Introduction

The purpose of this chapter is to explore the implications of some facts about prosody and intonation for efforts to create more general and higher quality speech technology. It will emphasize parallels between speech synthesis and speech recognition, because I believe that the challenges presented in these two areas exhibit strong similarities and that the best progress will be made by working on both together.

In the area of synthesis, there are now text-to-speech systems that are useful in many practical applications, especially ones in which the users are experienced and motivated. In order to have more general and higher quality synthesis technology it will be desirable (1) to improve the phonetic quality of synthetic speech to the point where it is as easily comprehended as natural speech and where it is fully acceptable to naive or unmotivated listeners, (2) to use expressive variation appropriately to convey the structure and relative importance of information in complex materials, and (3) to model the speech of people of different ages, sexes, and dialects in order to support applications requiring use of multiple voices.

Engineers working on recognition have a long-standing goal of building systems that can handle large-vocabulary continuous speech. To be useful, such systems must be either speaker-independent or speaker-dependent; if speaker-dependent, engineers must be trained using a sample of speech that can feasibly be collected and analyzed. Present systems exhibit a strong trade-off between degree of speaker independence on the one hand and the size of the vocabulary and branching factor in the grammar on the other.

Introduction

One of the most delightful features of a small symposium is that it allows for protracted discussions in which many people participate. Ample time for discussion was built into the symposium schedule throughout, but we allocated a special two-hour slot to challenge ourselves to identify the most significant problems capable of being solved in a five- to ten-year period. That they be solvable in that time frame challenges us beyond what we can see, but not beyond what we can reasonably extrapolate. That their solution be significant takes the discussion beyond questions of purely academic interest.

Furthermore, at the suggestion of one of the government representatives, we asked what applications should drive research (much as the application of natural language interfaces to database drove research in the 1970s and 1980s).

All attendees, including representatives of various governmental agencies, participated in this discussion.

To keep our thoughts large, we construed natural language processing (NLP) as broadly as possible, freely including such areas as lexicography and spoken language processing.

To direct the discussion without focusing it too tightly, we set forth the following questions:

1. What are the most critical areas for the next seven (plus or minus two) years of natural language processing? (“Critical” is taken to mean that which will produce the greatest impact in the technology.)

2. What resources are needed (such as people, training, and corpora) to accomplish the goals involved in that work?

3. What organization is needed (e.g., coordinated efforts, international participation) to accomplish those goals?

4. What application areas and markets should open up in response to progress toward those goals?

Introduction

If current natural language understanding systems reason about the world at all, they generally maintain a strict division between the parsing processes and the representation that supports general reasoning about the world. The parsing processes, which include syntactic analysis, some semantic interpretation, and possibly some discourse processing, I will call structural processing, because these processes are primarily concerned with analyzing and determining the linguistic structure of individual sentences. The part of the system that involves representing and reasoning about the world or domain of discourse I will call the knowledge representation. The goal of this chapter is to examine why these two forms of processing are separated, to determine the current advantages and limitations of this approach, and to look to the future to attempt to identify the inherent limitations of the approach. I will point out some fundamental problems with the models as they are defined today and suggest some important directions of research in natural language and knowledge representation. In particular, I will argue that one of the crucial issues facing future natural language systems is the development of knowledge representation formalisms that can effectively handle ambiguity.

It has been well recognized since the early days of the field that representing and reasoning about the world are crucial to the natural language understanding task. Before we examine the main issue of the chapter in detail, let us consider some of the issues that have long been identified as demonstrating this idea. Knowledge about the world can be seen to be necessary in almost every aspect of the understanding task.

Introduction

Although natural language processing (NLP) has come very far in the last twenty years, the technology has not yet achieved a revolutionary impact on society. Is this because of some fundamental limitation that can never be overcome? Is it because there has not been enough time to refine and apply theoretical work that has already been done?

We believe it is neither. We believe that several critical issues have never been adequately addressed in either theoretical or applied work, and that, because of a number of recent advances that we will discuss, the time is due for great leaps forward in the generality and utility of NLP systems. This paper focuses on roadblocks that seem surmountable within the next ten years.

