Despite the knowledge that the doorstop is a book, this sentence is strange. Lascarides and Copestake therefore also argue that logical metonymy is partially conventionalised, stemming from lexical specification of word use. Next I will attempt to establish a clearer formalisation of that conventionalisation.
Metonymic sentences with begin and finish seem largely
acceptable on a ``coming into existence'' interpretation -- i.e. one
in which the reconstructed event comes from the agentive role of the
qualia structure -- particularly if context aids in arriving at such
an interpretation.
John began the book (writing) / house
(building) / dictionary (compiling) / cigarette
(rolling) / highway (constructing) / door (making)
/ car (building) / cake (baking)
Thus the data in beg88 are certainly possible, although some may
require contextual reinforcement.
This conclusion is supported by the high proportion of agentive
metonymies in the corpus, as reflected by the Agentive section of
Table 5.1, which shows that 39.6% of the begin
metonymies and 29.5% of the finish metonymies are resolved
with respect to the agentive role of the complement NP. A
possible explanation of the general availability of the agentive
role in metonymies is due to the semantic uniformity of this role
across objects, as mentioned above.
Interpretations other than the coming-into-existence (agentive) interpretation seem to be largely impossible, except in the restricted cases indicated by the Telic section of Table 5.1, exemplified by the selection of sentences in beg33-beg37.
John began the sandwich. (eating) John began the beer. (drinking) John began the story. (telling) John began the book. (reading/writing) John began the cigarette. (smoking) John began the solo. (playing) John began the song. (singing)
These cases are highly conventionalised. That is, they occur relatively frequently in English, are highly idiosyncratic, and are likely learned as individual instances of a more general logical metonymy construction. Their usage is triggered by lexical specification of the event conventionally associated with a particular noun on a metonymic usage. As mentioned in Chapter 1, I assume that there is justification for postulating the lexicalisation of knowledge only when a particular linguistic phenomenon cannot be explained purely in terms of world knowledge. This applies to qualia structure: although general knowledge and experience of objects provide potential values for their telic roles, this information is only lexicalised when it is relevant to some linguistic phenomenon. In the case of logical metonymies, the distinction between the felicitous sentences in beg33-beg37 and the infelicitous sentences in beg74 cannot be explained solely on the basis of world knowledge of the relevant objects, since each of the events which are intended to be coerced from the complement noun are commonly associated with the denoted object and there is thus no pragmatic distinction to be made. Nor, as argued above, can this distinction be made on the basis of Pustejovsky & Bouillon's aspectual restrictions or Godard & Jayez's semantic constraints on the general metonymic process. Instead, accurate modeling of the data can only be achieved through selective lexicalisation -- lexicalisation only of highly conventionalised events which can be accessed in the metonymic process. Events which are not available in metonymic coercions form part of world knowledge of an object, but must not be lexicalised.
Thus although feasible metonymic interpretations generally correspond to one of Pustejovsky's agentive or telic roles (i.e. contextual information does not often override the core semantic information found in the qualia structure), these do not always provide a feasible metonymic interpretation, as in the sentences in beg74. The infelicity of the specified interpretations of this data can, however, be explained via the assumption that for these cases, there is no event specified in the telic role of the nouns in the NP.
Further evidence in support of this analysis comes from comparison of the sentences in beg33-beg37 to ones containing semantically related nouns, as shown in beg40-beg41.
?John began the apple. (eating) ?John began the orange juice. (drinking) ?John began the flyer. (reading) ?John began the pipe. (smoking)
Since it is reasonable to assume that the nouns in beg40-beg41 would have the same qualia values in their lexical semantic structure as their counterparts in beg33-beg34, the difference in acceptability between these sentences apparently cannot be explained without recourse to idiosyncratic lexical properties of individual words.
The approach outlined above is preferable to one which assumes that all nouns have a telic role specified and then rules some out with various constraints, since telic-role metonymy seems to be possible in only very limited instances, on highly conventionalised uses. However, even an account which assumes that such lexical defaults are optional and may or may not be specified for any given object must provide a justification for the absence or presence of that default on some basis independent of the metonymy data for individual verbs; otherwise the default simply serves as a marker for whether or not metonymy can occur when a particular verb appears with a particular object. We can provide this independent justification on the basis of aspectual verbs other than begin and of ``metonymic adjectives'' - i.e. adjectives which modify events rather than or in addition to objects, which therefore may require type coercion of the object.
