Michael Collins
Natural language processing researcher with a Google Research profile; former MIT faculty member
Michael Collins works on natural language processing and machine learning; Google Research's profile describes a focus on statistical parsing, structured prediction, and applications such as machine translation, dialog systems, and speech recognition, and notes that Collins is a fellow of the Association for Computational Linguistics. Collins's Columbia homepage, which lists teaching through Spring 2020, says Collins was a researcher at AT&T Labs-Research from 1999 to 2002 and an assistant and then associate professor at MIT from 2003 to 2010, joined Columbia University in January 2011, and is also a research scientist at Google. Collins is listed with Google DeepMind on InfAlign, an ICML 2025 paper on inference-aware language model alignment, and is a co-author of Measuring Attribution in Natural Language Generation Models, which presents the Attributable to Identified Sources (AIS) evaluation framework. Collins also publishes lecture notes on statistical NLP topics on the Columbia homepage.1243
Work
Measuring Attribution in Natural Language Generation Models (AIS) (external site: arxiv.org)Published research
Co-author of the paper that defines the Attributable to Identified Sources (AIS) framework and a two-stage annotation pipeline for judging whether generated text is supported by underlying sources; the Google Research profile lists the work among Collins's publications.32
Statistical NLP lecture notes (external site: cs.columbia.edu)Published research
Publishes notes on language models, hidden Markov models, PCFGs, IBM translation models, log-linear models, and the EM algorithm, and taught an online natural language processing course on Coursera in Spring 2013.1
Other documented roles
Roles the sources document but that are not confirmed as current.
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.