The Prospects for Building Intelligent Communication Systems

Patrick Demasco

Introduction

In this paper, I will discuss my Visions for the field of Augmentative
communication during the next decade. I will address the technological
perspective, and specifically focus on the prospects for applying
artificial intelligence to future generation systems. My views are
motivated by what I believe to be two major goals for the research and
development community:

o Improvement of the quality of communication by current users;

o Improvements to technology that will enable us to serve new
  populations (e.g., cognitively impaired)

To discuss any vision for the future of a specific discipline or
topic, it is necessary to establish some set of assumptions about what
the rest of the world will be like. In the case of augmentative
communication, it would be appropriate to consider a "Technological
Context". Specifically, we want to ask the question: "What technology
might be available for the AAC system user by the year 2000?".

A Technological Context

John Naisbitt in his 1982 book, Megatrends, forecasted ten future
directions for our society. Perhaps his most important (and most
obvious) prediction is the shift of the American economy from an
industrial base to an information base. He further states that an
information-based economy and the development of information
supporting technology (e.g., computers) are symbiotic in
nature. Consequently we can expect technological development to
continue at an extremely rapid pace (i.e., exponential) for some time
to come.

In trying to project what computing technology will be available to
AAC users, it is most reasoanble to look at what might be available in
portable computers. I use this as a benchmark for two reasons. First,
there is an increasing trend toward the use of portable computers as
communication systems. I believe this trend will continue in the
future. Second, if AAC manufacturers are to continue producing
dedicated hardware, then they will need to keep competive with
portable computer technology. It would be reasonable to project by the
end of this decade that portable computers will have the following
capabilities:

o Computational performance comparable to today's high-end
  work-stations in the range of 50 MIPS, 100mB physical memory, 1000 mB
  mass storage.

o Greater use of parallel processing in the CPU and in
  IO subsystems (e.g., display).

o Interactive high resolution color and gray scale displays,
  integration of digital video,and high quality audio output.

o Graphically-based multi-tasking operating systems.

o Extensive communication and networking facilities including
 wireless connections.

I would expect that these portable computers will typically be
networked to other computers (such as business, school or home
computers), information networks, and computer-based systems such as
home control and monitoring.

Visions

I believe that the field of Augmentative Communication will be able to
take advantage of technological advances in a number of ways:

Technology Integration

It is impossible to conceive that AAC and other assistive technologies
can be researched, developed and applied in isolation. Advances in
software technologies that will produce more modular systems will pave
the way for the integration of Communication, Mobility, Environmental
Manipulation, and other technologies that might exist in the user's
environments.

Graphics

The availability of high resolution interactive displays with color or
gray-scale capabilities in conjunction with display coprocessors (and
greater memory capacities) will provide the capability of developing
graphics based systems. Full motion video will also be available for
applications such as animated graphical languages.

Physical Interfaces

Increases in computing power will provide the opportunity to create
more powerful physical interfaces. Most likely, these interfaces will
be tightly integrated with human physiology. Technology may finally
begin to support reliable, moderate cost eye gaze input. Software
technology will allow users and clinicians to configure unique input
devices from the available physical input signals.

Intelligent Systems

The internal processing component within a communication system
integrates all of the physical components and provides the system's
operating strategy. This includes an internal representation of
language (i.e., vocabulary set) and the selection technique. The
opportunity now exists to incorporate Artificial Intelligence into the
processing component of AAC systems. I would like to address some of
the possibilities in the remainder of this paper.

The Promise of Artificial Intelligence

Within the Computer Science community, there has always been debate as
to what Artificial Intelligence is and whether it is
achievable. Robert Sokolowski (1988) in his essay Natural and
Artificial Intelligence asserts that part of the controversy revolves
around the intent of the word "Artificial".