The presence or absence of a particular metonymy holds across all eventive verbs, as shown in beg84. Similarly, the application of metonymic adjectives in beg86 display the same patterns.
John began/finished/started/enjoyed the book. (writing/reading) John began/finished/started/enjoyed the nails. (making/*hammering in)
John read a long book. (physical length large/num pages large/long to read) This is a fast book. (fast to write/fast to read)
John hammered in the long nails. (physical length large/*long to hammer in) This is a fast nail. (?fast to make/*fast to hammer in)
There is additional evidence from data involving temporal prepositions, which also require eventive (and hence temporal) arguments. Consider the sentences in beg97, which display the same patterns of interpretation as the previous examples.
After five books, John was considered an expert. (writing, reading) After fifteen nails, John needed a break. (making, ?hammering in)
These examples show that the presence or absence of a telic role for a
particular object is consistent across metonymic
constructions.
However, there are certain inconsistencies in the data. For example,
a sentence such as beg98a seems to have a natural
interpretation with a temporal preposition and in the enjoy
metonymy which is not available in the begin metonymy.
After three stories, the children wanted to go outside. (hearing, listening to) The children began the stories. (*hearing, *listening to) The children enjoyed the stories. (hearing, listening to)
Such examples will be considered below in the discussion of enjoy. The infelicity of beg98b will be shown to follow from an additional constraint which begin imposes on its NP complement in the metonymy construction.
My proposal is therefore that for certain artifacts a telic event is conventionalised (whether there is some reason for this conventionalisation other than frequency is presently unknown) and therefore specified lexically, available for access in metonymic constructions. For other artifacts such a telic event is not conventionalised or specified lexically and is not available in metonymic interpretation. Thus the nouns in beg74 will not have telic roles specified, while the nouns in beg82 and beg33-beg37 will.
{begin on + NP}
In contrast, the metonymic phrase {begin on + NP} generally seems to serve as a syntactic marker for pragmatic interpretation. Consider the examples from the BNC in beg102.
He parted Jean-Paul's thighs and began on his legs. (massaging) Perhaps we began on Elizabeth Bowen's The Death of the Heart. (discussing the book) Sometimes I will begin on a central motif such as a tree trunk right in the middle of a blank sheet. (drawing) Then she began on me. (undressing)
The intended interpretations of the {begin on + NP} phrases indicated in italics clearly can only be established in context. Therefore, I propose that this structure indicates only that something is being done with the NP object, leaving a more specific interpretation to be established using contextual information. It does not need to look to the qualia structure for a default interpretation, as context will provide the interpretation. This proposal is supported by the high percentage of cases of this construction in the BNC which could only be interpreted on the basis of the surrounding context, as indicated in Table 5.1. Any apparently default interpretations of {begin on + NP} sentences (such as a default interpretation of John began on the book as John began reading the book) should not be considered a reflection of lexical defaults, but rather of the most likely pragmatic interpretation of such sentences in isolation of a defining context.
Metonymic ``enjoy''
The numerical data I have introduced have not included an investigation of the use of the verb enjoy, due mainly to the high frequency of this verb in the corpus and the lack of effective semantically based extraction mechanisms. However, for completeness and since some discussion in the literature on logical metonymy (e.g. Lascarides & Copestake 1995) concentrates on enjoy, the occurrences of {enjoy + NP} were extracted and briefly examined. The conclusions drawn from this examination are that the use of enjoy is less constrained than that of begin and the other aspectual verbs.
Examples involving enjoy which seem perfectly felicitous can be contrasted with parallel examples involving begin, as in enjoy2-enjoy5.
John enjoyed the symphony. (listening to) *John began the symphony. (listening to)
John enjoyed the film. (watching) *John began the film. (watching)
The goat enjoyed the book. (eating) *The goat began the book. (eating)
Examples enjoy2 and enjoy5 show that there are interpretations available for metonymic enjoy which are not available for metonymic begin. Example enjoy3 suggests that the default interpretations for enjoy are more easily overrideable by preferences deriving in the discourse context than those for begin.
However, Lascarides and Copestake (1995) point out that the acceptability of certain enjoy sentences cannot be improved with the addition of context, similar to the begin sentences, as I mentioned in Section 5.5.2. Some examples are found in beg59. They argue that this kind of data supports the hypothesis that the process of type coercion is partially conventionalised, motivating a lexical encoding of exceptions.
