Sort by Structure: Language Model Ranking as Dependency Probing
This work addresses the environmental and computational inefficiency in model selection for structured prediction tasks in NLP, offering a practical solution for researchers and practitioners.
The paper tackles the problem of selecting the best pre-trained language model for dependency parsing tasks by proposing a probing method that measures recoverable dependency information from embeddings, achieving 79% accuracy in predicting the optimal model across diverse LM-language pairs with significantly reduced computational cost.
Making an informed choice of pre-trained language model (LM) is critical for performance, yet environmentally costly, and as such widely underexplored. The field of Computer Vision has begun to tackle encoder ranking, with promising forays into Natural Language Processing, however they lack coverage of linguistic tasks such as structured prediction. We propose probing to rank LMs, specifically for parsing dependencies in a given language, by measuring the degree to which labeled trees are recoverable from an LM's contextualized embeddings. Across 46 typologically and architecturally diverse LM-language pairs, our probing approach predicts the best LM choice 79% of the time using orders of magnitude less compute than training a full parser. Within this study, we identify and analyze one recently proposed decoupled LM - RemBERT - and find it strikingly contains less inherent dependency information, but often yields the best parser after full fine-tuning. Without this outlier our approach identifies the best LM in 89% of cases.