"One of the first things that must be clarified is the ambiguous word
artificial. This adjective can be used in two senses, and it is
important to determine which one applies in the term artificial
intelligence. The word artificial is used in one sense when it is
applied to flowers, and in another sense when it is applied to
light. In both cases something is called artificial because it is
fabricated. But in the first usage artificial means that the thing
seems to be, but really is not, what it looks like . The artificial is
the merely apparent; it shows how something else looks. Artificial
flowers are only paper, not flowers at all; anyone who takes them to
be flowers is mistaken. But artificial light is light and it does
illuminate. It is fabricated as a substitute for natural light, but
once fabricated it is what it seems to be. In this sense the
artificial is not merely apparent, not simply an imitation of
something else. The appearance of the thing reveals what it is, not
how something else looks.

If we accept the second notion of artificial as the appropriate sense
(as Sokolowski does), then artificial intelligence has indeed been
achieved. In the context of communication systems,
intelligentprocessing has been applied primarily in the area of rate
enhancement (e.g., prediction). There are however limits to the
methods used in current techniques that I believe will impede
significant progress in the future.

One important characteristic of intelligent systems is the acquisition
and utilization of knowledge. For example, a word prediction systems
might maintain knowledge about word or word transition
frequencies. Unfortunately, methods based on statistical tables can
suffer from combinatorial explosion when the dimensionality of tables
are increased. We need to develop more sophisticated models of
language knowledge, if we are to continue to make
progress. Specifically, we need to embrace the idea of systems based
upon heuristics that are dynamically modified by both the system's
experience, and by the user's preferences. I believe that Natural
Language Processing (a sub-discipline of AI) provides some of the
answers.

Applications of Natural Language Processing

Rate Enhancement

One of the most obvious applications of Natural Language Processing in
the coming decade will be in the refinement of existing rate
enhancement techniques and the development of new techniques that were
previously not possible. In our own work, we have chosen to pursue
both paths. We are looking at how word prediction can be improved
through the use of NLP principles, specifically the application of
Systemic Grammar (Yang, Mccoy, & Demasco, 1990). In addition, we have
developed a new technique called Compansion that allows the user to
select the root forms of sentence content words and outputs a well
formed sentence (Demasco, McCoy, Gong, Pennington, & Rowe,
1989). Common to both projects is an attempt to "understand" the user
by forming a semantic representation of his/her message. The
prediction project goes a step further and attempts to use pragmatic
information. For example, the relationship between the two individuals
contributes important information to the prediction process.

Message Formulation Assistance

I believe that one of the most exciting application areas will be the
use of AI to assist language impaired individuals in formulating
messages. We can consider strategies as simple as spelling checkers
that are directed at spelling errors that a language impaired
individual might make (rather than typographical errors). We might
also imagine systems that provide help with sentence formulation. A
system could constrain choices presented to the user based on previous
input and proper sentence formulation rules. For example, if the user
selected the words "JOHN AND" the system would not provide verbs and
prepositions to the user.

Supporting Interaction

The "Chat" system developed at the University of Dundee by Norman Alm,
Alan Newell and John Arnott (1987) provided a new way of looking at
the technological side of communication interaction. Vocabulary sets
have been tradtionally represented as having three dimensions: row,
column and level (or page). By considering conversational patterns,
they were able to add three new dimensions to the notion of vocabulary
organization: conversational phase, mood, and content mode (i.e.,
substance or filler). This basic idea of supporting communication
"interaction" opens the door to a vast range of other
possibilities. One possible application is in the area of message
repair. A predominant concern among those individuals studying
communication interaction is that of "Communication Breakdowns"
(Kraat, 1987). We might imagine an intelligent system that supports a
number of message repair strategies such as elaboration and
reformulation.

The Reality

Because of the demands that intelligent processing places upon
computational resources, I believe that the first applications of
Artifical Intelligence will not be in the actual operation of
communication systems, but in the areas that indirectly support the
communication process. Expert Systems (the most commercial of current
AI technologies) has already been applied to both the design of
vocabulary sets (Shein, 1988) and to the selection of interfaces (Hsi,
Agogino, & Barker, 1988). What are some of the other possibilities for
the future?

