Abstract
For augmentative and alternative communications (AAC), there exists an inverse relationship between input expressivity and input efficiency: A technique that selects characters one-by-one provides substantial freedom of expression but requires many actions to compose a sentence, while techniques that select from a preset vocabulary can enable quicker input that is limited to the vocabulary. This disclosure describes techniques of augmentative and alternative communication that leverage a large language model (LLMs) to simultaneously optimize both input expressivity and input efficiency. Upon the entry of a few characters of a sentence or command (e.g., using eye trackers, finger tapping, etc.), word and sentence suggestion keyboards are surfaced and populated with probable sentence completions determined by an LLM. The user completes their sentence or command by selecting words, sentences, or characters suggested by the LLM. The LLM can be generic or specialized to particular domains. The LLM can be set (contextualized) to a particular persona. An intuitive interface is thus provided that enables high input speed while retaining expressivity.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Guevara-Rukoz, Adriana; Yamagami, Atsuko; Yamashita, Atsushi; Arao, Satoru; Takebe, Yasuaki; and Iitsuka, Shuhei, "Efficient Augmentative Communication Using Large Language Models", Technical Disclosure Commons, (April 03, 2024)
https://www.tdcommons.org/dpubs_series/6853
COinS