Large language models have led to transformative new applications of language technology. But do they work for everyone? The training and evaluation of large language models has focused predominantly on a few language varieties, such as standard American English. Few attempts have been made to measure the effectiveness of these models across the range of language varieties used by speakers of other dialects. 

David Eisenstein, research scientist at Google, will briefly survey linguistic theories of dialect, and then describe recent research on:

• Recognizing dialect features using large language models
• Building multi-dialect datasets
• Measuring the robustness of large language models to dialect differences.

Join Online
Meeting ID: 94065217584
Password: 027712