The following two chapters review the requirements, both actual and potential, that language engineering (LE) applications place on lexical semantic resources. We have chosen to address the topic in two ways. First, in Chapter 4, we discuss some areas of LE application within which lexical semantics has or can have some role. The application areas addressed are machine translation, information retrieval, information extraction, summarisation and natural language generation. For each we give a brief overview of the application, survey the approaches to the application that have been undertaken to date, point to related application areas and techniques, and discuss the role that lexical semantics has or could play in the application.
Second, in Chapter 5 we review various LE component technologies that utilise lexical semantic information and that are being, or may be, used in a variety of different LE application areas. These technologies address such tasks as semantic word clustering, multiword recognition, word sense disambiguation, proper name recognition and parsing and coreference. Unlike the application areas identified in Chapter 4, these component technologies are of no intrinsic interest to an end user; rather they are building blocks from which LE applications are constructed. Their value lies in being resuable across applications. So, for instance, robust word sense disambiguation could benefit both machine translation systems and information retrieval systems. For each component technology reviewed we give a brief overview, survey existing approaches, discuss the role of lexical semantics in the component technology, and point to applications that are using or could utilise the technique.
Given the scope and rapidly changing nature of LE applications any such review is bound to be partial: the inclusion of some areas and exclusion of others is not meant to be a judgment of relative importance. Clearly, we believe the areas included are of some significance - in particular we have concentrated on applications of relevance to the market at the current date. But those not included - for example, spell checking, NL front ends to databases, grammar checking, dialogue systems, voice access, and telematics - have been left out due to the limited time and space available in compiling this report, and the limitations in expertise of those compiling it, not because they are not of interest or importance.