Discussion

Limitations

It should be clear that monitoring and revision involves more than the avoidance of ambiguities. [Levelt1989] discusses also monitoring on the conceptual level and monitoring with respect to social standards, lexical errors, loudness, precision and others. Obviously, our approach is restricted in the sense that no changes to the input LF are made. If no alternative string can be generated then the planner has to decide whether to utter the ambiguous structure or to provide an alternative logical form.

During the process of generation of paraphrases it can happen that for some interpretations no unambiguous paraphrases can be produced. Of course, it is possible to provide the user only with the produced paraphrases. This is reasonable in the case that she can find a good candidate. But if she says e.g., `none of these' then the paraphrasing algorithm is of no help in this particular situation.

In [Meteer1990] a strict distinction is made between processes that can change decisions that operate on intermediate levels of representation (optimisations) and others that operate on produced text (revisions). Our strategy is an example of revision. Optimisations are useful when changes have to be done during the initial generation process. For example, in [Finkler and Neumann1989,Neumann and Finkler1990] an incremental and parallel grammatical component is described that is able to handle under-specified input such that it detects and requests missing but necessary grammatical information.

Comparison and Implementations

In [Meteer and Shaked1988] strategies for paraphrasing are described. They propose an approach where during the repeated parse of an ambiguous utterance potential sources of ambiguity can be detected. For example when in the case of lexical ambiguity a noun can be associated with two semantic classes a so called `lexical ambiguity specialist' records the noun as the ambiguity source and the two different classes. These two classes are then explicitly used in the generator input and are realized, e.g., as modifiers for the ambiguous noun.

The only common knowledge source for the paraphraser is a high-order intensional logic language called World Model Language. It serves as the interface between parser and generator. The problem with this approach is that parsing and generation are performed in an isolated way using two different grammars. If an ambiguous utterance S needs to be paraphrased, S has to be parsed again. During this repeated parse all potential ambiguities have to be recognised and recorded by means of different `ambiguity specialists'. The problem here is that also local ambiguities have to be considered that are not relevant for the whole structure.

Furthermore, the general status of their work is not clear. For example, during generation they use a procedural grammar where it is assumed that all relevant linguistic information is specified in the input of the tactical component. Our work is much more general because it is independent of the grammar and the underlying parser and generator.

In [Levelt1989] and [Meteer1990] the need for monitoring or revision is discussed in detail although they describe no implementations. As far as we know our approach is the first implementation that solves the problem of revising a produced utterance in order to find an unambiguous alternative. Our strategies are implemented in Prolog. The underlying parser and generator are described in [Shieber et al.1990] and [van Noord1991]. We are using lexicalized unification-based grammars for German and Dutch.