Tag
This paper investigates whether Large Language Models exhibit the same usage-based linguistic productivity constraints (entrenchment and preemption) as humans, finding that models can reproduce coercion but fail to apply statistical preemption to avoid overgeneralization.