MT@BZ

In this project, we aim to conduct a preliminary study on machine translation (MT) at South Tyrolean institutions. Local institutions generally do not employ customized machine translation software, but they have high translation demands, to some extent repetitive and structured texts and specific needs related to the local variety of German. Customized MT could therefore represent an improvement into their translation process, especially because free online MT systems already belong to their tool-box.

Against this background, our first aim is to build aligned parallel corpora with local institutional documents to feed an MT engine at a later stage. To do so, we need a great amount of parallel segments, ideally several hundred thousand up to millions. In order to build the corpora, we seek already aligned texts (i.e. translation memories) from the institutions or collect and align them afterwards. In both cases, we contact the institutions and pursue a collaboration.

Secondly, we aim to evaluate the performance of one customized MT software when translating local institutional texts. To do so, we will pursue both an automatic and a manual quality assessment. State-of-the-art automatic scores, such as BLEU, chrF and TER give a preliminary overview on the quality of the machine translation. However, a fine-grained in-depth manual analysis of the mistakes sheds much more light on the weaknesses of the selected system. For this reason, we plan to annotate a corpus of machine translated decrees from the Autonomous Province of Bolzano as for their accuracy and fluency mistakes, e.g. mistranslations, omissions, punctuation and gender mistakes.

At the end of this experimental study, we supposedly will have gained the necessary preliminary knowledge to define a proper research project on machine translation for South Tyrolean institutions, which we aim to submit to international funding programs (e.g. Interreg).