The KIT-LIMSI Translation System for WMT 2015

Ha, Thanh-Le; Do, Quoc-Khanh; Cho, Eunah; Niehues, Jan; Allauzen, Alexandre; Yvon, François; Waibel, Alex

doi:10.18653/v1/w15-3012

Cited by 5 publications

(3 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Inspired by Refs. 37 and 38, the “[cls]” token used to indicate the beginning of a sequence in BERT is replaced by a language token to distinguish the different encoding methods. The sequential and reverse-sequential tokenizers share the same vocabulary of 26 lowercase letters and 10 numbers, differing in terms of their respective language token and order of text sequence.…”

Section: Methodsmentioning

confidence: 99%

Multilingual semantic fusion network for text recognition in the wild

et al. 2023

View full text Add to dashboard Cite

Most current approaches in the literature of scene text recognition train the language model via a text dataset far sparser than in natural language processing, resulting in inadequate training. Therefore, we propose a simple transformer encoder-decoder model called the multilingual semantic fusion network (MSFN) that can leverage prior linguistic knowledge to learn robust language features. First, we label the text dataset with forward, backward sequences, and subwords, which are extracted by tokenization with linguistic information. Then we introduce a multilingual model to the decoder corresponding to three different channels of the labeled dataset. The final output is fused by different channels to get more accurate results. In experiments, MSFN achieves cutting-edge performance across six benchmark datasets, and extensive ablative studies have proven the effectiveness of the proposed method. Code is available at https://github .com/lclee0577/MLViT.

show abstract

Section: Methodsmentioning

confidence: 99%

Multilingual semantic fusion network for text recognition in the wild

et al. 2023

View full text Add to dashboard Cite

show abstract

“…In particular, they leveraged high-quality post-edits performed by professional translators on 600 output sentences from both PB-SMT and NMT systems for English-to-German translation of TED talks. The NMT system of Luong and Manning (2015) was compared against a standard PB-SMT system (Ha et al 2015), a hierarchical SMT system (Jehl et al 2015) and a system combining PB-SMT and syntax-based SMT (Huck and Birch 2015). HTERessentially, TER with a human-in-the-loop (Snover et al 2006)and multi-reference TER was used, and results showed that NMT outperformed the other approaches in all metrics at a statistical significance level of p=0.01.…”

Section: The Changing Nature Of Mt System Designmentioning

confidence: 99%

Quality Expectations of Machine Translation

Way

2018

Machine Translation: Technologies and Applications

View full text Add to dashboard Cite

Machine Translation (MT) is being deployed for a range of use-cases by millions of people on a daily basis. There should, therefore, be no doubt as to the utility of MT. However, not everyone is convinced that MT can be useful, especially as a productivity enhancer for human translators. In this chapter, I address this issue, describing how MT is currently deployed, how its output is evaluated and how this could be enhanced, especially as MT quality itself improves. Central to these issues is the acceptance that there is no longer a single 'gold standard' measure of quality, such that the situation in which MT is deployed needs to be borne in mind, especially with respect to the expected 'shelf-life' of the translation itself. Machine Translation Use TodayThere are many estimates as to how much the translation industry is worth today, and how much it will expand over the coming years. For example, the size of the overall global language industry in 2015 was estimated at $38 billion, with estimates of up to $46 billion by 2016. 2 Thought leaders in this space have even begun to estimate the worth of the MT sector itself; in August 2014, TAUS stated that the MT industry was worth $250M. 3 This was a significant announcement for a number of reasons. First and foremost, it recognised that MT was already being used successfully for a number of usecases; secondly, it noted that while this estimate might be seen to be on the low side, for MT companies even a small slice of $250M was not to be sniffed at; 4 and thirdly, it pointed out that MT technology is a key enabler and a force multiplier for new 1 This concept is also applied to crowdsourced translation by Jimenéz-Crespo in this volume. 2 https://www.gala-global.org/industry/industry-facts-and-data 3 https://www.taus.net/think-tank/news/press-release/size-machine-translation-market-is-250million-taus-publishes-new-market-report 4

show abstract

“…Conditioning a multilingual model on the input language has been studied in NMT (Ha et al, 2016b;Johnson et al, 2017), syntactic parsing (Ammar et al, 2016) and language modeling (Östling and Tiedemann, 2017). The goal is to embed language information in real-valued vectors in order to enrich internal representations with input language for multilingual models.…”

Section: Language Embeddings and Typologymentioning

confidence: 99%

Multilingual Learning and Adaptation for Neural Language Models

Üstün¹

View full text Add to dashboard Cite

This thesis represents more than just an academic achievement. It marks a new chapter in my life, one that has been filled with lots of fun, fulfillment, and rewards that will endure for a lifetime. I would like to sincerely thank everyone who has supported me along the way.First and foremost, I thank my supervisors: Gertjan, Gosse, and Arianna.Gertjan, you were a constant source of support that I needed many times. As my supervisor, you always made me feel secure and confident, not only in my academic pursuits but also as a person. Your guidance on my research direction has played a pivotal role in this thesis. I will also always remember our attempt at a 40 km walk, even though I was not unable to finish it.Gosse, in addition to your continuous mentorship for my research and this thesis, your calm attitude and patience as a supervisor, were perfect examples for me to follow, especially as an ambitious and sometimes impatient student, that helped me become a better researcher. I also genuinely enjoyed our lunchtime chats, particularly about cycling, which made me excited to watch the Tour de France at the end of 4 years.Arianna, I'm very glad that I asked you to be my supervisor when you joined the department. Your knowledge and guidance in my research interests, your attention to precise details and pursuit for "excellence" have deeply influenced this thesis and my perspective on research. I also never imagined having a supervisor with whom I can speak Turkish which is an unexpected delight, especially with your kids. I believe we made a great team, and working together was a truly enjoyable experience. Once again, I thank my supervisors.Throughout my PhD, I had the opportunity to collaborate with a group of exceptional individuals. I would like to thank Asa Cooper Stickland, Alexandre Berard, Barbara Plank, Matthias Gallé, Rob van der Goot, Laurent Besacier, and Sebastian Ruder for their invaluable contribution to my research.I would also like to thank the members of my reading committee, Joakim Nivre, Jörg Tiedemann, and Jack Hoeksema, for reading my thesis and providing me with useful feedback which helped to improve this thesis.

show abstract

The KIT-LIMSI Translation System for WMT 2015

Cited by 5 publications

References 22 publications

Multilingual semantic fusion network for text recognition in the wild

Multilingual semantic fusion network for text recognition in the wild

Quality Expectations of Machine Translation

Multilingual Learning and Adaptation for Neural Language Models

Contact Info

Product

Resources

About