Parsing is the process by which grammatical strings of words are assigned syntactic structure. This structure, in turn is necessary for many types of text analysis, including the analysis of meaning. This chapter will provide an overview of current computational techniques for parsing and in and interpretingnatural language texts. The emphasis here, as the little suggests, will be on parsing, and much of the discussion of the semantic and pragmatic analyses will concern how these interact with the parser. The sections on higher-level analysis are merely intended to provide some context for the role of parsing in the task of understanding natural language.
Parsing is a critical step in many importantapplications. First, an early application which is still of major significance is machine translation (see chapter 4)CHAPTER4.doc. Almost as soon as it was understood that computers could manipulate symbols as well as numbers, a rush was on to translate texts automatically from one language to another. This enterprise led to some of the first parsers. Although the enthusiasm of the 1950s and1960s was later dampened by the realization that sophisticated analysis of meaning was also required, practical systems were produced which helped human translators perform the task more quickly. Despite the difficulty of fully-automatic, high quality translation, the tremendous benefit such systems could provide continues to fuel interest in parsing.
Another motivating application has been thedesire to allow people to interact with computer systems in natural language, as opposed to the cryptic or awkward interfaces to which they must currently submit themselves. Natural language interfaces would allow people with no special training or skills to profit from computer interaction. The most attractive form of natural language interface would provide access to the computer though everydayspeech. Unfortunately speech understanding introduces additional difficult acoustic and phonological problems which have tended to isolate speech research from parsing. Nevertheless, parsing will be an essential ingredient of spoken language interfaces.
Another application involves skimming a data source for information on a particular topic. Some prototype systems have been constructed that cantake input from newswire services, for instance. Here it is probably not practical to attempt to parse all the data ( this type of application would involve looking for certain keywords, followed by detailed parsing when appropriate cues have been found. There are many applications of this sort in political and industrial intelligence gathering (Winograd, 1983).
An area of artificialintelligence which could benefit greatly from effective parsing is computer-aided instruction (for applications to the learning of languages, see chapter 9)CHAPTER9.DOC. The students are probably having enough trouble learning the subject-matter without having to learn how to use an unnatural computer interface. In addition, natural languages answers given by students often provide useful clues, forinstances about uncertainty (e.g. through the use of modal hedges). Other types of interfaces, designed to avoid the problem of understanding natural language, restrict the vital channel of communication between the teacher and the student.
Word-processors have made a significant contribution to text preparation, not only through their editing capacities, but also through their access tospelling checkers (see also chapter 10)CHAPTE10.doc - chapter 10.doc. Parsing may eventually become widely used in word-processors (sentence processors) to find syntactic errors, or perhaps as the first step to finding semantic and pragmatic problems.
Finally, a problem of urgent social and economic concern is parsing the vast databases of the texts which are now available. Inexpensive computer...