Abstract
There have been many efforts to try to understand what grammatical knowledge (e.g., ability to understand the part of speech of a token) is encoded in large pre-trained language models (LM). This is done through `Edge Probing' (EP) tests: simple ML models that predict the grammatical properties of a span (whether it has a particular part of speech) using \textit{only} the LM's token representations. However, most NLP applications use \finetuned\ LMs. Here, we ask: if a LM is \finetuned, does the encoding of linguistic information in it change, as measured by EP tests? Conducting experiments on multiple question-answering (QA) datasets, we answer that question negatively: the EP test results do not change significantly when the fine-tuned QA model performs well or in adversarial situations where the model is forced to learn wrong correlations. However, a critical analysis of the EP task datasets reveals that EP models may rely on spurious correlations to make predictions. This indicates even if \finetuning\ changes the encoding of such knowledge, the EP tests might fail to measure it.
Abstract (translated)
URL
https://arxiv.org/abs/2109.07102
