Attention Meets Reachability: Structural Equivalence and Efficiency in Grammar-Constrained LLM Decoding

TL;DR

This paper investigates grammar-constrained decoding in large language models, establishing theoretical bounds, introducing structural ambiguity costs, and connecting these concepts to model architectures and optimization strategies.

Contribution

It provides a formal analysis of grammar-constrained decoding, introduces the structural ambiguity cost metric, and characterizes the complexity and efficiency of decoding algorithms.

Findings

SAC is bounded under right-recursion but grows quadratically under concatenation.

Any efficient online masking engine must incur quadratic work per token.

Existence of minimal-SAC grammar representatives within bounded rewrite families.

Abstract

We study grammar-constrained decoding (GCD) as a coupling between an autoregressive next-token distribution and a reachability oracle over a pushdown system compiled from a context-free grammar (CFG). We prove an oracle invariance theorem: language-equivalent grammars induce identical admissible next-token sets for every prefix, hence identical logit masks, yet can yield provably different compiled state spaces and online ambiguity costs. We give exact control-state blowup counts for the canonical $a^n b^n$ language under redundant nonterminal delegation, and introduce a left-to-right structural ambiguity cost (SAC) measuring incremental packed-parse-forest growth per token. For two equivalent grammars over all finite strings, SAC is $O(1)$ per token under right-recursion but $\Theta(t^2)$ per token and $\Theta(n^3)$ cumulatively under concatenation. We establish engine-independent lower bounds: any sound, retrieval-efficient, parse-preserving online masking engine must incur $\Omega(t^2)$ work per token on a specific constant-size CFG family, unconditionally within this model. We define decoding-cost equivalence classes of grammars and prove existence of minimal-SAC representatives within bounded rewrite families. Finally, we characterize the true conditional sampler via a Doob $h$-transform and derive sharp one-step KL and total-variation distortion bounds for hard-masked decoding in terms of survival-probability spread among admissible next tokens. We integrate these results with Transformer and Mixture-of-Experts architectures, derive latency envelopes in terms of vocabulary size, active state sets, and beam width, and connect SAC to instrumentation-based predictive performance models and automated grammar optimization.