[4.43e]
[i] They laughed John out of the room. [ii] #They laughed John into the room/down the hall. [iii] #They insulted John out of the room.
[4.61a] John hammered the metal flat/*safe/*red. [4.71b] *The feather tickled her silly. [4.73a] The feather excited her into a frenzy.
I argued in Section 4.5 that the constraints on the resultative contruction derive from general pragmatic principles. Grice's Maxim of Relevance (Grice 1975) is a heuristic governing discourse coherence which suggests that each constituent in a discourse must be as relevant as possible to the current discourse context. In the case of the resultative construction, this heuristic can be taken to mean that a rhetorical connection must exist between the elements in the construction in order for them to be felicitously combined. In this section I will demonstrate that this maxim can be viewed not just as a constraint on attachment of discourse constituents within a discourse but also on the coherence of the semantics expressed by particular constructions, i.e. within a single constituent, and that the recruitment of this maxim to apply intra-sententially can help to explain the impact of pragmatics on the felicity of resultative constructions. This approach has been demonstrated as a useful strategy for modeling the interaction of pragmatic reasoning with lexical semantics in the cases of co-predication and coordination (Lascarides et al. 1996).
As the resultative construction essentially conveys a cause and effect relation between the event expressed by the main predicate and the resultative phrase, it must be possible to establish a rhetorical connection between the two components in order for the cause-effect relation to be coherent. The relevant rhetorical relation in this case is Result (Hobbs 1985, Polanyi 1985). This causal link must be established on the basis of world knowledge inferences and reasoning about information in the discourse. Sentences like res9e.ii.-res9e.iii. are therefore normally ruled out because the Result relation between the cause and the effect in each case (e.g. some people laughing and John going into a room) cannot easily be supported. I argue that in a highly specific context which establishes the Result relation, these sentences would be entirely felicitous as they would satisfy the Gricean Maxim of Relevance (Grice 1975).
Consider again the discourse presented in Section 4.4.3:
[4.63]
The slide at the park had a section which had come loose. Several children had hurt themselves on the protruding edge. In order to prevent further injuries, John hammered the metal safe.
Under normal circumstances, hammering metal does not cause the metal to become safe. The context preceding w13c in this discourse, however, establishes the basis for the causal relation -- the metal is unsafe because an edge is protruding, so it follows that an event which stops the edge from protruding (e.g. hammering it) will cause the metal to be safe. Once this basis is established, the constraint on the resultative construction is met and the sentence will be judged felicitous in context.
Let us formalise this reasoning. Following Asher asher:93a
and Lascarides and Asher lasc_asher:93, the Result relation
is constrained by the Axiom stated below, where e 
represents the main
eventuality described in the constituent x. This Axiom states that if the
Result relation holds between two discourse constituents, then there
must be a causal relationship between the eventualities they describe.
In addition I assume a rule specific to the Resultative Construction,
Resultative Link, which formalises the pragmatic constraints
on the semantics of that construction.
A distinctive feature of the theory of discourse semantics used by Lascarides and Asher (lasc_asher:91, lasc_asher:93) (Discourse in Commonsense Entailment, or DICE, in combination with Segmented Discourse Representation Theory, SDRT (Asher asher:93a)) is that it models this accommodation process: if two constituents are attached together with a rhetorical relation R and R's coherence constraints aren't met by the constituents already, then the required semantic content can be added in a constrained manner on the basis of world knowledge (through a process known as SDRS Update (Asher and Lascarides asher_lasc:96, Copestake and Lascarides copestake_lasc:97)). Intuitively, this models the way speakers expect hearers to fill in gaps when they interpret discourse. In DICE, rules reflecting pragmatic maxims and capturing the influence of world knowledge, syntax and semantics on interpretation are used to compute rhetorical relations. Default knowledge is captured in formulas of the form A > B which mean approximately ``if A, then normally B'', which allows nonmonotonic inference to take place. Details of the logic underlying such formulas can be found in Asher and Morreau asher_morreau:91 and Asher asher:93b.
In the context under analysis, DICE would have to reason on the basis of world knowledge about what kinds of things people do to make edges safe to accommodate a causal relation as required by the Resultative Construction (cf. Asher and Lascarides asher_lasc:96). The context in w13a-w13b establishes that there is an unsafe protruding edge on a slide, and resolution of the definite description the metal (on the basis of knowledge such as Slide Material below) will result in an initial interpretation for w13c of John hammered the unsafe protruding metal slide edge safe. To accommodate the cause relation, additional world knowledge must be recruited and added to the discourse structure. Here, the knowledge that a non-protruding edge is normally safe (e.g. Safe Edge) and that hammering an edge can result in it being non-protruding (e.g. Making Protruding Edges Non-protruding) can be added to the discourse, which together would be enough to support the causal relation between hammering a protruding edge and it being safe.
These pragmatic constraints can also work to rule out examples which based on general world knowledge might be acceptable but which don't make sense in context. That is, if discourse information suggests that a causal link is not valid, the coherence constraints on Result will not be met (discourse information will conflict with world knowledge in this case but discourse information is more specific and therefore ``wins'' over world knowledge -- see Lascarides and Copestake lasc_copestake:95) and so the sentence will be judged infelicitous in context.