Dear excellent student,
Thanks very much for your interest in applying to do a Ph.D. or M.S. degree with me.
I receive many such inquiries, and apologize that I can't reply thoughtfully to all of them without taking too much time away from my current students. However, I hope your questions are answered below.
Best wishes to you in your application. I look forward to reading it this spring.
Q: Are you accepting new graduate students for next fall?
A: Yes, I am. (But note that admissions offers come from the department's admissions committee, not from me personally.)
Each fall, several new students are accepted whose primary interest is in language and speech processing or machine learning. I tend to interact technically with most of the students in this research area, since we are a cohesive group here and we encourage them to try doing research with different faculty members. (Indeed, that is a requirement of the CS department.)
The CS faculty in language and speech processing like to formally co-advise the new CS students at first, so that they will feel free to work with any of us. Students do not have to choose their dissertation advisor immediately. Some students do arrive with a clear preference for a particular research style or topic, but others prefer some exploration.
I would personally love to have more advisees from groups that are underrepresented in computer science, including women. The department strongly welcomes such applications.
Q: Are you accepting new graduate students for the spring?
A: We rarely do so, but it is possible under special circumstances. You may wish to consult with our department's Graduate Program Coordinators, Kim Franklin (kimfranklin at jhu dot edu) for the Ph.D. degree or Revelie Niles (rniles3 at jhu dot edu) for the M.S.E. degree.
Q: May I list you as my advisor of interest?
A: Sure, it's your application—say what you like!
Listing faculty members of interest can be helpful during Ph.D. application review. However, it is not a fateful choice that determines who will advise your dissertation. Mentioning me by name may make me more likely to see your application and interview you. But it does not guarantee that I would be your initial advisor, and omitting my name doesn't guarantee that I would not be your initial advisor. The initial advisor assignment listed in your offer letter will be based on a match of interests and on interviews. Also, Ph.D. students are allowed to change advisors during the program (although many do not). A Ph.D. student's advisor is responsible for supervising their work and for finding funding for them, so taking on a Ph.D. student is a big commitment.
Master's students are admitted by a departmental committee. They do not have initial research advisors, since not all of them write a thesis. They do have an academic advisor who helps them choose courses. Master's students who want to get involved in research have enough time to find supervised projects after they arrive, particularly if they impress their professors in classes or have useful prior experience.
Q: How do I apply?
A: The department has a page with about how to apply with some useful answers and links. The actual application is online here. There's a $75 fee and a December 15 deadline to apply for the following fall.
Q: What should I talk about in my statement of purpose?
A: I need to figure out: Would you be a great collaborator for me on any of the topics that I'm considering working on next? So it helps to know: What kinds of problems excite you, what abilities/knowledge/experience do you already have, and do you have a track record of success? What would it actually be like to work with you?
Applicants usually describe their goals, interests, work style, preparation, and past projects. Some also outline ideas for future projects.
Prof. Nathan Schneider (my grandstudent) has written an excellent long answer to this question.
Q: What can I do to improve my preparation?
A: You should try to get a solid grounding in the following four areas:
- Machine learning (including fundamental background in probability, statistics, and optimization). My lab often works on structured prediction, which sometimes requires approximate inference. You'll also want experience in designing and implementing deep learning architectures, and experience with using large language models.
- Theory of computation (regular languages and finite-state automata, context-free languages and push-down automata, etc.).
- Linguistic structure (either linguistic or NLP approaches to syntax, phonology, morphology, and/or semantics).
- Software engineering (including efficient implementations, elegant object-oriented design, documentation, testing, debugging, visualization, comfort with finding/using/modifying libraries). You'll want to be comfortable building complicated programs and running experiments.
To do this on your own, you will need to read the right books and papers. You may also benefit from looking at university courses that post their slides and assignments online. Finally, there are now MOOCs that teach NLP and ML. (Figuring out what to read can be hard, so I should list some of the best resources here.)
Most important, find a way to do research! You need to start looking at the field as one of its researchers. It's a big playground and it's time to learn the ropes — read papers, try out the equipment, figure out how to play safely, get good at standard tricks, and make up your own tricks.
