paper-with-me

Papers

Beyond Aggregate Scores: Behavioral Correctness Assumptions for Assessing Reference-Based Automatic Evaluation Methods

2026-09-04 · Maria Mahbub, Ashley Rice, Michael R. Munroe, Amidu Kamara, Amir Sadovnik arxiv

Automated reference-based evaluation methods play a critical role in assessing natural language generation systems. Existing meta-evaluation primarily measures agreement with human judgments or benchmark labels, providing limited insight into evaluator behavior under controlled conditions. We introduce behavioral correctness assumptions, a complementary framework for evaluating reference-based automatic evaluation methods. We define a taxonomy of correctness-preserving and correctness-altering assumptions and operationalize them through controlled response transformations that specify expected scoring behaviors. We evaluate diverse lexical, character-level, semantic, LLM-based, and hybrid evaluators and analyze their assumption-level behavior, stability, sensitivity, repeat-run variability, configuration sensitivity, and reproducibility. Our experiments reveal distinct behavioral trade-offs across evaluation paradigms: no evaluator satisfies all proposed correctness assumptions, and evaluators with similar aggregate performance can exhibit substantially different behavioral profiles. These findings demonstrate that behavioral correctness assumptions provide diagnostic information obscured by conventional aggregate meta-evaluation.

📄 PDF Abstract BibTeX arXiv:2609.05289

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Two Regimes of Chain-of-Thought Unfaithfulness: Behavioral Detection Fails Where Models Are Wrong

2026-07-26 · Suramya R. Angdembay, Dikshant Aryal, Nick Rahimi arxiv

Chain-of-thought (CoT) explanations support oversight only if they are faithful: the stated reasoning must actually produce the answer. Auditing black-box (behavioral) detection of unfaithful CoT against FaithCoT-Bench's…

Conformal Off-Policy Prediction in Contextual Bandits

2022-06-09 · Muhammad Faaiz Taufiq, Jean-Francois Ton, Rob Cornish, Yee Whye Teh 외

Most off-policy evaluation methods for contextual bandits have focused on the expected outcome of a policy, which is estimated via methods that at best provide only asymptotic guarantees. However, in many applications, t…

Conformal PredictionMulti-Armed BanditsOff-policy evaluationPrediction

The Asymmetric Harms of LLM Compression

2026-08-20 · Yuan Wu, Mairui Li, Lesia Semenova, Chudi Zhong arxiv

Large language models (LLMs) compression reduces deployment costs, but standard aggregate metrics like perplexity and accuracy often mask underlying behavioral shifts. In this work, we systematically evaluate 3 LLMs acro…

Beyond Local Accuracy: A Protocol-Level Identifiability Audit for Controlled LLM Reasoning Evaluation

2026-08-13 · Junhao Luo, Ning Huang, Ziqi Sha, Wenxuan Tang 외 arxiv

LLM benchmark scores can be precise even when the observation protocol does not identify the behavioral property they are intended to measure. In a controlled, solver-grounded setting, we formalize a protocol-level ident…

Improving Small Language Models for Code Generation with Reinforcement Learning from Verification Feedback

2026-05-28 · Egor Skopin, Evgeny Kotelnikov arxiv

Reinforcement learning with verifiable rewards (RLVR) trains language models using programmatically checkable signals such as unit-test outcomes, enabling direct optimization for functional correctness in code generation…

Reinforcement LearningCode Generation