Study Reveals How Language Models Learn Grammatical Substructure
Researchers have published a theoretical and empirical analysis of how neural language models learn the substructures of context-free grammars (CFGs), accepted to ICML 2026. The work proves that language modeling loss decomposes linearly over a grammar's top-level subgrammars, breaking down recursively into losses for irreducible components. The findings reveal a fundamental difference between how neural models and human children acquire grammatical structure, with implications for pretraining strategies.
A study accepted to the 43rd International Conference on Machine Learning (ICML 2026) investigates how neural language models behave with respect to the substructure of context-free grammars (CFGs), a formalism relevant to natural language syntax, programming languages, and arithmetic. The authors define the concept of subgrammars and prove that language modeling loss recurses linearly over top-level subgrammars, ultimately decomposing into losses for irreducible subgrammars. Empirically and under additional theoretical assumptions, parametrized models are found to learn subgrammars in parallel — a striking contrast to human children, who tend to master simpler grammatical substructures before more complex ones. The study also finds that subgrammar pretraining can improve final model performance, but only for models that are very small relative to the grammar's complexity. Additionally, alignment analyses show that pretraining consistently produces internal representations that better reflect the grammar's underlying substructure, even when performance gains are limited.
What's missing
The study focuses on synthetic or formal CFGs as a controlled setting; it is unclear to what extent the theoretical results and empirical findings generalize to the full complexity of real-world natural language, which is not strictly context-free. The paper does not address how these findings scale to large language models trained on natural text corpora, leaving open questions about practical applicability.
What different sources said
- arXiv cs.CLCenter
Unraveling Syntax: Language Modeling and the Substructure of Grammars
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.