Hi, I'm Parisa Rabbani

I am a PhD candidate in Computer Science at the University of Illinois Urbana-Champaign, working at the Conversational AI Lab with Prof. Dilek Hakkani-Tür. My research focuses on AI safety and LLM alignment, especially in dialogue settings. My work draws on both NLP and human-computer interaction to find what we should be measuring, and what we should be aligning to. Previously, I completed my B.S. in Computer Science at the University of Tehran.

AI Safety LLM Alignment Conversational AI Human-AI Collaboration

Research

Publications

See Google Scholar for a full list.

AI Safety and Alignment

Under Review

DialDefer: A Framework for Detecting and Mitigating LLM Dialogic Deference

Parisa Rabbani, P. Sahoo, R. Mathew, A. Mondal, H. Ketharaman, N.B. Bozdag, D. Hakkani-Tür. 2026.

Introduced a framework for measuring dialogic deference: how LLM judges shift from verifying statements to deferring to speakers. Benchmarked 4 LLMs across 9 domains (3k+ instances) and 280 Reddit conversations, finding an average deference shift of +18pp (2-4x higher in naturalistic settings). Developed fine-tuning mitigations that achieved +22pp accuracy and -24pp deference.

Paper Code

IWSDS'26 Oral

From Fact to Judgment: Investigating the Impact of Task Framing on LLM Conviction in Dialogue Systems

Parisa Rabbani, N.B. Bozdag, D. Hakkani-Tür. 2025.

Investigated how task framing affects LLM conviction in dialogue systems. Showed that reframing evaluation from statement verification to speaker judgment significantly reduced model conviction, revealing high sensitivity to social framing across multiple LLMs.

Paper Code Slides

Human-AI Collaboration

Preprint

Toward Socially-Aware LLMs: A Survey of Multimodal Approaches to Human Behavior Understanding

Z. Liu, Parisa Rabbani, V. Duddu, K. Fan, M. Lee, Y. Huang. 2025.

Surveyed multimodal approaches (speech, vision, physiological signals) to human behavior understanding and examined their potential for building socially-aware language models.

Paper

AIED'25

AI-Enhanced Speech-Language Intervention Documentation: Opportunities and Design Goals

Q. Zheng, A. Choudhry, Z. Liu, Parisa Rabbani, et al. AIED 2025.

Identified design opportunities and goals for AI-assisted documentation in speech-language intervention through interviews and co-design sessions with clinical practitioners.

Paper

CSCW'24

SOAP.AI: Documenting Human Behavior in Videos through Multimodal Generative AI

Q. Zheng, Parisa Rabbani, Y-R. Lin, D. Mansour, Y. Huang. CSCW Companion '24.

Developed a human-AI collaborative system for documenting child behavior in therapy videos using GPT-4V and Whisper, in collaboration with the AI Institute for Exceptional Education. Designed with 17 speech-language pathologists through an expert-in-the-loop approach with automatic prompt engineering from clinician preferences.

Paper Demo

Coursework

Selected Projects

CS 598 · Spring 2025

LLMs and Lost Translations

Evaluated GPT-4o and Gemini on 180 CrowS-Pairs sentences across 4 languages using embedding-based bias metrics. Found bias shifts in up to 90% of gender-related translations (Russian) and 90% of religion-related translations (Korean).

Paper

CS 598 · Fall 2024

Deciphering the Digital Zeitgeist

Studied the effect of AI integration on platform trust by analyzing 68k Reddit comments using SBERT classification and Poisson regression. Found a 26% decrease in competence distrust post-Meta AI launch, with tech subreddits exhibiting higher competence concerns vs. non-tech communities focusing on sincerity.

Paper

CS 467 · Spring 2024

The Many Faces of Computer Science

Visualized CS faculty diversity using ML-generated facial composites from 3,900+ images across 48 U.S. universities. Found 67% masculine representation overall; top-ranked institutions predominantly reflect white male features, while minority-serving institutions mirror their student body demographics.

Paper

Talks