At this week’s lab meeting Koustuv Sinha will give a talk on existing large language models and their need (or lack thereof) to learn syntax given the current training and evaluation methodologies.
- Thursday, January 21, 13:30–14:30 (Montreal time, UTC-5).
- Meetings are via Zoom. If you are interested in attending any of the meetings this semester, please take a moment now to register at this link. After approval, you will receive a confirmation email with the link to join.
Natural Language Understanding has witnessed a watershed moment with the introduction of large pre-trained Transformer networks. These models achieve state-of-the-art on various tasks, notably including Natural Language Inference (NLI). Many studies have shown that the large representation space imbibed by the models encodes some syntactic and semantic information. However, to really "know syntax", a model must recognize when its input violates syntactic rules and calculate inferences accordingly. In this work, we find that state-of-the-art NLI models, such as RoBERTa and BART are invariant to, and sometimes even perform better on, examples with randomly reordered words. With iterative search, we are able to construct randomized versions of NLI test sets, which contain permuted hypothesis-premise pairs with the same words as the original, yet are classified with perfect accuracy by large pre-trained models, as well as pre-Transformer state-of-the-art encoders. We find the issue to be language and model invariant, and hence investigate the root cause. To partially alleviate this effect, we propose a simple training methodology. Our findings call into question the idea that our natural language understanding models, and the tasks used for measuring their progress, genuinely require a human-like understanding of syntax.
Koustuv is a third year PhD student at Mila, co-supervised with Joelle Pineau and Will Hamilton. His primarily research direction involves understanding systematic generalization in natural language and graph structured data. Personal website here.