Large Language Models as Theories of Human Language

Dear Colleagues,

The rapid development of Large Language Models (LLMs) has created a productive yet contentious intersection between computational modeling and linguistic theory. While LLMs have demonstrated remarkable success across practical applications, their relevance for understanding human language competence remains unresolved. This Special Issue, Large Language Models as Theories of Human Language, aims to critically examine whether and how LLMs can inform theoretical linguistics, cognitive science, and philosophy of language. By situating current advances within long-standing debates—such as competence versus performance, symbolic versus statistical representations, and the role of innate structure—this issue seeks to clarify the theoretical implications of LLM-based approaches.

Focus:
The Special Issue focuses on the relationship between LLMs and theories of human language, particularly whether LLMs can serve as explanatory models of linguistic competence rather than merely engineering tools. It emphasizes the interface between empirical findings from neural language models and formal theories in linguistics, including syntax, semantics, morphology, and pragmatics.

Scope:
The scope is interdisciplinary, bringing together contributions from linguistics, cognitive science, artificial intelligence, and philosophy. It includes empirical studies comparing LLMs with human linguistic behavior, theoretical analyses of model architectures and representations, and methodological reflections on the use of LLMs in linguistic inquiry. Topics range from learnability and acquisition to interpretability, symbolic structure, and philosophical implications for theories of mind and language.

Purpose:
The purpose of the Special Issue is fourfold:
(i) To assess the extent to which LLMs provide insight into human language competence.
(ii) To encourage further experimental analysis of language structure in LLMs, such as long-distance dependencies.

(iii) To identify the limitations of current models in capturing cognitively plausible linguistic structures.
(iv) To foster dialogue between traditionally distinct research paradigms, including generative linguistics and data-driven machine learning. By doing so, the issue aims to advance a more nuanced understanding of what constitutes an adequate model of human language.

Relationship to Existing Literature

This Special Issue builds upon and extends a growing body of literature that examines the intersection of neural language models and linguistic theory. Prior work has explored correlations between deep neural network representations and grammatical structures (e.g., Hewitt & Manning, 2019; Tenney et al., 2019), as well as the capacity of such models to capture hierarchical dependencies (Linzen et al., 2016). At the same time, critical perspectives—most notably from Chomsky (2012) and Bever et. al (2023) —have emphasized the distinction between performance-based success and theoretical adequacy.

Existing Special Issues in Languages, such as those on tense and aspect, emergent sign languages, and morphosyntax, have demonstrated the value of focused, theoretically grounded collections that integrate empirical and conceptual insights. However, none have directly addressed the implications of LLMs for core questions in linguistic theory. This Special Issue supplements the existing literature by explicitly targeting this gap, offering a venue for systematic comparison between human and machine language systems, and encouraging methodological innovation in the use of LLMs as tools for hypothesis testing.

Moreover, it engages with broader debates in cognitive science regarding inductive bias, learnability, and the poverty of the stimulus (Pinker, 1994; Pullum & Scholz, 2002), situating LLMs within these discussions as both empirical artifacts and theoretical probes. By integrating perspectives across disciplines, the issue aims to move beyond polarized positions and toward a more integrative framework.

Submission Procedure

We request that, prior to submitting a manuscript, interested authors initially submit a proposed title and an abstract of 400–600 words summarizing their intended contribution. Please send it to the guest editors (Csaba.Veres@vu.edu.au) or to Languages editorial office (languages@mdpi.com). Abstracts will be reviewed by the guest editors for the purposes of ensuring proper fit within the scope of the Special Issue. Full manuscripts will undergo double-blind peer-review.

References

Bever TG, Chomsky N, Fong S, Piattelli-Palmarini M. (2023). Even deeper problems with neural network models of language. Behavioral and Brain Sciences. 46:e387. doi:10.1017/S0140525X23001619

Chomsky, N. (2012). The Science of Language: Interviews with James McGilvray. Cambridge University Press.

Hewitt, J and Manning, C.D. (2019). A Structural Probe for Finding Syntax in Word Representations. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4129–4138, Minneapolis, Minnesota. Association for Computational Linguistics.

Linzen, T, Dupoux, E and Goldberg, Y. (2016). Assessing the Ability of LSTMs to Learn Syntax-Sensitive Dependencies. Transactions of the Association for Computational Linguistics, 4:521–535.

Pinker, S. (1994). The Language Instinct. New York Morrow.

Pullum, G. K., & Scholz, B. C. (2002). Empirical assessment of stimulus poverty arguments. The Linguistic Review, 19, 9–50.

Tenney, I,  Das, D. and Pavlick, E. (2019). BERT Rediscovers the Classical NLP Pipeline. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4593–4601, Florence, Italy. Association for Computational Linguistics.

Prof. (Emeritus) Dr. Csaba Veres
Guest Editor

Manuscript Submission Information

Manuscripts should be submitted online at www.mdpi.com by registering and logging in to this website. Once you are registered, click here to go to the submission form. Manuscripts can be submitted until the deadline. All submissions that pass pre-check are peer-reviewed. Accepted papers will be published continuously in the journal (as soon as accepted) and will be listed together on the special issue website. Research articles, review articles as well as short communications are invited. For planned papers, a title and short abstract (about 250 words) can be sent to the Editorial Office for assessment.

Submitted manuscripts should not have been published previously, nor be under consideration for publication elsewhere (except conference proceedings papers). All manuscripts are thoroughly refereed through a double-anonymized peer-review process. A guide for authors and other relevant information for submission of manuscripts is available on the Instructions for Authors page. Languages is an international peer-reviewed open access monthly journal published by MDPI.

Please visit the Instructions for Authors page before submitting a manuscript. The Article Processing Charge (APC) for publication in this open access journal is 1600 CHF (Swiss Francs). Submitted papers should be well formatted and use good English. Authors may use MDPI's English editing service prior to publication or during author revisions.

• large language models
• symbolic and statistical language models
• syntactic probing in neural networks
• interpretability of LLM representations