Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.27720

Tracing Recurrent and Emergent Hallucination Patterns in Decomposed Text-to-SQL Generation

Hayden Moore·Abhinav Verma·Anshuman Dutt
View original source ↗Read source document ↗Export metadata ↓

Abstract

Large language models can achieve strong Text-to-SQL benchmark performance while still producing schema and query-logic errors. Most evaluations characterize such failures only in the final generated query, making it difficult to determine whether a related error was also observed in an earlier, simpler step or appears only at the final stage. We present a drill-down evaluation framework that decomposes Text-to-SQL queries into progressively more complex executable sub-queries and corresponding natural-language questions, allowing rule-based hallucination categories to be examined across a generation path. We distinguish recurrent hallucinations, in which the same rule-based category is observed in an earlier step and again in the final query, from emergent hallucinations, in which the category is observed at the final step without an earlier detection in the same path. Across five LLMs evaluated on an expanded BIRD-mini benchmark, several categories show appreciable recurrence probabilities, while clause abuse shows comparatively higher conditional emergence. Under the conversational-history condition, recurrence probabilities are generally higher and emergence probabilities are generally lower. These observations indicate that the two conditional measures do not move uniformly in the same direction under additional context and motivate trajectory-level analysis in addition to final-output evaluation. Code and evaluation artifacts are available in the project repository. https://github.com/HaydenMM/bird-mini-drilldown

Explore related subjects

cs.CL

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Hayden Moore, Abhinav Verma, Anshuman Dutt. 2026-10-20. Tracing Recurrent and Emergent Hallucination Patterns in Decomposed Text-to-SQL Generation. https://arxiv.org/abs/2610.27720

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

KEEP EXPLORING

Related papers

arXiv2026

PRiME: Pragmatic Reasoning In MEntal health

In this work, the task and domain interaction of reasoning and mental health are investigated. As large language models (LLMs) present exceptional text understanding and fluent text generation capability, we analyze whether they can reason in the high-stakes domain of mental health. We begin by introducing three datasets and three tasks leveraging the existing work in pragmatic reasoning. The analysis constitutes an evaluation of $12$ LLMs, including MentaLLaMA, which is a mental health reasoning and explanation generation model. The performance of the models remains on par with human evaluation, with the Mistral-7B model giving the best performance. We further analyze the performance of the models with error asymmetry ratio and conduct first-order attention rollout experiment with MentaLLaMA-7B. We present the results for the community reporting the architectural behaviour of the models for the input text and recommend further careful analysis for the high-stakes domain.

cs.CL↗
arXiv2026

Controlling Output Rankings in LLM-Based Search via Generative Engine Optimization

The way customers search for and choose products is changing with the rise of large language models (LLMs). LLM-based search, or generative engines, provides direct product recommendations to users, rather than traditional online search results that require users to explore options themselves. However, these recommendations are strongly influenced by the initial retrieval order of LLMs, which disadvantages small businesses and independent creators by limiting their visibility. In this work, we propose CORE, an optimization method that \textbf{C}ontrols \textbf{O}utput \textbf{R}ankings in g\textbf{E}nerative Engines for LLM-based search. Since the LLM's interactions with the search engine are black-box, CORE targets the content returned by search engines as the primary means of influencing output rankings. Specifically, CORE optimizes retrieved content by appending strategically designed optimization content to steer the ranking of outputs. We introduce three types of optimization content: string-based, reasoning-based, and review-based, demonstrating their effectiveness in shaping output rankings. To evaluate CORE in realistic settings, we introduce ProductBench, a large-scale benchmark with 15 product categories and 200 products per category, where each product is associated with its top-10 recommendations collected from Amazon's search interface. Extensive experiments on four LLMs with search capabilities (GPT-4o, Gemini-2.5, Claude-4, and Grok-3) demonstrate that CORE achieves an average Promotion Success Rate of \textbf{91.4\% @Top-5}, \textbf{86.6\% @Top-3}, and \textbf{80.3\% @Top-1}, across 15 product categories, outperforming existing ranking manipulation methods while preserving the fluency of optimized content.

cs.CL↗
arXiv2026

StructLens: A Structural Lens for Language Models via Maximum Spanning Trees

Language exhibits inherent structures, a property that explains both language acquisition and language change. Given this characteristic, we expect language models to manifest their own internal structures as well. While interpretability research has investigated how models compute representations mechanistically through attention patterns and Sparse AutoEncoders, the organization of the resulting representations is overlooked. To address this gap, we introduce StructLens, a framework to analyze representations through a holistic structural view. StructLens constructs maximum spanning trees based on the semantic representations in residual streams, inspired by tree representation in dependency parsing, and provides summaries of token relationships in representation space. We analyze how contiguous tokens are also nearby in representation space and find that middle layers show the strongest local-span organization. Moreover, analysis of pre-training checkpoints reveals that smaller local units become detectable earlier in pre-training, and larger units later. Our findings demonstrate that StructLens provides insights into how models organize token representations across layers and training. Our code is available at https://github.com/naist-nlp/structlens.

cs.CL↗