Latent Taxonomies: How Property Induction Develops in Language Model Pre-training

Key points

• This behavior, called category-based property induction, is well documented in humans, but whether and how it develops in language models trained only to predict the next token is unclear.1
• To shed light on this, we construct a hierarchical taxonomy of synthetic entities grouped by the properties they share.1
• We then inject facts from this latent taxonomy into the continued pre-training of Pythia and OLMo 2 models from 70M to 7B parameters, and track the development of inductive behaviors.1
• Contrary to the usual expectation that scale helps, we find that induction weakens with model size when the data budget is fixed, suggesting that larger models memorize individual facts rather than compress them into shared structure.1

Sources (1)

• [1]Latent Taxonomies: How Property Induction Develops in Language Model Pre-training
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 11, 06:38 PM

  We find that taxonomically aligned induction emerges from next-token prediction alone: a property learned for one entity transfers more strongly within its taxonomic category than across categories.

  This behavior, called category-based property induction, is well documented in humans, but whether and how it develops in language models trained only to predict the next token is unclear.

Extractive summary: sentences quoted from the sources.