The Rediscovery Hypothesis: Language Models Need to Meet Linguistics
arXiv:2103.01819 · doi:10.1613/jair.1.12788
Abstract
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.
References in corpus (6)
- Unsupervised Domain Adaptation by Backpropagation
- Tying Word Vectors and Word Classifiers: A Loss Framework for Language Modeling
- What do you learn from context? Probing for sentence structure in contextualized word representations
- A Mutual Information Maximization Perspective of Language Representation Learning
- Analogies Explained: Towards Understanding Word Embeddings
- When Do You Need Billions of Words of Pretraining Data?