Our Academic Partner Universitat Autònoma de Barcelona (UAB) used the KantanMT platform for numerous projects and courses in the University. In this blog post, we caught up with Professor Olga Torres-Hostench where she describes her experience of using our custom MT platform for her course.
KantanMT: Could you please tell our readers a bit about yourself and what made you select the KantanMT Academic Programme?
My name is Olga Torres Hostench. I am a member of the Research Group Tradumàtica (Translation and Technologies) and I teach translation and localization at UAB. Even though I had known about KantanMT for some years, I hadn’t had the time (isn’t this always the perfect excuse?) to show their platform in my localization classes.
However, last year I had the chance to carry out a research stay at Dublin City University in Dublin. As my research topic was related to machine translation post-editing, I decided to set up a meeting with KantanMT. That’s when I realised how good their Academic programme is, and shortly after, in summer 2016, UAB partnered with KantanMT. From the very beginning, my colleagues and I had great support from the Professional Services Team at KantanMT, with real-time demos and excellent training material.
Of course, visiting the KantanMT headquarters is not a requirement to close an academic agreement, but nonetheless it was nice to meet with such efficient, nice, ethical and generous people there, and experience how they work.
Moreover, the fact that the company is based in the Invent Building of the DCU campus itself helps KantanMT understand the needs of universities in general, including UAB.
In addition to the academic partnership, KantanMT also joined us at the 2nd International T3L Conference: Tradumatics, Translation Technologies & Localisation, hosted at UAB and organised by the Tradumàtica Research Group. At the conference, Laura Casanellas, KantanMT Product Manager, carried out a very interesting workshop where she showed the KantanMT platform to a fully booked class.
KantanMT: Can you tell us more about how you used the KantanMT platform at UAB?
We have been using KantanMT for the Master’s Course in Tradumàtica (Translation and Technologies). The main purpose of this Master’s course is to train the best professionals that will work in the localization industry, especially in Spain.
Our students use KantanMT with a clearly business-oriented focus. After the course, they are fully capable of managing the MT processes in any company or organisation. What these students value most from their experience with KantanMT is that the platform is easy to use and they can build and deploy their own MT engines, without having a computer science background.
Some of my students have also used KantanMT for their final MA dissertation, where training and research were combined. In this context, they particularly valued KantanBuildAnalytics and the PEX Editor as their favourite features in the whole platform.
KantanMT note: In the new User Interface, the KantanMT Rule Editor is a simplified version of the Pre-processor and Automatic Post-editing (PEX) editor. It can be found in the Training tab, where it can be used to apply rules to source, target and/or monolingual training data, and in the Translation tab, where it can be used to modify translation files (source and/or target).
Other students have been using the platform exclusively to conduct research. One of our PhD candidates, for instance, is currently carrying out her research on medical terminology for MT.
Finally, I have made excellent use of the KantanMT platform in one of our localization modules within the Translation and Interpreting course.
KantanMT: That sounds interesting. Could you please tell us a bit more about this module?
This module covers a lot of different elements related to localization. The part devoted to MT is only limited to a few sessions, but we manage to cover the basics of MT, including some history and evolution of the field, types of MT systems and some practical exercises.
Unfortunately, MT is often looked down upon by linguists and translators, who sometimes even discourage students from using MT for academic purpose. Therefore, in our course, not only do we teach the technological aspects of MT, but we also attempt to break some of these preconceptions. This isn’t always an easy task to undertake, especially considering the limited time we have.
In the module’s practical sessions with KantanMT, the students are given a project where they are asked to imagine they are translators for Global Voices, the leading online community for citizen media reporting. The students download the Global Voices free corpus from Opus and then choose some news to translate using MT. On the KantanMT platform, the students create an engine and train it with the downloaded corpus.
Interestingly, when the students analyse the MT quality of this engine with KantanBuildAnalytics, they see that the BLEU score is quite low, which means that the quality of the engine might not be very good. They learn that the corpus used is too small to train a performant engine. On the other hand, using a small corpus in class allows them to complete the training process in just one session and this is good for educational purposes. As they have a chance to fully understand all the basics involved in the process, by the end of the course the students see MT with growing interest and curiosity.
KantanMT note: It is very important to understand that SMT works on the basis of statistical probabilities of finding the best translation depending on the data pool used to train an engine. In general, this means that the higher the amount of good quality bilingual data used for training, the better the quality of the engine.
The students feel empowered because they can create a customised MT system starting from their corpus and they can see MT as a helpful resource that can act as an ally rather than a direct competitor. Moreover, undergraduate students appreciated the friendly and colourful interface.
To conclude, our experience with KantanMT has been so far very positive. I would like to highlight once again the different possibilities that the platform offers to different student profiles, from a quite basic use for beginners to a more advanced use for PhD students.