arXiv AI

Exposing Weaknesses in Emotion Recognition in Conversations

arXiv Computation and Language
Aug 28

Evaluating Language Models in Realistic Conversational Contexts

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%.

By Ilija Subasic, Andrew Rabinovich, Zhao Chen
arXiv Computation and Language
Sep 3

TalkFa: A Unified Benchmark for Farsi Dialogue Generation and Understanding

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.

By Neda Jamshidi, Kamyar Zeinalipour, Fahimeh Akbari, Monica Bianchini, Marco Maggini, Marco Gori
arXiv AI
Aug 26

When Less Is More: An Empirical Study of Minimal Responses in Counseling Dialogues and the Behavior of LLMs

The paper investigates the role of minimal responses—short, empathic utterances—in psychological counseling, noting that such brief replies are common in human dialogues but underrepresented in large language model (LLM) outputs. Using a two‑stage filtering approach and contextual verification with an LLM, the authors systematically analyze minimal responses across multiple counseling datasets. They find that while strong commercial LLMs can produce minimal replies when prompted, they often fail to judge when these replies are appropriate, and counseling‑specific models trained on synthetic data tend to generate longer, content‑rich responses instead.

By Zhiyang Qi
arXiv AI
Sep 2

VoiceLongMemEval: Do Assistants Remember How You Sounded?

VoiceLongMemEval (VLME) is a new benchmark that tests AI assistants on their ability to remember how users sounded by incorporating paralinguistic metadata—such as emotion labels, prosody descriptors, and voice events—into each conversational turn. The benchmark uses a three‑stage adversarial gate to ensure that models cannot succeed with transcript alone, revealing a significant affect gap: models gain 0.09 to 0.38 accuracy when provided with paralinguistic cues, and audio‑native models outperform standard ASR pipelines in extracting these signals. The dataset and code will be released upon acceptance.

By Ramit Pahwa, Parivesh Priye, Apoorva Beedu
arXiv AI
Aug 28

When Text Misleads: Inconsistent-Aware Reasoning for Audio-Grounded Dialogue

The paper introduces ContraTalk, a benchmark that tests whether dialogue models truly use acoustic cues or rely on transcript shortcuts. It formalizes cross‑modal disagreement, creates conflict and consistent QA examples, and proposes an Audio Twin representation to expose acoustic evidence to models. Experiments show that while text‑only LLMs perform well on consistent cases, they falter on conflict cases, and AudioLLMs only partially mitigate this issue.

By Yen-Ju Lu, Yuzhe Wang, Yaohan Guan, Xiluo He, Jiarui Hai, Mingrui Liang, Kaavya Chaparala, Thomas Thebaud, Laureano Moro-Velazquez, Najim Dehak, Jesus Villalba