...specification

Copestake's examples of this are the fact that a pile of small feathers can be referred to as feathers or down and that in Italian spaghetti is count while in English it is mass.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...properties

For example, the lexical process of grinding underlies the use of rabbit on an animal use in The rabbit ran across the field (count) and on a meat use in We had rabbit for dinner (mass).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...ji2c,

Note that there is structural ambiguity in both ji1b and ji2b. The prepositional phrase at the office can be either a NP or a VP modifier. We assume the VP attachment reading for the purposes of this discussion.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...context,

When one utters sentences such as those in ji21 (from Jackendoff 1983:49), there are constraints as to what each of the italicised elements can refer to.

 

Your coat is here. [pointing] He went thataway. [pointing] Can you do that/do this? [pointing/demonstrating] You shuffle cards this way. [demonstrating] The fish that got away was this long. [demonstrating]

These constraints are imposed by the lexical semantic contexts in which the elements appear. The hearer of such sentences can only interpret the anaphors in a way compatible with the type of entity required by the context. This ``entity type'' must correspond to a particular ontological category. This is evidence for use of the ontological cateogories as building blocks for conceptual structure - without such general categories there would be no way to constrain the entities that could fulfill function argument positions in structures capturing the meaning of particular expressions.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...phenomena.

See discussion in pollard_sag:94, pp. 275-277.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...roles.

Other researchers, e.g. dowty:91 and wechsler:91, deny that an explicit representation is necessary, instead arguing that judgements made on the basis of certain lexical entailments are sufficient for explaining linking.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...determined

Idiomatic expressions may in some cases violate general principles, and there may be some lexical exceptions which behave unpredictably.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...verbs.

An example of such an approach as applied to the causative alternation can be found in johnston:95. Furthermore, davis:95 introduces thorough inheritance-based proposals for constraining the relationship between a verb's semantic and syntactic arguments.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...ji15),

Note that on the interpretation I am interested in for the sentences in ji15, however, the PP behaves as an adjunct rather than a verbal argument. The location is perceived as the location where the event takes place rather than the location of entity referenced by the subject NP.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...syntactically.

Unless we assume extremely fine-grained syntactic categories which would miss important a generalisation about the relation between this data and its semantic basis.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...powerful.

In fact, in the case of the example ji26, there does seem to be syntactic evidence for including the launching/entraining causation distinction in the lexical semantics: the recipient dative construction is only possible with launching causation and not entraining causation, as shown in ji30.

 

Throw Bill the ball. *Dribble Bill the ball. *Drag Bill the ball.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...structure,

This of course only holds at a certain general level of detail, since for example the inferences from a function in the domain of possession would be different than in the spatial domain.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...property.

Note that we are only interested in the treatment of literal sentences. The word happiness could possibly be a metonymic reference to a thing - Jackendoff's theory correctly predicts that under such a type-shifting interpretation the sentence in ji22 would be grammatical, as in that case there would be no violation of the constraints on the argument of the go function.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...HPSG

HPSG is Head-Driven Phrase Structure Grammar, as defined by Pollard and Sag pollard_sag:87,pollard_sag:94.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...verbs.

Of course, some mechanism could be invented for this purpose, such as grouping non-alternating verbs together in the subsumption hierarchy or adding a binary feature specifying whether the verb alternates or not, but these options are not motivated from a theoretical standpoint given the semantic basis of these alternations (Pinker 1989).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...actors.

Note that the interpretation of snow as an actor in dav17b depends on world knowledge that snow falls and that through this falling snow can have certain effects. No such world knowledge of associated events is available for vase or blanket.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...entity.

In fact, the difference between ch-of-st-rel and creation-rel has implications for determination of aspect. Consider the contrast between dav16a and dav16b:

 

John baked a potato for 90 minutes/?in 90 minutes. John baked a cake ?for 90 minutes/in 90 minutes.

The sentence John baked a cake has a much stronger accomplishment sense than John baked a potato.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...move-displacement-rel,

A move-displacement-rel is a relation expressing that the actor moves along some path. See Section 4.7.2.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...1994:322).

I will ignore issues of quantification in this thesis. Therefore the clause of the Semantics Principle handling quantification is not included in d93.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...time,

I have not explored the representation of temporal information and will leave the precise definition of the time type unspecified.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...of)

This qualification derives from the fact that the -PP in these sentences is actually ambiguous between pseudo-complement and adjunctive readings. So the sentences can either be interpreted as indicating that Mary benefited from the event as a whole or that Mary benefited specifically from the cake/drawing. This issue will be addressed in Section 3.3.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...Susan.

