Hannaneh Hajishirzi is a professor in the Paul G. Allen School of Computer Science & Engineering at the University of Washington and a senior director of research at the Allen Institute for Artificial Intelligence (AI2). Her research focuses on natural language processing, large language models, and machine reasoning, with contributions to question answering, semantic parsing, and efficient model architectures.
Hajishirzi earned her PhD in computer science from the University of Illinois at Urbana-Champaign in 2010, where she worked on algorithms for graph-based problems and natural language understanding. She then completed postdoctoral research at the University of Washington before joining the faculty there in 2014. At AI2, she leads the Mosaic team, which develops open-source language models and tools for scientific reasoning.
Early Career and Education
Hajishirzi received her bachelor's degree in computer engineering from Sharif University of Technology in Tehran, Iran, in 2003. She moved to the United States for graduate studies, earning a master's degree from the University of Illinois in 2006 and a PhD in 2010. Her doctoral thesis addressed combinatorial optimization in natural language processing, applying neural network methods to semantic role labeling and coreference resolution.
After her PhD, she joined the University of Washington as a postdoctoral fellow, collaborating with researchers on machine learning approaches for text understanding. In 2014, she became an assistant professor, later receiving tenure and promotion to associate professor in 2020, and full professor in 2024.
Research Contributions
Hajishirzi's work spans several areas of NLP. She co-developed the Transformer-based architecture known as the Longformer, which extends attention mechanisms to handle long documents efficiently. This model, introduced in 2020, enables processing of texts up to thousands of tokens, addressing a key limitation of standard transformers.
She also contributed to the development of the UnifiedQA model, a question-answering system trained on multiple datasets to generalize across formats. Her team at AI2 created the Tulu series of open-source large language models, which are designed for instruction tuning and reasoning tasks. These models have been widely used in academic research and industry applications.
Hajishirzi has published over 100 papers in top conferences such as ACL, EMNLP, and NeurIPS. Her research on semantic parsing and knowledge graph reasoning has influenced how AI systems represent and query structured knowledge.
Awards and Recognition
In 2021, Hajishirzi received a National Science Foundation CAREER Award for her work on robust and interpretable NLP systems. She was named a Fellow of the Association for Computational Linguistics in 2023, recognizing her contributions to the field. Her paper on Longformer won the Best Paper Award at the 2020 Workshop on Representation Learning for NLP.
She has also been recognized for her mentoring, receiving the University of Washington's Distinguished Teaching Award in 2019. Her students have gone on to positions at major tech companies and research labs.
Current Work and Impact
At AI2, Hajishirzi leads the Mosaic initiative, which focuses on building efficient and transparent language models. The team released the OLMo (Open Language Model) series in 2024, providing fully open-source models with training data and code. This effort aims to democratize access to large language model technology, contrasting with proprietary systems from companies like OpenAI and Google DeepMind.
Hajishirzi's research also addresses AI safety and evaluation. She has developed benchmarks for measuring reasoning capabilities and factual accuracy in models, contributing to the broader understanding of generative AI limitations. Her work on retrieval-augmented generation has improved how models incorporate external knowledge.
She serves on the editorial boards of major NLP journals and has organized several workshops on efficient NLP. Her collaborations span academia and industry, including partnerships with Amazon Web Services and Intel on hardware-aware model design.
Selected Publications
- Beltagy, I., Peters, M. E., & Hajishirzi, H. (2020). Longformer: The Long-Document Transformer. arXiv preprint.
- Khashabi, D., et al. (2020). UnifiedQA: Crossing Format Boundaries With a Single QA System. Findings of EMNLP.
- Groeneveld, D., et al. (2024). OLMo: Accelerating the Science of Language Models. arXiv preprint.
Her citation count exceeds 20,000, reflecting the broad impact of her work on modern NLP research and applications.