Shay Cohen

Shay Cohen is a Reader at the Institute for Language, Cognition and Computation within the School of Informatics at the University of Edinburgh. He teaches advanced courses in Natural Language Understanding, Generation, and Machine Translation, as well as Accelerated Natural Language Processing and Foundations of NLP. His research focuses on the intersection of NLP and machine learning, particularly in text generation, parsing, representation learning, and applications of large language models, neural networks, and probabilistic grammars.

His recent work includes advising students on projects like RNA structure prediction via dependency parsing (DEPfold, ICLR), tracking evolution in commercial MT systems (EMNLP 2023), mitigating hallucinations with concept erasure (BlackBoxNLP 2022, EMNLP 2023), and using PCFGs for neural architecture search (ACL 2023). He has also explored polarization trends in review language (LRE) and developed tools like the Rainbow Parser for latent-variable PCFGs.

Shay co-organizes outreach programs like the Informatics Circle for teaching computer science to children and has led workshops on LLM scaling behavior (EACL 2024), representation learning (ACL 2017), and vector space modeling (NAACL 2015). He collaborates with institutions such as IBM, Google, and Amazon and has advised postdocs now at Telecom Paris, Naver Labs, and CNRS.