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.

Introduction

Document preparation systems might be now the most commonly used computer systems, ranging from stand-alone text processing individual machines to highly sophisticated systems running on mainframe computers. All of those systems internally use a more or less formal system for representing documents. Document representation formalisms are very different according to their goals. Some of them define the interface with the printing device, they are oriented towards a precise geometric description of the contents of each page in a document. Others are used internally in systems as a memory representation. Yet others have to be learned by users ; they are symbolic languages used to control document processing.

The trouble is that there are today nearly as many representation formalisms as document preparation systems. This makes it nearly impossible, first to interchange documents among heterogeneous systems, second to have standard programming interfaces for developping systems. Standardization organizations and large companies are now trying to establish standards in the field in order to stop proliferation of formalisms and facilitate document interchange.

This paper focuses in the last sections on three document representation formalisms often called ‘revisable formats’, namely SGML [SGML], ODA [ODA], and Interscript [Ayers & al.], [Joloboff & al.]. In order to better understand what is a revisable format, the paper starts with a look at the evolution of the architecture of document preparation systems.

Introduction

Increasing numbers of people are using computers for the preparation of documents. Many of these new computer users are not “computer types”; they have a problem (to produce a neatly formatted document), they know the computer can help them, and they want the result with a minimum of (perceived) fuss and bother. The terms in which they present the problem to the computer should be “theirs” – easy for them to use and understand and based on previous document experience.

Many powerful document preparation tools exist that are capable of producing high quality output. However, they are often awkward (some would say difficult) to use, especially for the novice or casual user, and a substantial amount of training is usually necessary before they can be used intelligently.

This paper presents the design of a document preparation system that allows users to make use of existing formatters and yet makes document entry relatively easy. The following topics are discussed:

• the requirements and overall design for such a system, and

• some of the issues to be resolved in constructing the system.

First, some terminology is clarified.

Terms and Concepts

We use Shaw's model for documents [Shaw80, Puruta82, Kimura84]. A document is viewed as a hierarchy of objects, where each object is an instance of a class that defines the possible components and other attributes of its instances. Typical (low level) classes are document components such as sections, paragraphs, headings, footnotes, figures, and tables.

For many years text preparation and document manipulation have been poor relations in the computing world, and it is only recently that they have taken their rightful place in the mainstream of computer research and development. Everyone has their own favourite reason for this change: word processors, workstations with graphics screens, nonimpact printers, or authors preparing their own manuscripts.

Whatever the reason, people in computing have suddenly found themselves using the same equipment and fighting the same problems as those in printing and publishing. It would be nice to say that we are all working happily together, but there are still plenty of disputes (which is healthy) and plenty of indifference (which is not). There is no doubt, however, that this coming together of different disciplines has brought new life and enthusiasm with it.

The international conference on Text Processing and Document Manipulation at Nottingham, is not the first conference to focus on this field of computing. It follows in the footsteps of Research and Trends in Document Preparation Systems at Lausanne in 1981, the Symposium on Text Manipulation at Portland in 1981, La Manipulation de Documents at Rennes in 1983, and the recent PROTEXT conferences in Dublin. We hope, however, that it marks the beginning of a regular series of international conferences that will bring top researchers and practitioners together to exchange ideas and share their enthusiasm with a wide audience.

As the papers for this conference started to come in, a number of themes began to emerge. The dominant theme (in number of papers) was document structures for interactive editing.

Wonderful Appearance–Wretched Content

With the advent of relatively inexpensive laser printers, computer output is being routinely typeset. It can be expected that there will be a revolution in the way business and technical documents are created, based on the use of low-cost typesetters. Easy typesetting and graphics is an extension of word-processing capability, which is already widespread. The essential feature of word processing is its ability to quickly reproduce a changed document with mechanical perfection. However, as the appearance improves, the quality of writing seems to fall in proportion. These forces are probably at work: (1) More people can (attempt to) write using better technology, and because writing is hard, novices often produce poor work. (2) With improved technology, projects are attempted that were previously avoided; now they are done, badly. These factors are familiar from programming, and suggest an analogy between creating a document and developing a program. The current word-processing situation corresponds to the undisciplined use of programming languages that preceded so-called “modern programming practices.”

Introduction

W is an interactive text-editor and document preparation facility being developed within the department of Computer Science at Manitoba. A working prototype of W, known as W-p, has been described elsewhere [King84]. W is a considerable development of that earlier system, but retains the same basic philosophy:

• W is an interactive, extensible, integrated editor and formatter;

• W adheres as closely as possible to the “what you see is what you get” (wysiwyg) philosophy;

• W encompasses a wide range of document items, incuding text, tables, mathematical formulae, and provision for general graphical items;

• W is portable and adaptable; that is, several versions of W are being produced to run on different architectures; although the user interface will differ in its detail, the underlying system will be common;

• W is user extensible in a variety of ways.

The remainder of this paper is organised as follows. Section 2 describes W from the user's viewpoint and gives some details of its implementation. For the most part, it is a review of material which is covered in greater depth in [King84].

Introduction

There is no doubt that ‘electronic publishing’ and other computer based tools for the production and dissemination of printed material open up new horizons for efficient communication. The problems currently faced by the designers of such systems are enormous. One problem area is the identification of duplicate material especially when there is more than one source generating similar documents. Abstracting is a good example here. Another problem area is the linkage between full text and bibliographic databases.

As part of its attempt to establish collaboration between different countries, the European Economic Community initiated the DOCDEL programme of research and a series of studies such as DOCOLSYS which investigate the present situation in Europe regarding document identification, location and ordering with particular reference to electronic ordering and delivery of documents.

The majority of DOCDEL systems likely to be developed fall under one of the following areas:

1. (a) Delivery from computer-based systems

2. (b) Delivery from non-computer based systems

3. (c) Electronic publishing

4. (d) On-demand publishing

How people read books

Books are rarely read completely linearly; mystery novels almost are, but how many people let their curiosity get the better of them and sneak a look at the last page to see who dunnit?. A text book is more likely to be dipped-into looking for a particular section, and a journal article tends to be read in full only if the reader considers it useful and relevant after having read the title, then the abstract, conclusion and finally the references [Maude85, Line82].

Page Description Languages

An imaging device, such as a typesetter, laser printer, or display, must have some way of knowing what image it is being asked to show. The two traditional means of providing it with that information have been to describe the image to the imager in terms of a bitmap or character map or describe the image to the imager by means of a sequence of control commands to the imager's electronics.

The bit-map or character-map schemes are the simplest and oldest. For example a line printer is provided with a character map (in this spot put the character in this spot put the character and so forth).

Introduction

This paper deals with a recently completed project to develop a publishing system which provides users with responsive methods for the collection of information, and flexible ways for producing different publications.

It concerns a major supplier of replacement car parts which uses a range of publications to enable dealers and individuals to identify parts to fit cars. While it is obviously a specialised application, it represents an important market sector and also demonstrates a number of issues concerning the structuring of data and passing control directly to the users.

Context

Unipart is the largest supplier in Europe of replacement car parts and accessories for all makes of cars. It has been a leader in the development of systems for the distribution of product information, and has been using computerised publication systems for a number of years. As an indication of scale, it provides 12,000 parts for 3,500 vehicles and produces 135 major publications.