Yonatan Bisk (he/him) יונתן ביסק (הוא)

Assistant Professor @ CMU     GHC 6703
Small robot clawCLAW Lab Website      includes Publications
My research is in Natural Language Processing (NLP) with a focus on grounding and embodiment. In particular, my interests are broadly: Modeling the semantics of the physical world, and Connecting language to perception and control.
twitter logo

Yonatan Bisk (white male wearing a black t-shirt in front of a window)
ybisk  ¯\_(ツ)_/¯  cs.cmu.edu

Hi! Here are a few common reasons people come to this website
Yonatan Bisk is an assistant professor of computer science in Carnegie Mellon's Language Technologies Institute. His group works on grounded and embodied natural language processing, placing perception and interaction as central to how language is learned and understood. Previously, he received his PhD from the University of Illinois at Urbana-Champaign working on unsupervised Bayesian models of syntax, before spending time at USC's ISI (working on grounding), the University of Washington (for commonsense research), and Microsoft Research (for vision+language).


Photo URL:
Applying to LTI vs RI vs MLD: The short answer is you should apply to the department whose curriculum aligns most closely with your educational and research goals. The department does not restrict who can serve as your advisor, but it does change when we get access to the applicant pool a little to add comments and whether my name is in the default drop down for visit days. You can select me during visit days regardless.
The following is a quick note about (cold) emailing professors. You have undoubtedly seen that many professors have notes on their websites that simply say “don’t email me.” I’d like to encourage you to selectively ignore that advice but understand why, when and how. (updated 12/29/2023)

Before we get too far in. Are you ...
a current or previous collaborator? Email me
a member of a historically marginalized group in academia? Email me
simply telling me you applied? Thanks, low probability of a reply
mass mailing every professor? Please no

Note: Gmail sometimes classifies student emails (particularly the fill-in-the-blank form emails) as spam and so I try to check once a month or so but I might just never see them if I've been too busy this month

Information about emailing professions (including some discussion on when and how)

Success in academia, like many aspects of society, is often a function of who you know and networking. Students of famous professors don’t need to email because they know their applications will get read carefully. Further, they may have had the opportunity to meet potential advisors at conferences and so they are already well connected when applying. If you don’t fit this description, as many of us don’t,

  • How are you supposed to break into the “in-group”?
  • How is someone from a smaller lab, a different discipline, or an underrepresented group supposed to change the field?
They have to email. They have to take the chance of getting no reply to at least put themselves out there. If this is you, please email me and everyone else.
Professors get lots of emails, that’s why you’re reading this. Many of them are generic and have nothing to do with their specific research. Let’s say a professor works 40hrs a week and works with 10 students/collaborators.
  • 40 - 10hrs of collaboration meetings = 30hrs.
  • 3hrs teaching, 6hrs preparing, & one office hour = 20hrs left.
  • Thesis committees, graduate admissions, hiring, rec letters, … = 10hrs.
  • Grant writing and staying current on the literature… = 0hrs.
All of these things actually take much longer than what I’ve listed and are worse around paper and grant deadlines, but we still haven’t budgeted any time to reply to emails from our collaborators or anybody else. This means, a professor likely is reading your email while eating lunch, on the bus on the way home, or at night when they are trying to figure out what they forgot to do today. So, a useful exercise might be to ask yourself:
  1. How much time does it take to reply to the email I’m sending?
  2. Is this email actually a DOS attack? [i.e. did it take me much less time to write than it will for them to reply?]

The following are a few broad classes of common emails we receive:
❌Generic “please accept me into your program” email -- DOS attack
  • We will read your application eventually anyway, I promise, I will.
  • We have to look through our budgets, look at existing obligations, etc
  • We have to read through your previous research papers to find a fit
This email is unlikely to get a reply

👍Tailored “please accept me - we do similar work” email -- Potentially balanced
  • We still can’t promise anything, we still have to investigate funding, you still need to apply and we still will go through your application and recommendation letters in detail.
  • You brought up a specific intersection in our research and yours. This requires substantially less homework for us to say something helpful and even if we don’t end up working together now, hopefully, we’ll cross paths and learn from each other in the future.
If there is a specific question, a reply is more likely

✋Very specific technical question
  • Love that you’re in deep in the same area as me, but these often are better answered by one of the other authors. Use your best judgment here.
No, maybe, it's complicated. I think we need to talk a little about how admissions works (YMMV at other institutions).

