A performance model based on uniform processing.

We are also interested in the uniform process as a basis for performance-oriented approaches like monitoring, revision, disambiguation or generation of paraphrases. We demonstrate that uniform grammatical processing increases the degree of flexibility of a natural language system, enabling it to select an utterance appropriate to the particular context. A suitable model of performance-oriented behaviour on the linguistic level is an interleaved approach to parsing and generation. We demonstrate that the uniform algorithm in combination with the item sharing method leads to a practical interleaved approach. In particular we present:

• A novel method for monitoring and revision. In situations of communication where the generation of ambiguous utterances should be avoided our method is able to compute an un-ambiguous utterance for a given semantic input. The underlying mechanism will be developed as a chart-based incremental generate-parse-revise strategy: substrings produced during generation are parsed to test whether they lead to some ambiguities. Detected ambiguities are handled immediately by means of revision using as much as possible of the previously computed structures. We demonstrate that such an enhanced technology enables a natural language system to reduce the risk of generating ambiguous sentences in a purposive and efficient way.
• A novel method for the generation of paraphrases. We also demonstrate how the same method that is used for monitoring and revision can be used for generation of paraphrases during language understanding. The idea here is that after parsing an utterance, then if this utterance has several readings, corresponding paraphrases are generated that reflect the semantic differences. The user is then asked to choose the one he intended.

The performance model takes full advantage of the uniform tabular algorithm and the item sharing method. The combination of uniform processing of reversible grammars and performance-oriented strategies are carried out in an easy and elegant way. This means that we are not only able to show - for the first time - that efficient uniform processing of reversible grammars is possible but also that systematic pursuit of uniformity in natural language processing - as followed in this thesis - achieves the necessary preconditions for a practical interleaving of parsing and generation.