This sentence is okay, however, on an interpretation in which the entire action of Adam baking a cake for Debbie has been performed for Susan's benefit. See Section 3.3.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...York.

Thanks to Janet Hitzeman for the suggestion of this data.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...Debbie.

Note that the structure I intend for this sentence is *[Adam gave [a book] [in the library] [to Debbie]] and not [Adam gave [a book [in the library]] [to Debbie]], in which the PP is a noun phrase modifier and which is grammatical.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...structure.

Note that this sentence differs from sentences which superficially resemble it, e.g. John worried about his homeland, in which no verb-internal argument to be modified by a pseudo-complement clearly exists. The difference is that an act of worrying entails a topic about which one worries, while an act of singing does not entail a song topic. That is, John worried 69#69, while John sang 70#70. Thus the fact that the object of the event in the worry sentence is not clear does not discredit a pseudo-complement analysis of verbs like sing which do have a semantically entailed product. The PP appearing with worry would be licensed through compatibility between the semantic relation type of worry and the semantics of the about PP.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...move-rel

See Section 4.7.2 for details of this relation, which is a subtype of act-rel.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...NP73#73

Gimbles is a marker for verbs which can appear in this construction.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...occur.

i.e. which lexical rule will apply - see Section 3.5; 3.5.3 in particular.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...types:

This discussion of adjunct types is mainly derived from the discussion in Kasper 1993.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...order.

See Kasper (1993) for a good overview of the cases of interaction among multiple adjuncts.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...selects,

Note that this in fact does not constrain adjunctive placement enough, improperly allowing lexical heads rather than phrasal heads to be modified by an adjunct. This would therefore not rule out phrases such as *The king in the bath of France or sentences like *John kicked in the park the ball. These sentences must be ruled out via the lexical entries of the prepositions which select for nominal/verbal heads: a head with an empty subcat list must be explicitly selected for in the mod field of the preposition.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...``head''.

Note that it is necessary to interpret the mod field as specifying a particular type of element with which an adjunct can combine, rather than necessitating that the adjunct modify a phrasal head. This is because the head of a phrase may not be the element in the phrase which the adjuncts actually modifies, as was discussed in Section 3.4.3.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...adjuncts.

In defining this rule, I have ignored linking issues stemming from the definition of the semantic relation type hierarchy. Unification of the semantics of the preposition with the semantic relation of the verb is not sufficient for ensuring grammatical sentences, due to issues of redundancy as discussed in Section 3.4.1. A pseudo-complement can only either fill in an unfilled argument in a relation (to rule out sentences such as John baked a cake for Mary for Mary), or extend the relation to one which has all of its arguments linked either to explicit surface elements or to discourse referents (to prevent the extension of a transfer-und-rel, e.g. John sent the letter directly to a transfer-to-ben-rel without the recipient component of the relation being specified either through the context or through the addition of a ``to'' pseudo-complement, e.g. to prevent ?John sent the letter for Mary without an understood recipient). These issues could potentially be resolved through linking constraints on the relations resulting from unification (see Davis davis:95) or through discourse interactions, but I leave them for future investigation.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...forms.

See verspoor:94 and pinker:89 for a fuller discussion of lexical rules used to capture syntactic alternations with corresponding semantic consequences which depend on a verb's semantics. Goldberg goldberg:95 also addresses this issue, discussing examples such as *John blew a kiss to Mary/*John gave a kick to Mary in contrast to John blew Mary a kiss/John gave Mary a kick. These differences in acceptability could stem from slight variation in the meaning of the double object form as compared with the standard dative form.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...res35.

My annotations: italics indicate an argument of the main verb which is also the subject of the resultative predicate; underlining indicates something that is not an argument of the main verb, but which is the subject of the resultative predicate. Additionally, in what follows I use a `*' to indicate ungrammatical sentences, `#' to indicate pragmatically infelicitous sentences, and `?' to indicate sentences of questionable grammaticality.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...res15.

Data from Di Tomaso ditomaso:96; see that paper for detailed discussion of Italian verbs of manner of motion and spatial prepositions.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...

Note here that on the interpretation of the Italian preposition a as the locative at, this sentence is acceptable.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...acceptable.

The uses could be for stylistic or literary effect.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...interpretations.

