Godard & Jayez (1993)

Godard & Jayez (1993) focus on constraints on the process of type coercion itself, rather than identification of the missing event introduced as a result of this process. They argue for an approach to type coercion in which differences in the coercive properties of aspectual verbs such as begin and stop (motivated with reference to French data but which transfer for the most part directly to English) are specified lexically, and in which coercion does not require type-shifting the NP complement, but is incorporated into a richly structured semantic representation for the aspectual verb. Their approach depends on the existence of candidate events for the missing event prior to the type coercion process, and as such does not provide a solution to the problem of constructing an interpretation for logical metonymies, under examination in this chapter.

Godard & Jayez do provide, however, a proposal for the constraints which the French commencer (=begin) imposes on its NP complement. These are:

(i)

The complement must be bounded. (whether a non-eventive NP or an event)

(ii)

The reconstructed event is an event in which the objected denoted by the NP is controlled by the entity denoted by the subject of commencer. (That is, the subject of the main verb is the entity which ``triggers and causally maintains'' (Godard & Jayez 1993:172) the event into which the complement object is coerced.)

(iii)

The reconstructed event should be a type of modification.

They use these constraints as the explanation for a range of excluded type coercions. The first constraint (i) explains the contrast in beg50 in that ``some cheese'' refers to an indefinite entity -- i.e. an unbounded entity -- while ``the cheese'' refers to a definite, discrete entity. It would, however, incorrectly predict that beg50c is ungrammatical. The boundedness constraint does appear to apply accurately to the logical metonymy construction, but not to the fully expanded {begin + VP} construction. The second constraint correctly rules out the sentences in beg51. The third attempts to explain why the sentences in beg52 are ruled out, expressing an intuition that the object in the NP usually comes into being, is consumed, or undergoes a change of state - hence, it is modified in some way. These sentences do not express such a modification.

 

Jean a commencé le fromage.
John began the cheese. *Jean a commencé du fromage.
John began some cheese. [Godard & Jayez 1993:(24)] Jean a commencé a manger du fromage.

 

*A ce moment Jean a commencé un grand mépris pour les politiciens.
At that moment John began a great contempt for politicians.[Godard & Jayez (31)] *L'acide a commencé la destruction du marbre.
The acid began the destruction of the marble.[Godard & Jayez 1993:(35)]

 

*Jean a commencé la pierre.
John began (moving) the stone. *Jean a commencé le désert de Gobi.
John began (to go through) the Gobi desert.

I will generalise from these constraints in my proposals for the logical metonymy constructions, adopting (i) and (ii) as these constraints appear to accurately reflect this data, but replacing (iii) with more specific claims about the reconstructed events.

Let us consider briefly the G&J (1993) proposal that the complement of begin must be bounded. While this is clearly not the case on a use of begin with a VP complement, it does seem to be true when begin appears with an NP complement. Thus we have the contrasts in beg61. Both (a) and (b) are felicitous and involve a finite set of books. While (c) also involves a finite set of books, it seems to be slightly less good, probably due to pragmatic inferences which specify that it is impossible for one person to read multiple books simultaneously and therefore impossible to begin reading multiple books in a single event. Both (d) and (e) involve unbounded sets; in (d) the reading would have to be a generic one (e.g. John began reading books when he was five years old) and therefore does not refer to any specific set of books, and (e) also does not refer to a specific set of books. The constraint proposed by G&J - that the NP complement must itself be bounded - accurately captures these data. P&B attempt to capture these same data by pointing out that in (d) the missing event is an activity. As we saw above, however, aspectual restriction does not seem to capture the metonymy data entirely accurately in general. In fact, it cannot rule out (e) either - in this case although the set of books in question is not specified, it is finite. Thus the event of reading some books is still a transition as it cannot go on indefinitely and cannot be ruled out with the aspectual constraint.

 

They/John began the book. They began the books. ?John began the books. *They/John began books. *They/John began some books.

  *John began books/houses/sandwiches.

The ``bounded NP'' constraint, then, also rules out the data in beg43.

There does, however, seem to be an aspectual constraint which governs the behaviour of {begin + NP}. The missing event in the metonymy seems to be constrained to be a single specific durative event, that is, not iterative or generic. This can be argued on the basis of beg75 and the contrast between (5.18b) and (5.18c).

  *John began the dictionary. (consulting)

In (5.20), consulting the dictionary is a transition. However, it is a point-like transition and a single point-like transition is unlikely to be begun. As discussed by Freed (1979), begin is one of a class of aspectual verbs which presuppose or entail certain temporal facts about the events named in their complements. She argues that begin picks out a particular temporal segment of its complement events. This segment includes part of what she calls the nucleus of an event, which is an activity which occurs during a stretch of time. It therefore makes sense that begin requires its complements to have some duration. For the instances that don't, either an iterative (in which a series of single events of the same type occurs continuously within a given stretch of time) or a generic (in which a series of single events of the same type occurs at different times) interpretation is inferred. Thus, when we say John began sneezing, we generally infer that John entered a time period in which he sneezed multiple times. Similarly, beg75 would have an iterative interpretation. In contrast, (5.18c) is likely to require a generic interpretation since pragmatics suggests that John cannot begin reading multiple books simultaneously, and so we must interpret the metonymic event as a series of events occuring at different times. (5.18b) has neither an iterative or generic interpretation, because it does not refer to a series of events, it rather specifies quantification over single begin reading the book events. By ruling out iterative and generic metonymic events, we rule out beg75 and a generic interpretation of (5.18c). The latter would only be allowed on a reading with quantification over single events, as in (5.18b). The pragmatic oddity of this reading accounts for the oddity of the sentence.

The Godard & Jayez constraint (iii) does not seem to adequately capture the logical metonymy data. In one of the most frequently cited logical metonymies and my first example, John began the book, reading surely does not involve a change of state or modification of the book. Godard & Jayez attempt to justify this by claiming that reading a book involves the imposition of an ``informational layer'' on the book, derived from interpretation of the text, which constitutes ``informational modification''. I find this explanation somewhat forced; in any case, the range of corpus data which will be presented below (Section 5.5.2) indicate that there is not a unifying semantic type (e.g. modification) of the derived events in logical metonymy interpretations.