Every school has an admissions committees that does the initial ranking and sorting of candidates. At LTI/RI this is done by faculty, at UIUC and UWashington, this is done ( at least in part ) by current graduate students. I will look at every application that mentions my name in the statement of purpose (in round 3 -- see below). For some faculty that might just be a few dozen, for others, it may be a large chunk of the applicant pool. I will also try and read every application that mentions keywords that are relevant to my research -- further increasing my load. This process involves reading your statements, your transcript/resume/CV, your letters of recommendation, and possibly skimming a paper you've written. This takes a long time and honestly, this means that we might not get to spend as much time on your application as you were hoping we would. Here's where your email helps or hurts:

If I get an email that meets the above criteria -- I give the application a closer look. That's it.

It's not nothing, but it might not be what you were hoping for. I do not have the ability to unilaterally admit you to the PhD/MLT program. If we are a really good match I can try and put my thumb on the scale, but once admitted, you're under no obligation to work with me (this is a good thing!). But that means that I can't be the only person who is willing to work with you either. This creates a preference for applications that are "broadly" strong and that several faculty agree would be successful if admitted.
When? After applications have been submitted. Because, again, the main thing you're accomplishing is getting a second look at your application. For more information, please read Admissions FAQ

Who? People you have an existing relationship with.
  • Did we meet at a conferece/workshop/summer-program/etc? Remind me/us
  • If there's a specific PI you are looking to impress -- your letter writer can also email them/us
  • No existing relationship? See next question about being an outsider
If you're feeling discouraged, see initial list of people who should email me regardless :)
Ask for a meeting? It doesn't hurt to ask, but meetings take even longer than emails so -- see time issues above

Meeting at a conference? Great idea! At a conference our schedule is already cleared for talks and social events so it's /much/ easier to find time to meet with prospective students.
That’s 100% fine, and you should say that. You can indicate where you have tried looking to learn more, and details about your interests. I take no offense to receiving an email, it’s when there’s a request attached to it that things get tricky.
You don’t and that sucks, I’m sorry. Normally you can assume they will try and take one or two per year but funding changes as do other commitments. A professor deciding to have a child, work with a start-up, create a new conference, or any number of other things might eat into their time and so they may not take anybody that year.
While flattering, that’s probably not true. Related to my own multimodal research area I can quickly name a bunch of really awesome other people. Here are a few in no particular order that I wish I had more time to work with:
Henny Admoni (CMU)
Joyce Chai (U Michigan)
Yejin Choi (U Washington)
Sonia Chernova (Georgia Tech)
Mirella Lapata (U Edinburgh)
Cynthia Matuszek (U Maryland Baltimore County)
Ellie Pavlick (Brown)
Alane Suhr (Berkeley)
Jacob Andreas (MIT)
Yoav Artzi (Cornell Tech)
Desmond Elliott (U Copenhagen)
Daniel Fried (CMU)
Jesse Thomason (U Southern California)
That’s legit, and please let me know. It doesn’t mean I can solve it, but with more information, I can at least try and be helpful.
Email me, let’s see what’s up. :)
If you have additional questions, feel free to email me. Thanks!
Regardless of when you're reading this, you already have started your application. A common mistake is to think that you can just sorta "pull things together" in your senior year. Hopefully after you read the sections below, it'll become clear why that's not the case, but here's some basic math. If you wanted to have a published paper included in your application in November, you'd need to have received the acceptance notice by early November, which is typically at least three months after submission (i.e. August). And if everything went perfectly, you'd still need at least 6 months to do and write up the work (i.e. January). This means, to do original research, you probably need to have started the Fall of your Junior year. But, a lot of professors (... I'm like this too typically) can't easily integrate you into a research project until you have taken prerequisite courses. This is why it's really hard to have a paper or more than one really strong letter. How the hell are you supposed to have taken all these classes, met these professors, gotten lucky with things working out, put together the app, etc all in time?! You probably can't, but that doesn't mean you're unqualified or shouldn't apply. I'm describing "an ideal" world, and we're all gonna just try our best!
It depends, but there are two factors that are potentially balancing each other out. 1. More time means ability to get better letters but 2. We compare BS to BS, and MS to MS. So is that MS really going to help or put you even further behind? You have to sort of take this on a case by case basis. If you didn't get a chance to do research in undergrad, you'll do as much research in your MS as you can and see where that puts you. That's the hand life dealt you. If you are a junior who just started research and you could graduate early or get a full year in, I'd recommend taking your time and getting those strong letters. In general, this isn't a race and nobody is going to check if you were 21 or 22 or 25 or ... when you applied. Just how you compare to other people with the same degree and opportunities.
We probably need to discuss how "rounds" work at LTI/RI. There is an admissions committee that is your first hurdle. They will go through a couple rounds of filtering before the rest of the faculty ever gets a chance to see your application. For sake of simplicity, here is a simplified caricature of the process in LTI and sort of RI (can't speak for others):
Stage Actions Participants
Round 1 50% of applicants are retained (admissions committee)
Round 2 50% of smaller pool are retained after closer inspection (admissions committee)
Round 3 General faculty is given access to applications. This is the first time most of the faculty will see applications and it's only ~a quarter of what was originally submitted. Importantly, it's also still way too many people. (full faculty)
Shortlist Through interviews, reading your apps carefully, contacting your letter writers, etc. we have to find a set of good matches for our labs and indicate who we want to advocate for. Most professors do not do interviews, BUT if a professor emails you about an interview -- respond! you have made it to this stage. (potential advisor)
Decisions Some small set of people (often those with more than one good potential match) are actually admitted. Not all of you are going to come, so there are some expected value calculations here to admit more than one good candidate. In LTI the whole faculty votes, in RI only the admissions committee. (Vote)

