Marina Danilevsky

Career

Danilevsky's IBM Research profile records a PhD in computer science from the University of Illinois at Urbana-Champaign in 2014, with research in data mining supervised by Jiawei Han. Danilevsky is now a senior research scientist and manager for core language and conversational technologies at IBM Research, working on generative language models, evaluation and human-in-the-loop approaches. The profile also lists patents on topics including cross-lingual information extraction, table recognition in PDF documents, explainable rules learned with deep learning, and relevance modeling and clarification in conversational systems.¹

Research

In 2020 Danilevsky was first author of "A Survey of the State of Explainable AI for Natural Language Processing", with co-authors including Ranit Aharonov and Yannis Katsis. Danilevsky also co-authored Label Sleuth (2022), an open-source, no-code system for labeling text and building text classifiers.⁵⁴

More recent work centers on evaluating retrieval-augmented generation (RAG). Danilevsky is the last author of MTRAG (2025), a human-generated benchmark of multi-turn RAG conversations published with its data in an Apache-2.0 repository, and co-authored the 2026 follow-ups MTRAG-UN and the SemEval-2026 shared task MTRAGEval. Related papers include InspectorRAGet, an introspection platform for RAG evaluation (NAACL 2025), and a CHI 2025 paper on the difficulty of creating conversational datasets for RAG applications.²³¹

Work and contributions

• MTRAG (Multi-Turn RAG Benchmark) Open data

  Last author of "MTRAG: A Multi-Turn Conversational Benchmark for Evaluating Retrieval-Augmented Generation Systems" (2025) and co-author of MTRAG-UN (2026); the IBM/mt-rag-benchmark repository, under the Apache-2.0 license, provides the human and synthetic conversation data and document corpora.²³

• Label Sleuth Published research

  Listed among the authors of "Label Sleuth: From Unlabeled Text to a Classifier in a Few Hours" (2022), which introduces a free, open-source, no-code system for labeling text and creating text classifiers.⁴

Timeline

1. 2014Completed a PhD in computer science at the University of Illinois at Urbana-Champaign on data mining.¹
2. Aug 6, 2018Listed on a granted US patent for a cross-lingual information extraction program.¹
3. Oct 2020Posted "A Survey of the State of Explainable AI for Natural Language Processing" as first author.⁵
4. Aug 2022Co-authored Label Sleuth, a no-code system for building text classifiers.⁴
5. Jan 2025Posted MTRAG, a multi-turn conversational benchmark for RAG systems, as last author.²
6. 2026Co-authored MTRAG-UN and the SemEval-2026 MTRAGEval task, both at ACL 2026.¹

Where they work

• IBM 29 people in the wiki

Related people

• Ruchir PuriAlso at IBM
• Avi SilAlso at IBM
• Sriram RaghavanAlso at IBM
• Ranit AharonovAlso at IBM
• Kate SouleAlso at IBM
• Rohan TaoriLanguage models · Evaluation and benchmarks
• Swabha SwayamdiptaEvaluation and benchmarks · Datasets and data
• Adina WilliamsEvaluation and benchmarks · Language models
• Kyle LoDatasets and data · Language models
• Jesse DodgeEvaluation and benchmarks · Datasets and data