Transformers, the Chomsky Hierarchy, and the Limits of Simulation
Abstract
This paper analyzes transformer-based language models through the lens of the Chomsky hierarchy. We argue that implemented transformers are best characterized as bounded finite-state transition systems (Type-3 by mechanism) under fixed architectural and precision constraints. While such systems can approximate higher language classes within bounded regimes, they do not instantiate pushdown stacks, linear bounded tapes, or Turing-complete symbolic machinery. The paper distinguishes simulation from instantiation and explores both formal and economic implications of this boundary, including compositional drift under invariant density and the runtime cost tradeoffs between probabilistic recomputation and deterministic enforcement.
Fields of Study & Subjects
semigroups and automata theory
Natural Language Processing Techniques
Syntax
Semantics
Linguistic Variation
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 1 March 2026
- Language
- EN
- Document type
- Preprint