He laughed himself to death. *He laughed himself dead.
He laughed himself to sleep. *He laughed himself sleepy/asleep.
He laughed himself out of a job. *He laughed himself jobless/unemployed. *He laughed himself out of the room/down the hall.
He laughed himself silly. He laughed himself faint/dizzy. ??He laughed himself tired.
They laughed John out of the room. #They tittered John out of the room. #They laughed John into the room/down the hall. #They insulted John out of the room.
He danced himself to fame. *He danced himself famous.
He danced his feet sore. *He danced his feet to soreness. ?He danced himself sore. *He danced himself crippled.
On the basis of these examples, it can be concluded that there are clearly specific lexical and semantic constraints on the resultative construction which must be identified and incorporated into any treatment which hopes to achieve a complete model of the constructions. There is no obvious way to incorporate such semantic constraints into the model given that their treatment concentrates purely on syntactic structure and does not involve lexical semantic features of individual words.
Even minor variations in some component of a felicitous resultative construction results in an infelicitous instance. The variations displayed in res9a-b and res10a are straightforward syntactic substitutions, where the resultative phrases have the same semantics but different syntactic form. Each variation conveys the same the result state. The general treatment give of resultatives cannot account for such differences, as they do not make any distinctions on the basis of the form of the resultative phrase. The type of the resultative (change of state or change of location) only affects the behaviour of the manner of motion verbs/verbs of sound emission and is therefore only an issue for these verbs. All other resultative constructions are seemingly allowed on their account, regardless of the form or type of the resultative phrase. Hence the approach also cannot account for the infelicity of the semantic substitutions in res9c-e and res10b, in which the resultative phrases are exchanged for syntactically identical but semantically distinct, albeit closely related, phrases of the same syntactic type, a verb is interchanged with a semantically related verb, or a reflexive of one sort is exchanged for another reflexive.
The idiosyncratic nature of these constructions suggests that the view of resultatives as phenomena governed solely by syntactic constraints is inadequate to fully account for their behaviour. However, a purely semantic treatment of resultatives would also not be able to explain these substitutions. Syntactic substitutions with identical resultative interpretations could not be distinguished on a semantic account. Additionally, the semantic distinctions involved in semantic substitutions would need to be extremely fine-grained in order to adequately model the data above.
How can the restrictions on the productivity of this phenomenon be explained? The conventionality of the resultative construction must be acknowledged. Rather than assuming the existence of any systematic semantic constraints which govern contrasts such as those in res9c.i vs. res9c.iii and res9e.i vs. res9e.ii, certain uses of the resultative construction seem to be restricted solely on the basis of partially lexicalised instances of the resultative construction, which, like idioms, allow little variation in their component parts. Thus some mechanism for encoding conventional constraints is needed.
Furthermore, pragmatics plays a role in the felicity of resultative constructions. Discourse coherence is essential. This was observed by Simpson (1983), who suggested that the only predicates consistent with a change of state can occur in the resultative construction. I suggest that this ``consistency'' is relative to the discourse context in which the predicate appears. Thus the constrast between example res9e.i and res9e.ii can be explained on the basis of discourse coherence. The discourse must support a link between the cause and the effect expressed in the resultative construction. In the case of res9e.iii, there is no obvious way in which laughing at someone can cause that person to go into a room or down a hall, while it is more clear how the laughing can cause that person to go out of the room where people are laughing at him. This example also shows clearly discourse coherence constraint interacts with conventionality: there are certain cause-effect links which must be conventionalised in world knowledge.
Lastly, syntactic form must play a role in the interpretation of resultative constructions, as the contrast between the i. and ii. sentences in res9a-b shows. Prepositional phrase and adjective resultative phrases seem to behave differently in the resultative construction, such that semantically equivalent resultative phrases in one syntactic form are felicitous in one sentence and infelicitous in others (also consider res10b.i-ii). Thus reference to the syntactic form of the resultative phrases is also critical to the modeling of the resultative construction.