π€ AI Summary
This study addresses the absence of human baselines and the reliance on manual task construction in evaluating entity tracking capabilities of language models by introducing a fair comparative benchmark grounded in natural narratives. Through multi-complexity narrative testing and controlled human behavioral experiments, we systematically assess entity tracking across language models of varying scales. Our findings reveal that models with merely 410M parameters achieve human-level performance, with larger models significantly surpassing it as scale increases. Furthermore, degradation in entity tracking is primarily driven by narrative complexity rather than length. This work reshapes our understanding of the emergence thresholds for core linguistic capabilities, demonstrating that reliable language comprehension emerges at model scales substantially smaller than previously anticipated.
π Abstract
Understanding language requires tracking entities across discourse - i.e., knowing where things are and how they change, even when not explicitly stated. Whether language models perform such tracking in a human-like fashion remains unclear, in part because existing evaluations rely on artificial tasks, far removed from natural language comprehension, and lack comparisons to humans. Here, we evaluate entity tracking in both language models and humans (N = 48) using naturalistic narratives at multiple levels of complexity. In humans, we find that entity tracking degrades specifically with narrative complexity, not narrative length. In language models, we find that human-level entity tracking is already present at 410 million parameters - well below the multi-billion parameter, code-specialised models identified by prior work - and improves with scale, with contemporary models far exceeding human performance. Together, these results demonstrate that entity tracking, a core component of language understanding, emerges at model scales far smaller than previously thought.