METIS-II: low resource machine translation

Carl, Michaël; Melero, Maite; Badía, Toni; Vandeghinste, Vincent; Dirix, Peter; Schuurman, Ineke; Μαρκαντωνάτου, Στέλλα; Sofianopoulos, Sokratis; Vassiliou, Marina; Yannoutsou, Olga

doi:10.1007/s10590-008-9048-z

Cited by 9 publications

(3 citation statements)

References 11 publications

(12 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Other strategies are the crowdsourcing of additional parallel data, or the use of large monolingual and comparable corpora to perform bilingual lexicon induction before training an MT system (Klementiev et al, 2012;Irvine and Callison-Burch, 2013;Irvine and Callison-Burch, 2016). The METIS-II EU project replaced the need for parallel corpora by using linguistic pre-processing and statistics from target-language corpora only (Carl et al, 2008). In a recent study applied to Afrikaans-to-Dutch translation, the authors use a characterbased "cipher model" and a word-based language model to design a decoder for the low-resourced input language (Pourdamghani and Knight, 2017).…”

Section: Monolingual Resourcesmentioning

confidence: 99%

Machine Translation of Low-Resource Spoken Dialects: Strategies for Normalizing Swiss German

Honnet,

Popescu-Belis,

Musat

et al. 2017

Preprint

View full text Add to dashboard Cite

The goal of this work is to design a machine translation (MT) system for a low-resource family of dialects, collectively known as Swiss German, which are widely spoken in Switzerland but seldom written. We collected a significant number of parallel written resources to start with, up to a total of about 60k words. Moreover, we identified several other promising data sources for Swiss German. Then, we designed and compared three strategies for normalizing Swiss German input in order to address the regional diversity. We found that character-based neural MT was the best solution for text normalization. In combination with phrase-based statistical MT, our solution reached 36% BLEU score when translating from the Bernese dialect. This value, however, decreases as the testing data becomes more remote from the training one, geographically and topically. These resources and normalization techniques are a first step towards full MT of Swiss German dialects.

show abstract

Section: Monolingual Resourcesmentioning

confidence: 99%

Machine Translation of Low-Resource Spoken Dialects: Strategies for Normalizing Swiss German

Honnet,

Popescu-Belis,

Musat

et al. 2017

Preprint

View full text Add to dashboard Cite

show abstract

“…Irvine and Callison-Burch using the combination of parallel and comparable corpus, and gain between 0.5 and 1.7 points of bleu scores improvement in some Hindi local languages into English (Irvine & Callison-Burch, 2013). While Michael Carl shows an improvement by using lemma and PoS Tag in Spanish, Dutch and Greece languages translation into English (Carl, et al, 2008), and Jeff MA modifies word alignment by using clustering and succeed to handle more than 90% of loss resulting from poor word alignment caused by small training corpus in Chinese into English translation (MA, Matsoukas, & Schwartz, 2011). While in this paper, by considering the absence of Sundanese tool and resources to generate PoS, lemma, and other linguistics features, we prefer to use the PoS label tagged manually to get a better translation result in PBMT.…”

Section: Related Workmentioning

confidence: 99%

Experiment on a phrase-based statistical machine translation using PoS Tag information for Sundanese into Indonesian

Suryani¹,

Widyantoro²,

Purwarianti³

et al. 2015

2015 International Conference on Information Technology Systems and Innovation (ICITSI)

View full text Add to dashboard Cite

This paper discusses the problem of Sundanese into Indonesian text translation, as one of low-resource language pair translation. The number of parallel corpus gives a significant impact on a statistical machine translation. Whereas to date, there are no Sundanese to Indonesian parallel corpus that ready to use. It is, therefore, we apply the PoS Tag rather than only surface form in the translation model to get a better translation result. This experiment was done to get an early result in Sundanese to Indonesian text translation and to identify problems arise on it. The result shows that the model using surface form and PoS Tag was slightly outperformed the model using only surface form. However, there are some problems faced in this experiment which are the large number of OOV caused by the limited number of parallel corpus and unproper phrase translation caused by some noise in the parallel corpus such as typos and inconsistency writing a word in Sundanese corpus.

show abstract

“…Other pioneer hybridizations include: the combination of translation memories with rule‐based MT (Heyn, ), the combination of example‐based MT and rule‐based MT, and the combination of translation memories and example‐based MT (Carl & Hansen, ). More advanced hybrid techniques include the combination of rule‐based and several types of corpus‐based approaches (Carl et al., ). Finally, most recent works on hybridization tend to involve statistical systems with linguistic knowledge (Costa‐jussà & Farrús, ), which seems to suggest that modern statistical MT may have a high influence on rule‐based MT.…”

Section: Introductionmentioning

confidence: 99%

How much hybridization does machine translation Need?

Costa-jussà

2015

Asso for Info Science & Tech

View full text Add to dashboard Cite

Rule-based and corpus-based machine translation (MT) have coexisted for more than 20 years. Recently, boundaries between the two paradigms have narrowed and hybrid approaches are gaining interest from both academia and businesses. However, since hybrid approaches involve the multidisciplinary interaction of linguists, computer scientists, engineers, and information specialists, understandably a number of issues exist.While statistical methods currently dominate research work in MT, most commercial MT systems are technically hybrid systems. The research community should investigate the benefits and questions surrounding the hybridization of MT systems more actively. This paper discusses various issues related to hybrid MT including its origins, architectures, achievements, and frustrations experienced in the community. It can be said that both rule-based and corpus-based MT systems have benefited from hybridization when effectively integrated. In fact, many of the current rule/corpus-based MT approaches are already hybridized since they do include statistics/rules at some point.

show abstract

METIS-II: low resource machine translation

Cited by 9 publications

References 11 publications

Machine Translation of Low-Resource Spoken Dialects: Strategies for Normalizing Swiss German

Machine Translation of Low-Resource Spoken Dialects: Strategies for Normalizing Swiss German

Experiment on a phrase-based statistical machine translation using PoS Tag information for Sundanese into Indonesian

How much hybridization does machine translation Need?

Contact Info

Product

Resources

About