I was trained as a computer scientist, but my research group has diverse interests in natural language processing, and we work on many problems in this area, from theoretical to empirical. We tend to focus on basic curiosity-driven research in areas that aren’t yet well understood, in order to improve the foundations of the field. We view NLP as an interdisciplinary field, and to do interdisciplinary research, you need a discipline. So I look for collaborators from mathematics, machine learning, computer science, and linguistics. I don’t expect you to be an expert in all of these fields—few people are—but I do expect you to learn something about all of them from a diverse of group of collaborators.
- Prospective PhD students
- Prospective MSc, MScR, or MPhil students
- Current students in Edinburgh
- Prospective interns and visitors
- Prospective visitors
- Prospective Postdocs
Prospective PhD students
I will take at most one new student in September 2020, and it is very unlikely that I will take any in 2021. Beyond that I may take one or two students per year. Good advising takes time, and taking more students than this would make it difficult for me to spend sufficient time with the students I’m already working with.
Interest in natural language processing is skyrocketing. There are some good things about this, but a downside is that I’m no longer able to respond personally to every prospective student who is interested in working with me. In fact, there’s a good chance that you’re reading this because I referred you to this page. I don’t do this to discourage you—please apply! I wrote this advice page to help you: it summarizes the advice that I used to discuss with applicants individually, and I keep it up-to-date.
Naming me as a potential supervisor on your application
Many prospective students ask whether they can name me as a supervisor. Yes, please list me as a potential supervisor if the research topics below look interesting to you. When you name me as a potential supervisor, your application will be routed to me during the admissions process.
You can name multiple potential supervisors on your application, and generally this is fine, since many of us in the EdinburghNLP group have overlapping interests, and in practice we often forward applications to our colleagues when we think they will be of interest. (But don’t list everyone indiscriminately—focus on those whose interests are most closely aligned to yours.)
General advice on the research proposal
To apply, you must write a research proposal. The School of Informatics advises you to contact prospective supervisors prior to applying, to discuss the topic of your proposal. Unfortunately, I cannot advise you on your research proposal due to the sheer number of applicants interested in natural language processing. But if you name me as a potential supervisor, then your application will then be routed to me and I will read your proposal during the normal admissions cycle, which takes place early in the calendar year.
Some applicants stress over the research proposal because they worry that they need to commit to a four-year plan before even starting the PhD. Don’t worry about this: your research proposal is not a contract. I realize that your interests will change over the course of a PhD, and even between the time of the application and your first day of a PhD program.
At the same, take the research proposal seriously! The proposal is an exercise in persuasion and scientific communication. It should convince the admissions committee that you can find interesting scientific questions and plausible ways to answer them. Pick a question that you think is interesting and show that it is worthwhile, unanswered, and possibly answerable within a few years. To show that it’s unanswered, use evidence from the current scientific literature. To show that it might be answerable, sketch a plausible 2–3 page plan for answering it. (A longer proposal is not more impressive.) I understand that your interests and plans will change over the course of a PhD, so nothing you write is set in stone. It’s just the opening of a conversation about your interests. I also realize that you may not have written anything like this before, and that not everyone has the same access to opportunities. I am looking for something that shows you’ve done the best you can with the opportunities available to you.
I’m open to a wide variety of proposals in my areas of interest. To get an idea of what these are, I recommend that you do some reading to see how my group approaches research problems. My current research interests are best reflected in the papers from the last few years—if you find something exciting in that set, then we likely have some interests in common. But below are some ideas to get you started.
Natural language understanding systems confront a menagerie of complex and subtle phenomena. Deep learning is a useful tool in building these systems, but it’s only one tool among many, and we don’t fully understand its power or its flaws. My group aims to improve our understanding; here are some examples:
Deep learning researchers claim that their models learn to represent linguistic properties of language without any explicit guidance. But recent work in machine learning suggests that deep models are simply very good at memorizing local correlations. What do these models really learn about language? The goal of projects in this area is to first understand what, when, and how deep learning models learn about language; we’re interested in both mathematical perspectives on learning and generalization, and linguistic or cognitive perspectives on what phenomena are or aren’t acquired from text or speech. We also want to understand what cultural biases or latent demographic information these models learn, in order to understand the social implications of their rapidly increasing use.
