Proceedings of the Third Conference on Machine Translation: Shared Task Papers 2018
DOI: 10.18653/v1/w18-6416
|View full text |Cite
|
Sign up to set email alerts
|

CUNI Submissions in WMT18

Abstract: We participated in the WMT 2018 shared news translation task in three language pairs: English-Estonian, English-Finnish, and English-Czech. Our main focus was the lowresource language pair of Estonian and English for which we utilized Finnish parallel data in a simple method. We first train a "parent model" for the high-resource language pair followed by adaptation on the related lowresource language pair. This approach brings a substantial performance boost over the baseline system trained only on Estonian-En… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
4
0

Year Published

2018
2018
2021
2021

Publication Types

Select...
5

Relationship

3
2

Authors

Journals

citations
Cited by 5 publications
(4 citation statements)
references
References 20 publications
0
4
0
Order By: Relevance
“…2.3.4 CUNI-KOCMI (Kocmi et al, 2018) The CUNI-KOCMI submission focuses on the low-resource language neural machine translation (NMT). The final submission uses a method of transfer learning: the model is pretrained on a related high-resource language (here Finnish) first, followed by a child low-resource language (Estonian) without any change in hyperparameters.…”
Section: Afrlmentioning
confidence: 99%
“…2.3.4 CUNI-KOCMI (Kocmi et al, 2018) The CUNI-KOCMI submission focuses on the low-resource language neural machine translation (NMT). The final submission uses a method of transfer learning: the model is pretrained on a related high-resource language (here Finnish) first, followed by a child low-resource language (Estonian) without any change in hyperparameters.…”
Section: Afrlmentioning
confidence: 99%
“…The framework T2T drops sentences from training corpus that are too long in order to allow bigger batch sizes, which lowers the training time. In our experiments, the threshold is set to 100 subwords, which is based on our previous findings (Kocmi et al, 2018b), where it was enough for most of the languages. Table 4.4: The amount of training corpus removed by filtering long sentences with more than 100 subwords (lower is better).…”
Section: Direct Transfer Drawbacksmentioning
confidence: 99%
“…Aside from the common back-translation (Sennrich et al, 2016a;Kocmi et al, 2018), simple copying of target monolingual data back to source has been also shown to improve translation quality in low-data conditions.…”
Section: Output Analysismentioning
confidence: 99%
“…We consider the multilingual model mBART (Liu et al, 2020) as well as all the WMT submissions that reported results on English Ø Kazakh. Of these baselines, only mBART and (Kocmi et al, 2018) use sacreBLEU which inhibits proper comparison with the rest of the models. We include them for completeness.…”
Section: Evaluating Translation Quality and Catastrophic Forgettingmentioning
confidence: 99%