Note that res8a and res8b differ greatly in interpretation. res8b clearly does not have a causative meaning. Rather it has an accompaniment/aiding meaning.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...state.

This point about the CAUSE component of meaning in the unergative resultative construction has also been made by (1996:4,7). Note that they do not explicitly address any semantic differences between manner of motion verbs and other verbs in the resultative costruction.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...at_the_store(John).

Levin and Rappaport Hovav (1995) do not provide any formal representation of meaning for the resultative constructions. Furthermore, they do not specify how path prepositional phrases, such as to the store, are reinterpreted as expressing a result state of being in some location. The discussion here assumes that this issue could be satisfactorily resolved on the account and that the representations used approximate the meanings assume.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...relation.

This difference in meaning is also implicit in the analyses Jackendoff (1990, ch. 10) gives of resultative constructions and of manner of motion verbs.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...nose.

This example is a modified version of a sentence spoken by Henry Thompson at lunch on 17/4/97: ``Make me laugh hot tomato soup up my nose and you'll regret it.'' Thanks to Claire Grover for spotting it and passing it on.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...res28.

Thanks to Joan Maling for the suggestion of this example.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...differ).

This observation is also made by (1996) who argue that ``the semantics of the construction is not constructed solely from the meaning of the verb itself'' (pg. 5).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...w2.

This distinction corresponds to a binding difference in Jackendoff (1990)'s Resultative Adjunct rule: the patient role in the meaning expressed by the resultative phrase may be bound to the patient of the verb (a `control' resultative) but need not be (a `raising' resultative).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...pieces.

I am assuming a literal interpretation of run here, not a metaphorical one.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...caused.

See Goldberg (1995) for further details about what it means for something to be directly caused. The notion is relevant here to the extent that pragmatics can play a role in determining direct causation.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...sort.

This motivation is also present in Jackendoff's (1990, ) work.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...use.

(1996) argue against the use of a lexical rule for the sense extension at issue here, instead proposing a monotonic inheritance network of semantic relation types which provides the basis for sense extension without recourse to addition of lexical entries. The lexical rule approach I have advocated also does not involve proliferation of lexical entries, and is in some sense equivalent to the inheritance-based approach (the lexical rule defines potential extensions of the verb meaning, as does the inheritance network). Some motivation for the approach comes from the data presented in Section 4.4.2 in example w17, repeated below for convenience, and similar examples in resalt41:

[4.57]

[b] The blacksmith hammered the metal for three days. [c] ?The blacksmith hammered the metal in three days.

 

John wiped the table for an hour/in an hour. John painted the picture for an hour/in an hour.

The verbal predicates in these sentences are argued by to be ambiguous between a resultative and non-resultative reading and hence possible with both a telic and an atelic aspect. But the telic variants of these sentences, w17c and the in an hour variants of resalt41, do not seem to have the standard resultative interpretation, as they lack the element of causation standard for their resultative counterparts (e.g. John wiped the table in an hour doesn't seem to mean John caused the table to be clean in an hour by wiping it but rather John completed the activity of wiping the table in an hour), and many of these telic variants are very odd w17c. Furthermore, as discussed in Section 4.4.2, telicity shifts may be triggered by world knowledge or adverbial adjunction, and are morely the result of general non-lexical processes which are independent of the resultative construction (Verkuyl 1989, Krifka 1989, Moens and Steedman moens_steedman:88). Lastly, I do not agree with the notion of lexical selection of the resultative phrase embedded by the matrix verb in this analysis given the discussion in Section 4.4.2.

I therefore believe that this is not a convincing argument in favor of the inheritance-based approach for the particular problem of resultatives and in fact may lead to inappropriate resultative interpretations of such uses of transitive verbs as given here. In contrast, I do believe it is appropriate for the manner of motion verbs, for encoding potential extensions of that meaning, as we will see in the next section. This is because these extensions are specific to a particular semantic class of verbs and do not involve a change in the core meaning expressed but rather merely an augmentation of that core meaning.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...verb.

The constructions as proposed by Goldberg (1995) do not make this assumption since the constructions themselves specify the thematic roles of the arguments in the construction. I have chosen not to do this because (a) the thematic specifications Goldberg proposes seem to be too rigid (see example resalt23 and preceding discussion) and (b) the cases for which the semantic relations between a verb and its arguments appear to change are not justifiably treated as instances of the resultative construction (to be discussed presently).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...usage.