Most of the work in my lab is trying to turn the real world into math — that is, to construct tasteful mathematical models of complex phenomena. So you'll need the ability to consider how the formulas behave and how the phenomena behave, and to discern the root cause of any discrepancies, so that you can truly fix the formulas while keeping them elegant. For your model to be of any use, you'll need to design efficient algorithms for estimating its parameters ("training") and drawing conclusions from the model ("inference"). Finally, you should develop a broad perspective on the kinds of problems that exist in the field, so that you can choose your scope of work: a good new technique is one that lets you solve at least one important problem, and hopefully can be applied to a whole swath of related problems.
Q: Is funding available? How does it work?
A: Yes! Ph.D. students in our department generally receive an offer of full funding (tuition, stipend, and health insurance) at the time that they are accepted. Typically, an NLP student would serve as a teaching assistant for 2–4 semesters (during his or her first two years), and receive funding to work on research grants the rest of the time. We try to ensure that students have the freedom to pursue research ideas of interest to them, within the broad scope of a grant.
Many of my students have brought their own funding in the form of outside fellowships (NSF, NDSEG, Hertz, Fulbright, HLTCOE, Bloomberg, Facebook, ...). We are grateful for such outside support, and we will usually augment your fellowship stipend a bit to reward you for securing such a fellowship. But more important is that having your own funding increases your research freedom.
If you are a master's student, see here.
Q: Can you read the attached material and evaluate my chances of admission?
A: I do welcome your application. However, reading hundreds of application folders in the spring already takes a tremendous amount of time. I can't read everyone's application in the fall as well!
Even if I could do so, my evaluation would not be very accurate without your letters of recommendation, or the ability to compare different applications and discuss them with my colleagues.
Thus, please just send your application materials through the usual channels. I look forward to reading your complete folder during admissions season in January and February.
As noted below, your application is allowed to include extra materials such as copies of publications or other work that you are proud to have written. This can be useful.
If you have a specific concern about your application, you are of course welcome to email me about it. The next answer may also be helpful to you.
Q: How will you or your department make the admissions decisions?
A: For the master's degree, the department's admissions committee reviews all applications and chooses the strongest applicants. I am usually not consulted. For more information, please contact Revelie Niles (rniles3 at jhu dot edu) in our department office.
Admissions for the Ph.D. degree are more complicated. If your application indicates a primary interest in Natural Language Processing, then your folder will be reviewed by a subcommittee consisting of Professors Dredze, Duh, Eisner (that's me!), Field, Khashabi, Khudanpur, Koehn, Lippincott, Van Durme, and Yarowsky — that is, the CS faculty who are actively involved with the Center for Language and Speech Processing. (See below for other relevant faculty.) Several senior graduate students also serve on this committee.
In this Ph.D. admissions subcommittee, we read applications carefully, pass them around (e.g., "You might be especially interested in this student"), and discuss them in detail. We also keep electronic notes on each applicant. Each application is initially read by at least 2 humans who are assigned to it. The highly-rated applicants are then discussed further. We interview the most interesting applicants by Zoom or in person.
After interviews, the faculty on the subcommittee hold a long and detailed meeting to discuss options and reach a consensus. In the end, a student may be admitted because one advisor would love to work with them, or because several advisors are potentially interested. We try to get a sense of who is likely to work with whom, and how they'd be funded if so. But this is just a feasibility check. Once they arrive, the admitted students are free to approach any advisor in the department; indeed, the department requires every Ph.D. student to try out advising relationships by doing research projects with two different professors. (See the sections "Important People" and "Qualifying Projects" here.)
What's the schedule? The NLP faculty work together in January to review Ph.D. applications. We do some Zoom interviews in January and early February. We then invite some students to campus for a two-day group visit in February or early March. This visit is a combination of further interviewing and recruiting, so that we can all figure out which advisors and which students are a good match. (Unfortunately, due to cost, in-person visits are normally limited to students who are already located in North America.) Most offers go out shortly after that, normally with an acceptance deadline of April 15 (there is an agreement among North American CS departments not to set earlier deadlines than that). Formally, the departmental admissions committee must approve any offers of admission.
Admissions offers may not all be made at the same time. For example, you might get a late offer if earlier offers are declined, or if new funding comes in, or simply because the advisor was slow to make a decision. If you have been interviewed, then feel free to continue talking to NLP faculty (and grad students) about whether you're likely to get an offer and what it would be like to come here.
