Abstract
This paper proposes an intelligent Korean word relay game (Kkeutmalitgi) designed specifically for language rehabilitation and diagnosis. Traditional word relay programs often discourage players because computers possess an unbeatable vocabulary. To address this, we implemented a system that considers real-world language usage patterns. By utilizing large-scale corpus data and a Min-Max algorithm, the system modulates its difficulty based on word frequency and "neighborhood size" (branching factor). This approach ensures the computer acts as a human-like opponent, making it a more effective tool for treating aphasia and supporting language development in children.
Introduction
Word relay games are excellent cognitive exercises for linguistic rehabilitation. However, the inherent advantage of a computer—the ability to store and recall hundreds of thousands of words—often results in a "superhuman" opponent that demotivates patients. For a game to be therapeutic, the computer must exhibit human-like behavior, understanding which words are common and how specific moves might "trap" a player. This research presents a framework that uses statistical language data to create a balanced and motivating environment for therapy.
System Architecture
The system is divided into three primary components: a user interface (including speech recognition and synthesis), a dictionary system, and a word recognition/generation engine. The dictionary system is built upon the Sejong Corpus (10 million words), from which nouns and their usage frequencies were extracted to create distinct dictionaries for recognition and generation.
Intelligent Word Selection and Difficulty Scaling
The core innovation of this system lies in its move-selection logic. Unlike simple random selection, this system calculates a word's difficulty using two primary metrics:
- Neighborhood Size (NS): This refers to the number of words in the dictionary that start with the final syllable of the current word. A low NS creates a "trap" for the opponent, increasing difficulty.
- Word Frequency: Words are categorized into High, Medium, and Low frequency based on the Sejong Corpus. Common words are easier for human players to retrieve, while rare words increase cognitive load.
To simulate strategic play, the system employs a Min-Max algorithm with look-ahead. Instead of just picking the next word, the computer evaluates the game state several turns in advance to choose moves that align with the user's selected difficulty level (Easy, Medium, or Hard).
Furthermore, to feel truly human, the computer includes a probabilistic failure rate. Even if a valid word exists, the system may "fail" to find it based on a probability threshold ($p$), allowing the human player a chance to win and maintain engagement.
Implementation and Results
The system was successfully implemented on both PC and PDA platforms. It features three distinct modes that adjust the weights of frequency and neighborhood size in the difficulty formula. Beyond gameplay, the system offers diagnostic potential by tracking user metrics such as average response time, the frequency level of words used, and word length, which can help clinicians monitor a patient's progress.
Conclusion
By incorporating real-world language characteristics and strategic algorithms, this intelligent word relay game provides a more natural and therapeutic interaction than standard software. Future work will focus on clinical validation with patient groups to establish statistical norms for language diagnosis and further refine the difficulty parameters.
Summary updated