Probing Across Time: What Does RoBERTa Know and When?
This work provides insights into the learning dynamics of language models, which can help researchers develop more efficient pretraining methods.
The paper investigates when RoBERTa acquires different types of knowledge during pretraining, finding that linguistic knowledge is learned quickly and robustly, while facts and commonsense are slower and domain-sensitive, and reasoning abilities are unstable.
Models of language trained on very large corpora have been demonstrated useful for NLP. As fixed artifacts, they have become the object of intense study, with many researchers "probing" the extent to which linguistic abstractions, factual and commonsense knowledge, and reasoning abilities they acquire and readily demonstrate. Building on this line of work, we consider a new question: for types of knowledge a language model learns, when during (pre)training are they acquired? We plot probing performance across iterations, using RoBERTa as a case study. Among our findings: linguistic knowledge is acquired fast, stably, and robustly across domains. Facts and commonsense are slower and more domain-sensitive. Reasoning abilities are, in general, not stably acquired. As new datasets, pretraining protocols, and probes emerge, we believe that probing-across-time analyses can help researchers understand the complex, intermingled learning that these models undergo and guide us toward more efficient approaches that accomplish necessary learning faster.