AI in Telecommunications

Introduction

AI has found a wide application in improving the effi­ ciency of the telecommunications infrastructure. Some of the world's first practical expert systems were employed to improve operations and maintenance of telecommunications networks and services. With the growth in complexity of networks, there will be ample opportunity for the application of AI to this future in­ frastructure. Several of the panel members will present their contributions in this direction.

Future telecommunication services are very complex, and if they are to be promulgated to a wide audience they will require a much easier user interface. AI holds the promise of overcoming these difficulties and playing a major role in the widespread promulgation of new ser­ vices. How do we construct new environments that are suitable for this task ? What are the key research issues in creating these service environments? What role will speaker understanding have in this progression ?

Others hold the hope of AI techniques overcoming the difficulties that different languages present. The future promise of translation of language as part of a commu­ nication network has profound implications for all of us. Whilst this is a long term goal, the implications are very far seeking. It is timely to review progress in this direc­ tion, and to assemble the issues that need addressing. With progress towards high rate international communi­ cation networks, there are great opportunities for adding high level services such as translation.

Akira Kurematsu

President, ATR Interpreting Telephone Laboratories Kyoto, Japan

Greg Vessonder and Jon R. Wright AT&T Bell Laboratories New Jersey, USA

Future information services offer the promise of de­ livering a wide range of information sources to a very large audience. Yet we are only now beginning to grap­ ple with the issues of how to present these services to new customers. Many hold the promise of AI techniques playing a large part in the new information society. New techniques for natural language processing and knowl­ edge acquisition are very important in this area.ln many ways AI may be one of the key enabling technologies for the creation of these new services. This panel will high­ light the research issues that are important for this new phase we are entering.

Given that we have some experience in the application of AI, it is timely to review the important issues in this application. What are the prerequisites to successful ap­ plication and what issues are important for the future? How will this experience contribute to better methods of building AI Systems? Clearly we would like to take the lessons of the past to give a direction to the future of AI in telecommunications.

This panel will present a view of the state of AI in telecommunications from the viewpoint of a number of international contributors. It presents a view of the cur­ rent status of AI in telecommunications, and examines the key issues for the near future.

The perspective of Automatic Telephone Interpretation

Akira K urematsu

President

ATR Interpreting Telephone Laboratories Kyoto , Japan

An automatic telephone interpretation system will transform a spoken dialogue from the speaker's language to the listener's automatically and simultaneously. Cre­ ation of such a system will require developing various constituent technologies: speech recognition, machine translation and speech synthesis. Technology to inte­ grate individual subsystems to form an automatic tele­ phone interpretation system is also important. A significant requirement for an automatic telephone inter­ pretation system is two-way spoken language interpreta­ tion between persons speaking different languages. The recent progress in the technologies on speech recogni­ tion, machine translation of dialogues, and speech syn­ thesis allows a promising forecast of the feasibility of an automatic telephone interpretation system in an goal­ oriented dialogue in a specific domain. The perspective of an automatic telephone interpretation system will be discussed in this panel.

Bellcore's R&D Program in AI

Adam lrgon

Bell Communications Research New Jersey, USA

A wide range of issues drive Bellcore's R&D program in AI, from communicating with customers to provid­ ing new types of service to day-to-day management of the network. Communicating effectively and efficiently with customers is one problem motivating a program in natural language processing, which includes work on the lexicon, parsing, and generation, and on the acquisition, analysis, management, and delivery of massive docu­ ment collections. As new telecommunications services become increasingly complex, predicting their behavior becomes correspondingly crucial. This is one problem motivating work in logic programming, including de­ ductive databases and extending the expressive power of logic programming languages. A program in neu­ ral networks and connectionist learning, including the development of neural network chips, addresses issues that cannot be handled easily by traditional methods. Moreover, the ubiquity of software in to day's telecom­ munications business is leading us to address a range of software engineering problems through AI. Finally, we observe that the American telecommunications industry now has fewer people who are expected to learn how to do more and more complex tasks, which motivates work in Intelligent Tutoring Systems with emphasis on ITS authoring. We explore Bellcore's current R&D effort in these and other areas and identify some opportunities and issues yet to be addressed.

A Decade of Applying AI at AT&T

Greg Vessonder & Jon R. Wright AT&T Bell Laboratories New Jersey, USA

AT&T has been involved in applying artificial intelli­ gence for over a decade. Our first commercial applica­ tion of artificial intelligence was the expert system ACE, a diagnostic system that assists analysts in the local tele­ phone loop network. Since that time we have developed systems that are in regular use for both telecommuni­ cations and non-telecommunications applications. Dur­ ing this period, we passed through several generations of techniques and tools used for building AI systems.

We will describe several systems that we have devel­ oped during the past decade and discuss some of the lessons we have learned during that period.