Pustejovsky & Bouillon (1995) use Godard & Jayez (1993)'s analysis to develop constraints on type coercion in terms of the aspectual properties of the reconstructed event and the raising/control properties of the eventive/coercing verbs. Their analysis attempts to explain the logical metonymy data in terms of the linguistic criterion of aspectual class rather than the vague semantic notion of modification adopted by Godard & Jayez. In doing so, Pustejovsky & Bouillon are able to give a uniform treatment across the different syntactic frames of eventive verbs, as the constraints they propose are claimed to hold for all uses of these verbs, not just the metonymic uses. I will show below, however, that their analysis cannot account for the full range of logical metonymy data and is too strong in its claim of application across all syntactic frames in which the eventive verbs appear.
The Pustejovsky & Bouillon account relies on a structured representation of events, in which subevents are represented and the ``focus'' of the event is marked as the head of the event structure. The event structure provides ``a configuration where events are not only ordered by temporal precedence, but also by relative prominence'' (Pustejovsky & Bouillon 1995, p. 141). Pustejovsky & Bouillon provide examples of headed transitions, proposing that build is a left-headed event structure since the event-focus is on the first subevent, the activity of building, while arrive is a right-headed event structure as the focus is on the actual arrival rather than the pre-arrival process, and break is lexically unspecified with respect to headedness. The relevant event structures are shown in iwcs1, with the asterisk indicating the headed sub-event. Left-headed structures correspond to accomplishments, while right-headed structures correspond to achievements.
arriving
arrived 
2
breaking broke 2
Another element of the Pustejovsky & Bouillon (1995) proposal is the distinction between control and raising aspectual verbs. Using tests proposed by Perlmutter (1970), they argue that begin is ambiguous between a control use and a raising use. They build on the insight provided by Godard & Jayez that logical metonymies are restricted to control uses of begin.
The constraints on aspectual/eventive verbs proposed by Pustejovsky & Bouillon make use of these two elements, and can be summarised as follows:
*John began books/houses/sandwiches. John began the highway. (*driving on) / dictionary. (*consulting)
Sentences like those in beg51 are ruled out because these are raising uses of begin, so NP coercion is impossible. Sentences like those in beg52 are ruled out simply because the appropriate values are not available in the qualia structures for stone and desert.
The details of the Pustejovsky & Bouillon account of aspectuality in
relation to qualia structure are rather vague and would need to be
fleshed out for a full evaluation of their approach. They provide no
clue as to what event structure is specified for verbs in the lexicon,
or how the event structure is built up as a verb is combined with an
argument. In the sample representation they offer for
book (Pustejovsky & Bouillon 1995, p. 151), the events in the
qualia structure, read and write, are specified in the
lexical entry as transition events, and no indication is given of how
the aspectual type of these events might be shifted (e.g. to account
for the aspectual differences between reading a book and
reading books). Their formalisation therefore ignores the
influence of the form of the noun on the aspectual nature of the event
which may be associated with the NP, although they assume that the
aspectual class of an eventuality can be correctly identified in their
theoretical discussion. This lack of clarity means that the event
headedness of particular eventualities, a critical part of the
Pustejovsky & Bouillon approach, is difficult to establish with
certainty.
The definition of event headedness is also a problem simply because it is often not clear what the internal structure of an event is or how to establish on which subevent the ``focus'' is. Reading a book is a left-headed transition on the Pustejovsky & Bouillon account, suggesting that it is composed of two subevents: the reading of the book event, and the have read the book state. The focus of the predicate read, then, is on the first of these, the process of reading. Arriving is a right-headed transition on their account. Presumably, the two subevents of arriving are the doing something to bring on an arrival event and the arrival event. The focus here is on the actual arrival. But what about consulting a dictionary? Analogous to the two events described, one would postulate two subevents for this event: the consulting of the dictionary event, and the have consulted the dictionary event. Is this a left-headed or a right-headed event, or possibly something which is neutral between the two? The problem with relying on event headedness as a criterion is that it depends on intuitions about event structure which in many cases are difficult to determine, and difficult to isolate from the context in which the event is described. Pustejovsky and Bouillon do not attempt to define clear tests for determining event structure linguistically, let alone defining an automatic mechanism for determining event structure in the syntax-semantic processor. A fundamental question emerges: if event-headedness is essentially a representation for a verb's aspectual class, what does it buy us over other (simpler) aspectual representations? The answer to this question remains unclear to me.
Even if we assume that somehow the formalism works out such that the eventuality types are appropriately represented, the constraint that begin may only appear with left-headed transitions on a control reading is too strong. Each of the sentences in beg55-beg56 have a structure reflecting a clear control use of begin (on the basis of Perlmutter (1970)'s tests). In beg55, we find in each case a control use of begin with an eventive complement which is an activity. In beg56, we again have control uses of begin, this time with eventive complements which are achievements, i.e. right-headed transitions.
Mary forced John to begin building houses. Mary forced John to begin running. Mary forced John to begin discussions.
Mary forced John to begin leaving the pub. Mary forced John to begin destroying the book.
Pustejovsky & Bouillon base their claim on examples such as those in beg57.
*John began to find the house key. ?John began to arrive.
Mary forced John to leave the pub. Mary forced John to destroy the book. *Mary forced John to find the house key. *Mary forced John to arrive.
However, these sentences pair a control use of begin with an event for which the subject cannot be a controller. This distinction is clearly exhibited by the contrasts in beg58. Thus the sentences in beg57 are actually instances which require a raising interpretation of begin. For discussion of the constraints on this interpretation, see Section 5.5.2.
The Pustejovsky & Bouillon constraint that type coercion is not possible with a raising construction is supported by observations by Jacobson jacobson:90 as reported in Pollard & Sag (1994:141-142), which suggest that raising verbs never allow NPs instead of their VP complements. This constraint is therefore an instance of a more general constraint on raising verbs. The constraint on the complement of aspectual verbs on control uses, however, does not seem to be supported by the data. An alternative explanation for the type coercion phenomena must be sought.