Choosing a new Ph.D. student to work with for the next 5 years (and beyond) is a major commitment. These are the most important decisions we make as faculty, and we have to get them right. When studying your folder, we are primarily interested in your potential to do independent research:
Past research. This is the best evidence of research potential, so please let us know in detail about any original research that you may have done already. Emphasize what was new and important about the problem and what creative or unusual steps you took to solve it. You may wish to include copies of published or unpublished papers.
Intellectual qualities. We also look for other evidence that you are skilled, creative, and persistent at solving problems. We take letters of recommendation very seriously — and if we are considering you seriously, we will probably contact your recommenders to discuss your intellectual qualities. Thus, the most useful recommendations come from professors or researchers who have discussed ideas with you and know how your mind works.
(So choose as recommenders the people who have worked most closely with you, rather than famous or important people who will only be able to say vague things about you. Your different recommenders might know different things about you, so that the set of 3 letters will illustrate your various projects and skills. If you have trouble choosing among recommenders, it is okay for your application to have more than 3 letters.)
Of course, we also consider your grades, since most strong researchers are also able to do well in classes. (However, doing well in classes does not prove that you will have the creativity and initiative to find new problems and new solutions.) If your grades are mixed, please tell us why.
We don't like to rely too much on the GRE, because it is just an artificial one-day exam. In fact, the GRE is optional and many students don't submit it. Very high GRE scores are most useful if your recommendations and grades come from a lower-ranked institution: your high GRE will reassure us that you will shine as brightly here as you did there. Surprisingly low GRE scores on an otherwise strong application may just be a fluke, so they do not disqualify you, but they will make us check your application for other signs of weakness. Most of our applicants do not take the GRE subject test unless they want to establish that they know CS despite having a non-CS major.
Match of interests. One of the most important factors in a Ph.D. is whether the student and advisor get excited about the same problems and solutions. Students who work well with me are mathematical sophisticates who love to design algorithms, and who love language and have been fascinated for a long time by the nuances of human language and the capacities of human intelligence.
Relevant academic background. We sometimes do take exceptional students whose interest in NLP exceeds their background in it. However, we are very interested to learn about your past coursework, class projects, or original research in natural language processing, machine learning (including data mining, probability, or statistics), linguistics, or search/optimization. A good background in any of these areas will help you start doing research here immediately, and will give you a useful perspective as you take classes in the other areas.
Technical skills. While recognizing that different people have different strengths, I look for evidence of certain skills that are relevant to research in my lab:
- Programming ability, part 1 — strength at building complicated systems and otherwise making software work well.
- Programming ability, part 2 — strength at designing new algorithms or data structures.
- Mathematical ability — strength at formalizing ideas, proving theorems, and reading mathematically dense papers. This may be indicated by strong grades (or advanced coursework) in pure or applied math or theoretical CS.
- Modeling ability — creativity and judgment in devising formal approaches that capture qualitative aspects of real-world phenomena.
- Linguistic ability and interest — a sensitivity to the nuances of sentences or words (their internal structure, meaning, and sound or written expression). This may be indicated by coursework in linguistics, a serious interest in writing, knowledge of multiple languages, etc.
- Writing, speaking, and teaching ability — Basic skills that you will need to succeed as a researcher.
Quality of technical discussion. If I'm your advisor, we'll be having lots of intense technical discussion over several years. Many of your research ideas, as well as mine, will be born in such discussions. Furthermore, I'll probably ask you to write them up afterwards in an email.
You will also spend a lot of time throwing ideas around with the other grad students. So it is important that you are articulate (in English), energetic, and interesting to talk to.
Therefore, before recommending you to the admissions committee, the CLSP faculty will want to spend several hours talking with you, either in person or (for foreign students) over the phone. We want to see that you will pick up new ideas, draw connections to things you already know, ask good questions, and reply with ideas of your own.
Q: What is a "center"? How do the centers and departments at JHU work together?
A: JHU has a long tradition of interdisciplinary collaboration. This wonderful collegial culture is what attracted me to JHU. Department boundaries are not very important. The faculty know one another well, and work together across departmental lines on research, grants, curriculum, admissions, hiring, advising, and examining students. It is common for students to take courses and do research with faculty in other departments. Furthermore, a recent remarkable $350 million gift from Michael Bloomberg is funding the creation of 50 interdisciplinary senior professors, each with appointments in two schools of the university.