See Chapter 5 for a discussion of the conventional nature of logical metonymy constructions, another example of the role of conventionality.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...distraction.

I owe this example to a BBC presentation of Anne Brontë's The Tenant of Wildfell Hall.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...HREF="node49.html#chapter_pp">3).

In Chapter 3, the term pseudo-complement is primarily used to refer to certain dative prepositional phrases, such as for Mary in John sang a song for Mary

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...dancing.

Whether one of these sentences is ambiguous depends on whether the preposition heading the goal phrase can behave strictly as an adjunct, strictly as a pseudo-complement, or as both. If it can behave as both, the sentence will be ambiguous.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...resalt18a.

I assume in these examples that the prepositional phrases modify the main verb rather than a noun phrase.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...verb.

Grice's Maxim of Quantity (Grice 1975) states that a statement should provide as much information as necessary for adequate interpretation of the meaning, and no more. In this context, an AP which expresses exactly the information already expressed by the semantics of the verb would be redundant and therefore would violate this maxim. Hence the oddness of a sentence such as The ice melted liquid.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...burned).

This is similar to a phenomenon discussed in Jackendoff (1990) in which a Theme internal to the verb semantics can be made explicit via a PP headed by with. In that case, the with-adjunct (pseudo-complement in my terminology) is subject to a non-redundancy condition: the noun phrase in the PP must be distinct from the implicit value of the Theme as encoded in the lexical semantics.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...precisification

This is a term borrowed from Manfred Pinkal's (1995) discussion of underspecification, meaning essentially ``made more precise''.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...mentioned.

For the purposes of this chapter, I am ignoring subtle differences which might exist between the use of infinitival VP complements and progressive (-ing) forms of the VP complement of aspectual verbs. See Freed (1979) for discussion of this issue.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...bounded.

Boundedness is given a technical definitions by Godard & Jayez in terms of Krifka's model of aspectual predicates, essentially corresponding to the atelic/telic distinction. Bounded events or objects are viewed as having a set terminal point, while amorphous events/objects do not.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...possible.

This criterion in fact complements the constraint expressed by Godard & Jayez that the object denoted by the NP must be controlled by the subject of begin - if the subject of begin is not a controller because begin is being used as a raising verb, then clearly the NP cannot be controlled by the subject.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...argument.

For example, such an event structure can be built up in terms of the transition network ontology proposed by Moens and Steedman (1988) for capturing the influence of tenses, adverbials, and argument type on the aspect of an eventuality. How this network might interact with the Pustejovsky and Bouillon lexical representations is unclear, since they depend on phrasal or sentential-level properties of a sentence referring to an event.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...data.

See http://www.hd.uib.no/cd-info.html for additional information about the LOB and http://info.ox.ac.uk:80/bnc/ for more information about the BNC.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...HREF="node139.html#iwcs_table1">5.1.

A tally of one half was allocated to each of two possible options if the intended interpretation of the metonymy was not entirely clear from the context. For example, begin the psalms could mean begin reading the psalms or begin singing the psalms.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...reinforcement.

For example, in beg49, the addition of context in the form of a more explicit subject in (b), for which certain world knowledge exists, gives (b) a clear interpretation which is not so readily available in (a).

 

John began the door. (??) The carpenter began the door. (making)

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...constructions.

Note, however, that there exist analyses of other phenomena in terms of qualia structure which depend on the telic role to be filled in in cases for which I would claim it isn't. Johnston and Busa (1996), for example, argue for an analysis of the interpretation of nominal compounds which depends on information in the telic role. So the interpretation of bread knife as a knife for cutting bread stems from a cutting event represented in the telic role of knife. I would clearly not want to allow the coercion of John began the knife to John began cutting the knife. This particular example could be explained by a restriction that the events coerced from an object must involve that object as UNDergoer (as is in fact represented in the lexical entries for begin presented in Section 5.5.3), but it is unclear how conflicts arising from similar examples could be reconciled. This remains a point for future investigation.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...interpretation,

Due to the assumption of the conventional nature of qualia structure, I must depend on a pragmatic account of enjoy for those cases in which it doesn't pattern with begin. In this section I address most of those cases, but still must account for apparent ``default'' interpretations of enjoy metonymies which do not have complement NPs with a conventionalised telic role. These ``default'' interpretations must be seen as stemming from pragmatic defaults rather than lexical defaults, that is from reasoning about the most likely interpretation of these sentences given world knowledge. So a sentence like John enjoyed the pipe will be assigned, based on lexical processing, an underspecified interpretation of John enjoyed doing something with the pipe. This interpretation would then be made more specific on the basis of knowledge about pipes. A ``default'' interpretation would come from knowledge of the most common activity that is performed with a pipe, e.g. smoking.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...specified.

