A dynamically updated surprisal-based curriculum, run over a small ELC-BERT model, improves BabyLM 2024 EWOK and GLUE scores over official baselines but hurts BLiMP.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
A surprisal oracle for when every layer counts
A dynamically updated surprisal-based curriculum, run over a small ELC-BERT model, improves BabyLM 2024 EWOK and GLUE scores over official baselines but hurts BLiMP.