Noah Smith is a professor in the Paul G. Allen School of Computer Science & Engineering at the University of Washington, where he leads research in natural language processing (NLP). He is recognized for contributions to machine learning and deep learning, particularly in the development and evaluation of large language models. Before joining the faculty, he worked as a senior researcher at the Allen Institute for AI (AI2), where he contributed to projects on semantic parsing and model interpretability.
Smith earned his PhD in computer science from the University of Maryland in 2006, under the supervision of Philip Resnik. His doctoral thesis, "Adaptation of Statistical Machine Translation Models," laid groundwork for later work in domain adaptation and structured prediction. He subsequently held a postdoctoral position at the University of Toronto before becoming an assistant professor at Carnegie Mellon University in 2007.
Research Contributions
Smith's early work focused on statistical machine translation and syntactic parsing. He co-authored the paper "Discriminative Training and Maximum Entropy Models for Statistical Machine Translation" (2006), which introduced a discriminative framework that improved translation quality. In 2011, he published "Structured Sparsity and the Generalization of the LASSO" with his student, which advanced the theory of sparsity in high-dimensional learning.
A significant portion of his career has been dedicated to neural network methods. His 2016 paper "A Convolutional Neural Network for Modelling Sentences" (with Yejin Choi and others) demonstrated how CNNs could capture compositional semantics. He also worked on transformer architectures, contributing to the 2018 study "Deep Contextualized Word Representations" (with Matthew Peters and colleagues), which introduced ELMo, a model that influenced later large language models like BERT.
Work at AI2 and the University of Washington
From 2014 to 2019, Smith held a joint appointment at AI2, where he led the Semantic Scholar team. Under his guidance, Semantic Scholar became a widely used academic search engine that employs NLP to extract structured data from scientific papers. He also contributed to AI2's Aristo project, which aimed to answer science exam questions, and to the AllenNLP library, an open-source NLP toolkit that has been adopted in research and education.
At the University of Washington, Smith has supervised numerous PhD students who have gone on to prominent positions in academia and industry. He has also been involved in the development of the OpenAI-affiliated research community, though he maintains an independent academic stance. His teaching includes graduate courses on NLP and deep learning, and he has served on program committees for major conferences such as ACL, EMNLP, and NeurIPS.
Awards and Recognition
Smith received a National Science Foundation CAREER Award in 2010 for his research on structured prediction. In 2017, he was named an ACL Fellow for his contributions to statistical NLP and machine translation. He has also been recognized with best paper awards at ACL 2013 and EMNLP 2015, and his work on ELMo was part of the NAACL 2018 best paper. In 2020, he was elected to the board of the Association for Computational Linguistics.
Recent Directions
In recent years, Smith has focused on the societal implications of generative AI. He has published analyses on the reliability and bias of large language models, including a 2023 paper in the journal Nature Machine Intelligence titled "Challenges in Evaluating Large Language Models." He has also advocated for open research practices, releasing code and datasets for his projects. As of 2024, he continues to teach and lead the UW NLP group, which collaborates with industry labs such as Google DeepMind and Anthropic on interpretability research.
Selected Publications
- Smith, N. A. (2006). Adaptation of Statistical Machine Translation Models. PhD thesis, University of Maryland.
- Peters, M., Neumann, M., Iyyer, M., Gardner, M., Clark, C., Lee, K., & Smith, N. A. (2018). Deep Contextualized Word Representations. NAACL-HLT.
- Smith, N. A. (2011). Structured Sparsity and the Generalization of the LASSO. AISTATS.
- Smith, N. A., & Eisner, J. (2006). Discriminative Training and Maximum Entropy Models for Statistical Machine Translation. ACL.