Peer-reviewed research
Using Large Language Models to Accelerate Communication for Users with Severe Motor Impairments
Shanqing Cai et al., Nature Communications, 2024 · DOI 10.1038/s41467-024-53873-3
Read the originalRead on October 16, 2026
- What it is
- A study of SpeakFaster, a text entry system built on large language models that predicts whole phrases from abbreviated input. The authors measured it in an offline simulation, in a small pilot with people who were not augmentative communication users, and in tests with two users with ALS.
- How much weight it carries
- Peer-reviewed. The offline simulation measures motor actions, not time, and the pilot had 19 non-AAC participants on a phone. The headline speed gains are for two users with ALS typing by eye gaze, so they do not describe typical typists.
Facts taken from this source
- In an offline simulation, SpeakFaster, a text entry system built on large language models, saved 57 percent more motor actions than traditional predictive keyboards.
“saving 57% more motor actions than traditional predictive keyboards in offline simulation” (Abstract)
- For participants who do not use augmentative communication, overall text entry speed with SpeakFaster was similar to conventional smart keyboards, while the motor action savings remained substantial.
“On non-AAC participants, the overall speed of text entry was similar to conventional smart keyboards, while the motor action savings was still substantial.” (Article text, non-AAC pilot)