I have pointed out at several points in this chapter that the resultative construction seems to be governed in part by conventionalised constraints. A framework is needed in which the conventionalised nature of the resultative construction can be captured. To serve this purpose, I will adopt certain principles of the Construction Grammar approach to grammar. These are outlined below with respect to the analysis of the resultative construction proposed by Goldberg goldberg:95. I will in addition assume the existence of standard compositional interpretation mechanisms, which will be applied whenever possible to maintain maximum flexibility and generality in the interpretive process (e.g. to handle standard adverbial modification).
The Construction Grammar approach assumes that form-meaning correspondences are the basic units of language. Within this framework, constructions exist independently of the particular words which instantiate them. Each construction has a specific syntactic configuration which is associated with a specific semantics (and possibly certain pragmatic properties) -- each construction specifies the semantic roles of different syntactic positions and relations between the roles, and the semantics of the words which appear in the construction must fuse with the semantics of the construction itself. This approach is similar in spirit to Jackendoff's (jack:90, jack:96) analysis of certain phenomena, including the resultative.
The analysis
of the resultative construction given by Goldberg (1995) does not
treat all of the constructions discussed elsewhere under this name as
uniform. In fact, she isolates four different constructions (cf.
Jackendoff 1990), which are divided into two groups: the
Intransitive Motion Construction which is a subpart (in terms of
syntactic and semantic specifications) of the Caused Motion
Construction, and the Intransitive Resultative Construction
which is a subpart of the Resultative Construction. The four
constructions and their specifications are summarised in resalt3,
where OBL stands for oblique, solid lines between
semantic roles and roles in the PREDicate's role array
indicate that the semantic role must be fused with an
independently existing verbal participant role, and roles
represented in bold are profiled arguments -- that is,
entities in a verb's semantics that are ``obligatorily accessed and
function as focal points within the scene, achieving a special degree
of prominence (Langacker 1987)'' (Goldberg 1995, p. 44).
The I 
label on the arrow between
constructions indicates that the construction at the head of the arrow
is a subpart of the construction at the tail.
The existence of these various constructions is motivated on the basis of data in which the semantic interpretation of a sentence as a whole cannot be attributed to the meaning of the main verb or any composition of the meanings of the sentential components. I will not delve deeply into the arguments Goldberg presents that the meaning of these constructions does not straightforwardly come from the meaning of the words in the constructions (cf. the discussion of the interpretation of manner of motion verbs in Section 4.3.2). I simply cite one of her strongest arguments against the adequacy of strictly lexical compositionality for determining this meaning (Goldberg 1995, p. 154-155):
[I]t is fallacious to argue that because we may be able to pragmatically infer the meaning of a construction, its existence is therefore predictable rather than conventionalised. Such reasoning is based solely on a model of interpretation, yet we must also account for production.The relevance of this point is that it may be possible for a purely compositional analysis of the resultative construction(s) to account for the interpretation given to of one of these constructions, but it cannot explain how the constructions are licensed, given that the resultative phrase is not normally considered to be subcategorised by the verb (as it does not play a clear role in the verbal relation) yet does not behave as an adjunct (as it influences the ``internal'' semantic relation expressed in the main clause of the sentence). In fact, this criticism can be applied to the Levin and Rappaport Hovav (1995) analysis of unergative resultatives, as that analysis does not explain why it is possible to predicate something of an entity which isn't an argument of the verb. This point poses a further challenge in the attempt to model these constructions: there must be some theory of why particular syntactic configurations are possible. The Construction Grammar approach provides this explanation directly through the assumption of form-meaning pairs which license and constrain particular syntactic patterns.
The interpretations Goldberg associates with the various constructions should be clear from the representations she postulates for them, as presented in resalt3. They can be summarised as in resalt4 (compare with the meanings identified in Section 4.3.1 and summarised below in resalt2).
Caused Motion: NP causes NP to move to Result Location Intransitive Motion: NP moves to Result Location Resultative: NP causes NP to become Result State Intransitive Resultative: NP becomes Result State
Each different construction, then, has a meaning independent of the others and therefore different thematic constraints. Furthermore, each has a meaning independent of the words which instantiate it. Instantiation-specific meanings result from the merging of the meaning of the construction with the meaning of the main verb, which is seen as an instance of the main semantic relation (CAUSE-MOVE, MOVE, etc.) or as specifying the means by which that semantic relation is accomplished. For example, in John hammered the metal flat, hammer specifies the means by which the CAUSE-BECOME relation is achieved and so the meaning of this instantation of the resultative construction is John caused the metal to become flat by means of hammering.
Another property of this account is that certain semantic restrictions on the instantiation of these constructions can be shown to fall out from the thematic constraints encoded in the construction definitions. The sentences in resalt5 are ruled out because the object of the verb must be construable as a patient in this construction, but cannot be construed as such in those sentences. The objects fail the traditional test for patienthood (*What Sue did to the monster was watch it/*What happened to the book was Sue kept it) and so the sentences are not compatible with Goldberg's Resultative Construction.
*Sue watched the monster to sleep.
on interp.: Sue caused the monster to fall asleep by
watching it
*Sue kept the book dirty.
on
interp: Sue caused the book to become dirty by keeping it
Furthermore, the restriction that among the only unergative verbs which can appear in the ``unaccusative resultative construction'' (using the label give to the particular syntactic construction of NP+Verb+ResP) are the manner of motion verbs falls out from the definition of the Intransitive Motion construction: the predicate must be an instance or means of MOVE-ing, which manner of motion verbs clearly are and very few other verbs are. This leaves, however, the verbs of sound emission which can appear in this construction unaccounted for.
