The paper introduces UPHELD, a large benchmark of human-to-human dialogues written by professional script writers, featuring realistic turn densities and over 36,000 per-turn human annotations. It evaluates existing automatic metrics and LLM-as-a-judge methods, finding them unreliable against expert human judgment. Using UPHELD, the authors develop a Mixture-of-Judges framework that improves correlation with human assessments by about 30%.
arXiv:2607. 12085v1 Announce Type: new Abstract: Evaluating retail conversational agents requires methods beyond lexical-overlap metrics to assess intent alignment, factuality, helpfulness, clarity, tone, and overall response quality.
Large Language Models (LLMs) as judges across various scenarios such as assessing model responses is becoming an increasingly accepted paradigm. However, existing judgment approaches often rely on trained judgers using fixed preference data, which tend to overlook diverse user preferences and struggle to adapt to real-world human-AI dialogue scenarios.
The paper reports the first Turing test for speech‑to‑speech systems, gathering 2,968 human judgments on conversations between nine state‑of‑the‑art S2S systems and 28 humans. None of the evaluated systems passed the test, highlighting a clear gap in human‑likeness. The authors diagnose the failure with an 18‑dimension taxonomy, finding that paralinguistic cues, emotional expressivity, and conversational persona—not semantic understanding—are the main bottlenecks, and they propose an interpretable model for automatic human‑vs‑machine discrimination.
The paper evaluates the use of large language models (LLMs) as judges for assessing conversational voice agents, comparing human judgments with GPT‑4.1 and GPT‑5 across telecom and retail interactions. It examines agreement, metric‑level correlations, and consistency across three evaluation configurations (p0, p1, p2) to determine how reliably LLMs can judge conversational quality and safety. The study finds that LLM‑based evaluation can be effective but its reliability varies by metric and configuration, suggesting a hybrid approach where LLMs handle scalable assessment while humans focus on metrics requiring contextual interpretation.
arXiv:2609.24516v1 Announce Type: new Abstract: In recent years, large language models (LLMs) have emerged as a popular alternative for evaluation. Often referred to as LLMs as judges (LLJs), these s...
The paper introduces a de‑identified corpus of 33 in‑person cognitive assessment conversations, comprising 8,250 utterances annotated for three speaker roles and 56 dialogue acts. The authors benchmark large language models on fine‑grained dialogue‑act classification and next‑patient‑utterance generation, finding that instruction tuning and reasoning‑aware fine‑tuning improve performance but that models still struggle with closely related dialogue acts. The corpus and benchmark are presented as tools to measure interaction structure in cognitive assessments and to support future research on conversational markers, clinician education, and validated simulated patients.
The paper introduces a three‑layer checklist-and-judge framework to evaluate interpreter agents that mediate live conversation across languages. It assesses semantic, pragmatic, and cultural‑social dimensions—naturalness, intent, and social appropriateness—rather than just fidelity, in both single‑turn and multi‑turn settings. Extensive validation shows that conventional MT metrics miss failures in stronger interpreters, and that context, structured instructions, and cultural cues influence communicative success.
arXiv:2609.05806v1 Announce Type: new Abstract: Emotion Recognition in Conversations (ERC) aims to identify speakers' emotions in multi-turn dialogue. Accurate emotion recognition can support a wide...
The paper introduces LongJudgeBench, a benchmark designed to evaluate large language models (LLMs) acting as judges for long-form text generation. It highlights that long-form evaluation requires complex, document-level assessments beyond simple length, such as organization, coverage, depth, consistency, and scenario-specific quality. Experiments show a significant reliability gap among current LLM judges, indicating instability across scenarios and limited effectiveness of rubrics or references.
TalkFa introduces a unified benchmark for Farsi dialogue generation and understanding, comprising three datasets: WIKI‑FADIAL (4.2K Wikipedia‑grounded dialogues), DAILYDIALOG‑FA (6.6K dialogues with dialogue‑act and emotion annotations), and PLAYDIAL‑FA (2.1K theatrical dialogues with sentiment labels). All dialogues are curated through multi‑stage review by native speakers, ensuring high quality. Experiments show that LoRA fine‑tuning improves generation performance with less data, while specific models excel on classification tasks, and human evaluation confirms the benchmark’s reliability.
The paper introduces BASSE, a multilingual meta‑evaluation dataset containing 2,040 human‑rated abstractive summaries produced manually or by five LLMs with four prompts. Annotators scored each summary on coherence, consistency, fluency, relevance, and 5W1H using a 5‑point Likert scale. Benchmarking shows proprietary LLM‑judge models best align with human judgments, followed by criteria‑specific automatic metrics, while open‑source judge LLMs perform poorly.
arXiv:2601. 22888v4 Announce Type: replace-cross Abstract: More than 80% of the 1.