# Noah Smith > ~1980– · Computer Scientist, NLP Researcher > > **Recorded contribution:** NLP research; AI2; structured prediction; computational social science ## How to use this dossier Read for a causal chain, not a hero story: inherited problem → contribution → mechanism → downstream capability → limit. Then close the page and complete the reconstruction exercise from memory. ## 1. Historical orientation Noah A. Smith is a computer scientist known for statistical natural-language processing, structured prediction, computational social science, and careful analysis of NLP methodology. He taught at Carnegie Mellon and became a professor at the University of Washington and senior director at the Allen Institute for AI. His work spans parsing, language variation, political text, evaluation, and scientific communication. ## 2. The problem inherited Language is structured and socially situated, yet many statistical models treated predictions independently or evaluated them on narrow datasets that obscured population and measurement assumptions. ## 3. The central contribution Smith advanced probabilistic and structured approaches to language and connected NLP methods with social-science questions and rigorous evaluation practice. ## 4. Reconstruct the mechanism 1. Represent an output as a connected structure such as a tag sequence or parse tree. 2. Define features or learned scores over local and global parts of that structure. 3. Use dynamic programming or approximate inference to find a high-scoring valid output. 4. Evaluate both predictive error and whether the dataset operationalizes the claimed linguistic or social construct. ## 5. What changed downstream - Structured prediction became a durable bridge between linguistic constraints and machine learning. - Computational social-science work expanded NLP's objects of study while sharpening questions about validity and ethics. ## 6. Attribution, limits, and uncertainty - Smith's contributions are distributed across many collaborations rather than reducible to one registry slogan. - Text-derived social conclusions can confuse platform behavior with populations, prediction with explanation, and linguistic proxies with latent human attributes. ## 7. Reconstruction lab Build a tiny part-of-speech tagger with a transition score and emission score. Decode by dynamic programming, then test it on a dialectal or domain-shifted sentence and document the measurement failure. Compare token accuracy with whole-sentence accuracy and inspect whether frequent tags dominate the metric. Add a social variable to a text-classification task and test whether it is a genuine explanatory factor or a proxy for topic and sampling. Report confidence intervals and an error slice selected before viewing results. This joins Smith’s technical and interdisciplinary lessons: structured models make dependencies explicit, while rigorous claims require measurement choices that survive domain, population, and researcher degrees of freedom. ## 8. Evidence trail - [Noah A. Smith](https://nasmith.github.io/) — University of Washington - [Linguistic Structure Prediction](https://www.cs.cmu.edu/~nasmith/LSP/) — Morgan & Claypool / author copy - [Noah Smith publications](https://aclanthology.org/people/n/noah-a-smith/) — ACL Anthology --- *Research checked 2026-08-09. Dates, roles, and claims about living people are historical snapshots. Linked sources remain the authority; this dossier is original instructional synthesis.*