Goldberg (1995) also identifies general semantic constraints which help to account for the apparent idiosyncrasy of these constructions. These are summarised resalt6 below.
Constraints on the Caused Motion construction
These constraints restrict the instantiation of the constructions. Thus the constraints in resalt6a can be shown to account for the differences in resalt50-resalt8; in particular via the agent/natural force constraint resalt50 and the constraint that the motion must be directly caused by the action (the motion in resalt7b is secondary to the shooting and there is no obvious causation between laughing and causing someone to get into his car). Similarly the constraints in resalt6b can account for the differences in resalt9-resalt10: resalt9 due to the animate agent constraint and resalt10 due to the end-of-scale constraint.
The sound waves blasted the dust off the table. *The loudspeaker blasted the dust off the table.
Pat shot the bullet across the room. *Pat shot Sam across the room. (unacceptable on the interpretation that Pat shot Sam and the bullet forced him across the room)
They laughed the poor guy off the stage/out of the auditorium. ?They laughed the poor guy into his car.
She slept herself sober. *The feather tickled her silly.
He drank himself drunk/sick/dead. ?He drank himself a little sick. *He drank himself funny/happy.
Several of these constraints, however, could be explained by more general pragmatic principles which govern the use of these constructions. I would like to suggest that rather than assuming constraints on the semantics of the construction itself, it is preferable to assume that these constraints apply at the pragmatic level, to determine the felicity of the utterance within a discourse context, as the constraints derive from general pragmatic principles rather than construction-specific properties. I will explain how in what follows.
As suggested in Sections 4.3.3 and 4.4.3, the causation expressed in the resultative constructions must be coherent within the discourse context. The relationship between the causing eventuality and the caused eventuality must be clear from either world knowledge or discourse context. I refer the reader to example w13 and the preceding discussion, and the analysis of that example found in Section 4.7. Pragmatic reasoning there is seen to be triggered by requirements derived from the (fixed) interpretation dictated by the construction.
Furthermore, the constraints as expressed by Goldberg are defeasible. Consider for example resalt19, in which the instrument of the causation is specified rather than the agent, in apparent violation of the constraint in resalt6a.1. Relevant world knowledge or an appropriate discourse can seemingly license violation of Goldberg's proposed constraints. This defeasibility is a typical property of pragmatic constraints, in contrast to semantic constraints which should not be dependent on pragmatic reasoning.
The feather excited her into a frenzy. The work pushed him to the brink of insanity.
These observations then lead to a reinterpretation of the constraints on these constructions in terms of pragmatic principles. Grice's Maxim of Quantity (Grice 1975), for example, provides the basis for the direct causation constraint in resalt6a.2. This Maxim states (i) make your contribution as informative as is required for the current purposes of the exchange and (ii) do not make your contribution more informative than is required, that, for example, a speaker will not use a construction whose primary semantic content is caused motion along a path if he wishes to focus on an aspect of an event other than caused motion. Verbs which express a primary effect other than motion, like the sense of shoot in resalt7b, would therefore not be used in this construction. Thus I would argue that the direct causation constraint is actually nothing more than a restatement of this pragmatic principle, specified to the case of the Caused Motion construction. The other constraints in resalt6 (with the possible exception of resalt6b.4 which seems to be an idiosyncratic syntactic constraint) can similarly be argued to follow from the need to firmly establish the causal relation during pragmatic reasoning.
Although the Construction Grammar approach is highly appealing, not all syntactic patterns are suitable for treatment via constructions (see Jackendoff jack:96 for a summary). Instances where the meaning of a sentence can be accounted for entirely in terms of lexical composition and where its syntax is licensed by other generative aspects of the grammar should probably not be handled via constructions. There certainly seem to be relevant instances in which a constructional approach can be overrestrictive, or at least misses some more general syntactic or semantic process underlying the generation of a particular phrase. Consider the sentences in resalt11, for example.
John danced mazurkas across the room. John walked the dog to the store. John swam laps to exhaustion. The children played leapfrog across the park.
These sentences correspond approximately to Goldberg's Intransitive Motion construction in terms of their basic meaning, yet do not fit this construction syntactically due to the presence of a verbal noun phrase complement. That noun phrase complement makes the activity expressed in the verb more specific -- John didn't just dance in haphazard fashion across the room, he danced mazurkas -- but the basic interpretation of motion along a path is preserved. Syntactically, these examples fit the Caused Motion Construction, but the noun phrase object can hardly be construed as a theme, since the noun phrase object in each instance is not the entity involved in the motion expressed by the verbal predicate (mazurkas and leapfrog do not change position), so this construction cannot be applied either. Goldberg would be forced to assume a constructional variant of the Intransitive Motion Construction which licenses the noun phrase complement. This, however, seems unnecessary when a straightforward compositional analysis can account for both types of sentences, as will be shown in Section 4.7.2.
Other examples which Goldberg's analysis would have difficulty accounting for are found in resalt100-resalt101.
The pebbles rolled smooth. John ran to exhaustion.
*Mildred exercised into the room.
In resalt100, a manner of motion verb appears with a non-locative resultative phrase. The interpretation is not an instance of the Intransitive Motion construction (CAUSE-MOVE), but rather the Intransitive Resultative construction (CAUSE-BECOME). Because of Goldberg's requirement of compatibility between the predicate expressed by the verb and that expressed by the construction, however, it is impossible for a motion verb to be licensed by the Intransitive Resultative construction. resalt101, in contrast, would be compatible with the Intransitive Motion Construction since it contains a MOVE verb and therefore would incorrectly be predicted to be grammatical. Again the problem hinges on the compatibility between the predicate expressed by the verb and that specified by the construcition.
