Graduate School of Arts and Sciences

October 28, 2024

Abigail Arnold | Graduate School of Arts and Sciences

Josh Broderick Phillips, B/MS ‘24 in Computational Linguistics (CL), first met Lotus Goldberg, Professor of Linguistics and Vice Chair of the Linguistics and Computational Linguistics programs, when he took her Syntax I course as a sophomore at Brandeis University. Intrigued by the subject, he continued to take linguistics courses with Goldberg, studying additional advanced syntax content as well as linguistic typology. “It was a great treat to have Josh in class,” said Goldberg. At a linguistics program event, she approached him and asked him if he was interested in working in her lab. The journey that began in an introductory class led to Broderick Phillips working as Goldberg’s head research assistant during his senior year as a Linguistics undergraduate and his additional fifth year as a CL MS student.

Together, the two worked on a big project: building a large dataset of the Verb Phrase Ellipsis construction in multiple varieties of English, with detailed syntactic labeling. They used transcribed speech as a source for the construction, in which a verb and its objects are left out but can be inferred from context (an example, Goldberg said, is “Josh and I had an interview today, and Sarah will tomorrow”). This dataset will ultimately be a resource for other researchers in syntax to study the construction, as well as for computational linguists who work in Natural Language Processing. Broderick Phillips worked on all aspects of the project, including collecting examples, writing detailed instruction documents for annotators who would label the examples, supervising and adjudicating the annotators’ work, and supporting Goldberg in writing a proposal for a National Science Foundation grant. He also led the computational portion of the project, carrying out extensive data management and software configuration tasks.

Broderick Phillips found that his coursework and his research work in Goldberg’s lab complemented each other well. He said that his linguistics and syntax coursework prepared him to understand examples of the construction and judge whether examples fit the outlined criteria, as well as to “be able to imagine all of the different possibilities we could encounter” when looking at natural speech. What he learned in the course Natural Language Annotation for Machine Learning, which is required for CL MS students and focuses on linguistic corpus development for machine learning, helped him a great deal in the labeling process. “A lot of program alumni bring this course up as super relevant for understanding what actually goes into creating data and what quality data looks like,” he said.

Goldberg and Broderick Phillips joined the Graduate School of Arts and Sciences to talk more about their experience working together and what students can learn in the Computational Linguistics program.

Lotus, what is your approach to mentorship and what do you like about working with Brandeis graduate students? Josh, how did you find the support you received from Lotus?

Goldberg: I love getting to work with Brandeis graduate students as colleagues. I don’t want to give someone more than they’re ready for, but when someone like Josh is very ready, it’s very enjoyable and satisfying to work with a grad student in this way. It’s very satisfying to see them take the lead on things and come into their own. I also like being able to collaborate on research with students. Last summer, I had three students in the lab, including Josh, and the four of us worked together intensively, developing annotation guidelines and pulling a lot of our results together for an NSF grant proposal. I do some explaining of what we’re talking about and what I’m looking for, but I like it when students can do their own learning and bring it back to me and the group. It’s wonderful to get to work with students in this way because it’s such a treat for me – and I know that the work they are doing will benefit their own career and intellectual development just as much.

Broderick Phillips: I loved collaborating directly with someone who’s been in the field so long and is an expert in their field of research. It wasn’t a huge lab, so everyone worked together as one group of people. Everyone’s ideas were taken into account, and you could bring what you wanted to the table. I had Lotus as a resource to ask theoretical questions – she knows off the top of her head every paper that’s ever been written in this area! I learned all the syntax I know and a lot of my ability to write well from Lotus. Working in the lab was similar to coursework but let me take my knowledge further.

Goldberg: Yes, even though this isn’t a course, I see students learning and growing. It’s lovely.

Josh, what have you been doing since graduating? How did your experience in the Computational Linguistics program prepare you for it?

Broderick Phillips: I am an AI engineer at MITRE. I’m part of a group of around fifty people with a variety of backgrounds, a lot from linguistics or computational linguistics. My department works on a lot of large language models and data collection, and we also collaborate with different departments. We also get to read research. So I am in an atmosphere where we create our own knowledge, which is very similar to my experience at Brandeis. My role is more focused on the technical side than on theory and data collection, but all the skills I learned at Brandeis are relevant.

While I was a student, I interned at Babel Street. I worked on their linguistics team and did a lot of data collection and creation of annotated data sets. This helped lead to my role at MITRE because I was able to talk about the skills I developed in thinking critically about these things, which MITRE values more than just pure technical ability.

In the Computational Linguistics program, we have industry receptions every semester. When I joined Babel Street, the team had three alums from the Brandeis CL MS program who I met at one reception; I was able to reach out to them, which led to the position. They typically hire a student intern from the program every year. And for MITRE, Ben Wellner, who is a program faculty member, also works there; I met him through coursework and reached out to him to learn about openings at the company. The connections I made through the program were very valuable and much better than applying out into the ether!

Goldberg: When we designed the program, we wanted the faculty to have a blend of people who work in academia and people who are very senior in industry. The latter would teach at Brandeis but also work at companies and be able to help students make connections.

Broderick Phillips: Now I work directly with Ben on one project, and I am coming back to the industry reception myself this semester. Hopefully we can continue the chain!

What makes the Computational Linguistics program special?

Broderick Phillips: You get to work very closely with all of the faculty. Despite their small numbers, the faculty manage to teach a huge variety of courses, so you can easily get to know them that way. Basically every student ends up in a lab, and it’s easy to get involved.

Goldberg: Yes, nothing Josh has described is unusual in our program. We try to make the available opportunities clear to students early on – when they arrive at Brandeis for the master’s program or are advanced undergraduates – because we want to get them involved. Not only is there not a barrier to working in a lab, we want to encourage it! And it’s very common for students to find internships and jobs through our industry receptions and connections, as Josh did.

Broderick Phillips: Neither linguistics nor computational linguistics were on my radar at all when I came to Brandeis. I took an introductory linguistics class as an elective as a freshman and ended up majoring in it and discovering computational linguistics through that as well. It’s easy to get involved in the community once you start.