Do Self-Supervised Speech Models Exhibit the Critical Period Effects in Language Acquisition?

2025-08-28

AI Summary

This study investigates whether self-supervised speech models (S3Ms) exhibit the critical period effect observed in human language acquisition—specifically, whether delayed L2 exposure impairs L2 learning while delayed L1 exposure enhances L1 retention. Method: Using child-directed speech data, we precisely manipulate the onset time of L2 input and the offset time of L1 input, then evaluate model performance on phoneme discrimination tasks. Contribution/Results: Contrary to human developmental patterns, S3Ms do not display a canonical critical period. Instead, delaying L2 input significantly improves L2 phoneme discrimination, whereas truncating L1 input induces L1 forgetting. These findings challenge established theories of language acquisition and reveal a fundamental divergence between neural network speech learning dynamics and human cognitive development. The results provide novel empirical evidence and theoretical insights for modeling developmental trajectories in AI systems and advancing our understanding of artificial language acquisition mechanisms.

Technology Category

• Natural Language Processing: Speech
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

This paper investigates whether the Critical Period (CP) effects in human language acquisition are observed in self-supervised speech models (S3Ms). CP effects refer to greater difficulty in acquiring a second language (L2) with delayed L2 exposure onset, and greater retention of their first language (L1) with delayed L1 exposure offset. While previous work has studied these effects using textual language models, their presence in speech models remains underexplored despite the central role of spoken language in human language acquisition. We train S3Ms with varying L2 training onsets and L1 training offsets on child-directed speech and evaluate their phone discrimination performance. We find that S3Ms do not exhibit clear evidence of either CP effects in terms of phonological acquisition. Notably, models with delayed L2 exposure onset tend to perform better on L2 and delayed L1 exposure offset leads to L1 forgetting.

Problem

• Investigating Critical Period effects in self-supervised speech models
• Examining L2 acquisition difficulty with delayed exposure onset
• Analyzing L1 retention patterns with varying training offsets

Innovation

• Self-supervised speech models trained
• Varying L2 onset L1 offset
• Evaluated phone discrimination performance