Large language models have learned to use language.

Area of Science:

• Computational Linguistics
• Cognitive Science
• Artificial Intelligence

Background:

• Traditional language evaluation methods may be insufficient for assessing advanced AI.
• Large language models (LLMs) exhibit complex linguistic behaviors.

Purpose of the Study:

• To explore the implications of LLM language abilities for language science.
• To advocate for new frameworks for evaluating language knowledge.

Main Methods:

• Conceptual analysis of LLM capabilities.
• Review of existing language evaluation paradigms.
• Discussion of the post-Turing test landscape.

Main Results:

• LLMs possess sophisticated language use abilities.
• Current evaluation methods may not capture the full extent of LLM language competence.
• The field is entering a post-Turing test era, necessitating new assessment strategies.

Conclusions:

• Recognizing LLM language learning is crucial for advancing language science.
• New theoretical and empirical approaches are needed to evaluate language knowledge beyond traditional benchmarks.