The HAITrans team was involved in three presentations at the 44th Translating and the Computer conference hosted by the International Association for Advancement in Language Technology (AsLing) in Luxembourg on 24-25 November 2022.

On 24 November, HAITrans members Miguel Rios and Raluca Chereji presented on a project conducted together with Prof. Dragoș Ciobanu and ZTW Senior Scientist Alina Secară. In their presentation titled "Impact of Domain-Adapted Multilingual Neural Machine Translation in the Medical Domain", Miguel and Raluca outlined the quality improvements brought about by fine-tuning a Multilingual Neural Machine Translation (MNMT) model with in-domain data for the medical domain, compared with an out-of-domain MNMT model, for the English-Romanian language pair. Miguel and Raluca used automatic metrics and human error annotation based on a terminology-specific error typology to show that the in-domain MNMT model outperformed the out-of-domain MNMT model across most metrics, producing fewer terminology errors across most error categories surveyed.

On 25 November, in the morning, Raluca Chereji and Justus Brockmann, on behalf of the entire research group, presented on “The use of speech technologies in translation, revision, and post-editing machine translation (PEMT)”, providing an overview of HAITrans research projects that investigate applications of automatic speech recognition and automatic speech synthesis. The results of these investigations indicate that speech synthesis can support revision and PEMT, especially in terms of improved attention to errors. Future efforts include investigating the impact of speech synthesis on post-editors’ cognitive load, among other things. Three doctoral research projects within the HAITrans group will further investigate applications of speech recognition and synthesis in different scenarios.

On the same day, Dragoș Ciobanu chaired a panel discussion on the “Past, present and future of speech technologies in translation”. Dragoș was joined by panelists Julián Zapata, Carlos Teixeira, Marcin Feder, and Alina Secară. The panelists discussed how speech recognition and synthesis technologies have reached levels of maturity that have recently boosted their prominence. Examples in the area of translation include experimental translation and PEMT set-ups that rely on touch and voice rather than mouse and keyboard, new university courses that integrate the use of speech recognition in translator training, or a real-time speech-to-text plus machine translation solution being deployed to increase accessibility at the European Parliament. Accessibility was identified as another major driver behind the increasing prominence of speech technology, with new legal accessibility requirements and the large volume of content they affect warranting interest in the use of automated speech-to-text and text-to-speech solutions.