In general, JHU's interdisciplinary strategy has been to hire related people in multiple departments, who can work together across department boundaries. This has allowed us to build large, world-class interdisciplinary groups in selected specialties, without overwhelming any individual department.
A center is an administrative mechanism that helps these people to work together. It is a group of faculty and grad students who have similar interests even though they are from multiple departments.
The Center for Language and Speech Processing (CLSP) was established in 1992. The professors and students associated with CLSP all have offices close together in the same building, and the center has its own administrative staff and compute cluster. We run a weekly speaker series, weekly student talks, an annual 6-week research workshop and summer school, a full set of courses, several reading groups, social events, etc.
As a grad student, you'll certainly interact with plenty of people outside of CLSP. In particular, you will get a full education in your department. We don't want you to be narrow. But CLSP will be your intellectual home base. Most people in the Computer Science department have "dual citizenship"—they are full citizens of both the department and some center. Some of the other centers deal with areas related to NLP, such as vision, robotics, genomics, and big data. They have many interesting talks and activities of their own, which you are welcome to attend.
The Human Language Technology Center of Excellence (HLTCOE) is closely associated with CLSP. You can think of it as an extension of CLSP. It was created in 2007 with funding from the U.S. government, to focus on a wide range of problems of national importance in human language technology. It employs or funds many researchers and publishes many papers. Officially, the HLTCOE is a separate center in a different building. However, the government chose JHU for this center because CLSP is here, so there is a great deal of overlap in the people and research. Most of the CLSP faculty have some involvement with HLTCOE. In fact, several of the CLSP faculty have HLTCOE as their primary affiliation, but they are cross-appointed as faculty in the CS or ECE departments, and they teach courses and advise Ph.D. students.
Q: Which department should I apply to? Does it matter?
A: Yes, it matters because you have to get your degree through one department. Each department sets its own degree requirements.
If you are in the CS department, you will be required to take CS classes and write a dissertation with substantial computational content. You are also more likely to end up with a job in a CS department after graduating. It is okay if your undergraduate degree (like mine) was in something else, but you should know enough CS to do well in this department.
If you are an electrical engineer or linguist without much programming experience, then the ECE department or Cognitive Science department may be a more natural home. You will still get to talk to me, since there is much interaction among all the faculty and students in the Center for Language and Speech Processing.
If you are interested in speech or signal processing, then Profs. Arora, Khudanpur, Dehak, Hermansky, Andreou, or Elhilali would be a more appropriate advisor than I would. Although most of them are in the ECE department, you are free to apply to either ECE or CS as you prefer, since they also have secondary appointments in CS and can advise students in either department. See below for their webpages.
Q: What kinds of problems do you work on? What is your approach?
A: That's no secret: you can find plenty of information on my home page (including the links "what/why," "research summary," and "publications"). As you already know, I tend to work on algorithmic and modeling problems at the intersection of computational linguistics and machine learning. (Also, the design of our declarative programming language and its runtime system.) But I have broad interests and can get excited about problems in almost any area. Technical correspondence is welcome. I'm driven by the desire to "really" or deeply understand whatever I'm working on, which often means trying to identify key obstacles and formalize elegant general solutions. Good papers by other people make me very happy, and rather than fight them to be the first to fill obvious gaps, I prefer to maximize my value to the field by identifying important problems where I can make a contribution that others won't.
Q: What are you like as an advisor?
A: My students are my closest collaborators, and my goal is for them to be equal collaborators as soon as possible, both in finding problems and in coming up with solutions. My personal style is therefore pretty informal and centers on technical (and social) discussion, as explained here. I do ask that the work should move along quickly and be of high interest and high quality. My students learn from one another as well, and have even gotten together to write papers or teach courses without me. I take my students seriously, like them personally, meet with them regularly and frequently, and try to find interesting problems to work on with them. In general I try to give them the attention, technical help, and career advice that they need. I also believe that people should behave decently toward one another.
Q: Who are other relevant faculty at Johns Hopkins?