TL;DR -- Most faculty won't see the full pool of applications, only those in Round 3. This set becomes available to us in ~mid to to late January. Participating in Round 3/Shortlist are optional. Some advisors may simply wait to see who shows up in the Fall. I look and try to indicate preferences well in advance.

If you're feeling discouraged, see initial list of people who should email me regardless :)
Hopefully this helps: Graduate Application Support Program for personalized feedback. There are many similar programs advertised every year. You're hoping to show that you really understand the subfield, have experience, and know why you're applying (including who you mention in your SoP). It can feel like being broad makes you more flexible, but it can also mean nobody is particularly excited about your application. It's a balance. You don't need or want to lay out the title of your thesis in your SoP, but we also don't want something like "I like all of robotics", because, well, we can't do much with that.

Also, ask for examples from other people: e.g. Nelson Liu (UW), Rishi Bommasani (Stanford), CS SoP collection, or Nathan Schneider's Advice
TL;DR -- Professors with whom you have done research.

The Impossibly Strong Packet
The strongest letter is one that speaks to your ability to do the job you're applying for. In this case, that job is research in a PhD program. So the "perfect" set of letters is three letters, all from people who the reviewer considers their peer, all giving detailed comments about your work together, your ability to work independently, ability to come up with interesting research questions/next steps, and how you perform compared to previous students they have worked with. This is a lot to ask for and most people will not have this. The general rule of thumb here is to get letters "as close" to this as possible. So your situation will very but here is a rough set with some notes.

Famous Professor Did they actually work closely with you? Did you impress them? A strong letter from them is a big win but they have advised a lot of people, so if your letter says very little about you, or your project, they will likely have to say you're in the bottom X% of applicants, and that's not useful.
Professor (research) This is the "normal" case. Ideally this would be directed research that lead to a publication or submission, but a capstone project might be ok too. Keep in mind, that this person should mention the details above.
Professor (instructor) A letter that simply says you did well in class, does not add anything to your application that wasn't already covered by your transcript. This letter doesn't hurt you, but it also doesn't help. It's a no-op. If there was a semester long class research project, maybe we have something to work with here.
Internship / Manager (with Phd) These letters are tricky. What you do not want is a letter that says you were a good employee. You're applying to be a researcher, not an employee so this does not help and can hurt. This person obviously can't say they would take you as their own student, so what we need is a comment about how you compare to other PhDs with whom they have worked or who went on to do PhDs (and where those people did PhDs). This person has a PhD and so they should also compare you to people with whom they went to school. These comparisons should be explicit.
Internship / Manager (without PhD) A weaker version of the one above
A graduate student Try to avoid if at all possible. They simply have much less experience, fewer people/institutions to compare you to, probably have not served on admissions,... so their high praise does not translate that well.
If you have been asked to interview, congrats! This is big, regardless of the result. This is not a "technical" interview in the way that tech companies will have you go to a whiteboard, etc. This interview serves a few purposes:
  1. Is the SoP really you? An opportunity for you to prove you really understand the work you did, the field you're hoping to pursue, and can hold a research conversation. Think of this a bit like an extension of your statement of purpose, except where you probably got help crafting that, this is pure you and so an opportunity to set yourself apart as insightful.
  2. Do we have a similar vision? We might broadly be interested in similar topics, but when we start to discuss the next 3-5 years, do we have compatible visions for what some of the big open questions are that we think are worth tackling?
  3. Do we get along? The PhD averages a bit over 5 years. That's a long time to be in meetings with someone you don't click with. Hopefully, in this call we'll start bouncing ideas off each other, building on them and feel like there's a nice rhythm.
It's complicated. In this space, we are looking for two things:
  1. Research experience/knowledge
  2. Leveraging available resources
Experience/Knowledge this is relatively straight forward. If you are first author on a paper at an impressive venue (and your letter corroborates that you were in fact the project lead), then we know you are someone who can come up with research questions and execute on them. Pretty simple. If you are second, third, ... author these tend not to be that meaningful. Unless the letter says otherwise, we will assume that you helped but were mostly following directions. Relatedly, a common mistake is trying to get "lots of little papers" in random venues. Don't do this. This is directly at odds with demonstrating knowledge of the area. Specifically, if you are well versed in the state-of-the-art, then you know where it is being published and a single paper in that venue means a lot more than lots of papers in a 3rd tier. Similarly, a strong workshop paper at an important venue means more (and will probably give you an opportunity to meet relevant people) than a 3rd tier venue.