For a system to understand a text and answer questions about it, the system must distill the meaning of the text into a set of facts (semantic parsing). We can represent these facts as a graph: entities and events become nodes, and relationships between them become edges. We now have datasets that pair text with such graphs, and we’d like to learn a semantic parser from this data, so we need to model graphs.How do we design and use deep probabilistic models of graphs? To really answer this question, we’ll need to combine insights from recent empirical work with a web of untested theoretical work, and we’d like to systematically explore the space.
I’m interested in many other things involving language, structure, and learning. To get a sense of what some of those things are, you can read some of my recent papers. I am happy to discuss specific questions or ideas related to my published work—if you write to tell me that you liked one of my papers, without any questions of your own, then the only response I can give you is to refer you back to this page. Again, I don’t do this to discourage you—it’s just the best I can do.
Across all topics, I am especially interested in multilingual approaches that work well across different languages. Meaning can be expressed in many very different ways.
If you’re interested in natural language, but not these specific topics, you may want to contact other members of EdinburghNLP or browse PhD topics they have proposed. If you aren’t genuinely interested in language, then my research group is not a good fit for you, and you should contact a different supervisor in the School of Informatics.
How to apply
You cannot apply by emailing your documents to me. You must use the university’s common application system, which you can find in the application guidance links below. I advise students in two different programs:
- PhD students in the Institute for Language, Cognition, and Computation. This is a three-year research-only program. Read the application guidance.
- PhD students in the Centre for Doctoral Training in Natural Language Processing. This is a new four-year program combining research and some formal taught elements. Read the application guidance
You can apply to multiple programmes concurrently. (But please only apply to programmes that you are genuinely interested in. Applicants who submit to many unrelated programmes across schools and colleges are routinely rejected because they don’t appear to be committed to any particular area of study, and a PhD definitely requires that commitment.) Read more about the application process here.
When to apply
I only review applications in winter admissions cycle. For a September 2020 start, the application deadline is November 29, 2019. The deadline is not hard, but applying by then ensures that the school will consider you for funding, which is crucial since we only make funded offers (unless you have a full outside scholarship; read about self-funding above).
If you have full funding from a fellowship that will pay for your tuition and a living stipend, that’s great! The purpose of your fellowship is to enable you to pursue your intellectual interests, and if we have interests in common, I’m happy to discuss them with you. Having a fellowship makes your admission easier because it isn’t contingent on funding, but it is still crucial that I have capacity to take a new student and that we have mutual research interests. If you have a fellowship, but I am at capacity, or you want me to advise you on a topic that is very far from my interests, I will decline—such an arrangement wouldn’t benefit either of us.
I do not take self-funded PhD students who pay out of their own pockets. This would be unethical on at least two counts: (1) it gives unequal access; and (2) the university benefits from the labor of PhD students through their research and publications, and PhD students should be paid for this.
Practically, this means that we only offer admission to students we can fund, or who are able to obtain a fellowship. I am acutely aware that this means that I can’t work with many good applicants, but I will not make exceptions.
Chances of admission
I cannot advise you on your chances of admission, which depend on many factors outside your control or even mine. PhD admissions is a very stochastic process, so I advise you to apply widely.
Prospective MSc, MScR, or MPhil students
Do not contact me if you are applying for one of these degrees. I have no influence over who is admitted or whether they will work with me.
Longer explanation: Taught MSc programmes in Edinburgh are very intensive and you will not have time on the side to join a research project. A third of the overall credits for the MSc come from an individual project, supervised by a member of staff, and you will need to devote all of your effort to your project. Projects are assigned early in semester 2, and though the assignment considers your preferences, there is no guarantee that you will get your preferred project due to the very large numbers of students. Because of all the uncertainties around this process, I can make no guarantee that you will be able to work with me if you join our taught MSc programme. As a practical matter, MSc admissions are handled centrally and I have no influence over them.
Our department does not admit students directly to MPhil or MScR degrees.
Current students in Edinburgh
If you’re interested in working with me and you are a current research student in a PhD or CDT, please get in touch!
If you are an MSc, MInf, or undergraduate honours student, the most likely way for us to work together is during your thesis project. You are welcome to contact me during the allocation phase for these projects. However, allocation is not decided by me, so I cannot make any guarantees.
Prospective interns and visitors
I do not have any openings for internships. Please do not send me an unsolicited application.
I will not host visiting scholars who I do not know personally.
If you are already visiting the university I am happy to talk about research.
I do not currently have any openings for postdocs. You are welcome to contact me if you expect to be looking for a postdoc in the future. It’s a good idea to do this half a year or even a year in advance, since it often takes time (and luck) to find funding. We may be able to work something out in less time, but there are few guarantees in research funding.