A Word-Finding Algorithm with a Dynamic Lexical-Semantic Memory for Patients with Anomia Using a Speech Prosthesis

Kenneth Mark Colby, Daniel Christinaz, Santiago Graham, Roger C. ParkisonAAAI 1980

Abstract

Anomia, a common symptom of aphasia, leaves patients unable to retrieve specific words despite having the underlying concept clearly in mind. This paper describes an Intelligent Speech Prosthesis (ISP) that utilizes a dynamic lexical-semantic memory to assist in word retrieval. By leveraging residual knowledge—such as partial orthographic information or semantic associations—the system employs a hierarchical search algorithm to locate target words. Crucially, the system features a dynamic reorganization mechanism that adapts the lexicon to the user's personal associations, improving retrieval efficiency over time.

Introduction

Originally presented at the first AAAI conference in 1980, this research addresses the debilitating "tip-of-the-tongue" state experienced by individuals with anomia, often resulting from stroke or brain injury. While these patients often retain the conceptual framework of their intended message, the neurological path to the specific word-sign is obstructed. Clinical evidence suggests that patients frequently maintain "generic properties" of the lost word, including its first and last letters, syllable count, or related semantic concepts. The ISP was developed as a portable, microprocessor-based tool to transform these fragmented clues into successful word retrieval.

System Design and Methodology

The core of the ISP is a Word-Finding Algorithm that operates on a structured, adaptable lexicon.

Lexical Data Structure

The lexicon is organized into Topic Areas (TAs), such as "Food," "Body," or "Clothing." Within these areas, words are stored in linked lists. A significant innovation is the "Goes-With-Word" (GWW) association, which links words based on common or idiosyncratic relationships (e.g., linking "steak" with "meat" or "cow").

To find a word, the user provides up to five parameters based on their residual knowledge:

  1. Topic Area
  2. First Letter
  3. Last Letter
  4. Middle Letter String
  5. Associated Word (GWW)

The algorithm processes these clues through a hierarchical fallback logic designed to maximize efficiency:

  • Case 1 (Specific): The system searches within the designated Topic Area using all orthographic constraints and the GWW association.
  • Case 2 (Broadened): If Case 1 fails, the system ignores the GWW association but remains within the Topic Area.
  • Case 3 (Exhaustive): If the word is still not found, the system searches the entire lexicon using only the letter constraints.

Dynamic Memory Reorganization

A key "intelligent" feature of the ISP is its ability to learn from successful retrievals. When a word is located via an exhaustive search (Case 3), the system automatically updates its internal structure. It adds the word to the user’s specific Topic Area and creates a new GWW link. This mimics human memory reorganization, ensuring that subsequent searches for the same word are faster and more aligned with the user's personal semantic map.

Lexicon Development

The initial system was seeded with approximately 1,800 words, selected from high-frequency discourse lists and standard word-association norms. The researchers also incorporated data from studies on semantic confusions in aphasic patients to ensure the initial links were clinically relevant. This "starting" lexicon is intended to evolve into a specialized vocabulary tailored to the individual's environment and conversational habits.

Results and Observations

Preliminary evaluations of the ISP highlighted several key findings regarding its utility as an assistive device:

  • Computational Efficiency: Despite the hardware limitations of the early 1980s, exhaustive searches of the 1,800-word lexicon were completed within seconds.
  • Constraint Effectiveness: Providing just two or three clues (e.g., "Food" + "Starts with S") typically reduced the candidate list to fewer than 20 words, a manageable number for user selection.
  • Accessibility: The interface was designed for diverse input methods, including keyboard and eye-movement tracking, accommodating patients with severe motor impairments.
  • Therapeutic Potential: The authors posited that the ISP could serve as a rehabilitation tool, where the act of using the device might stimulate natural recovery of word-finding abilities through repeated exercise.

Conclusion and Future Work

The ISP represents an early and significant application of artificial intelligence in assistive technology. While the initial 1,800-word lexicon was sufficient for daily needs, the authors identified scaling the architecture to a full adult vocabulary as a primary challenge. Additionally, the system's reliance on accurate user input remains a limitation; if a patient provides incorrect orthographic clues, the search will fail. Future work involves extensive field testing to quantify the impact on communication speed and quality of life for patients with aphasia.

Summary updated

BibTeX
@inproceedings{colby1980word,
  author = {Kenneth Mark Colby and Daniel Christinaz and Santiago Graham and Roger C. Parkison},
  title = {{A Word-Finding Algorithm with a Dynamic Lexical-Semantic Memory for Patients with Anomia Using a Speech Prosthesis}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1980},
}
Figure