Rather than presenting new results, this paper identifies the problems that we believe must block widespread use of computational linguistics, and that can be solved within five to ten years. These are the problems that most need additional research and most deserve the talents and attention of Ph.D. students. We focus on the following areas, which will have maximum impact when combined in software systems:

1. Knowledge acquisition from natural language (NL) texts of various kinds, from interactions with human beings, and from other sources. Language processing requires lexical, grammatical, semantic, and pragmatic knowledge. Current knowledge acquisition techniques are too slow and too difficult to use on a wide scale or on large problems. Knowledge bases should be many times the size of current ones.

2. Interaction with multiple underlying systems to give NL systems the utility and flexibility demanded by people using them. Single application systems are limited in both usefulness and the language that is necessary to communicate with them.

3. […]

Introduction

This chapter concerns a dispute about the relationship of sentences to the events they describe, and how that relationship is manifested in sentences with adverbial modifiers. The two sides to the argument might be called the “Davidsonian position” and the “situation semantics position”; the former being chiefly represented by Donald Davidson's well-known paper “The Logical Form of Action Sentences” (Davidson, 1980) and the latter by John Perry's critique of Davidson's view, “Situations in Action” (Perry, unpublished manuscript).

The issue turns on Davidson's analysis of how a sentence such as (1) is related to a similar sentence with an adverbial modifier, such as (2).

1. (1) Jones buttered the toast.

2. (2) Jones buttered the toast in the bathroom.

Stated very informally, Davidson's position is this: (1) claims that an event of a certain type took place, to wit, a buttering of toast by Jones, and that (2) makes a similar claim but adds that the event took place in the bathroom. Put this way, an advocate of situation semantics could find little to complain about. Perry and Barwise themselves say rather similar things. The dispute is over the way that (1) and (2) claim that certain events took place. Davidson suggests that the event in question is, in effect, a hidden argument to the verb “butter”. As he would put it, the logical form of (1) (not analyzing the tense of the verb or the structure of the noun phrase) is not

1. Buttered (Jones, the toast)

but rather

1. ∃X(Buttered(Jones, the toast, x)),

where the variable x in (4) ranges over events.

Introduction

The present work investigates the contrastive discourse functions of a definite and a demonstrative pronoun in similar contexts of use. It therefore provides an opportunity to examine the separate contributions to attentional state (Grosz and Sidner, 1986) of two linguistic features – definiteness and demonstrativity – independently of pronominalization per se. The two pronouns, it and that, have clearly contrastive contexts of use, explained here in terms of distinct pragmatic functions. Certain uses of it are claimed to perform a distinctive cohesive function, namely, to establish a local center (that modifies rather than replaces the notion of a center). The crucial distinction between a local center and the Cb (backward-looking center) of the centering framework (cf. Sidner, 1983; Grosz et al., 1983; Grosz et al., 1986; Kameyama, 1986) is that there is only a single potential local center rather than an ordered set of Cfs (forward-looking centers). The local center is argued to constitute a reference point in the model of the speech situation in a manner analogous to 1st and 2nd person pronouns. In contrast, a deictic function is posited for apparently anaphoric uses of that whereby the attentional status of a discourse entity is changed, or a new discourse entity is constructed based on non-referential constituents of the linguistic structure. Because it is impossible to observe attentional processes directly, I present an empirical method for investigating discourse coherence relations. I analyze statistically significant distributional models in terms of three types of transitions in the cognitive states of conversational participants: expected transitions, unexpected transitions, and transitions with no relevant effect.

SIGNALS

When describing the behaviour of a real-time process, we may wish to include instantaneous observable events that are not synchronisations. These signals may make it easier to describe and analyse certain aspects of behaviour, providing useful reference points in a history of the system. For example, an audible bell might form part of the user interface to a telephone network, even though the bell may ring without the cooperation of the user. This is incompatible with our existing view of communication.

