Advising and Admissions

Current NYU Students

If you're working on a PhD and are interested in collaborating or seeking an advisor, contact me.

I only rarely agree to supervise undergraduate or master's students, and I only recruit students who have already taken some coursework in computational linguistics or NLP and done extremely well. If that's you, send me an email six weeks before the start of the term with a CV, a transcript, and a couple of sentences about your plans and research interests.

If your goal is to get started with independent research in NLP, you should take NLU (DS-GA 1012) or MLLU (LING-UA 52) as soon as you can and before you reach out to me for one-on-one research supervision. I designed both courses to walk you through your first major NLP research project, and the prerequisites and the non-project-related assignments in these courses are relatively light.

Prospective Graduate Students

Temporary warning: I am unlikely to have capacity to take on a new student in Fall 2021, except as a co-advisee with another faculty member. I generally recruit students interested in topics related to crowdsourcing and evaluation in NLP, and the use of large neural network models in linguistics. In addition, Linguistics is severely limiting PhD admissions for Fall 2021. Talk to me before applying to the Linguistics PhD.

I can advise graduate students in the Department of Linguistics (PhD), the Center for Data Science (MS, PhD), and the Department of Computer Science at Courant (MS, PhD). If you have questions about my group that you'd like to discuss before you apply, contact me. (Writing to just say hello doesn't help your application, though.) As above, if you're applying for an MS, I won't commit to supervising you as a research student until after you've taken coursework in NLP at NYU.

In the interest of fairness (and my sanity), I don't hold interviews or admissions-related meetings with prospective students until after we have received and reviewed everyone's applications. I never hold interviews with MS or undergraduate applicants.

Admissions for positions in NLP and computational linguistics are are very competitive. There are no hard rules and I am interested in students with unusual backgrounds. Under almost all circumstances, though, the applicants who I advocate to admit will (i) have already published work in my subfield and (ii) have at least two detailed recommendations from researchers who regularly publish in my subfield.

At the PhD level, the Linguistics program offers students a full five-year fellowship, while funding for CS and Data Science students, though guaranteed, often comes through grants for research on specific areas, which flow through advisors. This leads to somewhat different expectations for admission. Linguistics will admit students without a close fit to an advisor, so it's important that the applicant already be quite independent and have a good fit to the department overall. In CS and Data Science, fit to the department is less important, but it's crucial for applicants to name specific potential advisors and to demonstrate (i.e., through reference letters and published/publishable written work) that they're ready to work on problems that those advisors are likely to be interested in (and able to write grants for). Admission rates for all three programs are similar, so you should apply to whichever best fits your record and your interests, though if you're undecided between CS and Data Science, go for Data Science. For students broadly interested in cognitive science, this page offers some useful information about the available programs at NYU.

Under NYU Arts and Science rules, you can't apply to more than one of these programs in the same year.

Prospective Postdocs and Visitors

I don't have any specific postdoc openings ready now, but that can change quickly. If you'd be interested in working with me as a postdoc, get in touch.

I'm able to host visiting students and faculty, but only those who have published work on a research topic that's narrowly of interest to the lab.


Variational Autoencoders

My paper on this topic with Luke Vilnis was done during a Google internship, and we were not able to take any code or data with us at the end of the internship. If you need help applying VAE language models in new areas, my coauthor Luke has some notes that we are allowed to share, but your best bet is to look at any of the many good papers on the topic that have come out since ours.