| Title | : | LLMs for accessibility and science |
| Speaker | : | Dr. Subhashini Venugopalan (Technical Lead, Google) |
| Details | : | Mon, 15 Jul, 2024 11:00 AM @ SSB 334 |
| Abstract: | : | Being unable to speak or type due to a debilitating condition like ALS is the reality for many individuals with impaired speech. While augmentative and alternative communication (AAC) devices provide essential support, there is much to be done to enable such individuals to engage in spontaneous conversation to express themselves and be better understood. In this talk I will present two applications that demonstrate the potential of Large Language Models for such users. First, on how LLMs can enhance the speed and ease of text entry for eye-gaze typers and people living with ALS. Second, on how LLMs can be used as a proxy for human evaluation to determine quality of speech recognition models, where we use LLMs for evaluating the comprehensibility of the predicted transcripts with respect to ground truth. LLMs also have a huge potential to help with many aspects of science. If time permits, I will share information on a dataset that we release to help evaluate the ability of LLMs to address queries that rely on the comprehension of information contained in figures and tables from scientific papers. |