?John enjoyed the path/tunnel/Gobi desert/dictionary/door. ?John enjoyed the doorstop. ?John enjoyed Route 280.
The range of metonymic facts for enjoy therefore differs from that of the aspectual verbs, and must be explained within the framework I have suggested for logical metonymy on the basis of the begin data. This implies that cases in which enjoy does not pattern with begin, such as enjoy2-enjoy3, must be explained.
Note first that the majority of cases which are infelicitous for begin are also somewhat difficult to interpret with enjoy (e.g. beg59). These are instances for which I would argue that the telic role of the complement noun phrases is not lexically specified. In this way, just as for the corresponding begin sentences, the oddity of this data is explained. There will be no default interpretation for these sentences proposed by the lexicon, hence their interpretation will depend on pragmatic reasoning. So the interpretation of these examples would likely be eased with reference to a rich context supporting a particular interpretation; if the oddity does not go away for particular instances I postulate that this results from the complexity of the pragmatic processing necessary to establish the interpretation. I will not, however, investigate this hypothesis.
The slight difference in the grammaticality judgements of the
corresponding begin and enjoy sentences in beg59 (* vs.
?), and the differing judgements of the examples in enjoy3 stem
from the same source, which I will elaborate in
Section 5.5.4. I will suggest there that the lexically
proposed interpretations of the begin sentences (including those
instances where there is no lexically proposed interpretation because
there is no value specified in the qualia structure of the complement
NP) are indefeasible, that is that they cannot be overridden by
contextual or pragmatic information. This is an idiosyncrasy of the
behaviour of the aspectual verbs in metonymic environments. The verb
enjoy is not subject to the same constraint and can therefore be
influenced by pragmatic reasoning. The sentences in beg59 are odd
because they have no lexically-derived interpretation (on the
assumption that the telic role is not represented in the qualia
structure of the complement nouns) and possibly because it is
difficult to establish a ``neutral'' context to enable their
interpretation,
but they are not ungrammatical because they do not violate any
linguistic or discourse constraints. The sentence enjoy3a is
fully felicitous because the subject NP the goat provides enough
contextual information to enable a pragmatically induced
interpretation which overrides the lexical default of reading.
Examples enjoy2-enjoy5 remain as problematic. Notice that both of these examples involve ``event-objects''. Recall that these are NPs which have a natural interpretation as an event (e.g. the performance of the symphony, the playing of the film), but which can also be referred to as an object. I have not been able to come up with any unrelated (i.e. not involving an event-object) examples in which there is such a clear grammatical difference between begin and enjoy and for which that difference cannot be explained on the basis of pragmatic reasoning. This suggests that there is something unusual about metonymies with this type of NP.
Figure 5.1: The Pustejovsky & Bouillon (1995) proposal for the lexical
entry for sonata
Pustejovsky & Bouillon (1995) propose that the structure of the event-object sonata should be as in Figure 5.1, in their representational notation. In this structure, both the event aspect and the object aspect of the event-object are represented, and the telic role has a compound value which relates the two aspects. This lexical entry indicates that the object sonata is performed, and this performance is the event sonata, whose function is to be listened to. The structures for symphony and film can be considered analogous.
The sentence enjoy2a, then, does not convey
John enjoyed watching the film-object (which could be paraphrased
as John liked staring at the film reel), but rather John
enjoyed watching the film-event or John enjoyed watching the
playing of the film-object. The contrast between enjoy2a
and enjoy2b can be explained if begin does not allow the
event coerced from the qualia structure of its complement NP to itself
involve an event. The difference between these examples therefore
indicates a semantic difference between enjoy and begin
which must be lexically specified. Specifically, it appears
that the NP complement of the coercing form begin must be
constrained to refer to an object, while the complement of the
coercing form of enjoy is not subject to that
constraint..This difference will be
reflected in the lexical entries proposed in Section 5.5.3.
The sentence enjoy2a can now be explained by assuming that the
metonymic coercion occurs relative to the event interpretation of the
event-object symphony, which will not be possible for begin
since it must coerce relative to the object interpretation.
Returning to the data above in beg98, then, it appears that temporal prepositions pattern with enjoy in allowing coercion relative to an event with elaborated qualia structure. Therefore, beg98a can be interpreted as After listening to the telling of three stories, the children wanted to go outside. The event of telling is in the telic role of the object interpretation of story, and the event listening is in the telic role of the event interpretation. The coercion in this example occurs relative to the event interpretation.