Large language models (LLMs) exhibit remarkable similarity to neural activity in the human language network. However, the key properties of language underlying this alignment—and how brain-like representations emerge and change across training—remain unclear. We here benchmark 34 training checkpoints spanning 300B tokens across 8 different model sizes to analyze how brain alignment relates to linguistic competence. Specifically, we find that brain alignment tracks the development of formal linguistic competence—i.e., knowledge of linguistic rules—more closely than functional linguistic competence. While functional competence, which involves world knowledge and reasoning, continues to develop throughout training, its relationship with brain alignment is weaker, suggesting that the human language network primarily encodes formal linguistic structure rather than broader cognitive functions. Notably, we find that the correlation between next-word prediction, behavioral alignment, and brain alignment fades once models surpass human language proficiency. We further show that model size is not a reliable predictor of brain alignment when controlling for the number of features. Finally, using the largest set of rigorous neural language benchmarks to date, we show that language brain alignment benchmarks remain unsaturated, highlighting opportunities for improving future models. Taken together, our findings suggest that the human language network is best modeled by formal, rather than functional, aspects of language.

Large language models (LLMs) exhibit remarkable capabilities on not just language tasks, but also various tasks that are not linguistic in nature, such as logical reasoning and social inference. In the human brain, neuroscience has identified a core language system that selectively and causally supports language processing. We here ask whether similar specialization for language emerges in LLMs. We identify language-selective units within 18 popular LLMs, using the same localization approach that is used in neuroscience. We then establish the causal role of these units by demonstrating that ablating LLM language-selective units – but not random units – leads to drastic deficits in language tasks. Correspondingly, language-selective LLM units are more aligned to brain recordings from the human language system than random units. Finally, we investigate whether our localization method extends to other cognitive domains: while we find specialized networks in some LLMs for reasoning and social capabilities, there are substantial differences among models. These findings provide functional and causal evidence for specialization in large language models, and highlight parallels with the functional organization in the brain.

SentSpace is a modular framework for streamlined evaluation of text. SentSpacecharacterizes textual input using diverse lexical, syntactic, and semantic features derivedfrom corpora and psycholinguistic experiments. Core sentence features fall into three primaryfeature spaces: 1) Lexical, 2) Contextual, and 3) Embeddings. To aid in the analysis of computed features, SentSpace provides a web interface for interactive visualization and comparison with text from large corpora. The modular design of SentSpace allows researchersto easily integrate their own feature computation into the pipeline while benefiting from acommon framework for evaluation and visualization. In this manuscript we will describe thedesign of SentSpace, its core feature spaces, and demonstrate an example use case by comparing human-written and machine-generated (GPT2-XL) sentences to each other. We findthat while GPT2-XL-generated text appears fluent at the surface level, psycholinguistic normsand measures of syntactic processing reveal key differences between text produced by humansand machines. Thus, SentSpace provides a broad set of cognitively motivated linguisticfeatures for evaluation of text within natural language processing, cognitive science, as wellas the social sciences.