Sharon Goldwater: Want to work with me?

Interest in NLP has exploded recently. This is exciting but it means that I get a very large number of requests for collaboration and supervision, and I simply no longer have time to answer all of them individually. If you are interested in working with me, here is what you need to know.

Prospective interns, visitors, MSc students

Current UG and MSc students

Current UG students

Sorry, but I'm not supervising any honours/MInf projects in 2019-20.

Incoming/current MSc students

Please do not contact me about research or project supervision until at least halfway through your first semester, when you have a more realistic picture of your time commitments. Our programme is very intensive, and many MSc students overestimate how much time they will have available. In the meantime, if you want to get a better idea of what NLP research is going on in the School, you are welcome to attend the weekly Edinburgh NLP group meeting.

Due to the large number of MSc students currently interested in NLP projects, I have recently been focusing mainly on group supervision: taking a cluster of students on closely related projects and holding group supervision meetings. Therefore, I am very unlikely to take on a student with a self-proposed project, unless it happens to align with my planned group. If you are considering a self-proposed project that is well-defined and clearly within my main research interests (please take a look at my papers to see), you can get in touch with me about it sometime after the first half of Semeter 1 (see above). Otherwise, please look elsewhere.

Please note, however, that the allocation of MSc students to supervisors is heavily constrained and not under my control. You will be assigned to one of your top few preferences, but there is no guarantee you will get your top choice.

Prospective PhD students

If I have referred you to this page, it's not to discourage you: please do apply! Unfortunately I just don't have the bandwidth to reply individually to prospective students or to comment on your research proposal, so I have summarized the advice I found myself repeating frequently.

If you're reading this before contacting me and you plan to apply, it is probably still a good idea to send me an email with a very brief (1 paragraph) outline of your research interests and a CV. In a few cases I may let you know that it's not a good fit, saving you the time of completing a full application. If not, I will try to acknowledge that I have received your email (but can't promise to do so) and to note down your name so that I can keep an eye out for your full application.

Deadlines, programmes, and what to expect of a UK PhD

I'm no longer taking students for 2019, but I am likely to take a student for 2020 entry. There are normally multiple deadlines, but you should definitely aim for the December deadline if you're serious, because we start evaluating applications at that point, and both funding and supervisor availability can easily be gone by the time later deadlines roll around.

I can potentially take students on either of two PhD programmes:

Both of these programmes are shorter, and have less coursework, than (say) a PhD programme in North America, and both expect students to get started working on their thesis research from day one (the main difference being that this will be part-time in the first 1-2 years of the CDT programme, while you are also doing some coursework that supports or complements your research).

In practice, what that means is that most accepted students already have a Master's degree or equivalent as well as some research experience and a pretty good idea of what they plan to focus on. This is especially true of the three-year programme. There can be exceptions, but if you're considering applying to work with me, please read the following section about what I look for in prospective PhD students.

You can apply to one or both programmes, but please only apply to a programme if you are really interested in it. There can be different funding sources available for each programme, so if you think you might be interested in either one, it is a good idea to apply to both.

What I look for in an application

Evidence that you have good programming skills and enough mathematical/machine learning background to do the kind of research you are interested in.

This mainly applies to linguists and cognitive scientists. I have lots of students from those backgrounds, but you should have at least some experience with computational methods and either machine learning or statistical/empirical approaches. The extent of mathematical background needed will depend on what you are hoping to do. Also, if you are hoping to do cognitive modeling, you should have some previous experience either with cognitive modeling or machine learning.

Evidence that you are interested in language, or at least in a scientific approach to building speech and language systems.

This mainly applies to computer scientists and engineers. I am often contacted by students with these backgrounds who seem to be mainly interested in machine learning or particular engineering applications that involve text or speech. If you are mainly motivated by engineering (e.g., doing what it takes to make the accuracy number go up), you are probably not a good fit for my group. I am far more interested in understanding the principles that underlie both human language and computational language processing systems, and I really am excited about language itself. So I would like to see evidence that you share at least some of these interests, and I am unlikely to take on CS/engineering students who don't already have some experience with NLP.

Evidence of research experience: include your CV!

At minimum, you should have completed an undergraduate or master's thesis or research internship. Realistically, most students I take on have more experience than that (e.g., have been involved in more than one research project or have published papers). The application form may not explicitly ask for one, but please upload a 1-2 page CV that highlights your research experience and any publications you have (including submissions under review). Not all students have the opportunity for an experience that would lead to a published paper, but I'm looking for evidence that you have made the most of the opportunities you've had.

Evidence that your research interests align with mine.

In addition to reading the advice above, I hope you've already looked at some of my publications to get a better idea of the kinds of things I'm interested in and the methods I use. If you think your interests are a good match to mine, the best way to convince me of that is via your research proposal. See below for advice on writing it.

Evidence of good communication skills.

Doing research involves communicating about research. Your CV and research proposal are examples of your communication skills. Use them wisely.

Writing your research proposal

It may help to keep in mind that your research proposal serves several purposes as far as the admissions committe is concerned:

It indicates the general direction of your research interests.

This helps identify potential supervisors and whether you would be a good fit. However, your proposal is not a binding contract! Your direction might change as you start working on the research, or your supervisor might have an idea based on what you write that's a bit different but might work out better for both parties. So, please don't stress too much about the area.

Occasionally students have two rather different ideas for potential topics. Please try to pick one to focus your proposal on (to satisfy the other goals, described below) but it's okay to mention briefly in a paragraph at the beginning or end the other topic you are interested in, in case I or another supervisor might be more interested in that one.

It shows that you can formulate an interesting scientific question and a plausible approach to tackling it.

Quoting from Adam Lopez: "Pick a question that you think is interesting and show that it is worthwhile, unanswered, and possibly answerable within a few years." You can use evidence from the current scientific literature to show that it's unanswered and to motivate your approach to answering it. To show that it might be answerable, try to sketch a plan for answering it, including (as appropriate) the models, datasets, and evaluation methods.

Aim for no more than 3-4 pages altogether for the proposal. (You might have 1-2 additional paragraphs if you also describe your previous experience; see below). Longer is not better; we have a lot of proposals to read through and being able to express yourself clearly and concisely is important.

It provides evidence regarding the other points listed under 'what I'm looking for' (research experience, communication skills, etc.)

In particular, if you have previous research experience, it is probably worth writing 1-2 paragraphs describing the main goals and findings of that research, and your role in it. In some cases, your previous research may lead directly into your research proposal. But even if it doesn't, I like to see what you have been involved in before and how well you're able to explain it. Please clearly highlight this as a separate section (distinct from the main research proposal) and keep it short.