to SFB-Homepage
maintained by: Anke Weinberger (2000-09-20).
Computerlinguistics
University of Saarbrücken
For an analysis of reinterpretation it is important to determine the reinterpretation potential of an expression, i.e., whether it may be reinterpreted at all, and, if so, which of its constituents are affected by reinterpretation. Previous analyses of reinterpretation (Pulman 1997; de Swart 1998; Koller, Niehren and Striegnitz 2000) determine reinterpretation potential in terms of syntactic constituent structure. E.g., NP-VP structures like in (1) would be assigned reinterpretation potential in that the VP may be reinterpreted.
This strategy can be extended to more complex examples like (2). It exhibits a semantic mismatch between the noun semantics (a property of groups) on the one hand and the AP and VP semantics (properties of individuals) on the other hand.
We assume reinterpretation potential for an AP-N structure, too, which allows reinterpretation of the AP. Then our interpretation for (2) follows immediately, where both the AP and the VP are reinterpreted (as properties of groups). This interpretation is that the majority of teams whose members are red-haired lost their members by death.
However, such a purely syntactic characterization of the reinterpretation potential of an expression does not take into account that sometimes expressions have the same syntactic constituent structure, but differ in their reinterpretation potential. In particular, reinterpretation cases in specific (professional) contexts, e.g., Nunberg's notorious (3), exhibit a reinterpretation potential that is different from the one of (1), even though (1) and (3) have the same constituent structure:
For (3), one must assume reinterpretation of the N' in the Det-N' constituent structure to derive the desired interpretation in which the quantification is over consumers of ham sandwiches but not over sandwiches: the sentence is true if the majority of ham sandwich consumers complained, even if they did not eat the majority of the sandwiches. A similar problem arises in the aspectual domain. Consider e.g. (4) and (5), which have the same constituent structure, viz., an adjunction structure where an adverbial adjoins to a sentence:
(4) is reinterpreted in that the adverbial does not measure the length of Charlotte's leaving itself but of the aftermath of this leaving (i.e., Charlotte's subsequent absence).
This reinterpretation avoids an aspectual conflict: the modified Charlotte left is a bounded expression (roughly, one that introduces eventualities with inherent boundaries), whereas durative adverbials like for two hours may modify only unbounded expressions.
But if we put down the reinterpretation potential of (4) to its constituent structure, this begs the question of how to block reinterpretation potential for (5). This sentence cannot be reinterpreted analogously (in the sense that Charlotte left and her subsequent absence took place yesterday).
Thus, syntactic characterizations of the reinterpretation potential of an expression are by themselves not fine-grained enough.
To distinguish between examples like (4) and (5), one might adopt de Swart's (1998) approach: she proposes restricting the assignment of reinterpretation potential to only those expressions that actually have semantically mismatching constituents. This would immediately block the unwanted reinterpretation potential for (5).
However, there are two areas where this strategy may turn out problematic. On the one side, semantically well-formed expressions, too, may call for reinterpretation to reconcile them with world or background knowledge. E.g., the semantically well-formed (6) is only reinterpreted (so that the N' of the subject NP refers to recordings of opera singers), to fit in (6) with our knowledge that opera singers are highly unlikely to occupy shelves:
On the other side, anaphoric references may enforce reinterpretation of sentences that by themselves do not exhibit any semantic mismatches between their constituents:
E.g., only the anaphoric reference of (7b) is responsible for the need to reinterpret (7a). Therefore, precluding reinterpretation potential for a sentence like (7a) since it exhibits no semantic mismatches is too rash: from the perspective of larger texts such sentences have reinterpretation potential, too.
Thus, it seems advisable not to make the determination of the reinterpretation potential of an expression conditional on a check for semantic mismatches between constituents of the expression.
The analysis of reinterpretation which we propose in this paper derives the reinterpretation potential of an expression largely from its syntactic constituent structure. This basic strategy is complemented by lexical means of determining reinterpretation potential, which makes possible a more fine-tuned approach to reinterpretation phenomena that naturally extends to the problematic examples which were discussed in this abstract.
|
to SFB-Homepage
|
created by: Anke Weinberger (2000-09-20). maintained by: Anke Weinberger (2000-09-20). |