The formalisation presented here relies on the semantic representation introduced in Chapter 2 and the techniques for implementing lexical rules outlined in Chapter 3.
Constructions, as form-meaning pairs at the phrasal rather than lexical level, do not have any exact correlates in the standard form of Head-driven Phrase Structure Grammar (HPSG) as presented in Pollard and Sag (1994). However, the problem of licensing and integrating non-subcategorised elements (adjuncts in particular) into the representation of a phrase has been addressed by work such as van Noord and Bouma (1994) , and adopted in Chapter 3. Here I will also adopt the basic mechanism proposed in that work: lexical rules define different types of adjunction, altering the subcat list of a verb and the content associated with that verb as appropriate for each particular type of adjunction. Thus we can have a lexical rule specific to the Resultative Construction, which adds the noun phrase object if necessary and adds the resultative phrase, and modifies the representation of the semantics associated with the verb phrase as a whole. This approach avoids changes to the fundamental constructs of HPSG which would otherwise be required to capture the phrasal properties of constructions.
As a brief reminder, the lexical rules are formalised in terms of
recursive constraints on lexical categories which are processed using
delayed evaluation techniques. Information provided by a parser then
triggers the application of rules, thus allowing the linguistic
context to influence the lexical entry associated with a particular
use of a word. This conception of the lexical rules thus allows them
to capture constructions in the Goldberg (1995) sense -- the verb
itself does not undergo a meaning shift or acquire a permanent
additional sense encoded in the lexicon; instead a use of the verb in
a particular syntactic frame triggers the application of a rule which
specifies the appropriate meaning of that use.
This lexical rule will have two syntactic variations (one for each of
the intransitive and transitive base syntactic forms which a verb
might have before a resultative phrase is added), and three subcases
which specify distinct methods of semantic integration: one subcase in
which the resultative phrase is an Adjectival Phrase and two subcases
for the Prepositional Phrases, one for path PPs and the other for
locative PPs. The rule assumes in its present form that the semantic
relations between a verb and its arguments remain the same in the
resultative construction as they are in a non-resultative use of the
verb. This assumption means that sentences such
as the (a) sentences in resalt42-resalt43 are not licensed by
this construction. In these sentences the relationship between the
verb and the direct object differs from the standard relationship as
evidenced by the (b) sentences. (Sentences from Goldberg 1995, p.
154.)
Sam sawed/tore/hacked/ripped a piece off the block. Sam sawed/tore/hacked/ripped the block.
Sam rinsed/cleaned the soap out of her eyes. Sam rinsed/cleaned her eyes.
However, (1995) argue that these cases are not instances of the resultative construction, instead resulting from an alternate projection of the arguments of verbs of removal into the syntax. I accept their argumentation and do not perceive these as counterexamples for my proposal. Other apparent counterexamples, e.g. resalt44, stem from independently allowed intransitive variants of the matrix verb (1995). In these cases the construction adds the direct object to the intransitive variant.
Fred drank the teapot dry. [L&RH (1995), 65:(73)] Fred cooked the stove black. [Jackendoff (1990), 227:(38b)]
The distinction between path and locative PP types for the purposes of the lexical rule definitions exists because of a fact about the resultative construction identified by Goldberg (1995): the resultative construction allows PPs which cannot express a path independently of this construction to appear as the resultative phrase, with a path interpretation. So a locative preposition like inside can acquire a path interpretation in the resultative construction but cannot be used as a path in other contexts, as suggested by resalt30. As a result, locative prepositions are coerced into a path interpretation by the lexical rule. Specifically, the location they express will be incorporated as the PLACE argument of a path. This coercion follows from a relationship of endpoint focus (Brugman 1988) which exists between the meaning of the locative term and its directional interpretation -- the location expressed in the locative PP is the endpoint of a path.
John squeezed the rubber ball inside the jar.
*Inside the room he ran, quick as lightning.
on directed motion interp: He ran into the room quickly
There is an additional constraint on this path interpretation, in that the resultative phrases in this construction are interpreted as specifying a change of state/location with a clear end point. This is unlike the path PPs we find with manner of motion and sound emission verbs since they do not necessarily convey a change of location to a specific location. Thus we find a contrast in resalt28, but not in resalt29.
John sneezed the tissue to the other side of the room. *John sneezed the tissue along the wall.
John ran to the other side of the room. John ran along the wall/towards the store.
So all paths in the resultative construction must be instances of a to path rather than a toward or from path: that is, the path must lead to a specific endpoint. Locative prepositional phrases are straightforwardly added as the PLACE argument of a to path.
The lexical rule will perform the following functions:
The semantics given for the verb phrase corresponding to the construction will be virtually identical in each (sub)case. Differences will exist only in whether components of this meaning come from subcategorised or unsubcategorised elements, and in how precisely the resultative phrase is integrated. The lexical rules can be defined as in ResLRtop, building on the rules previously defined in Section 3.5.3.
The top level rule controls the addition of the resultative phrase and
the object NP if necessary, prior to the addition of any adjuncts to
the verb's subcategorisation list. The three clauses of
add_ResP handle the semantic integration of the three types
of resultative phrases, embedding the semantic relation normally expressed by
the main verb as a means component of the
cause-eff-means relation which is returned as the internal
semantics to be associated with the Resultative Construction. This
relation essentially conveys, for the structure
{NP 
V NP 
ResP}, the expected semantics NP acts
on NP , causing NP to {go to some location expressed in
ResP/change to some state expressed in ResP} by means of V-ing.
add_adj_top(Head 
,
SubcatIn 
,
SubcatOut 
,
SemanticsIn 
: 
,
SemanticsOut ,
Operator-adjsIn 
,
Operator-adjsOut 
) :-
test_subcat(SubcatIn , SubcatMid 
),
add_ResP(SubcatMid , SubcatMid2 
, Situation 
, SemanticsMid 
),
add_adj(Head , AdjunctList 
,
SemanticsMid , SemanticsOut ,
Operator-adjsIn , Operator-adjsOut ),
append(SubcatMid2 , AdjunctList ,
SubcatOut ).
Add an NP to an intransitive Subcat frame;
A transitive Subcat frame remains unchanged.
test_subcat( 
, 
).
test_subcat( 
,
).
Add different kinds of Resultative Phrases
Case one: Path PP
Case two: Locative PP (Coerce to Path interpretation)
Case three: Adjective Phrase
These rules depend on various representational assumptions which build on the representation introduced in Chapter 2. The lexical semantic structure for path and locative prepositions is expected to follow the examples given in Section 3.5.1 d62 and d65. The semantics of adjectives are assumed to be subtypes of a subtype of und-rel and state, be-rel, which predicate a property of an undergoer. The lexical entry for the adjective flat, for example, is found in resalt25 and has a semantics of flat-rel. This is essentially equivalent to the logical form flat(x).
The resultative interpretations which stem from these representations follow from inferencing over the effect values: if something goes to somewhere (go-rel plus to-path) or changes to a specified state (ch-st-result-rel), that thing can be inferred to be at that place or in that state.