A: Quite a number of faculty are part of the interdepartmental Center for Language and Speech Processing. Click on their webpages below to read about their interests. We work together in many ways, including with one another's students, and most of us are in interdisciplinary office space in Hackerman Hall. You should also check out our impressive Machine Learning Group at JHU.
- Computer Science department:
- Electrical and Computer Engineering department:
- Sanjeev Khudanpur (has joint appointment in CS)
- Najim Dehak
- Mounya Elhilali
- Andreas Andreou
- Jesús Villalba
- Laureano Moro-Velazquez
- Cognitive Science department (which includes linguistics; I have a joint appointment):
- Other potentially relevant faculty:
- Human Language Technology Center of Excellence (works closely with CLSP)
- Machine Learning Group (and other centers linked from there)
- Justin Halberda (Psychological and Brain Sciences)
- Steven Gross (Philosophy)
- Justin Bledin (Philosophy)
Q: What is JHU's ranking?
A: This is the wrong question. The standard advice is that you should choose your route to a Ph.D. by choosing the best advisor for you, and then you try to go work with that advisor, wherever he or she happens to be.
(Realize that top professors may end up at a variety of good universities, not necessarily the #1-ranked university. Where they end up is affected by many factors: departmental specializations, geographic preferences, family considerations, and where faculty positions happened to be open in the year they were applying for jobs.)
More generally, your goal is to emerge from the Ph.D. as a recognized leader in your subfield. It is helpful if the university has not just your advisor but other strong faculty in the same general area. This means that you will be part of a larger community of faculty and fellow students who are interested in the same things as you. (1) This exposes you to a lot more ideas, as well as more feedback and guidance on your own ideas. (2) You will stay in touch with these people throughout your career. (3) Larger communities can accomplish things as a group — attract large shared grants, shape a good curriculum, invite prominent speakers, etc. (4) Having multiple relevant faculty around gives you a backup plan if things don't work out with your original choice of advisor.
By contrast, the overall department ranking is not especially relevant. See "Academic Rankings Considered Harmful." In particular, note that departmental rankings are highly correlated with size. Smaller departments (like JHU CS) can be made up mostly of well-known and productive faculty, so they score well on per capita measures. The trouble is that most rankings are based largely on surveys of reputation, which reward large departments — a smaller department has to focus on a smaller number of research areas, so most survey respondents will not work in any of these areas and so will not be familiar with the department. You can read more about rankings here and here.
But to answer your question directly:
- For language and speech processing research, JHU is a major player and is generally considered one of the top few places in the world. We also have a strong machine learning group, and our Cognitive Science department has the #1-ranked graduate program in linguistics (as of 2010).
- As a university, JHU is ranked #13th in the world by U.S. News and World Report's Best Global Universities (as of 2025) and #16th in the world by The Times Higher Education World University Rankings (as of 2025). Several of its schools and specialties are ranked #1 in the U.S. Every year for the past 35 years, it has been the #1 recipient of federal research funding. JHU was the nation's first research university and pioneered the Ph.D. model of graduate education in the U.S.
- The undergraduates are also excellent and a pleasure to teach. JHU's undergraduate program is ranked #6 by U.S. News (as of 2025) and #5 by U.S. high school guidance counselors (as of 2018). About 2/3 of the undergraduates get involved in research during their degree.
- JHU's engineering school is growing extraordinarily rapidly in AI, with plans to hire 80 new tenure-track faculty in AI and data science over a 5-year period (about 40 of these will be in CS), as well as 30 new Bloomberg Distinguished Professors who do interdisciplinary AI. This will make JHU CS one of the largest CS departments at a private U.S. university and is likely to increase our ranking. Currently the CS department as a whole is ranked about #21 in AI and #24 overall, as We have historically been small. That said, our measures of research productivity are unusually high (number of Ph.D. students per professor, number of publications per professor).
Obviously, NLP is one of the department's great strengths, and draws extremely strong applicants. Experienced U.S. students who are applying to grad school in NLP will usually apply to JHU, and if we admit them, they usually accept.
Q: Should I do a Ph.D.? How does grad school work?
A: Here's a good overview for you — definitely worth reading. There is plenty of other similar advice online. Here are my thoughts on whether to do a Ph.D. immediately.
Q: What is the relationship between the master's degree and the Ph.D. degree?
