Every year, Johns Hopkins University holds the Frederick Jelinek Memorial Summer Workshop (about page here). This human language technologies (HLT) workshop has taken place almost every year since the mid-90s. This year, 2019, is the sixth year that it’s held in honor of Frederick Jelinek. It’s abbreviated JSALT – which, I have to say, doesn’t make much sense to me. Jelinek Summer Annual (workshop) on Language Technologies is the best I can come up with to reconcile the full name and the abbreviation.
This year, it is held in Montreal. I couldn’t imagine a better place to spend the summer. It’s usually at a university in the U.S., but this year it’s held at École de Technologie Supérieure (ETS). One of the organizers did his Ph.D. at ETS and took advantage of his contacts there to make JSALT happen in Montreal. Last December, the Hopkins organizers accepted proposals from professors all over the world who had an idea for a research direction during the workshop. The five following projects were selected:
- Speaker Detection in Adverse Scenarios with a Single Microphone
- Distant Supervision for Representation Learning
- Improving Translation of Informal Language
- Using Cooperative Ad-hoc Microphone Arrays for ASR
- Neural Polysynthetic Language Modeling
The participants of the six-week-long workshop are mostly professors and graduate students. They reserve a few spots for undergrads, who go through an application process at JHU. I had been applying for other research positions throughout the semester, but nothing was working out. I really wanted to do something in language technology. As a last resort, I applied to a summer research position at Rice in the Networking Lab. I interviewed with the professor and was offered a position but hesitated to accept. Ironically, a few days later, I received an email to the ECE undergrads from that same professor advertising JSALT. I had never heard of it, but it sounded perfect. I applied in early March. I knew chances were slim, as it is an excellent research experience with fairly competitive pay. I remember the day I received an acceptance email – I remember thinking how fortunate I was and how big of the step JSALT would be in my academic career.
I was placed on my first-choice team, the “Speaker Detection in Adverse Scenarios” team led by Paola Garcia from JHU. To grossly oversimplify, we are interested in recognizing speakers in audio recordings that aren’t well-behaved. For example, in recordings that were made far away from the speaker of interest or in recordings involving a few speakers talking over each other. The senior members on the team are:
- Jesus Villalba (Johns Hopkins University)
- Diego Castan (SRI)
- Alex Crista (Ecole Normale Superieur, France)
- Herve Bredin(LIMSI, France)
- Jun Du (USTC, China)
- Najim Dehak (Johns Hopkins University)
- Emmanuel Dupoux (team affiliate, École des Hautes Études en Sciences Sociales and Facebook)
The graduate students:
- Marvin Lavechin (LSCP)
- Phani Nidadavolu (Johns Hopkins University)
- Saurabh Kataria (Johns Hopkins University)
- Leo Galmant (Université Paris-Sud, France)
- Lei Sun (USTC, China)
- Marie-Philippe Gill (ETS, Canada)
- Sajjad Abdoli (ETS, Canada)
- Mohammad Esmaelipour (ETS, Canada)
- Koji Okabe (Biometric Research Laboratoires NEC Corporation)
- Ling Guo (Biometric Research Laboratories NEC Corporation)
- Bar Ben-Yair (Israel)
Lastly, are the wee undergrads: Sizhu Chen from UCSD, and myself. Here’s the team description for more details.
The organizers hold a two-week summer school preceding the workshop for the students. It is intended to get students familiar with the directions of research in the workshop and the state-of-the-art technologies in HLT. The general schedule was a lecture in the morning and a lab/tutorial in the afternoon. It’s nice to work with people from the different teams because we are coming from such diverse backgrounds. Some are familiar with only Linguistics, others only Machine Learning and Computer Science, and others sit right between language and technology.
