Evaluating pictogram prediction in a location-aware augmentative and alternative communication system

Area of Science:

• Augmentative and Alternative Communication (AAC)
• Natural Language Processing (NLP)
• Human-Computer Interaction (HCI)

Background:

• Pictogram-based communication systems are crucial for individuals with communication impairments.
• Predicting the next pictogram accurately can significantly enhance communication efficiency.
• Existing prediction mechanisms often lack context-specific adaptation, limiting their effectiveness.

Purpose of the Study:

• To compare the performance of location-aware pictogram prediction mechanisms against a location-agnostic baseline.
• To evaluate the impact of sentence reuse on the effectiveness of different prediction strategies.
• To determine the optimal approach for pictogram prediction across varying user behaviors.

Main Methods:

• Developed and simulated three pictogram prediction mechanisms: All-purpose (All), location-specific (Spec), and mixed (Mix).
• The All approach used a single language model across all locations.
• The Spec approach utilized location-specific language models, while the Mix approach combined general and location-specific models.

Main Results:

• Both Mix and Spec approaches significantly outperformed the All approach when sentence reuse exceeded 50% and 75%, respectively.
• No significant differences were observed between location-aware and All approaches under low sentence reuse conditions.
• The Mix approach demonstrated a statistically significant advantage over the Spec approach in low sentence reuse scenarios.

Conclusions:

• Location-aware pictogram prediction is beneficial but contingent on user sentence reuse patterns.
• The mixed approach offers a balanced strategy, performing well across different reuse levels.
• Future research should explore adaptive strategies to optimize pictogram prediction for diverse user needs.