In my representation in Chapter 2, I have not allowed for compound events within the qualia structure of a noun and so this precise structure could not be implemented directly under the proposals there. A similar effect would be achieved by allowing eventive nouns to have qualia structure. There would thus be two lexical entries for the event-objects, one corresponding to the event and the other to the object. The latter would have a telic role of playing/performing while the former would have a telic role of watching/listening. The coercion of the eventive form of the NP could proceed in precisely the same manner as with the object forms.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...constraint.

I have not attempted to explain the origin of this constraint. It may result from the fact that begin must be a control verb in the metonymic construction, in contrast to enjoy (cf. the infelicity of *Mary forced John to enjoy reading the book). Whether an explanation can be found on this basis remains a question for future work.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...entry.

The fact that there is also one less word may also cause preference for {begin + NP} over {begin on + NP}, but this is unlikely to make much difference.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...Corpus

Information about the British National Corpus is available on the web at http://info.ox.ac.uk:80/bnc/.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...computation.

A description of various implementation efforts within the HPSG framework can be found at http://ling.ohio-state.edu/HPSG/Implementation.html.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...language,

A word in the source language which corresponds to multiple words in the target language need not be considered ambiguous in the source language, only underspecified with respect to some feature which the target language discriminates. Colour words provide a good example of this phenomenon, since languages differ as to how specific their colour words are. In English we divide dark colours into black, brown, blue, and green, while in other languages all these colours might be lumped together under one word essentially conveying the meaning dark colour.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...parsing,

This should actually read mainly derived through parsing, since syntactic relationships can depend on the semantics of the sentence, such as in the case of prepositional phrase attachment ambiguities.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...1996).

However, NLU systems face more difficulty than NLG systems when dealing with multi-sentence texts because of the pragmatic reasoning involved with interpreting multiple sentences, particularly for resolving anaphors and establishing the connections between a set of sentences. NLG systems can avoid these issues by having pre-defined text plans and canned text.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...sense.

A similar proposal has been made for `standard' dictionaries. kilgarriff:97a reports on how the Longman Dictionary of Contemporary English (3rd edition, ldoce:95) and the Collins COBUILD English Dictionary (New edition, collins:95) incorporate word frequency information derived from corpora.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...lexicon.

Empirical support for this position is provided by Briscoe (1990), who show that logical metonymy is utilised only in instances where either the default interpretation is intended, or in contexts which are rich enough to override the default. It is not used when the context is not rich enough to override the default which would arise from the use of logical metonymy, in favor of a construction which more explicitly provides an interpretation. The data I introduce in Chapter 5 also follow this pattern.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...1990)

WordNet is a public-domain lexical knowledge base. It consists of a hierarchy of lexicalised concepts, which correspond to sets of synonymys. It is organised around semantic relations including synonymy and antonymy, hypernymy and hyponymy, and meronymy and holonymy. Its development was based on psycho-linguistic considerations of concept relations.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...disambiguated.

The idea of using a context window dates back to lesk:86, who proposed that the correct sense of a word is the sense with the greatest number of overlaps with senses of other words in a 10-word window of context.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...wilks_stevenson:97).

Since this work was done on LDOCE which is constrained to a limited vocabulary, the co-occurrence data is of restricted size and reflects relations between specific senses. How the technique could scale up to analysis of unrestricted texts is unclear, given that these can contain hundreds of thousands of words which are potentially highly ambiguous.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...data.

Note that the acquisition procedure outlined here suffers from the zero-data problem, in that there may be potential possible logical metonymies which are not represented in the corpus. This is particularly a problem for logical metonymy since it is such an infrequent phenomenon. A corpus unfortunately also cannot provide negative, i.e. ungrammatical, instances of a phenomenon. The strategy outlined makes use of the positive instances of logical metonymy but treats the lack of an instance of the phenomenon as evidence for a negative instance, which is likely to be too restrictive. There is no obvious way to get around this since this phenomenon does seem to be governed by conventionality and an NLP system will have no way of learning the conventions without reference to a corpus. The best solution is to base the acquisition on as large a corpus as possible.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...knowledge

See http://www.cyc.com for information on the CYC project.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.