André Martins was the first dual degree student to finish a Ph.D. in Language Technology (LTI), both at Instituto Superior Técnico of the Universidade Técnica de Lisboa (IST/UTL) and Carnegie Mellon University (CMU). After his graduation in 2012, he joined the Machine Learning research team at Priberam, a spin-off company from IST focused in language technology development. In 2015 he started working as Head of Research at Unbabel, a Translation company that combines neural Machine Translation with professional human editing, to deliver high quality translation at massive scale.
- Unbabel is a Web translation service firm that combines artificial and human intelligence to deliver faster translations with human quality. What is the importance of areas such as machine learning or big data, and how are they managed within the company?
Machine learning is a key component of Unbabel’s products. For example, we use it to build neural machine translation systems that are tailored to our customers’ data. The better the quality of these systems, the less post-editing humans need to do, increasing the efficiency of the whole translation process. Another example is our quality estimation systems, which estimate how reliable a translation is before sending it to the customer, using neural models learned from customers’ data and translators’ post-edits. In both cases we clearly observe that the systems improve their performance continuously, as the amount of available data keeps increasing.
- Since the CMU Portugal Program has a strong focus in these areas, what are your main expectations for this partnership?
CMU is very strong in machine learning and natural language processing. We already collaborate with some faculty professors there, and we have hosted some of their students in Summer internships. We expect this partnership to enhance these connections even further in the future, possibly via joint research projects.
- How do you envision that a stronger partnership with academia can help Unbabel reach its goals?
We’ve believed since day one that a strong partnership with top international academic institutions is key to achieve our ambitious goals. We participate in projects involving some of the best universities in Europe in natural language processing. We participate in innovation actions on which we hosted several students and professors from these institutions, and we have AI Summer research internships involving students from all over the world. For us, the partnership is important since we learn from the latest achievements in machine learning and ensure we are an active part of this research community. For the academic partners, it is beneficial since they can test their tools in real-world problems and have access to data that is generally not available at universities.
- How does specialized training in ICT fit into your HR strategy?
For our AI teams, specialized training is very important. For senior-level positions we hire doctorates with experience in natural language processing, machine learning, and related areas. Besides, across the board, we provide training in-house, via online courses, participation in conferences and workshops, and an internal mentorship program.
- How would you shortly (one word/sentence) describe the Engineering strategy at Unbabel?
Data-driven, product-oriented, AI-fueled.