semantics of
give/send could be paraphrased as ``cause to
possess'' whereas Jackendoff's semantics of these verbs is ``cause to
receive''. In some sense these are almost equivalent, in that it can in most
cases be inferred from a GO function that the person
at which the ``go-ing'' entity ends up possesses that entity. But the
Davis/Green/Pinker semantics has stronger entailments and therefore
runs into difficulties.
Firstly, it is only an intended effect of a send event that the entity to which something was sent possess it. This is not an entailment of the verb. Consider dav23.
John sent a letter to Mary, but it got lost in the post and she never received it.
In this case to say that Mary possesses the letter is patently false, although the sending event is completed. It is more accurate to say that the effect of sending is that the letter started along an abstract path (ending at Mary) than that the effect is that Mary possesses the letter.
Secondly, things do not have to be sent to entities which are capable of possessing things. Consider dav24.
John sent a package to New York. John gave $100 to the cancer research fund.
For dav24a, it can be inferred that the package will be acquired by someone in New York but this is not expressed by the lexical semantics: there is no participant in the sentence which necessarily comes to possess the package. Similarly for dav24b, in what sense can a fund possess a contribution? Such cases are more appropriately analyzed by Jackendoff's abstract GO function, leaving further reasoning for pragmatics.
Thirdly, the semantics in dav22 completely ignores the contribution of the to prepositional phrase. It contributes a path which is not evident in the representation in dav22. I will argue in Chapter 3 that the semantics of prepositional phrases should in many cases not be ignored, even for verbs which obligatorily subcategorise for them. This has also been suggested by gawron:86. Doing so misses generalisations which can be made about the PPs, and fails to capture a relationship between the verbs which subcategorise for these PPs.
For these reasons, I will introduce a relation go-rel under fig-grnd-rel with subsorts mot-rel and ch-poss-rel (change-possession-relation). The former subsort will be defined as in Figure 2.2 and will correspond to the spatial GO function. The latter will be defined as in dav25 and will correspond to the possessional GO function. Other subsorts could be added for other semantic fields, as necessary. The relation poss-rel will remain in the hierarchy for verbs (such as have) which really do convey this relation, but will not be used for change-of-possession relations.
