Transparency Helps Reveal When Language Models Learn Meaning
Wu, Zhaofeng, Merrill, William, Peng, Hao, Beltagy, Iz, Smith, Noah A.
–arXiv.org Artificial Intelligence
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.
arXiv.org Artificial Intelligence
Mar-4-2023
- Country:
- Oceania > Australia
- North America
- Dominican Republic (0.04)
- United States
- New York (0.04)
- Washington > King County
- Seattle (0.04)
- Minnesota > Hennepin County
- Minneapolis (0.14)
- Massachusetts
- Middlesex County > Cambridge (0.04)
- Hampshire County > Amherst (0.04)
- Louisiana > Orleans Parish
- New Orleans (0.04)
- Illinois > Cook County
- Chicago (0.04)
- Florida > Palm Beach County
- Boca Raton (0.04)
- California > Santa Clara County
- Stanford (0.04)
- Europe
- Spain (0.04)
- Germany (0.04)
- Portugal > Lisbon
- Lisbon (0.04)
- Netherlands > North Holland
- Amsterdam (0.04)
- Asia
- Laos (0.04)
- China > Hong Kong (0.04)
- Japan > Kyūshū & Okinawa
- Kyūshū > Miyazaki Prefecture > Miyazaki (0.04)
- Genre:
- Research Report > New Finding (0.93)
- Technology: