Methods of Input Segmentation for Simultaneous Speech Translation
Název práce v češtině: | Způsoby segmentace vstupu pro účely simultánního strojového překladu mluvené řeči |
---|---|
Název v anglickém jazyce: | Methods of Input Segmentation for Simultaneous Speech Translation |
Klíčová slova: | NLP|Simultánní strojový překlad|Segmentační metody |
Klíčová slova anglicky: | NLP|Simultaneous machine translation|Segmentation methods |
Akademický rok vypsání: | 2021/2022 |
Typ práce: | diplomová práce |
Jazyk práce: | angličtina |
Ústav: | Ústav formální a aplikované lingvistiky (32-UFAL) |
Vedoucí / školitel: | doc. RNDr. Ondřej Bojar, Ph.D. |
Řešitel: | Mgr. Václav Ryšlink - zadáno a potvrzeno stud. odd. |
Datum přihlášení: | 23.02.2022 |
Datum zadání: | 15.06.2022 |
Datum potvrzení stud. oddělením: | 27.06.2022 |
Datum a čas obhajoby: | 07.09.2022 09:00 |
Datum odevzdání elektronické podoby: | 21.07.2022 |
Datum odevzdání tištěné podoby: | 25.07.2022 |
Datum proběhlé obhajoby: | 07.09.2022 |
Oponenti: | Mgr. Peter Polák |
Konzultanti: | Mgr. Aleš Tamchyna, Ph.D. |
Zásady pro vypracování |
During simultaneous machine translation, we are required to generate partial translations even before the sentences are completely finished. Furthermore, to minimise the latency we would like to output those partial translations as frequently as possible in order to keep the listeners synchronised with the speaker.
The goal of the thesis is to explore and compare both fixed and adaptive methods according to which we can make the decision to translate or keep waiting for additional input when interpreting. The ultimate evaluation measure will be the quality and latency (estimated automatically) of a machine translation system deployed at the different segmentation options. The experiments should be conducted on the ESIC dataset with manually transcribed original English speeches supplemented with Czech and German interpretations. More specifically, the work should involve: - surveying current literature addressing simultaneous machine translation and speech segmentation topics - considering the concept of meaningful translation units for automatic simultaneous speech translation - building or fine-tuning neural machine translation models most suitable for the task - evaluating different segmentation methods for simultaneous translation based on translation quality and latency |
Seznam odborné literatury |
ZHANG, Ruiqing, et al. Learning adaptive segmentation policy for simultaneous translation. In: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). 2020. p. 2280-2289.
KOO, Youngeun, et al. Towards a Linguistically Motivated Segmentation for a Simultaneous Interpretation System. In: Proceedings of the 34th Pacific Asia Conference on Language, Information and Computation. 2020. p. 129-137. ZHAO, Jinming, et al. It is Not as Good as You Think! Evaluating Simultaneous Machine Translation on Interpretation Data. arXiv preprint arXiv:2110.05213, 2021. THUNES, Martha. The concept of ‘translation unit’ revisited. 2017. JUNCZYS-DOWMUNT, Marcin, et al. Marian: Fast neural machine translation in C++. In Proceedings of ACL 2018, System Demonstrations, pages 116–121, Melbourne, Australia, July 2018. Association for Computational Linguistics. |