Commonsense knowledge, like knowing that “not everything that looks like fruits is edible” or “rain makes things wet“, is essential for humans to live and communicate. To better understand humans and take more reasonable actions, learning and incorporating commonsense is essential for AI. The seminar will cover the latest research on commonsense modeling for natural language processing. We will talk about casting common sense tasks as NLP problems; review existing methods and benchmarks; assess our current knowledge about commonsense capabilities of language models, and investigate ways to incorporate commonsense into downstream tasks.
- Dozent*in: Thomas Arnold
- Dozent*in: Nafise Sadat Moosavi