The number of occurrences of logical metonymies involving begin in the corpora is very low, suggesting that this is not a pervasive phenomenon, at least for this particular verb. Only three relevant instances of begin metonymies were found in the LOB, and only 164 were found in the BNC. More significant are the relative figures: there are 40,407 sentences containing verbal forms of begin in the BNC. After phase 1 of the analysis of the BNC 4,470 sentences containing begin followed by a noun phrase remained as potentially metonymic. Only 3.67% of these were actually metonymic -- 0.41% of all sentences containing verbal forms of begin. The LOB figures are even lower. Out of 172 sentences containing begin followed by a noun phrase in the LOB, only 1.7% of these were metonymic.
For finish, the frequency of the phenomenon is much higher. Out
of 11,072 uses of finish as a verb, there were 2,799 occurrences of
finish followed by a noun phrase (25.3%). Out of a sample of
940 of these which were analysed, more than a third (319) were
metonymic. The distribution of the metonymy types for both of these
aspectual verbs as compared to that for begin on is, however,
interesting to look at. This information is summarised in
table 5.1.
This table shows the number and proportion, relative to the whole set of metonymies for the particular aspectual verb in question (as indicated by the column heading), of each different metonymic interpretation found in the corpus. They are divided into Agentive, Telic, and Context, indicating where the missing event needed in the interpretation seems to come from, assuming the existence of qualia structure. The noun phrase following the eventive verb in the interpretations listed either corresponds to the noun phrase actually found in the corpus examples (those in regular font) or to a generalisation over the types of noun phrases which were found on that interpretation in the corpus (those in small capitals). So, for example, the value 34 in the first row of agentive interpretations for begin indicates that 34 instances of begin WRITTEN_OBJECT (such as begin the book and begin the diary) occurred in the BNC on a writing interpretation. The OTHER AGENTIVE category includes all examples specifying the creating of an object in the way appropriate to that object, such as begin (digging) the tunnel and begin (painting) the portrait, while the OTHER TELIC category groups together very low frequency occurrences of other interpretations which seem to come from the telic role of the noun's qualia structure rather than from context.
It is clear from the table that begin and finish metonymies tend to be interpreted on the basis of information in the qualia structure of their complement noun rather than through the influence of context, as is evident from the high total percentages (95% and 95.6%, respectively) of begin and finish metonymies which are resolved on the basis of qualia information. Context plays a much larger role for the interpretation of begin on (accounting for the interpretation of 64% of the metonymies), suggesting a fundamental difference between the aspectual verbs and begin on with respect to the interaction of the verb with the meaning of the complement noun. This will have implications for the treatment of these verbs as proposed in Section 5.5.3.
An inspection of table 5.1 clearly slows that the range of possible metonymies is actually quite small. If we consider the size of the BNC, the fact that telic metonymies occur for only approximately 20 different categories of nouns is striking. The BNC contains a varied collection of texts, covering many topics and therefore certainly referring to far more than 20 categories of nouns. We would expect a far wider range of metonymies on the assumption of a fully specified qualia structure for all nouns: in terms of the Pustejovksy qualia-based theory, every noun should be able to appear in the logical metonymy construction. Although a corpus cannot provide negative examples of the use of the construction -- that is, ungrammatical or infelicitous examples -- and therefore does not give conclusive evidence in favour of the restrictiveness of telic role-centred interpretation of metonymies and does not allow us to develop a theory solely on the basis of this corpus data, the data retrieved in this analysis do suggest that the use of the construction is limited to only a few specific cases. In addition, the use of the results of linguistic analysis (e.g. judgements on infelicitous data) allows us to overcome this ``zero-data'' problem (see Chapter 6 for some discussion of this) by providing evidence of ungrammatical cases and enabling the development of a theory about this phenomenon. Agentive role-centred metonymies occur for a wider range of objects, which share the property of being artifacts or at least having a clear way in which they come into existence. The information in the agentive role is therefore much more uniform across objects than that in the telic role: all agentive events are creation events. These facts will form the basis of our proposals about logical metonymy in the next section.