Fichier de travail (INPUT) : ./DUMP-TEXT/2-6.txt
Encodage utilisé (INPUT) : utf-8
Forme recherchée : translation|traduction|机器
_________________________________________________________________________________________________
Ligne n°34 : ... POSTED ON AUG 31, 2018 TO AI Research- Ligne n°35 : Unsupervised machine translation: A novel approach to provide fast,
Ligne n°36 : accurate translations for more languages ...
Ligne n°35 : ... Unsupervised machine translation: A novel approach to provide fast,- Ligne n°36 : accurate translations for more languages
Ligne n°37 : Marc'Aurelio Ranzato ...- Ligne n°41 : Automatic language translation is important to Facebook as a way to
Ligne n°42 : allow the billions of people who use our services to connect and ...
Ligne n°43 : ... communicate in their preferred language. To do this well, current- Ligne n°44 : machine translation (MT) systems require access to a considerable
Ligne n°45 : volume of translated text (e.g., pairs of the same text in both English ...
Ligne n°46 : ... and Spanish). As a result, MT currently works well only for the small- Ligne n°47 : subset of languages for which a volume of translations is readily
Ligne n°48 : available. ...- Ligne n°50 : Training an MT model without access to any translation resources at
Ligne n°51 : training time (known as unsupervised translation) was the necessary ...
Ligne n°50 : ... Training an MT model without access to any translation resources at- Ligne n°51 : training time (known as unsupervised translation) was the necessary
Ligne n°52 : next step. Research we are presenting at EMNLP 2018 outlines our recent ...
Ligne n°55 : ... is equivalent to supervised approaches trained with nearly 100,000- Ligne n°56 : reference translations. To give some idea of the level of advancement,
Ligne n°57 : an improvement of 1 BLEU point (a common metric for judging the ...
Ligne n°62 : ... majority of the 6,500 languages in the world for which the pool of- Ligne n°63 : available translation training resources is either nonexistent or so
Ligne n°64 : small that it cannot be used with existing systems. For low-resource ...
Ligne n°67 : ... unrelated text in Urdu – without having any of the respective- Ligne n°68 : translations.
- Ligne n°70 : This new method opens the door to faster, more accurate translations
Ligne n°71 : for many more languages. And it may only be the beginning of ways in ...- Ligne n°75 : Word-by-word translation
Ligne n°76 : The first step toward our ambitious goal was for the system to learn a ...
Ligne n°77 : ... bilingual dictionary, which associates a word with its plausible- Ligne n°78 : translations in the other language. For this, we used a method we
Ligne n°79 : introduced in a previous paper, in which the system first learns word ...
Ligne n°94 : ... physical world; for instance, the relationship between the words “cat”- Ligne n°95 : and “furry” in English is similar to their corresponding translation in
Ligne n°96 : Spanish (“gato” and “peludo”), as the frequency of these words and ...
Ligne n°103 : ... information, we can infer a fairly accurate bilingual dictionary- Ligne n°104 : without access to any translation and essentially perform word-by-word
Ligne n°105 : translation. ...
Ligne n°104 : ... without access to any translation and essentially perform word-by-word- Ligne n°105 : translation.
Ligne n°107 : ... Two-dimensional word embeddings in two languages (left) can be aligned- Ligne n°108 : via a simple rotation (right). After the rotation, word translation is
Ligne n°109 : performed via nearest neighbor search. ...
Ligne n°111 : ... Two-dimensional word embeddings in two languages (left) can be aligned- Ligne n°112 : via a simple rotation (right). After the rotation, word translation is
Ligne n°113 : performed via nearest neighbor search. ...
Ligne n°115 : ... Translating sentences- Ligne n°116 : Word-by-word translation using a bilingual dictionary inferred in an
Ligne n°117 : unsupervised way is not a great translation — words may be missing, out ...
Ligne n°116 : ... Word-by-word translation using a bilingual dictionary inferred in an- Ligne n°117 : unsupervised way is not a great translation — words may be missing, out
Ligne n°118 : of order, or just plain wrong. However, it preserves most of the ...
Ligne n°126 : ... English. Equipped with a language model and the word-by-word- Ligne n°127 : initialization, we can now build an early version of a translation
Ligne n°128 : system. ...
Ligne n°130 : ... Although it’s not very good yet, this early system is already better- Ligne n°131 : than word-by-word translation (thanks to the language model), and it
Ligne n°132 : can be used to translate lots of sentences from the source language ...- Ligne n°135 : Next, we treat these system translations (original sentence in Urdu,
Ligne n°136 : translation in English) as ground truth data to train an MT system in ...
