Slav Petrov

Academic and Professional Background

Petrov's educational journey spans three countries and reflects the international nature of his career. He completed a Diplom in Computer Science and Business Administration at Freie Universität Berlin before pursuing doctoral studies in Computer Science at the University of California, Berkeley. He later studied at Duke University as part of his earlier academic path. From 2010 to 2016, he served as an Adjunct Professor at New York University, where he taught Statistical Natural Language Processing — a role that ran concurrently with his industry work at Google, where he joined in August 2009.

Role at Google DeepMind

At Google DeepMind, Petrov holds the position of Vice President of Research and co-leads Post-Training for Gemini, the company's flagship large language model. Post-training is a critical phase in the development of large language models, involving techniques that refine a model's behavior, instruction-following capabilities, and alignment after initial pretraining. His team is globally distributed, reflecting the scale and complexity of the work involved. Prior to the formation of Google DeepMind, Petrov led NLP research efforts that fed directly into Google products such as Search, Ads, Translate, Assistant, and Chrome.

His publication record spans over a decade and covers a wide range of topics within natural language processing, including machine translation, parsing, and attribution in natural language generation. He is a contributor to landmark papers associated with Google's large model efforts, including work on PaLM, PaLM 2, Gemini, Gemini 1.5, and Gemma 2, as well as the Natural Questions benchmark dataset — a widely used resource for question answering research.

Research Contributions and Recognition

Petrov's research output includes contributions across syntactic parsing, machine translation, and the development of evaluation frameworks for language generation. One notable recent area of work involves measuring attribution in natural language generation models — examining whether model outputs can be traced back to verifiable sources, a concern that has grown in importance alongside the deployment of large generative systems. This work, published in Computational Linguistics in 2023, reflects a broader interest in making language model outputs more reliable and interpretable.

Beyond his paper awards at ACL and NAACL, Petrov received a 10-year Test-of-Time Award at ACL 2023, recognizing research from a decade earlier that proved enduring in its influence on the field. His research areas, as listed in Google's research directory, span Natural Language Processing, Machine Translation, Machine Intelligence, and General Science.

Background and Other Achievements

Petrov has lived in Bulgaria, Germany, and the United States in roughly equal measure, and currently is based in Berlin. His early competitive achievements include a World Championship at RoboCup 2004 — an international robotics competition — which predates his career in NLP research. He completed his Diplom at Freie Universität Berlin, and his time in Germany appears to have bookended his career, with Berlin serving as his current base even as he leads research for one of the world's most prominent AI organizations. He has expressed a personal loyalty to the Bulgarian national soccer team, particularly in matches against Germany.