Skip to search boxSkip to navigationSkip to main content

Subspace Chronicles: How Linguistic Information Emerges, Shifts and Interacts during Language Model Training

Research Output:
Conference Article in Proceeding or Book/Report chapter
Article in proceedings
Peer-review

Open access

Publication Information

Output type

Research Output:
Conference Article in Proceeding or Book/Report chapter
Article in proceedings
Peer-review

Original language

English

Pages from-to (Number of pages)

Pages 13190-13208

Publication milestones

  • Published - 06/12/2023

Publication status

Published - 06/12/2023

Place of publication

Singapore

Publisher

Association for Computational Linguistics, United States

Publication IDs

  • Scopus: 85183298663

Host publication title

Findings of the Association for Computational Linguistics: EMNLP 2023

Abstract

Representational spaces learned via language modeling are fundamental to Natural Language Processing (NLP), however there has been limited understanding regarding how and when during training various types of linguistic information emerge and interact. Leveraging a novel information theoretic probing suite, which enables direct comparisons of not just task performance, but their representational subspaces, we analyze nine tasks covering syntax, semantics and reasoning, across 2M pre-training steps and five seeds. We identify critical learning phases across tasks and time, during which subspaces emerge, share information, and later disentangle to specialize. Across these phases, syntactic knowledge is acquired rapidly after 0.5% of full training. Continued performance improvements primarily stem from the acquisition of open-domain knowledge, while semantics and reasoning tasks benefit from later boosts to long-range contextualization and higher specialization. Measuring cross-task similarity further reveals that linguistically related tasks share information throughout training, and do so more during the critical phase of learning than before or after. Our findings have implications for model interpretability, multi-task learning, and learning from limited data.

Publication metrics

PlumX

Citations
8
Captures
18

Related Event

Title

Conference on Empirical Methods in Natural Language Processing

Event type

Conference

Degree of recognition

International event

Date

06/12/2023 - 10/12/2023

Location

Resorts World Convention CentreSingapore