Weak Supervision Challenges
Nathan discusses the difficulties of training models with weak supervision, emphasizing the need for additional lexical information to improve accuracy. Jonathan highlights the issue of models memorizing templates rather than generalizing to new syntactic structures, raising concerns about the effectiveness of current approaches in semantic parsing. Both agree on the importance of further investigation into these memorization effects, which undermine the foundational goals of compositionality in language understanding.In this clip
From this podcast
NLP Highlights
91 - (Executable) Semantic Parsing, with Jonathan Berant
Related Questions