Findings of the WMT 2022 Shared Tasks in Unsupervised MT and Very Low Resource Supervised MT
27480Authors
nan
Abstract
AbstractWe present the findings of the WMT2022Shared Tasks in Unsupervised MT and VeryLow Resource Supervised MT with experiments on the language pairs German to/fromUpper Sorbian, German to/from Lower Sorbian and Lower Sorbian to/from Upper Sorbian. Upper and Lower Sorbian are minoritylanguages spoken in the Eastern parts of Germany. There are active language communitiesworking on the preservation of the languageswho also made the data used in this Shared Taskavailable.In total, four teams participated on this SharedTask, with submissions from three teams for theunsupervised sub task, and submissions fromall four teams for the supervised sub task. Inthis overview paper, we present and discuss theresults.
Keywords
Full Content
Proceedings of the Seventh Conference on Machine Translation (WMT), pages 801 - 805 December 7-8, 2022 ©2022 Association for Computational Linguistics Findings of the WMT 2022 Shared Tasks in Unsupervised MT and Very Low Resource Supervised MT Marion Weller-Di Marco and Alexander Fraser Center for Information and Language Processing LMU Munich {dimarco,fraser}@cis.uni-muenchen.de Abstract We present the findings of the WMT2022 Shared Tasks in Unsupervised MT and Very Low Resource Supervised MT with experi- ments on the language pairs German to/from Upper Sorbian, German to/from Lower Sor- bian and Lower Sorbian to/from Upper Sor- bian. Upper and Lower Sorbian are minority languages spoken in the Eastern parts of Ger- many. There are active language communities working on the preservation of the languages who also made the data used in this Shared Task available. In total, four teams participated on this Shared Task, with submissions from three teams for the unsupervised sub task, and submissions from all four teams for the supervised sub task. In this overview paper, we present and discuss the results. 1 Introduction For a large majority of the world’s languages, only limited resources are available to train and provide NLP tools. The need for parallel data in a (super- vised) translation scenario aggravates this problem further. The Shared Tasks in Unsupervised MT and Very Low Resource Supervised MT aim at promot- ing the research on translating low and very low resourced languages. Following the Shared Tasks in the two previous years (Libovický and Fraser, 2021; Fraser, 2020), we continue to cooperate with the Sorbian commu- nity, namely the Sorbian Institute1 and the Witaj Sprachzentrum (Witaj Language Center)2 for this year’s Shared Task. We offer all translation direc- tions between the languages Upper Sorbian, Lower Sorbian and German, for both supervised and un- supervised translation. Upper and Lower Sorbian are minority lan- guages spoken in the eastern part of Germany in the 1https://www.serbski-institut.de/en/Institute/ 2https://www.witaj-sprachzentrum.de/ federal states of Saxony and Brandenburg. With only 30k and 7k native speakers, there are only few resources available. However, as western Slavic languages, Upper and Lower Sorbian are closely related to Polish and Czech and can thus make use of the comparatively large data sets available for those languages. In this year, four teams participated in the Shared Task, resulting in three to four submissions for each language pair for both the supervised and unsupervised variants. 2 Tasks and Evaluation In contrast to the previous Shared Tasks, all language combinations between Upper Sorbian, Lower Sorbian and German are considered, result- ing in the six following translation pairs: • Upper Sorbian ↔ German • Lower Sorbian ↔ German • Upper Sorbian ↔ Lower Sorbian Factoring in the variants supervised and unsuper- vised translation for each language pair, there is a total of 12 translation pairs. For the evaluation, we follow the strategy em- ployed in the previous Shared Task and use BLEU scores (Papineni et al., 2002) and chrF scores (Popovi´c, 2015) as implemented in sacreBLEU (Post, 2018).3 Furthermore, we evaluate the sub- missions using …
Content truncated to 500 words
Labeled Text
AbstractWe present the findings of the WMT2022Shared Tasks in Unsupervised MT and VeryLow Resource Supervised MT with experiments on the language pairs German to/fromUpper Sorbian, German to/from Lower Sorbian and Lower Sorbian to/from Upper Sorbian. Upper and Lower Sorbian are minoritylanguages spoken in the Eastern parts of Germany. There are active language communitiesworking on the preservation of the languageswho also made the data used in this Shared Taskavailable.In total, four teams participated on this SharedTask, with submissions from three teams for theunsupervised sub task, and submissions fromall four teams for the supervised sub task. Inthis overview paper, we present and discuss theresults.
Publication Details
- Paper ID:
- 27480
- Year:
- 2022
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan