John Mendonça was one of the twelve Ph.D. candidates selected under the first edition of the Affiliated Ph.D. Programs initiative launched in 2020. The Portuguese student received his MSc in Electrical and Computer Engineering from Instituto Superior Técnico in 2020, with a major in Systems, Decision and Control and a minor in Computer Engineering. During his master’s he worked with DefinedCrowd, developing methods to detect crowdsourcing fraud. Now enrolled in the 2nd year of his Ph.D. in Language Technologies at Instituto Superior Técnico, he is working in the MAIA research project at INESC-ID promoted by the ICT Company Unbabel. In Portugal he is advised by Isabel Trancoso and at CMU by Alon Lavie at the Language Technologies Institute.
In my free time I juggle between being a full-time foodie, playing videogames, and staying up to speed on geopolitics.
I also try to run every day, which is great as it helps keep me in shape despite being a foodie.
You are part of the first CMU Portugal Affiliated Ph.D. Program cohort. Why did you apply to this initiative?
The original grant I was receiving did not cover my full studies so finding a grant that would cover the 4 years was a priority. Given that I was planning on going to CMU to work with my co-supervisor for a year, the CMU Portugal Affiliated PhD program was an obvious choice.
You are being supervised in Portugal by Professor Isabel Trancoso. How did this opportunity come along?
I first met Professor Isabel as the instructor of the Speech Processing course during my masters. I’ve always wanted to work with Machine Learning and the Speech Processing course offered very interesting applications of machine learning. My Master Thesis soon followed and was supervised by Professor Isabel as well. The thesis went well, so in the end she offered me a position at INESC-ID as her PhD student.
What’s your thesis about?
The main goal of the thesis is to develop a framework to evaluate dialog systems. One of the issues of current evaluation metrics is that they focus on a single quality of dialog or on the “overall” quality, disregarding the fact that dialog quality is multifaceted (uses context, makes sense, interesting, etc.). To this end, this thesis explores how dialog dynamics intrinsically work, and propose metrics that evaluate dialog subqualities.
Your research work is being developed with Unbabel under the CMU Portugal Large Scale project MAIA. How did this opportunity appear?
At the time of my thesis final submission, the MAIA project was already ongoing, but with an opening for the work package pertaining to dialog evaluation, at INESC-ID (one of the partners of the project). Professor Isabel is also a collaborator of the project so the whole process was very straightforward.
One of the best aspects of working under MAIA is that I have the opportunity to work with other likeminded Ph.D .students
with research topics that align very well with my own.
How is this research collaboration working out?
We help each other often and have been supported on multiple occasions by other participants of the MAIA project. We meet every week to share our progress and discuss any issues we have. All in all, this leads to a great collaborative environment, which obviously encourages our research.
What are the main advantages of pursuing a Ph.D. in collaboration with a company such as Unbabel?
Collaborating with an ICT company is always a bonus to any PhD as it helps ground our research to the real world. This is especially important if you are interested in working in the industry after your studies. Unbabel’s role in the MAIA project also includes the provision of multilingual dialog datasets, which is a very important part of any contemporary Machine Learning research.
This summer you will go to CMU under Alon Lavie supervision, what are your expectations?
Being at CMU for a year opens plenty of opportunities to engage in parallel research topics with other CMU students and Unbabel-Pittsburgh. We have a well-defined workplan to follow at CMU, following up on initial experiments conducted so far, but we expect to collaborate with other students on related topics as well. On a personal note, American culture has always been fascinating to me, so living there for a year will definitely be an interesting experience.
What advice would you leave for potential Affiliated Ph.D. candidates?
In my experience the application process is straight forward, and is very much aligned with the FCT grant. The main advice I can give is to prepare a very strong workplan with your supervisor in Portugal and at CMU.