A communication tool for people with disabilities: lexical semantics for filling in the pieces
Kathleen F. McCoy, Patrick W. Demasco, Mark A. Jones, Christopher A. Pennington, Peter B. Vanderheyden, Wendy M. Zickus · 1994 · Proceedings of the First Annual ACM Conference on Assistive Technologies (Assets '94) · doi:10.1145/191028.191058
Summary
This paper presents a prototype communication tool designed to reduce the input burden for people with severe speech and motor impairments (SSMI) who use word-based augmentative communication systems. The core problem addressed is that even with word-prediction and word-based selection, generating grammatically correct English sentences requires significant effort — each word is essentially one keystroke, with additional keystrokes needed for morphological endings and function words like articles and prepositions. The researchers proposed a system that allows users to input only uninflected content words (nouns, verbs, adjectives) in a compressed, telegraphic form, while the system automatically fills in the missing grammatical elements. The technical challenge is substantial: without a full syntactic parse tree — since the input is telegraphic rather than standard English — the system must construct a semantic representation of the intended utterance from minimal cues. The paper focuses on the knowledge representation and processing required to derive meaning from these compressed inputs and then generate complete, grammatically correct English sentences.
Key findings
The system demonstrates that lexical semantic knowledge — information about word meanings, their relationships, and the roles they can fill in sentences — can be leveraged to reconstruct full sentences from telegraphic input. By analyzing the semantic properties of the content words provided by the user, the system infers the grammatical relationships between them, determines which function words (articles, prepositions) are needed, and applies appropriate morphological inflections. The approach effectively shifts the linguistic burden from the user to the system, allowing someone with severe motor impairments to communicate in complete sentences while only selecting the core meaning-bearing words. This represents an early application of natural language generation techniques specifically adapted for the constraints of augmentative communication, where the input is inherently incomplete and ambiguous compared to standard text processing scenarios.
Relevance
This paper is a pioneering contribution to the intersection of natural language processing and augmentative and alternative communication (AAC). The concept of expanding telegraphic input into full sentences anticipated features that would later become common in AAC devices and predictive text systems. For practitioners, the work highlights an important principle: accessibility tools should minimize the physical effort required from users with motor impairments while maximizing the expressiveness of their output. The research also raises questions still relevant today about balancing system automation with user control — when a system fills in grammatical elements, it must accurately capture the user's intended meaning. Modern AAC systems and AI-powered communication tools continue to grapple with these same trade-offs between efficiency and accuracy.
Tags: augmentative and alternative communication · natural language processing · lexical semantics · text generation · motor impairment · speech impairment · keystroke reduction