It has been noted many times in the literature that there seem to exist semantic constraints on the verbs which can appear with dative PPs. Jackendoff (1990:196), for example, proposes that only verbs of creation or preparation can appear in the double object form of for-beneficiary sentences. This follows from the fact that ``singing'' involves the physical formation of a song, ``baking'' involves bringing into existence baked goods, etc. and that only these type of verbs appear in this form. This notion of semantic constraints on the double object form of -datives also extends to -datives: only certain verbs can appear in the double object form with a dative interpretation. Levin levin:93 identifies the classes of alternating verbs as summarised in Table 3.2.
| Dative alternation (-datives) | Benefactive alternation (-datives) |
 |  |
 |  |
What is striking upon observation of that table is that the alternating verbs group into clear sets of semantically related verbs. In fact all of the verbs listed in the table, on their (di)transitive uses, express relations which are subtypes of cause-eff-rel. More specifically, the dative-alternation verbs all express the transfer of something somewhere or to someone (e.g. if you slide or throw or send a book, you are causing it to go somewhere), while the benefactive-alternation verbs divide into two sets: Jackendoff's creation and preparation verbs (which correspond to cause-create-rel and cause-ch-st-rel, respectively); and get-type verbs which are transfer verbs like the dative-alternation verbs, except that the location to which the object is transferred is encoded in the verb semantics to be the location of the entity referred to in the subject NP (e.g. John got a present (for Sally) where John first acts to transfer the present to his possession).
The conclusion that can be drawn from this is that only verbs expressing specific semantic relations are compatible with PPs expressing particular meanings. This point has also been argued by Wechsler wechsler:95, who suggests that complement PPs express a meaning which must be compatible with the type of the verb. Here I extend this notion to pseudo-complements as well. The constraints on which verbs can be extended by particular pseudo-complements (or appear with particular complement PPs) can therefore be captured in the semantic relation hierarchy. A possible structure for that hierarchy, as far as - and -dative verbs are concerned, appears in Figure 3.1.
Figure: Extended segment of the rel sort hierarchy
(them = thematic)
There are several things to note about this hierarchy. First is the
use of the thematic attribute them-arg (thematic
argument). The value of this attribute is constrained at the level of
cause-eff-rel to unify with the entity which is the
undergoer of the causation. This constraint is inherited by
each subtype at lower levels in the hierarchy, and ensures that the
thematic elements of the verbal relation are all relativised to this
same entity. The fact that this constraint is relative to a
particular relation type is justified by arguments made by Nam
nam:95 that when certain prepositional phrases modify an
n-ary function, they introduce properties restricting the
argument of that function (the thematic argument in my
terms). Thus different verbal relations, with different argument structures,
will have different thematic arguments and this can be efficiently
captured in the hierarchy.
The second point to note is the use of defaults. Here I am attempting to prevent arbitrary extensions of the thematic attribute associated with a particular semantic relation through grounding of the thematic type, while maintaining generality over the relationships between the verbal relations. Thus it is clear that the relation transfer-to-ben-rel (intended to accommodate sentences such as John gave the presents to Mary for the kids) is an extension of the relation transfer-to-rel and therefore should be (and is) a subtype of this latter relation. These relations both inherit a thematic value of effect from cause-eff-rel, but in the case of transfer-to-rel the thematic value must be restricted to a particular kind of effect (go-rel) which should not be extended with any other thematic information. A thematic value of effect is, however, compatible with other thematic information and could in theory be extended by pseudo-complements contributing other thematic information. In order to rule this out, the relation transfer-to-rel is given a default (node-specific) thematic value of effect_only, with which any pseudo-complement PPs must be compatible in order to modify a verb of this type. The subtype transfer-to-ben-rel, by similar reasoning, is given a default thematic value of effect_for-to_only, while its non-default value (effect_for-to) is an extension of the non-default value effect inherited from transfer-to-rel. If the (indefeasible) thematic value of transfer-to-rel had been specified as effect_only, it would have been impossible to capture the relationship between it and transfer-to-ben-rel due to the incompatibility between effect_only and effect_for-to_only. The use of defaults enables this intuitive relationship to be captured without losing the constraints which exist for individual verbal relations.
When the semantics of a PP is integrated with the verbal relation it modifies, it must unify with the default thematic value of that relation, or a subtype of that relation with which it is compatible. The effect of this is to ground the PP via unification (i.e. if the PP expresses an effect theme and it combines with a transfer-to-rel verb, it will be ground to type effect_only), and to only allow the PP to combine with verbs of compatible type (so a for-to PP might combine with a transfer-to-rel verb but the result would be a relation of type transfer-to-ben-rel).