Ligne n°135 : ... Next, we treat these system translations (original sentence in Urdu,- Ligne n°136 : translation in English) as ground truth data to train an MT system in
Ligne n°137 : the opposite direction, from English to Urdu. Admittedly, the input ...
Ligne n°137 : ... the opposite direction, from English to Urdu. Admittedly, the input- Ligne n°138 : English sentences will be somewhat corrupt because of translation
Ligne n°139 : errors of the first system. This technique was introduced by R. ...
Ligne n°141 : ... of MT systems (for which a good number of parallel sentences are- Ligne n°142 : available), and it was dubbed back translation. This is the first time
Ligne n°143 : this technique has been applied to a fully unsupervised system; ...
Ligne n°147 : ... fluent sentences, we can combine the artificially generated parallel- Ligne n°148 : sentences from our back translation with the corrections provided by
Ligne n°149 : the Urdu language model to train a translation system from English to ...
Ligne n°148 : ... sentences from our back translation with the corrections provided by- Ligne n°149 : the Urdu language model to train a translation system from English to
Ligne n°150 : Urdu. ...
Ligne n°153 : ... sentences in English to Urdu, forming another data set of the kind- Ligne n°154 : (original sentence in English, translation in Urdu) that can help
Ligne n°155 : improve the previous Urdu-to-English MT system. As one system gets ...
Ligne n°160 : ... Top: a sentence in English is translated to Urdu using the current- Ligne n°161 : En-Ur MT system. Next, the Ur-En MT system takes that Urdu translation
Ligne n°162 : as input and produces the English translation. The error between “cats ...
Ligne n°161 : ... En-Ur MT system. Next, the Ur-En MT system takes that Urdu translation- Ligne n°162 : as input and produces the English translation. The error between “cats
Ligne n°163 : are crazy” and “cats are lazy” is used to change the parameters such ...
Ligne n°169 : ... In our research, we identified three steps — word-by-word- Ligne n°170 : initialization, language modeling, and back translation — as important
Ligne n°171 : principles for unsupervised MT. Equipped with these principles, we can ...
Ligne n°175 : ... The first one was an unsupervised neural model that was more fluent- Ligne n°176 : than word-by-word translations but did not produce translations of the
- Ligne n°176 : than word-by-word translations but did not produce translations of the
Ligne n°177 : quality we wanted. They were, however, good enough to be used as ...
Ligne n°177 : ... quality we wanted. They were, however, good enough to be used as- Ligne n°178 : back-translation sentences. With back translation, this method
- Ligne n°178 : back-translation sentences. With back translation, this method
Ligne n°179 : performed about as well as a supervised model with 100,000 parallel ...
Ligne n°186 : ... been applied to unsupervised MT. In this case, we found that the- Ligne n°187 : translations had the correct words but were less fluent. Again, this
Ligne n°188 : method outperformed previous state-of-the-art unsupervised models. ...- Ligne n°209 : German-to-English translation examples show the results of each method:
- Ligne n°211 : German-to-English translation examples show the results of each machine
Ligne n°212 : translation method ...
Ligne n°211 : ... German-to-English translation examples show the results of each machine- Ligne n°212 : translation method
- Ligne n°214 : German-to-English translation examples show the results of each machine
Ligne n°215 : translation method ...
Ligne n°214 : ... German-to-English translation examples show the results of each machine- Ligne n°215 : translation method
Ligne n°221 : ... future improvements. In the short term, this will certainly help us- Ligne n°222 : translate in many more languages and improve translation quality for
Ligne n°223 : low-resource languages. But the learnings gained from this new method ...
Ligne n°228 : ... unlabeled data and perform tasks with very few, if any, of the expert- Ligne n°229 : demonstrations (translations, in this case) that are currently
Ligne n°230 : required. This work shows that it is at least possible for the system ...
Ligne n°237 : ... https://www.facebook.com/plugins/like.php?href=https://code.fb.com/ai-r- Ligne n°238 : esearch/unsupervised-machine-translation-a-novel-approach-to-provide-fa
Ligne n°239 : st-accurate-translations-for-more-languages/&width=450&layout=standard& ...
Ligne n°238 : ... esearch/unsupervised-machine-translation-a-novel-approach-to-provide-fa- Ligne n°239 : st-accurate-translations-for-more-languages/&width=450&layout=standard&
Ligne n°240 : action=like&size=small&show_faces=false&share=true&height=35&appId=2502 ...
Ligne n°257 : ... 08.03.2017- Ligne n°258 : Transitioning entirely to neural machine translation