The Bringing together Language and Behavior for Large-scale Analytical Breakthroughs Laboratory (Blablablab)

Software

• M3: Multi-modal, multilingual, multi-attribute demographic inference. You can also get it pypi as a pip-installable package!
• GenderPerformr: Infers gender performance for any kind of online username, name, handle, etc.
• Social Support: Infers how supportive, neutral or unsupportive is an online reply
• Citation Function: Infers why an author cited another paper
• Immigration Framing: Infers which frames are used in a short text message about immigration. Neural models available via HuggingFace!
• Social Relationships: Infers the type of social relationship between two Twitter users based on dialog and network structure
• Gif-Reply Models: Multimodal models that will pick an appropriate gif reply for a given message.
• Pepe the Gif-Reply Bot: A Slack App that you can deploy to automatically reply to messages on your workspace with gifs
• Certainty Estimator (also on pip): A neural library for estimating the certainty/uncertainty of a statement along with which aspects are certain or uncertain (this works much better than hedges.
• Intimacy Estimator (also on pip): A neural library for estimating the intimacy level of questions in conversation. Very useful for looking at social distance and social norms in conversation!

Data

• Citation Function Data: labeled training data of citations by their rhetorical function.
• Gender-Labeled Online Conversations: 100M online dyadic conversations from Reddit, Wikipedia, and StackExchange, labeled by gender salience. To respect the privacy and dignity of individuals, this dataset is available for non-commercial research purposes online; please email the lab PI to obtain access.
• Reply Supportiveness Ratings: ~9K ratings of replies to authors rated on a 5-point Likert scale for how supportive (or unsupportive) they are. Items are balanced across Reddit, Wikipedia, and StackExchange interactions, as well as by length (equal amounts of short, medium, and long comments).
• MultiCiteNew citation dataset with multiple intents per context and variable sized contexts expressing that intent (e.g., multiple sentences are needed to understand why an author cited that paper).
• Gif-based dialog 1.56M conversation turns on Twitter where the replying user has responded with a gif. Gifs are canonicalized to an image hash, which is re-usable for matching new gifs.

Masters and Undergraduate Students

For current students, due to the technical work that we do in the lab, I typically require students to have taking some class on NLP or advanced Machine Learning to give them the requisite skills. Without those classes, we end up teaching you many of the same techniques in a less principled way which takes more time. If feel like you already have significant research experience (just not in NLP or ML), please explain what you've done. Students are typically expected to join our group's research meetings. During the school year, we'll typically have one meeting a week that is also with the project's co-supervisor (a PhD student or postdoc). Interest students should apply through this form

PhD students

I admit roughly one PhD student per year. Sometimes students are co-admitted or co-advised, so the number of admits can vary.

For prospective PhDs, I especially like students who come with a strong computational background with some experience in social science. There are no set criteria, but you're much better off towards admissions if you've contacted me (or your advisor has) and let me know of your interests and goals. Make sure to look over these pages carefully; the match should be pretty strong. A PhD student is very costly - in time and money - and I select students for my research group carefully.

If you're a current PhD student outside of CSE or SI, I'm open to collaborations. One of the best things about SI and CSE are the interdisciplinary environments and I'm potentially open to hosting students outside my home departments (but inside UM) in lab or co-advising on projects where it makes sense. Regardless, I'd love to hear from you and you're always welcome to come take my classes

PhD Students not at the University of Michigan

If you're a PhD student somewhere else and want to work with me (while being external), this could happen under the right circumstances. Typically, your advisor at your primary institution and I would co-advise you on a specific project. I typically only do these kinds of arrangements when I know your advisor (more common) or when the collaborative project make sense (rare). To get this started, have your advisor email me (not you directly) about what the project is.

Non-PhD External Students not at the University of Michigan

I unfortunately rarely work with highschoolers, undergraduates or master students who are not physically at the University of Michigan. I still get emails from external students asking if we could together on something remotely and I really would love to, but my priority is to advise the current students at UM given the limited bandwith I have for advising. Your best bet to work with me is to get admitted to one of our programs and then drop me an email.

Postdocs

I would love to have you all in my lab but this is generally dependent on funding (but seriously, I would take you all if I could). Email me if you think you're a good match and tell me why and we might be able to figure something out. That said, at the moment, I'm not currently actively seeking postdocs (due to funding, of course). If you're coming with your own funding, that changes everything, so drop me a line then.