NeurIPS 2020
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Combining neural retrieval with generation for knowledge-intensive NLP tasks.
#3 most influential paper at NeurIPS 2020 — Paper Digest (March 2026 list)
Read paper 
About
I’m Scott Wen-tau Yih, a Research Scientist at FAIR, Meta SuperIntelligence Labs in Seattle. I am also an Affiliate Professor at the University of Washington. My current research focuses on multi-agent systems and memory for automated research, with a particular interest in recursive self-improvement: systems that learn from their experiments and improve how they pursue the next question.
This work builds on my research in information retrieval, question answering, and factuality in language models. I was named an ACL Fellow in 2024.
Current research
Agents that build on each other’s work
How can a group of agents turn individual attempts into cumulative progress? I study systems that coordinate exploration, experimentation, and verification through shared, persistent records of their work.
Memory for learning and discovery
Memory is central to retaining useful knowledge, revisiting earlier findings, and choosing what to try next. My work spans memory within language models and memory shared across agents. I’m interested in how these capabilities can support research over many rounds of experimentation.
Automated research and recursive self-improvement
My longer-term aim is to develop research systems that can propose ideas, evaluate them against evidence, and use the results to improve their methods. Grounding those systems in scientific literature connects this direction to my earlier work on retrieval and factuality.
Selected papers
EMNLP 2020
Dense Passage Retrieval for Open-Domain Question Answering
Learning dense representations to retrieve passages for open-domain question answering.
#1 most influential paper at EMNLP 2020 — Paper Digest (March 2026 list)
Read paper NAACL-HLT 2013
Linguistic Regularities in Continuous Space Word Representations
Showing that word embeddings capture linguistic relationships through simple vector operations.
#1 most influential paper at NAACL-HLT 2013 — Paper Digest (March 2026 list)
Read paper ICLR 2015
Embedding Entities and Relations for Learning and Inference in Knowledge Bases
DistMult represents entities and relations with a compact bilinear model for knowledge-base completion.
5699 citations — Google Scholar (as of )
Read paper EMNLP 2023
FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
Evaluating long-form generations by checking individual factual claims against reliable sources.
#4 most influential paper at EMNLP 2023 — Paper Digest (March 2026 list)
Read paper NAACL 2024
REPLUG: Retrieval-Augmented Black-Box Language Models
Augmenting black-box language models with retrieved evidence without changing the language model itself.
#1 most influential paper at NAACL 2024 — Paper Digest (March 2026 list)
Read paper EMNLP 2015
WikiQA: A Challenge Dataset for Open-Domain Question Answering
A benchmark of real questions and Wikipedia answer sentences for open-domain question answering.
#6 most influential paper at EMNLP 2015 — Paper Digest (March 2026 list)
Read paper KDD 2026
Best Paper Award — Datasets and Benchmarks track
CRAG-MM: Multi-modal multi-turn comprehensive RAG benchmark
A benchmark for retrieval-augmented answers in multimodal, multi-turn conversations, with egocentric images and realistic retrieval challenges.
Read paperACL-IJCNLP 2015
Outstanding Paper
Semantic Parsing via Staged Query Graph Generation: Question Answering with Knowledge Base
STAGG builds query graphs in stages to answer natural-language questions using a knowledge base.
#8 most influential paper at ACL-IJCNLP 2015 — Paper Digest (March 2026 list)
Read paper CoNLL 2011
Best Paper
Learning Discriminative Projections for Text Similarity Measures
Learning compact text representations that improve semantic similarity for cross-lingual retrieval and advertising relevance.
Read paperAll publications · Google Scholar
Background & service
Before joining Meta, I was a Principal Research Scientist at the Allen Institute for Artificial Intelligence (2017–2019). I worked at Microsoft Research from 2005 to 2017, most recently as a Senior Researcher. I received my PhD from the University of Illinois at Urbana-Champaign in 2005.
My papers have received the KDD 2026 Best Paper Award in the Datasets and Benchmarks track, the CoNLL 2011 Best Paper Award, and an ACL 2015 Outstanding Paper Award. I have served as program co-chair for CEAS 2009, CoNLL 2014, and EMNLP 2021, and as a senior area chair for ACL, NAACL, EMNLP, EACL, ICLR, ICML, and NeurIPS.