A: The master's degree here consists of 8 courses plus a research project (or just 10 courses, if you prefer). The Ph.D. begins with the same 8 courses plus 2 research projects, and continues with an oral examination and a dissertation.
The requirements therefore make it straightforward to switch from the master's program to the Ph.D. program. However, this would require a separate application. A few master's students in CS have managed to transfer into the Ph.D. program; they were doing research as well as the best Ph.D. students of the same year.
You should decide before applying which degree you are interested in. (If you would prefer a Ph.D., but would consider our master's program as a second choice, then please say this explicitly in your application, or follow this advice.)
If you already have a master's degree and come here for a Ph.D., you may be able to count some of your previous coursework toward our requirements. Detailed information about our graduate requirements is here.
See also the next question.
Q: I have been admitted into the MSE program; can I do research with you?
A: There are often research opportunities for strong master's students. As noted above, one way to fulfill the degree requirements involves a master's thesis or master's project, typically taking a year or more. For example, you would start the project in your second semester, move into high gear over your first summer, and finish by the end of your second summer. It is also possible to fulfill some of your coursework requirements with the special course "Independent Research."
A professor must agree to supervise the research. The project idea might come from either you or the professor, with appropriate discussion. It might be a new topic (which requires especially strong skills), or an existing project where you would work with Ph.D. students.
The usual path is that you would take some courses in your first semester or two, and then approach one of your professors about doing research. Some classes require term projects, which is a good way to get started on exploring a research idea.
Sometimes MSE students apply to transfer to the Ph.D. program. This is occasionally successful. The MSE requirements are essentially the same as the first 2 years of the Ph.D. requirements, so such a student is able to continue as if he or she had been in the Ph.D. program all along.
Step 1. Enroll in the MSE program and do very well in your classes (comparably to the Ph.D. students who work on NLP). Then one of the professors may be willing to supervise you for a master's thesis.
Step 2. Do great research, working with a professor. If your supervised research shows that you are again as smart and productive as the Ph.D. students, then your advisor may encourage you to apply to the Ph.D. program.
Step 3. In autumn of your second year, apply to transfer to the Ph.D. You now have 2 serious advantages over other applicants: (1) there is less guesswork in reading your application since you are a known quantity, and (2) you would hit the ground running as a Ph.D. student, since you will already have taken relevant courses and established a working relationship with our faculty. It is still possible that you won't be admitted, since there are hundreds of other applicants, some of whom will be temptingly smart and well-qualified. (See above for what we're looking for.) So you would be wise to apply to other Ph.D. programs at the same time. If you did well at steps 1 and 2, then you should at least get into some Ph.D. program: whether or not you continue at JHU, your JHU background will serve you well and will be taken seriously by other schools.
Q: Is funding available for MSE students?
Not usually. The MSE is a comparatively short degree (≤ 2 years) that is typically followed by a well-paid job.
Occasionally, an advisor who has extra funds may be able to offer a research assistantship to a qualified master's student. This will typically be a student who is already involved in research, rather than a new student.
The department may also sometimes hire master's students as course assistants. This can be interesting experience. However, it is compensated with a normal hourly wage, which is far from enough to cover the full expense of the degree.
(In contrast, Ph.D. students can get teaching assistantships that provide full funding, just like research assistantships. These are intended to subsidize them over the course of a much longer degree. Indeed, Ph.D. students are required to do some teaching as part of their training.)
Q: Dear Professor: My interest is in radiotopic barbavision and I think you would be the perfect advisor for me.
A: Please do not email all of the professors in the United States. Spam wastes everyone's time.
Q: I have unusual circumstances, or have a comment on your research, or my question is not answered above. May I email you?
A: Certainly. Occasionally I am slow about reading or answering mail from new people, but I will eventually reply. You may first wish to consult this advice from another professor.
If you have an adminstrative question about our admissions procedures or your application, please instead contact our department's Graduate Program Coordinator, Kim Franklin (kimfranklin at jhu dot edu). For the master's degree, contact Revelie Niles (rniles3 at jhu dot edu).
http://cs.jhu.edu/~jason/advice/prospective-students.html
Jason Eisner - jason@cs.jhu.edu (suggestions welcome) | Last Mod $Date: 2024/10/29 15:34:04 $ |