Leveraging Resources "cool cool" this all stinks you think to yourself. People at my school don't publish in those venues, advisors don't have graduate students to advise/help me, ... we try to take this into account. All the same advise still holds, you do want to target the primary venues, etc. but lots of programs will expect more research output from someone in a big top research university and lab. That doesn't mean you're off the hook if you don't have those resources, you have to demonstrate an equal level of commitment and interest in the field. This might mean being the first undergraduate that professor has written a submission with. It might mean taking way more courses than usual to create (for example) a sort of NLP minor out of a mixture of CS, Linguistics, and Math classes. We need something that demonstrates a mature knowledge of the field, ability to conduct research, and that if provided with more resources you would thrive. An additional important note here: We do not compare Bachelors to Masters. Different bars.
Both of these contribute to your general professionalism. I don't recommend spending a lot of time editing and improving, but part of research is science communication.

Website If you look at my students (or this website!) you'll see a simple solution: github.io. There are plenty of templates out there. This lets you include images, videos, links to details about your projects and papers which otherwise don't really fit into the SoP.

Google Scholar If you have papers that have been indexed by Google Scholar, make a google scholar profile. Like a website, this is a simple way to find your work, see your co-authors, and see if it is already impactful. It is basically, just another -- here's a thing grad students do, look at me, I'm acting like a grad student -- kind of thing.

LTI Video The video has changed as of Fall 2023 so if you're searching YouTube for examples like this MLT application by Kayo Yin or this PhD application by Billy Li, these no longer apply. Now a random question will pop-up and you'll answer within the browser. Should you do it? 100%. What's the point? It's another opportunity (like the SoP) to demonstrate that you have thought deeply about research and grad school. Doing well here can be really impactful because it shows that you know this stuff so well that even in off-the-cuff settings you perform, not just in polished documents that (might) have been proofed by others.
Two important things here:
  1. Different Departments
    All departments make their decisions independently and on different schedules. This is often a difference of several weeks. So not hearing from one department has no bearing on any others and often very strong candidates who would have a "better" home in another department, may be passed on by one department and a strong admit in another.
  2. LTI PhD/MLT
    There are multiple (very long) admissions meetings within LTI. Ideally, we would keep all of these a secret and there would be one big announcement of everything all at one time. Unfortunately, that's not possible Our own students, for example, are always able to figure out what we're up to. So, some decisions get shared earlier than others. For example, students who are current MLTs and applying to the CMU PhD will probably be told almost immediately by their current advisors, while external applicants might see a delay. Based on the last two years, I recommend giving it a week. Sorry for the mess, hopefully we can streamline this in the future.
This is above my pay grade -- university level policies, etc etc but
  1. SCS Language Policies
    https://www.cs.cmu.edu/academics/faq (TOEFL score of 100, Duolingo 120, ...).
  2. Are there exceptions?
    Very university specific. IIRC, at UIUC the requirements were set by the state government so we had no flexibility. At CMU, I believe some PhD programs /might/ try and do an interview as an exception if everything else in the record is /really/ strong and maybe a potential advisor has previously worked/corresponded with you etc. But this is very much the exception and rarely succeeds in my limited experience.
Internships and Visiting Researchers
TL;DR -- I probably won't reply. We get a lot of "let me visit" emails.

One thing you've probably already been able to divine from this document is that time is hard to come by. This means that while an offer of a self-funded internship does simplify things a little for us, taking another student, even for a few months, is still an expense. I, therefore, treat internships the same as any application. It is only fair if posted publicly and reviewed competitively. This gives everyone equal footing. For the last couple of years, we have run an internship program: LTI internships

Several other institutions and large labs do as well: As with admissions, we do review these carefully, in rounds, with interviews in the final round, but most of us are not going to follow up on a cold email about interning.

You know what I am going to say. A collaboration is like taking on another student, but you are more likely to get a reply under certain conditions:
  • You have papers that I recognize in a related area
  • You are more senior than a typical intern / visitor would be
  • You have a primary advisor who would also be working with us
  • You are funded by your home institution
  • You have a pre-existing plan that is well scoped
We all do genuinely love doing research. We do also genuinely like networking with new people. It's exciting to bounce ideas off one another, get a new perspective on an old problem, etc. The issue is running out of time in the day/week/year/life... so the more of these desiderata you can take care of, the more we can jump directly into the fun stuff with you and the easier it is for us to scope the time commitment (e.g. this will end before I start teaching).