This chapter is organized as follows. In the next two sections we discuss the modular status of natural language systems and the consequences of grammar reversibility for the system's design. We will show that the use of a uniform grammatical component leads to more compact systems and helps to identify a clean linguistic and conceptual separation. On the other hand, this kind of modularity implies a serious problem, namely that the conceptual component cannot control completely whether and how the linguistic system will realize a given semantic structure.
In section 5.3 we will argue that in order to maintain a modular design additional mechanisms are necessary to perform some monitoring and revision of the generator's output. We will argue that the best way in realizing these mechanisms is by means of a tight integration of parsing and generation. Before, however, in section 5.4 the new competence-based performance model is presented we discuss in section 5.3 previous approaches that also take an integrated view of parsing and generation.
In order to make clear how the new model is used to solve particular problems, we present in section 5.5 a fundamental generation strategy, namely that of not producing ambiguous output. The idea here is that the generator has to run its output back through the understanding system to make sure it's unambiguous. In particular we demonstrate how reversible grammars are used to make such strategy effective and efficient.
In section 5.6 we also apply this method during the understanding mode of a NLS for the purpose of disambiguation by means of the generation of paraphrases. 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.
An important property of both methods is that the nature of the underlying parsing and generation strategy is not important, i.e. the strategy can be used with any parsing- or generation strategy. However, an obvious restriction for both is that monitoring only takes place when the generator has finished computing a first string. Therefore, the underlying monitoring strategy could also be denoted as a non-incremental generate-parse-revise strategy.
The basic strategies followed in the non-incremental approach also allow the specification of a monitoring strategy that interleaves generation and parsing more tightly in such a way that monitoring can take place even during generation. In section 5.7 we present such an approach. As it will be clear such an incremental approach is much more flexible and natural. The method we are presenting takes full advantage of the uniform algorithm. In particular, the top/down generation approach followed in the uniform algorithm as well as the item sharing approach makes the incremental monitoring and revision method a practical one.