Filters








8 Hits in 3.9 sec

Reliability and Learnability of Human Bandit Feedback for Sequence-to-Sequence Reinforcement Learning

Julia Kreutzer, Joshua Uyheng, Stefan Riezler
2018 Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)  
We present a study on reinforcement learning (RL) from human bandit feedback for sequence-to-sequence learning, exemplified by the task of bandit neural machine translation (NMT).  ...  Finally, improvements of over 1 BLEU can be obtained by integrating a regressionbased reward estimator trained on cardinal feedback for 800 translations into RL for NMT.  ...  This work was supported in part by DFG Research Grant RI 2221/4-1, and by an internship program of the IWR at Heidelberg University.  ... 
doi:10.18653/v1/p18-1165 dblp:conf/acl/RiezlerKU18 fatcat:auiizdjbnjeunfsxpx57vg3sbq

Neural Machine Translation: A Review

Felix Stahlberg
2020 The Journal of Artificial Intelligence Research  
Statistical MT, which mainly relies on various count-based models and which used to dominate MT research for decades, has largely been superseded by neural machine translation (NMT), which tackles translation  ...  The field of machine translation (MT), the automatic translation of written text from one natural language into another, has experienced a major paradigm shift in recent years.  ...  I also thank the surveys editor Dragomir Radev for his guidance through the entire publication process, and all the anonymous reviewers for their detailed feedback.  ... 
doi:10.1613/jair.1.12007 fatcat:ryepzeon7nfnbgujk2kiacbrlm

Reliability and Learnability of Human Bandit Feedback for Sequence-to-Sequence Reinforcement Learning [article]

Julia Kreutzer, Joshua Uyheng, Stefan Riezler
2018 arXiv   pre-print
We present a study on reinforcement learning (RL) from human bandit feedback for sequence-to-sequence learning, exemplified by the task of bandit neural machine translation (NMT).  ...  Finally, improvements of over 1 BLEU can be obtained by integrating a regression-based reward estimator trained on cardinal feedback for 800 translations into RL for NMT.  ...  This work was supported in part by DFG Research Grant RI 2221/4-1, and by an internship program of the IWR at Heidelberg University.  ... 
arXiv:1805.10627v3 fatcat:kj7iskaodjc2tenogl3q5g7cjq

FINDINGS OF THE IWSLT 2020 EVALUATION CAMPAIGN

Ebrahim Ansari, Amittai Axelrod, Nguyen Bach, Ondřej Bojar, Roldano Cattoni, Fahim Dalvi, Nadir Durrani, Marcello Federico, Christian Federmann, Jiatao Gu, Fei Huang, Kevin Knight (+11 others)
2020 Proceedings of the 17th International Conference on Spoken Language Translation   unpublished
The evaluation campaign of the International Conference on Spoken Language Translation (IWSLT 2020) featured this year six challenge tracks: (i) Simultaneous speech translation, (ii) Video speech translation  ...  A total of 30 teams participated in at least one of the tracks. This paper introduces each track's goal, data and evaluation metrics, and reports the results of the received submissions.  ...  The permitted data for constrained submissions were: For English ASR: • LibriSpeech ASR corpus (Panayotov et al., 2015) , For English→Czech Translation: • • All the data for English-German track by WMT  ... 
doi:10.18653/v1/2020.iwslt-1.1 fatcat:cxogd5mojjdihl7gsitizte5am

Preference Learning for Machine Translation

Patrick Simianer
2018
preferences in an interactive post-editing scenario, learning precisely adapted machine translation systems.  ...  However, it is possible to build useful translating machines when the target domain is well known and the machine is able to learn and adapt efficiently and promptly from new inputs.  ...  Acknowledgments "The difference between the right word and the almost right word is the difference between lightning and a lightning bug. ž  ... 
doi:10.11588/heidok.00025488 fatcat:mvnd4rzwcfgr3fef2e6gaqupqa

Reinforcement Learning for Machine Translation: from Simulations to Real-World Applications

Julia Kreutzer
2020
The participants were university students with fluent or native language skills in German and English.  ...  news to TED translations from English to German.  ...  and an instruction. • Read the source sentence and the translation. • Follow the instruction by either marking the incorrect words of the translation by clicking on them or highlighting them, correcting  ... 
doi:10.11588/heidok.00028862 fatcat:jrsiseo4prf4pa3f7nnbp24wkq

Code-switched inspired losses for spoken dialog representations

Pierre Colombo, Emile Chapuis, Matthieu Labeau, Chloé Clavel
2021 Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing   unpublished
English, German, Italian, no labeled code-switching datasets for spoken dia- French and Spanish).  ...  The which is dedicated to the task-independent DA. task of mII consists of identifying the index of DA for German For German, we rely on the the inconsistent sentences  ... 
doi:10.18653/v1/2021.emnlp-main.656 fatcat:msosztszkrcchpgukue6ovizoa

Étudiants du Master Linguistique, spécialité LASTIC de l'Université d'Or-léans Comité de Programme de TALN Comité de Relecture de TALN 2017 7 et INRIA Béatrice Daille, Laboratoire d'Informatique Nantes Atlantique (LINA)

Hyun Kang, Hélène Couderc, Laetitia Delay, Siqi Fan, Lorraine Gaspard, Mélanie Lefeuvre, Sara Masaud, Stéphanie Nogueira, Camille Pertin, Cathy Querineau, Céline Vaschalde, Jidong Xie (+24 others)
unpublished
We would like to thank the anonymous reviewers for their helpful comments.  ...  Acknowledgments This work was supported by the CRC 991 "The Structure of Representations in Language, Cognition, and Science" funded by the German Research Foundation (DFG).  ...  The pay-offs of preprocessing for German-English Statistical Machine Translation. In M. FEDERICO, I. LANE, M. PAUL & F.  ... 
fatcat:gzcm2fdocnhqljnrm53jhxabea