Kuzman Ganchev

Timeline

1. 1999–2003

   A Swarthmore computer science degree

   Born in Sofia, Bulgaria, he leaves in 1989 for Zimbabwe and moves in 1995 to New Zealand, where he attends high school. He comes to the US in 1999 to study at Swarthmore College and graduates with a BA in computer science in 2003.

2. 2003–2010

   Doctoral work at Penn

   After graduating, he works at StreamSage Inc. in Washington, DC, a company that analyzes, searches and indexes audiovisual content, and in 2004 he starts at the University of Pennsylvania. He interns at TrialPay in 2007 and at Bank of America in 2008, spends 2008–2009 as a visiting collaborator at the Institute for Parallel Processing of Information of the Bulgarian Academy of Sciences, and graduates in 2010 with a PhD in computer and information science. His doctoral thesis covers machine learning with linear models for computational linguistics.

3. 2010

   The posterior regularization framework

   With João Graça, Jennifer Gillenwater and Ben Taskar, he publishes "Posterior Regularization for Structured Latent Variable Models" in the Journal of Machine Learning Research in 2010. The paper presents a probabilistic framework for structured, weakly supervised learning that places constraints on the posterior distributions of models with latent variables, and tests it on cross-lingual dependency grammar induction, unsupervised part-of-speech induction and bitext word alignment.

4. 2010–2011

   Google Research and the Atanasoff Award

   In 2010, he joins the New York office of Google Research as a research scientist, and a post by Slav Petrov on the Google Research blog congratulates him as runner-up for the John Atanasoff award. In 2011, he is a laureate of the John Atanasoff Award from the President of Bulgaria for his work in machine learning and natural language processing.

5. 2016

   Globally normalized transition-based networks

   He is one of eight authors, with Daniel Andor, Chris Alberti, David Weiss, Slav Petrov and Michael Collins among them, of "Globally Normalized Transition-Based Neural Networks" at ACL 2016. The feed-forward model reaches state-of-the-art part-of-speech tagging, dependency parsing and sentence compression results, and the paper is presented in the Outstanding Papers Session.

6. 2022

   Attributed question answering

   In 2022, he is among the authors of "Attributed Question Answering: Evaluation and Modeling for Attributed Large Language Models", with Bernd Bohnet, Dipanjan Das, Slav Petrov, Michael Collins and others. The paper formulates Attributed QA as a first step toward language models that attribute the text they generate, proposes a reproducible evaluation framework and benchmarks a broad set of architectures.

7. 2023

   An ACL Test-of-Time award

   The 2013 paper "Universal Dependency Annotation for Multilingual Parsing", which he co-authored with Ryan McDonald, Joakim Nivre, Slav Petrov and others for the short-paper track of ACL 2013 in Sofia, wins the ACL Test-of-Time Paper Award in 2023. The award recognizes papers for their long-lasting impact on natural language processing and computational linguistics, and the winners are announced at ACL 2023.

8. As of October 2026

   Research at Google in New York

   He has worked at Google in New York since 2010. In 2024, he co-authors DOLOMITES, a benchmark of 519 methodical long-form tasks elicited from experts in 25 fields, with Google DeepMind colleagues including Dipanjan Das, Mirella Lapata and Chris Alberti.