Transparency Helps Reveal When Language Models Learn Meaning
Zhaofeng Wu, William Merrill, Hao Peng, Iz Beltagy, Noah A. Smith
Abstract
Many current NLP systems are built from language models trained to optimize unsupervised objectives on large amounts of raw text. Under what conditions might such a procedure acquire meaning? Our systematic experiments with synthetic data reveal that, with languages where all expressions have context-independent denotations (i.e., languages with strong transparency), both autoregressive and masked language models successfully learn to emulate semantic relations between expressions. However, when denotations are changed to be context-dependent with the language otherwise unmodified, this ability degrades. Turning to natural language, our experiments with a specific phenomenon—referential opacity—add to the growing body of evidence that current language models do not represent natural language semantics well. We show this failure relates to the context-dependent nature of natural language form-meaning mappings.- Anthology ID:
- 2023.tacl-1.36
- Volume:
- Transactions of the Association for Computational Linguistics, Volume 11
- Month:
- Year:
- 2023
- Address:
- Cambridge, MA
- Venue:
- TACL
- SIG:
- Publisher:
- MIT Press
- Note:
- Pages:
- 617–634
- Language:
- URL:
- https://aclanthology.org/2023.tacl-1.36
- DOI:
- 10.1162/tacl_a_00565
- Cite (ACL):
- Zhaofeng Wu, William Merrill, Hao Peng, Iz Beltagy, and Noah A. Smith. 2023. Transparency Helps Reveal When Language Models Learn Meaning. Transactions of the Association for Computational Linguistics, 11:617–634.
- Cite (Informal):
- Transparency Helps Reveal When Language Models Learn Meaning (Wu et al., TACL 2023)
- PDF:
- https://preview.aclanthology.org/starsem-semeval-split/2023.tacl-1.36.pdf