The objectives of computational linguistics

Computational linguistics is the study of computer systems for understanding and generating natural language. In this volume we shall be particularly interested in the structure of such systems, and the design of algorithms for the various components of such systems.

Why should we be interested in such systems? Although the objectives of research in computational linguistics are widely varied, a primary motivation has always been the development of specific practical systems which involve natural language. Three classes of applications which have been central in the development of computational linguistics are

Up to now, we have restricted ourselves to determining the structure and meaning of individual sentences. Although we have used limited extrasentential information (for anaphora resolution), we have not examined the structure of entire texts. Yet the information conveyed by a text is clearly more than the sum of its parts – more than the meanings of its individual sentences. If a text tells a story, describes a procedure, or offers an argument, we must understand the connections between the component sentences in order to have fully understood the story. These connections are needed both per se (to answer questions about why an event occurred, for example) and to resolve ambiguities in the meanings of individual sentences. Discourse analysis is the study of these connections. Because these connections are usually implicit in the text, identifying them may be a difficult task.

As a simple example of the problems we face, consider the following brief description of a naval encounter:

The most evident linguistic problem we face is finding an antecedent for ‘it’. There are four candidates in the first sentence: ‘dawn’, ‘Valiant’, ‘Zwiebel’, and ‘torpedoes’. Semantic classification should enable us to exclude ‘dawn’ (*‘dawn sinks’), and number agreement will exclude ‘torpedoes’, but that still leaves us with two candidates: ‘the Valiant’ and ‘the Zwiebel’ (which are presumably both ships of some sort).

The poor cousin

As we noted in the first chapter, language generation has generally taken second place to language analysis in computational linguistics research. This imbalance reflects a basic property of language, namely, that there are many ways of saying the same thing. In order for a natural language interface to be fluent, it should be able to accept most possible paraphrases of the information or commands the user wishes to transmit. On the other hand, it will suffice to generate one form of each message the system wishes to convey to the user.

As a result, many systems have combined sophisticated language analysis procedures with rudimentary generation components. Often generation involves nothing more than ‘filling in the blanks’ in a set of predefined message formats. This has been adequate for the simple messages many systems need to express: values retrieved from a data base, error messages, instructions to the user.

More sophisticated systems, however, have more complex messages to convey. People querying a data base in natural language often begin by asking about the structure or general content of the data base rather than asking for specific data values (Malhotra 1975); we would like to extend natural language data base interfaces so that they can answer such questions. For systems employing lengthy sequences of inferences, such as those for medical diagnosis (e.g., Shortliffe 1976), user acceptance and system improvement depend critically on the ability of the system to explain its reasoning.

The role of syntax analysis

Syntax analysis performs two main functions in analyzing natural language input:

What is the objective of semantic analysis? We could say that it is to determine what a sentence means, but by itself this is not a very helpful answer. It may be more enlightening to say that, for declarative sentences, semantics seeks to determine the conditions under which a sentence is true or, almost equivalently, what the inference rules are among sentences of the language. Characterizing the semantics of questions and imperatives is a bit more problematic, but we can see the connection with declaratives by noting that, roughly speaking, questions are requests to be told whether a sentence is true (or to be told the values for which a certain sentence is true) and imperatives are requests to make a sentence true.

People who study natural language semantics find it desirable (or even necessary) to define a formal language with a simple semantics, thus changing the problem to one of determining the mapping from natural language into this formal language. What properties should this formal language have (which natural language does not)? It should

1. *be unambiguous

2. *have simple rules of interpretation and inference, and in particular

3. *have a logical structure determined by the form of the sentence

We shall examine some such languages, the languages of the various logics, shortly.

Of course, when we build a practical natural language system our interest is generally not just finding out if sentences are true or false.

What is a speech act?

Speech act theory has its roots in the work of Wittgenstein, who in Philosophical Investigations proposed an analogy between using language and playing games. His basic point was that language is a form of rule-governed behavior, much the same as game-playing, employing rules and conventions that are mutually known to all the participants.

The field of speech act theory is usually considered to have been founded by Austin (1962) who analyzed certain utterances called performatives. He observed that some utterances do more than express something that is true about the world. In uttering a sentence like “I promise to take out the garbage,” the speaker is not saying anything about the world, but is rather undertaking an obligation. An utterance like “I now pronounce you man and wife” not only does not say anything that is true about the world, but when uttered in an appropriate context by an appropriate speaker, actually changes the state of the world. Austin argued that an account of performative utterances required an extension of traditional truth-theoretic semantics.

The most significant contribution to speech act theory has been made by philosopher John Searle (1969, 1979a, 1979b), who was the first to develop an extensive formulation of the theory of speech acts.

What has been accomplished

Kamp represents the first step in a very ambitious program of research. It is appropriate at this time to reflect upon this program, how far we have come, and what lies in the future.

KAMP represents not merely an attempt to devise an expedient strategy for getting text out of a computer, but rather embodies an entire theory of communication. The goal of such a theory could be summarized by saying that its objective is to account for how agents manage to intentionally affect the beliefs, desires and intentions of other agents. Developing such a theory requires examining utterances to determine the goals the speakers are attempting to achieve thereby, and in the process explicating the knowledge about their environment, about their audience, and about their language that these speakers must have. Language generation has been ehosen as an ideal vehicle for the study of problems arising from such a theory because it requires one to face the problem of why speakers choose to do the things they do in a way that is not required by language understanding. Theories of language understanding make heavy use of the fact that the speaker is behaving according to a coherent plan. Language generation requires producing such a coherent plan in the first place, and therefore requires uncovering the underlying principles that make such a plan coherent.

Introduction

This chapter discusses in detail a typical example that requires KAMP to form a plan involving several physical and illocutionary acts, and then to integrate the illocutionary acts into a single utterance. This example does not reflect every aspect of utterance planning, but hopefully touches upon enough of them to enable an understanding of the way KAMP works, to illustrate the principles discussed in earlier chapters of this book, and to provide a demonstration of KAMP's power and some of its limitations. It is important to bear in mind that the implementation of KAMP was done to test the feasability of a particular approach to multiagent planning and language generation. Since it is not intended to be a “production” system, many details of efficiency involving both fundamental issues and engineering problems have been purposely disregarded in this discussion.

KAMP is based on a first-order logic natural-deduction system that is similar in many respects to the one proposed by Moore (1980). The current implementation does not take advantage of well-known techniques such as structure sharing and indexing that could be used to reduce some of the computational effort required. Nevertheless, the system is reliable, albeit inefficient, in making the necessary deductions to solve problems similar to the one described here.

Introduction

This chapter examines some of the special requirements of a knowledge representation formalism that arise from the planning of linguistic actions. Utterance planning requires the ability to reason about a wide variety of intensional concepts that include knowledge per se, mutual knowledge, belief, and intention. Intensional concepts can be represented in intensional logic by operators that apply to both individuals and sentences. What makes intensional operators different from ordinary extensional ones such as conjunction and disjunction is that one cannot substitute terms that have the same truth-value within the scope of one of these operators without sometimes changing the truth-value of the entire sentence. For example, suppose that John knows Mary's phone number. Suppose that unbeknown to John, Mary lives with Bill — and therefore Bill's phone number is the same as Mary's. It does not follow from these premises that John knows what Bill's phone number is.

The planning of linguistic actions requires reasoning about several different types of intensional operators. In this research we shall be concerned with the operators Know (and occasionally the related operator Believe), Mutually-Know, Knowref (knowing the denotation of a description), Intend (intending to make a proposition true) and Intend-To-Do (intending to perform a particular action).