The Rediscovery Hypothesis: Language Models Need to Meet Linguistics
This addresses a foundational debate in NLP about the role of linguistics in language models, but it is incremental as it builds on existing probing methods.
The paper investigates whether linguistic knowledge is necessary for language models' performance, showing that compressed models retain linguistic probing scores and proposing an information-theoretic framework to relate language modeling to linguistic information.
There is an ongoing debate in the NLP community whether modern language models contain linguistic knowledge, recovered through so-called probes. In this paper, we study whether linguistic knowledge is a necessary condition for the good performance of modern language models, which we call the \textit{rediscovery hypothesis}. In the first place, we show that language models that are significantly compressed but perform well on their pretraining objectives retain good scores when probed for linguistic structures. This result supports the rediscovery hypothesis and leads to the second contribution of our paper: an information-theoretic framework that relates language modeling objectives with linguistic information. This framework also provides a metric to measure the impact of linguistic information on the word prediction task. We reinforce our analytical results with various experiments, both on synthetic and on real NLP tasks in English.