System Configuration

The notion of "device prescription" is rapidly becoming obsolete due
to the increasing "customizability" of communication
systems. Clinicians in the coming decade will have systems of enormous
flexibility available and will most likely have a much greater amount
of quantitative data about their client. They will be faced with
tremendously complex decisions to make about the system
configuration. Fraser Shein's work is important in that his system is
not assisting the clinician in making a selection from a fixed set of
choices. It assists the clinician in the design process based on the
assumption of a flexible vocabulary set. This approach must be
extended to all aspects of augmentative communication systems,
especially the physical interface.

Intelligent Tutoring

Sometimes the creation of "better" communication systems is
accompanied by greater complexity that results in more demands for
systems training. In addition, if we are successful in extending
augmentative communication to cognitively impaired individuals, I
believe we will see a need for extensive (and repetitive) training
protocols. Unless our service delivery system makes some dramatic
changes in the coming decade, we may face a "crisis" in the delivery
of system training. The development and delivery of technology could
be held back by our inability to adequately train users. While I think
that the ultimate answer is a re-assessment of the way in which we
currently deliver AAC technology related services, it may be possible
to use technology as a way to extend services to the end
user. Intelligent tutoring systems could assist a clinician or teacher
in training the user how to use an AAC system.

Research Priorities

I would like to discuss four research priorities for the coming
decade. Two of them are technical and two are non-technical:

1.  It is only in the last 2 or 3 years that we have even tried to
    apply Artificial Intelligence to Augmentative Communication. We
    must continue to explore what techniques within the AI community
    will be useful to AAC. These efforts should be directed towrds
    the application of existing AI research to the AAC problems.

2.  David Waltz (1988) in his essay The Prospects for Building Truly
    Intelligent Machines states (among other things) that we need to
    make significant advances in software engineering in order to
    construct the complex programs required for intelligent
    behavior. I believe that this field could take advantage of its
    small size and collaborate in the development of code . We need to
    stop reinventing the wheel and investigate methods for sharing
    software components.

3.  The data that has come out of the work done in communication
    interaction must be reexamined from the perspective of intelligent
    systems. I think that this would help to provide some guidance for
    the technical research as well as generate some new ideas for what
    would be desirable intelligent behavior from an AAC system.

4.  Finally, I think research must be conducted in the area of user's
    attitudes towards system usage. By embedding "intelligence" into
    communication systems we run the risk of the "locus of control"
    being shifted from the user to the machine. While this may be
    somewhat reasonable in the context of a training session, it is
    essential that the machine maintain the user's control during
    actual communication.

References

Naisbitt, J. (1982). , Megatrends: ten new directions transforming our
lives, New York: Warner Books, Inc.

Sokolowski, R. (1988). Natural and artificial intelligence. Daedalus,
(117)1, 45-64.

Waltz, D. (1988). The prospects of building truly intelligent
machines, Daedalus, (117)1, 191-212

Kratt, A. (1987). Communication Interaction Between Aided and Natural
Speakers. Madison: Trace Research and Development Center.

Alm, N., Newell, A., Arnott, J. (1987). A communication aid which
models conversational patterns. Proceedings of the 10th Annual
Conference on Rehabilitation Technology. San Jose, CA. 127-129.

Yang, G., McCoy, K. Demasco, P. Prediction Using a Systemic Tree
Adjoining Grammar. Proceedings of the 13th Annual RESNA Conference,
Washington, DC, submitted.

Demasco, P., McCoy, K., Gong, Y., Pennington, C. Rowe,
C. (1989). Towards More Intelligent AAC Interfaces: The Use of Natural
Language Processing. Proceedings of the 12th Annual RESNA
Conference. NewOrleans, LA, 141-142.

Hsi, S., Agogino, A., Barker, M. (1988) Validation and testing program
of the expert system ADIS: Assitive Device Selector. Proceedings of
the Third International Conference on Rehabilitation
Engineering. Montreal Canada, 74-75.

Shein, F. (1988) A prototype expert system for the design of a visual
keyboard. Proceedings of the Third International Conference on
Rehabilitation Engineering. Montreal Canada, 382-383.