The MT-summit 2019 was held on the 21st, 22nd and 23rd of August in Dublin.
Before the main conference, there were 2 days of workshops that provided a vision of the new challenges in machine translation such as multilingualism, literature, low-resource languages and the importance of using MT in crisis situations. In addition, there were tutorial sessions in order to provide more insight into neural networks, statistics and how to choose a good test set.
The main conference lasted 3 days and there were 3 tracks: researchers, translators and users trying to join all the communities using machine translation. Nowadays, the quality of neural machine translation is much better and it is more integrated in the translation industry. The trend is that translators can post-edit but they must also have an excellent command of their own native language in order to give value to the machine translation content.
Some of the hot topics this year were: spoken translation, transfer learning, domain-specific models, low resource language translation and neural post-editing.
Pangeanic oral presentation
Pangeanic presented its study on “incremental adaptation of neural machine translation for professional post-editors” which is already available in the proceedings page 219. This work was carried out in cooperation with the research group PRHLT of the Polytechnic University of Valencia under the umbrella of the hybrid NMT platform project. This work consists of incrementally updating the neural machine translation models under an online learning paradigm applied to a real translation scenario. The results showed a reduction in the required amount of human effort needed when post-editing the outputs of the system, improvements in the translation quality and a positive user perception of this system.
The projects NEC TM and iADAATPA were presented by Pangeanic. The NEC TM Data consortium’s objective is to organise unexploited national bilingual assets that can be used as open data and general data for machine learning, in order to lower translation costs at a national level and across member states. It gathers translation memories from previous national contract awards from Member States and helps them to centralise these language assets with the fast-performing NEC TM database. iADAATPA is a platform which extends automatic translation and provides a European-wide marketplace for MT Providers to supply domain-adapted engines. ADAPT center also presented a poster about the evaluation of the machine translation engines trained in the iADAATPA framework, showing translation improvements in the domain-specific data provided by the European Member State institutions.
Pangeanic was one of the sponsors of the conference, where we had our own table to inform about our NLP technologies.
The event concluded with the closing ceremony announcing the next MT-Summit, which will be located in Seattle, USA.
Also, the locations of the EAMT 2020 and AAMT 2020 were announced. They will take place in Lisbon and Orlando, respectively.