Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.03329

SyntaxBench: A Statistical Diagnostic Framework for Character-Level Reasoning in Large Language Models

Mohsen Larni·Sobhan Ebrahimi Azar·Pouyan Nahed·Kazem Taghva
View original source ↗Read source document ↗Export metadata ↓

Abstract

Large language models are increasingly used where small syntactic errors matter, yet character-level reasoning is still evaluated mostly through isolated probes and aggregate accuracy. We introduce SyntaxBench, a diagnostic benchmark and statistical evaluation framework for character-level reasoning. It contains five core tasks, character counting, letter containment, palindrome detection, edit distance, and longest-string selection, plus index_to_span, a harder substring-extraction stress test. The five core tasks use paired English and character-length-matched random-string inputs. index_to_span documents share a 200-500 word band and are not character-length matched. All six tasks use zero-, one-, and four-shot prompts. We evaluate eight open-weight models from 2B to 32B parameters across 11 reasoning-mode configurations. The framework reports exact-match and relaxed accuracy, Cohen's kappa, paired McNemar tests with odds ratios, bootstrap confidence intervals, Kendall's tau, class-conditional metrics, tokenization analysis, and multiple-comparison-corrected tests. Three findings stand out. First, tokenization shapes accuracy: random strings are more character-visible than English strings (1.892 vs. 3.169 characters per token), and character-counting accuracy falls as English words occupy more tokens. Second, reasoning mode is not uniformly helpful: Gemma4-31B is nearly unchanged across modes on the near-saturated tasks, while Qwen3.6-27B is worse with thinking on palindrome detection (0.952 non-thinking vs. 0.886 thinking at four-shot). Third, index_to_span remains largely unsolved; the best four-shot exact-match accuracy is 6.75%. Character-level evaluation needs controlled inputs, paired tests, and analyses of tokenization and reasoning mode rather than aggregate accuracy alone.

Explore related subjects

cs.CLcs.AIcs.LG

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Mohsen Larni, Sobhan Ebrahimi Azar, Pouyan Nahed, Kazem Taghva. 2026-10-02. SyntaxBench: A Statistical Diagnostic Framework for Character-Level Reasoning in Large Language Models. https://arxiv.org/abs/2610.03329

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

ETHER: Aligning Emergent Communication for Hindsight Experience Replay

Hindsight Experience Replay (HER) enhances sample efficiency in goal-conditioned reinforcement learning (RL) by relabelling failed trajectories with goals that were actually achieved. However, HER implicitly assumes access to a goal relabelling function and a predicate function that determines whether a goal has been satisfied. These assumptions break down in instruction-following tasks, where goals are expressed in natural language and differ from the state space. We formalize this as the Hindsight Reinforcement Learning problem, which shows the need to jointly learn these functions alongside the RL policy. To address it, we propose ETHER (Emergent Textual Hindsight Experience Replay), an agent that leverages Emergent Communication to learn the goal-relabelling and predicate functions. ETHER uses a referential game (RG) to train a speaker and a listener to develop a grounded, artificial language describing environment states. It partially aligns this emergent language with instruction language using co-occurrence patterns between task instructions and RL observations. We prove that the relabelling and predicate functions that ETHER derives from the RG avoid the degenerate solutions of the Hindsight RL problem, namely trivial predicates and collapsed relabelling functions. Experiments on BabyAI's PickupDist task show that ETHER's learned RG speaker and listener can function as the goal relabelling and predicate functions of HER, improving sample efficiency despite imperfect language alignment. Our work bridges Emergent Communication and goal-conditioned RL, opening the door to wider applications of HER.

cs.CL↗
arXiv2026

Automatic register identification for the open web using multilingual deep learning

This article presents multilingual deep learning models for identifying web registers -- text varieties such as news reports and discussion forums -- across 16 languages. We introduce the Multilingual CORE corpora, which contain over 72,000 documents annotated with a hierarchical taxonomy of 25 registers designed to cover the entire open web. Using multi-label classification, our best model achieves 79% F1 averaged across languages, matching or exceeding previous studies that used simpler classification schemes. This demonstrates that models can perform well even with a complex register scheme at multilingual scale. However, we observe a consistent performance ceiling across all models and configurations. When we remove documents with uncertain labels through data pruning, performance increases to over 90% F1, suggesting that this ceiling stems from inherent ambiguity in web registers rather than model limitations. Analysis of hybrid texts (those combining multiple registers) reveals that the main challenge lies not in classifying hybrids themselves, but in distinguishing hybrid from non-hybrid documents. Multilingual models consistently outperform monolingual ones, particularly for languages with limited training data. Zero-shot performance on unseen languages drops by an average of 7%, though this varies by language (3--8%), indicating that while registers share features across languages, they also retain language-specific characteristics.

cs.CL↗
arXiv2026

Evaluating the Retrieval Robustness of Large Language Models

Retrieval-augmented generation (RAG) generally enhances large language models' (LLMs) ability to solve knowledge-intensive tasks. But RAG could also lead to performance degradation due to imperfect retrieval and the model's limited ability to leverage retrieved content. In this work, we evaluate the robustness of LLMs in practical RAG setups (henceforth retrieval robustness). We focus on three research questions: (1) whether RAG is always better than non-RAG; (2) whether more retrieved documents always lead to better performance; and (3) whether document order impacts results. To facilitate this study, we establish a benchmark of 1,891 samples spanning five datasets across three task categories, each with documents retrieved using both sparse and dense retrievers. We introduce three robustness metrics, each corresponding to one research question. Our experiments across 11 LLMs show that models achieve generally high retrieval robustness, but robustness varies substantially across tasks, suggesting that the decision to adopt RAG remains a case-by-case consideration. We further examine four additional prompting strategies that vary how models interact with retrieved documents. We find that Qwen and GPT models suffer notable robustness declines when reasoning is disabled, even on single-hop QA tasks, and that providing retrieved documents as tool responses improves Claude models but hurts Qwen and GPT models, highlighting potential issues of the GPT models regardless of their best overall robustness under vanilla prompting.

cs.CL↗