The relation types in Figure 2.2 require further explanation. They are based, as should be clear by now, on the entailments that hold of the participants in each particular relation. Davis draws mostly on Dowty (1991) and Wechsler wechsler:91 for the basic entailments which he identifies.
I begin with the relations defined in terms of act and und. The proto-role act corresponds generally to an actor in an event, and und to an undergoer. These are specified to more particular entailments in subsorts of act-rel and und-rel. In vol-rel the act participant is required to be volitional with respect to the corresponding event, in incr-th-rel the und participant is required to be an incremental theme, and in ch-of-st-rel the und participant is entailed to undergo a change of state.
A relation in which one entity causally affects another is represented by a cause-rel relation, for which the causer corresponds to the act proto-role and the affected entity to the und participant role. This is a subtype of the act-und relation because it requires (at least) two participants and associates particular entailments with those participants. The subsort cause-eff-rel applies to causative verbs which express that some event is caused by the causing event. For these verbs, the entity affected by the causing event is also the actor of the caused event. The representation Davis assumes of this follows pinker:89 who in turn follows jack:90: an event characterised by some relation rel is embedded under an effect attribute. Davis gives the example (1995, ch. 3:38,40) in dav2 of the relation cause-jump-rel, a subsort of cause-eff-rel for which the embbedded event is a jump-rel event.
Chris jumped the horse over the gate.
Similarly, cause-means-rel is meant to characterise relations which entail a particular event which is interpreted as the means by which the expressed causation is achieved.
The notion-rel derives from a suggestion by wechsler:91 that there are verbs which require a sentient participant which necessarily has a notion of (perceives) the other participant, but does not require the perceived object to be sentient with respect to the event. Consider an example from Wechsler (1991, ch. 2, ex. 81), also cited by Davis (1995, ch. 3:36), shown in dav1. Although normally when two people marry each person has a notion of the other, it is not entailed by the semantics of the verb but rather seems to be constrained by world knowledge.
The duke married the two year old princess. #The two year old princess married the duke.
This relation subsumes the semantics of mental state, perception, and volitional action verbs, such as remember, like, see, hear, chase, and murder.
The influence-rel stems from the discussion of control phenomena in pollard_sag:94, as introduced in Section 2.3, and is meant to reflect the semantics of verbs like urge and persuade, for which one participant influences another in some way. The entailments associated with these roles are left rather vague by Davis, because there do not seem to be many clear necessary entailments associated with the influence and influenced roles. This is suggested by dav4 (Davis 1995, ch. 3:47), which shows that the urger/persuader need not be sentient or have a notion of the urged/persuaded individual.
The letter/police persuaded/urged Sandy to leave.
The relations rooted at fig-grnd-rel are intended to correspond
to the thematic tier proposed by Jackendoff (1983, 1990) and
introduced in Section 2.2.1
(page 
). That is, these relations capture the
motion and location of participants in an event. Davis thus encodes
Jackendoff's two tiers through the distinction between
act-rels/und-rels and fig-grnd-rels. The two
tiers are brought together, to capture both actor/patient relations
and motion/location relations in a single lexical structure, in types
which inherit from both kinds of relations. This merging is motivated
by Davis in terms of linking: for transitive verbs such as pass
and cross motion is associated with the argument in subject
position while for causative verbs such as throw and deflect
motion is associated with the noun in object position (Davis 1995,
ch. 3:43), suggesting that both the entailments of motion and
causation must be taken into consideration for accurate mapping of
arguments to syntactic position.
The fig-grnd-rel captures all figure/ground relations. Its subtype mot-rel, motion relation, entails relative motion of two participants -- the object in motion is denoted by the value of the feature fig, the stationary object by the value of grnd which must be a path. Davis adopts Jackendoff's representation of paths and places. The semantics of enter, for example, is represented by Davis as in dav3. The enter-rel is a subtype of mot-rel, and indicates that the entity denoted by the value of fig enters the place denoted by the value of in.
The type mot-rel can therefore be seen to correspond directly to Jackendoff's go function. The type self-mot-rel, which inherits from both mot-rel and act-rel and constrains the value of fig to be structure-shared with the value of act, captures the semantics of verbs which express that one of the participants in the event acts agentively and moves.
Verbs of impingement, e.g. hit, poke, tap, express contact of their subject with their object as a result of motion. So impinge-rel inherits from both mot-rel and influence-rel to capture both the contact and motion aspects of this meaning. The feature structure associated with the sort impinge-rel in Figure 2.2 expresses these two aspects and requires the actor to the participant that moves, and the influenced entity to be the point of contact.
Verbs of launching vs. entrained motion, such as those discussed in Section 2.2.2 with respect to example ji26, are captured by Davis in terms of the sorts cause-mot-rel and entrained-mot-rel, respectively. Both specify that an effect of motion is caused, and the latter relation additionally indicates that the causer accompanies the causee along its path.
The relation loc-rel is for location relations, for which Davis argues (1995, ch. 3:57-61) there do not seem to be strong entailments for the locatum and location. In fact, Davis suggests that individual verbs idiosyncratically specify which of its arguments corresponds to fig (locatum) and which to grnd (location). In some cases this assignment must even be determined pragmatically.
Possession (e.g. as expressed by have, own, acquire) can be considered a kind of location relation -- the possessed entity is located at the possessor. Similarly, the whole/part relations expressed by verbs of inclusion (contain, include, etc.) also correspond to location -- the part is located at the whole. Thus the relations poss-rel and wh/pt-rel are subsorts of loc-rel. The entailments associated with fig and grnd are in these cases, however, stronger than for simple location relations.
The last sort in Figure 2.2 which needs to be discussed is acc-ev-rel. This sort introduces a feature acc-ev, which corresponds to `accompanying event'. This feature is instantiated in a relation which entails that two specific subevents occur simultaneously. An example of such a relation is the exch-rel, for verbs of exchange such as buy and sell. In the case of these specific verbs, one subevent involves the transfer of goods and one subevent involves the transfer of money. This is represented by Davis (1995, ch. 3:63) for the case of sell as in dav5a. Here the annotations specify what participants in the selling event the various semantic roles correspond to. The cause-poss-rel is a subsort of caus-eff-rel, as indicated in the sort hierarchy.
Note the correspondence between Davis' representation and Jackendoff's (1990:191) representation of the same verb in dav5b. Jackendoff's cause function is captured by the fact that exch-rel is a subsort of caus-poss-rel. The go function is captured by the subevent embedded under Davis' effect attribute, and Jackendoff's exch (a modifying function in his system meaning ``in exchange for'') introducing a transfer of money event corresponds to the embedding of a cause-poss-rel event under Davis' acc-ev attribute.
Figure 2.4: Summary of functions in Jackendoff (1990)
Figure 2.5: Correspondences between Jackendoff functions and Davis relations
