Malihe Alikhani
I am an Assistant Professor at Northeastern University’s Khoury College of Computer Sciences. I also serve as a resident Visiting Fellow at the Brookings Institution focusing on AI policy, the Ethics Chair of the Association for Computational Linguistics, and a member of the Northeastern Ethics Institute.
I develop safe and fair AI systems that enhance communication, decision making, and knowledge sharing across disciplines and populations. My work bridges academia, applied AI research, and public policy to ensure that AI serves society responsibly and effectively. I have served as a science and technology advisor in Congress and collaborated with the United Nations in Africa on initiatives using AI to support anti violence efforts and promote equitable access to education and health.
Building on this vision, as the lead of the Contextual AI Lab, where we focus on building embodied systems that operate meaningfully within context across language, embodiment, and society to support safer, fairer, and more productive interactions between humans and machines. By integrating insights from cognitive and social sciences with machine learning, our models capture the richness of human interpretation and support collaborative meaning construction.
Through this work, I aim to shape a future that expands the human capabilities for learning, health, creativity, and wellbeing, a future where intelligent systems foster these as rights.
email: m.alikhani AT northeastern.edu
Selected Recent Publications
Accounting for Sycophancy in Language Model Uncertainty Estimation, Anthony Sicilia, Mert Inan, Malihe Alikhani, the Association for Computational Linguistics: NAACL 2025.
Evaluating Theory of (an uncertain) Mind: Predicting the Uncertain Beliefs of Others in Conversation Forecasting, Anthony Sicilia and Malihe Alikhani, 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics.
Deal, or no deal (or who knows)? Forecasting Uncertainty in Conversations using Large Language Models, Anthony Sicilia, Hyunwoo Kim, Khyathi Raghavi Chandu, Malihe Alikhani, Jack Hessel, The 62nd Annual Meeting of the Association for Computational Linguistics, 2024.
Studying and Mitigating Biases in Sign Language Understanding Models, Katherine Atwell, Danielle Bragg, Malihe Alikhani, The 2024 Conference on Empirical Methods in Natural Language Processing.
Including Signed Languages in Natural Language Processing, Kayo Yin, Amit Moryossef, Julie Hochgesang, Yoav Goldberg, and Malihe Alikhani, The Joint Conference of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, 2021. Best Theme Paper Award
HumBEL: A Human-in-the-Loop Approach for Evaluating Demographic Factors of Language Models in Human-Machine Conversations, Anthony Sicilia, Malihe Alikhani, The 18th Conference of the European Chapter of the Association for Computational Linguistics, 2024.
PAC-Bayesian Domain Adaptation Bounds for Multiclass Learners, Anthony Sicilia, Katherine Atwell, Malihe Alikhani, Seong Jae Hwang, The Conference on Uncertainty in Artificial Intelligence (UAI), 2022. Best Paper Award
AI & Education
This research track develops computational frameworks for inclusive, cognitively-grounded educational AI across three areas:
-
Cognitively-Grounded Human-AI Interaction, introducing positive friction mechanisms that improve task success and trust calibration through strategic delays and user theory-of-mind modeling (TACL'25)
-
Code as Multimodal Communication, treating programming as a spatiotemporal modality requiring cross-modal coherence between natural language instructions, code structures, and visualizations, with which we can study how learners navigate ambiguity in data visualization tasks through interactive clarification dialogues (EMNLP'25)
-
Inclusive Educational Technologies, advancing native ASL integration in LLMs (SignAlignLM, ACL'25), prosodic intensification and cross-modal coherence metrics (EMNLP'21, ACL’22), AR-based collaborative STEM learning through the BRIDGE Project with Gallaudet University, and deployed multimodal instruction-giving systems (Alexa TaskBot 3rd place, NAACL’24, COLING ‘24)
AI Alignment & Atunement:
Whose goals and safety matter?
In this research track we study the extent to which:
-
AI outputs align with human judgments for presuppositions in text, across different trigger types, context windows, and demographics (ACL ‘25).
-
Alignment and decision-making in AI systems is impacted by sycophancy, or “yes-man” behavior (ArXiv)
-
Lack of theory of mind leads to sycophantic generations, how can we prevent them with sycophancy-robust uncertainty prediction separating learner confidence from correctness (NAACL ‘25, ACL’25)
AI Policy
In this research track we study the extent to which:
-
AI outputs align with human judgments for presuppositions in text, across different trigger types, context windows, and demographics (ACL ‘25).
-
Alignment and decision-making in AI systems is impacted by sycophancy, or “yes-man” behavior (ArXiv)
-
Lack of theory of mind leads to sycophantic generations, how can we prevent them with sycophancy-robust uncertainty prediction separating learner confidence from correctness (NAACL ‘25, ACL’25)
AI-powered discovery:
Code, communication, and the evolution of scientific work
In this research track we study the extent to which:
-
AI outputs align with human judgments for presuppositions in text, across different trigger types, context windows, and demographics (ACL ‘25).
-
Alignment and decision-making in AI systems is impacted by sycophancy, or “yes-man” behavior (ArXiv)
-
Lack of theory of mind leads to sycophantic generations, how can we prevent them with sycophancy-robust uncertainty prediction separating learner confidence from correctness (NAACL ‘25, ACL’25)
AI for Peace
In this track, we study how AI systems can be deployed for harm reduction in real-world scenarios. We also study the harms that exist in current AI systems, with a particular focus on:
-
Automatically-generated conflict detection dashboard using large language models, which scrapes local news articles and monitors election-related events, in collaboration with United Nations Development Programme
-
Building more ethical language technologies by reducing bias in sign language recognition systems (EMNLP ‘24) and surveying deaf and hard of hearing users on sign language technologies (upcoming)
Moderating social media across multiple platforms and languages (EACL ‘23) and utilizing style transfer to reduce offensiveness in social media conversations (COLING ‘22)