Shay Cohen

Chancellor's Fellow (≈ Assistant Professor)

Institute for Language, Cognition and Computation
School of Informatics
University of Edinburgh


In my research, I aim to discover computational methods for reasoning about natural language and linguistic structure.

Prospective students who might be interested in working with me: please see the note here. Also, see the new data science Ph.D. programme.

About me

My broad interests are in the intersection of computational linguistics and statistical learning. I am interested in developing ways to computationally reason about structured objects. Such structured objects are prevalent in natural language processing, for example, in syntactic analysis.

Much of my research has relied on capturing the syntax of natural language using probabilistic grammars -- grammars which originate in linguistics and formal language theory, and which have been augmented with a probabilistic interpretation. My most recent work treats the problem of predicting syntactic structures from text using spectral methods (in the supervised or unsupervised settings) and other linear algebraic methods. Most recently, we applied these ideas to machine translation.

Here is a project page.

Click here for a bio.

A new course, Topics in NLP, will be taught next semester (Spring 2015). See information below under Teaching.

I am writing a book titled "Bayesian analysis in NLP" for Morgan and Claypool. The current table of contents is here. If you see something that captures your eye, please send me an email.


Publications
Teaching
  • Spring 2015: Topics in Natural Language Processing (INFR11113). Course website. Course on PATH. Click here for a synopsis.
  • A tutorial about Spectral learning algorithms for NLP (NAACL, 2013). Similar tutorial with overlapping material at CMU (June, 2014).
  • Seminar at Columbia - Bayesian analysis for NLP (Spring, 2013).
  • A course at IBM about Probability and Structure in NLP (May, 2011)
Students and Post-docs
  • Shashi Narayan (post-doc, starting 2014)
  • Chiraag Lala (M.Sc., 2014)
Code
  • dageem - code for unsupervised grammar induction using logistic normal prior. Link on github.com. Download zip. New version (1.01) is out on August 19, 2014.
Contact information

scohen [strudel] inf.ed.ac.uk

10 Crichton Street
Informatics Forum 4.26
Edinburgh EH8 9AB
United Kingdom

Phone: +44 (0) 131 650 6542