Pardis Sadat Zahraei

NeurIPS 2026 · Oral Presentation

AgentAbstain: Do LLM Agents Know When Not to Act?

Xun Liu^*, Yi Evie Zhang^*, Vira Kasprova, Parisa Rabbani, Pardis Sadat Zahraei^†, Tianyu Zhang^†, Ali Ebrahimpour-Boroojeny^†, Varun Chandrasekaran^†

*Equal contribution (co-first author)  ·  †Equal contribution (co-second author)

We introduce AgentAbstain, the first benchmark testing whether tool-using LLM agents know when not to act, built from 263 paired tasks across 42 sandbox environments. Across 17 frontier models, even the best agent reaches only 59.5% paired accuracy, and abstention ability is largely independent of general task-solving skill.

EMNLP 2026 Main Conference · Oral Presentation

The Alignment Veto: How Safety Training Suppresses Cultural Knowledge in LLMs

Pardis Sadat Zahraei, Gokhan Tur, Dilek Hakkani-Tür, Ehsaneddin Asgari

Across 16 MENA countries, 26 models, and 1.53M human survey responses, we show that safety training suppresses rather than erases cultural knowledge: at the moment of refusal, a model's internal logits often match human survey data better than its stated answer. This suppression is unequal across countries, and sparse autoencoder analysis traces it to a specific DPO-stage feature.

🏆 Outstanding Paper · COLM 2026 & ACL 2026 KnowFM Workshop

Emergent Unfaithfulness: How Alignment Training Causes Language Models to Silently Override Task Faithfulness

Pardis Sadat Zahraei, Janvijay Singh, Gokhan Tur, Dilek Hakkani-Tür

We show that alignment training can paradoxically cause models to silently override instructions. We find that this behavior emerges as models become more capable—a form of reverse scaling law, where the better the model, the less faithful it becomes.

Under Review · ICLR 2027

Asymmetrically Unfair Decision Making and Reasoning in LLMs

Pardis Sadat Zahraei, Faeze Brahman, Dilek Hakkani-Tür, Gokhan Tur, Maarten Sap

We introduce FairStress, a benchmark of 300 hiring-style decisions across 29 domains that tests LLMs under varying pressure to favor one candidate over another. We find anti-stereotypical overcorrection: models hold accuracy steady when identity is explicit, but a single added sentence makes them abandon correct answers favoring the majority far more often than the minority.

Under Review · NAACL 2027

Right by Chance, Wrong with Confidence: What Flat Accuracy Hides in LLM Post-Training

Pardis Sadat Zahraei, Emre Can Acikgoz, Dilek Hakkani-Tür, Gokhan Tur

We show that flat accuracy across post-training stages can hide large shifts in whether models are truly correct or just confidently wrong. Tracking this calibration structure across SFT, DPO, and final training reveals changes that accuracy numbers alone cannot show.

Under Review · NAACL 2027

EiCAP: Benchmarking and Enhancing Emotional Intelligence in LLMs through Psychologically Grounded Multi-Turn Dialogue

Nizi Nazar, Pardis Sadat Zahraei, Dilek Hakkani-Tür, Natasa Milic-Frayling, Ehsaneddin Asgari

A psychologically grounded benchmark for evaluating emotional intelligence in LLMs through multi-turn dialogue — probing how models recognize, reason about, and respond to emotional cues across extended interactions.

Findings · ACL 2026

Prior Beliefs Prejudice LLM-as-Judge: Evidence from Persuasion Evaluation

Pardis Sadat Zahraei, Xiaoning Wang, Beyza Bozdag, Gokhan Tur, Dilek Hakkani-Tür

LLM-as-Judge frameworks are systematically biased by prior beliefs — using persuasion evaluation as a testbed, we demonstrate that what a model already "believes" shapes its assessment of argument quality, raising serious concerns for scalable oversight pipelines.

Findings · ACL 2025 · Lightning Talk @ GeBNLP

Translate With Care: Addressing Gender Bias, Neutrality, and Reasoning in LLM Translations

Pardis Sadat Zahraei, Ali Emami

The TWC dataset (3,950 scenarios) tests translation between gendered and genderless languages. All tested models default to masculine pronouns in professional contexts; fine-tuning on TWC significantly reduces these biases, outperforming proprietary LLMs.

Findings · EMNLP 2024

TuringQ: Benchmarking AI Comprehension in Theory of Computation

Pardis Sadat Zahraei, Ehsaneddin Asgari

The first benchmark for LLM reasoning in theoretical computer science. Fine-tuning Llama3-8B on TuringQ improves both theoretical reasoning and performance on related tasks like algebra.

EACL 2024 · Oral Presentation

WSC+: Enhancing The Winograd Schema Challenge Using Tree-of-Experts

Pardis Sadat Zahraei, Ali Emami

LLMs are good at answering Winograd Schema questions — but can they write them? We introduce Tree-of-Experts (ToE), a prompting method that pushes valid WSC generation from 10% to 50%, and use it to build WSC+ — 3,026 new questions including novel ambiguous and offensive categories. GPT-4 tops the leaderboard at 68.7% — still far behind humans at 95.1%. And surprisingly, models aren't better at evaluating questions they wrote themselves.

Preprint

Detecting Bias and Enhancing Diagnostic Accuracy in Large Language Models for Healthcare

Pardis Sadat Zahraei, Zahra Shakeri

BiasMD (6,007 Q&A pairs) and DiseaseMatcher (32,000 clinical Q&As, 700 diseases) enable EthiClinician — a fine-tuned model surpassing GPT-4 in ethical reasoning and clinical judgment.

Preprint · Survey

Generative AI for Character Animation: A Comprehensive Survey

Mohammad Mahdi Abootorabi, Omid Ghahroodi, Pardis Sadat Zahraei, et al.

A comprehensive survey of generative AI applied to character animation — facial animation, motion synthesis, datasets, trends, open challenges, and future directions.