Extending Brain-Computer Interface Access with a Multilingual Language Model in the P300 Speller.

Area of Science:

• Neuroscience and Biomedical Engineering
• Human-Computer Interaction
• Computational Linguistics

Background:

• Brain-computer interfaces (BCI) offer communication restoration for individuals with advanced neuromuscular diseases.
• Language models have significantly improved BCI typing speed and accuracy, primarily in English.
• Cross-linguistic generalizability of these BCI enhancements remains largely unexplored due to linguistic diversity.

Purpose of the Study:

• To adapt and evaluate a language model-based BCI classifier for non-English languages.
• To demonstrate the generalizability of advanced BCI methods across different linguistic structures.
• To assess the performance consistency of BCI systems in Spanish and Greek compared to English.

Main Methods:

• Adapted an English-based language model classifier for use with Spanish and Greek.
• Conducted online experimental trials with 30 healthy native speakers for each language (English, Spanish, Greek).
• Measured BCI system performance using information transfer rate (bits/minute).

Main Results:

• No significant differences in BCI performance were observed across the three languages.
• Average information transfer rates were comparable: English (66.20 bits/min), Spanish (61.97 bits/min), and Greek (60.89 bits/min).
• The adapted language model-based classifier demonstrated robust cross-linguistic applicability.

Conclusions:

• Language model-based BCI methods are generalizable across languages with diverse alphabets and grammar.
• This cross-linguistic adaptability can significantly expand access to BCI communication technologies.
• Potential for broader BCI adoption, including in underserved populations and developing regions.