Project Title: Compansion: A Technique that Applies Natural Language
Processing to Augmentative Communication

Name: Patrick Demasco, MS; Kathleen McCoy, PhD; Mark Jones, MS; Chris
Pennington, MS; Stephanie Snyder, BS

Address: Applied Science and Engineering Laboratories University of
Delaware/ A.I duPont Institute PO Box 269 Wilmington, DE 19899

Telephone: (302) 651-6830

Sponsor: National Institute on Disability and Rehabilitation Research,
U.S. Department of Education, 400 Maryland Ave. SW, Washington, DC
20202, Director: William Graves, Ed.D.

Nemours Foundation A.I. duPont Institute PO Box 269 Wilmington, DE
19899 Research Director: Charles Hartzell, Ph.D.

References: Jones, M., Demasco, P., McCoy, K., & Pennington,
C. (1991). Knowledge Representation Considerations for a Domain
Independent Semantic Parser. In J.J. Presperin (Ed.), Proceedings of
the Fourteenth Annual RESNA Conference (pp. 109-111). Washington, D.C:
RESNA Press.

Key Words: Augmentative and Alternative Communication, Rate
Enhancement, Natural Language Processing.

Category: Independent Living Aids - Communication Methods and Systems

Purpose

The goal of this project is to increase the communication rate of
physically disabled individuals via natural language processing
techniques. We have developed a technique called Compansion which
takes as input a compressed message (i.e., uninflected content words)
from the disabled individual, and generates a syntactically and
semantically well-formed sentence. At the same time, we wish to do
this by placing as little a burden on the user as possible. Thus, we
are not interested in a simple coding system where sentences have been
stored and are indexed by their content words.

Progress

The present system has a vocabulary of over 1000 words. It handles
most tenses, produces a variety of sentence constructions, and has the
capability to infer the verb or subject in some situations.The
structure of the semantic parser has been radically overhauled. It now
relies on a much more flexible knowledge structure than the template
matching that it previously employed. In addition, a new version of
the generator is being employed which provides a greater number of
possible sentence constructions.

Methodology

Input to our system are the uninflected content words of an utterance;
thus, many function words such as determiners (e.g., the, a) and
prepositions (e.g., of, in) will be left out. The system is
responsible for filling in missing words as well as correctly
conjugating the verb and forming a syntactically correct utterance. We
attempt to form an utterance whose word order most closely reflects
the word order given in the original input. For example, if the system
is given "APPLE EAT JOHN," we would like the system to produce the
sentence, "THE APPLE IS EATEN BY JOHN."  In order for the system to
generate a well-formed utterance, it employs a semantic parser to form
a semantic representation of the input words. In this example, the
parser recognizes that EAT can be a verb which prefers an animate
ACTOR and an inanimate/food OBJECT in order to correctly infer the
semantic relationships between these input words. The resulting
semantic representation (along with a specification of the original
word order) is then passed to the translation component. The
translator is responsible for enhancing the semantic terms with
language-specific information and converting them into a
representation format compatible with the final phase of processing, a
sentence generator. The generator attempts to form a syntactically
correct sentence that retains the general order of the original input
words (if possible).

Results

The system is much more robust. As well as being generally more
flexible, it can now gracefully handle unknown words, and even make a
reasonable attempt at determining the meaning of the unknown word from
the surrounding words. Areas for further improvement include
developing a syntactic pre-processor that will determine modification
bindings as well as determining the scope of conjunctions by examining
the order of the input. Also, we intend to allow for more complex
sentence constructions, make use of discourse information, and we are
continuing our collaboration with Semantic Compaction and Prentke
Romich to transfer this technology into a "scaled-down" system. In
addition, we are working on expanding the coverage of the system to
include metaphorical expressions.