BCIwiki (bciwiki.com) — The study was published in IEEE Transactions on Bio-Medical Engineering on May 15, 2026. P300-based speller brain-computer interfaces provide promising communication solutions for individuals with severe motor impairments such as amyotrophic lateral sclerosis (ALS), but existing systems are constrained by slow typing speed and limited efficiency.
Researchers Hong J., Rao P., Wang W., and Najafizadeh L. present ChatBCI-Assist, an intent-based P300 speller integrating a locally deployed large language model (LLM), an efficient graphical user interface, and an adaptive stopping strategy for key selection. The LLM, fine-tuned on an ALS-specific communication corpus using low-rank adaptation (LoRA), produces context-aware, prefix-constrained word and phrase predictions in real time. The team also introduces semantic spelling tasks and semantic information transfer rate (SITR), a metric for evaluating intent-based communication efficiency.
Online experiments show that ChatBCI-Assist achieves record performance: an average information transfer rate of 105.2 bits/min, a character-level mutual information rate of 52.9 bits/min, 19.7 characters per minute in copy-spelling tasks, 30.7 characters per minute in semantic spelling tasks, and a semantic ITR of 147.1 bits/min. User experience evaluations indicate reduced workload and improved usability compared to traditional copy-spelling paradigms.