In some cases, suitable environmental assumptions will allow us to describe such behaviour within the existing timed failures model. However, if we intend that these signals should be used to trigger other events or behaviours, then we must extend our semantic model to include an element of broadcast communication: some output events may occur without the cooperation of the environment.

In our model, signal events will occur as soon as they become available, and will propagate through parallel combination. A process may ignore any signal â performed by another process, unless it is waiting to perform the corresponding synchronisation a. If this is the case, then both â and a will occur. Of these, only the signal will be observed outside the parallel combination; it makes no sense to propagate a synchronisation.

REAL-TIME SYSTEMS

As computing devices become faster and more powerful, we find ourselves increasingly dependent upon systems which are difficult to understand and prone to failure. The failure of a commercial banking system or a company database may be expensive and inconvenient. The failure of an aircraft control system or a railway signalling network may result in injury or death. As the consequences of system failure become ever more severe, we must find ways to make these applications of computing technology safer and more reliable.

Over the past twenty-five years, mathematical techniques have been developed for the specification and implementation of computing systems. Formal methods have been used in the design and analysis of transformational systems—in which results are computed from a given set of inputs—and have been shown to reduce design costs and improve reliability. However, many of the systems in which safety is a primary concern are real-time systems, and cannot easily be viewed in a transformational setting.

Real-time systems maintain a continuous interaction with their environment and are often subject to complex timing constraints. They may also be required to perform several tasks concurrently. To reason about such systems we require a mathematical formalism that supports a treatment of timed concurrency. In this thesis we explore and extend one such formalism, the theory of Communicating Sequential Processes, first introduced by Hoare (1985).

OTHER APPROACHES

A wide variety of formal methods have been proposed for the specification and development of real-time systems, based upon process algebras, temporal logics, and timed programming languages. Although much research has been carried out, a consensus has yet to emerge concerning the applicability of the various formalisms to different types of system. A successful development method is likely to involve some combination of the features mentioned above. A notation that is well-suited to requirements capture is unlikely to be an efficient programming language, and vice versa.

Quantitative Temporal Logics

Hooman and Widom (1989) present a compositional proof system relating an occam-like language to a quantitative temporal logic, similar to the one developed by Koymans and de Roever (1983). Although the system description language is somewhat limited, it is clear that quantitative temporal logics are useful assertion languages; Jackson (1990) shows how such a logic may be employed as a specification language for timed CSP. It would be interesting to see the proof system applied in the development of a large, complex system.

Shasha et al. (1983) use a quantitative temporal logic to prove the correctness of a carrier-sense broadcast protocol, similar to the one described in chapter seven. By assuming a simplified version of the service provided by the physical layer, and an internal specification of the data link layer, the authors are able to establish that certain desirable properties hold of the network.

ABSTRACTION

If we wish to produce a readable specification of a large system, then we must take care to present our description in a clear, structured fashion. At each level of abstraction, we identify the interfaces between system components and conceal any events which are not of interest. We express our specification as a series of service specifications, each describing the service provided by a particular component of the system. In this way, we may refine a description of the service provided by a system towards a satisfactory implementation.

The hiding operator provides the mechanism for abstraction in timed CSP; the expression P\ A denotes a process that behaves as P, except that

• events from A occur as soon as they become available

• only events from outside A are observed

In section 5.2.8 we gave an inference rule for this operator that was easy to derive, but difficult to apply. We can achieve a significant reduction in complexity if we separate the concerns of concealment and scheduling. To this end, we define a predicate actA which holds of any A-active behaviour:

Definition 6.1 actA(s, X) ≙ [0, end(s, X)) × A ⊆ X

A behaviour (s, X) is A-active if all events from set A occur as soon as they become available. If we wish to establish that P\A satisfies a specification S(s, X), it is sufficient to show that

• S(s, X) holds for all of the A-active behaviours of P

• S(s, X) is unaffected by the concealment of events from A

The second condition is satisfied if the truth of the specification is unaffected by the removal of A's events from the trace and refusal.