SkepticalSam
A Unified Assessment of the Poverty of the Stimulus Argument for Neural Language Models
A Unified Assessment of the Poverty of the Stimulus Argument for Neural Language Models
Published: 12 May 2026 · Updated: 13 July 2026
Read the original sourceThe Critique
Uses perplexity-based preference (not actual linguistic competence). Dismisses multimodal grounding too quickly. "Above-chance" (65-74%) ≠ "acquired" (children reach near-adult mastery). Tests only GPT-2 style transformers.
Why It Matters
Paper positioned as evidence against linguistic nativism. Gap between above-chance performance and human-like syntactic competence being systematically elided.