{"uri":"/language-weaver/the-neural-mt-weekly/issue-126-learning-feature-weights-for-denoising-parallel-corpora","content":{"metadata":{"seo":{"description":"Issue #126 - Learning Feature Weights for Denoising Parallel Corpora | Iconic Translation Machines","breadcrumbTitle":"Article","title":"Issue #126 - Learning Feature Weights for Denoising Parallel Corpora","twitter":{"Image":"https://webassets.rws.com/images/data-creation-1200x610_tcm228-195130.jpg","Description":"Issue #126 - Learning Feature Weights for Denoising Parallel Corpora | Iconic Translation Machines","Title":"Issue #126 - Learning Feature Weights for Denoising Parallel Corpora","Type":"website","card":"summary"},"og":{"Image":"https://webassets.rws.com/images/data-creation-1200x610_tcm228-195130.jpg","Url":"https://www.rws.com/language-weaver/the-neural-mt-weekly/issue-126-learning-feature-weights-for-denoising-parallel-corpora/","Description":"Issue #126 - Learning Feature Weights for Denoising Parallel Corpora | Iconic Translation Machines","Title":"Issue #126 - Learning Feature Weights for Denoising Parallel Corpora","Type":"website"},"link":[{"rel":"cannonical","url":"https://www.rws.com/language-weaver/the-neural-mt-weekly/issue-126-learning-feature-weights-for-denoising-parallel-corpora/"},{"rel":"alternate","hreflang":"en-gb","url":"https://www.rws.com/language-weaver/the-neural-mt-weekly/issue-126-learning-feature-weights-for-denoising-parallel-corpora/"},{"rel":"alternate","hreflang":"x-default","url":"https://www.rws.com"}]},"showBreadcrumb":true,"noIndex":false,"customNavigation":{"title":"Language Weaver","type":"Default","logo":{"image":{"friendlyFileSize":"412 B","title":"RWS Icon - Large","isSVG":false,"mediaType":"Image","url":"https://webassets.rws.com/images/icon-large-1_tcm228-194189.png","fileSize":412,"mimeType":"image/png"},"mobileImage":{"title":"RWS Icon - Large","isSVG":false,"mediaType":"Image","friendlyFileSize":"412 B","url":"https://webassets.rws.com/images/icon-large-1_tcm228-194189.png","fileSize":412,"mimeType":"image/png"},"alternateText":"RWS Icon - Large","hasMobileImage":false},"links":[{"link":{"text":"Find a solution","isExternal":false,"isVideo":false},"subLinks":[{"link":{"text":"I'm interested in…","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/language-weaver/neural-machine-translation/","internalLink":{},"text":"Neural machine translation","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/government-solutions/","internalLink":{},"text":"Multilingual content intelligence","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/global-enterprise/","internalLink":{},"text":"Making customer support multilingual","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/legal-and-compliance/","internalLink":{},"text":"Enabling Multilingual e-Discovery","isExternal":false,"isVideo":false}}]},{"link":{"text":"By products","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/language-weaver/cloud/","internalLink":{},"text":"Cloud","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/edge/","internalLink":{},"text":"Edge","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/connectors/","internalLink":{},"text":"Connector Hub","isExternal":false,"isVideo":false}}]},{"link":{"text":"By use case","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/language-weaver/global-enterprise/","internalLink":{},"text":"Enterprise","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/legal-and-compliance/","internalLink":{},"text":"Legal","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/government-solutions/","internalLink":{},"text":"Government","isExternal":false,"isVideo":false}}]}]},{"link":{"text":"Service and Support","isExternal":false,"isVideo":false},"subLinks":[{"link":{"text":"Expert Services","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/partners/","internalLink":{},"text":"Partners","isExternal":false,"isVideo":false}},{"link":{"url":"/partners/list/","internalLink":{},"text":"Partner Directory","isExternal":false,"isVideo":false}},{"link":{"url":"/support/professional-services/","internalLink":{},"text":"Professional Services","isExternal":false,"isVideo":false}},{"link":{"url":"https://www.rws.com/support/training/?productSolution=LanguageWeaver","text":"Training ","isExternal":true,"isVideo":false}}]},{"link":{"text":"Support","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"https://gateway.sdl.com/CommunityHome","text":"Support Gateway","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"https://community.rws.com/product-groups/linguistic-ai/cloud/","text":"Community ","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"/support/newsletter-signup/","internalLink":{},"text":"Sign up for support updates","isExternal":false,"isVideo":false}}]},{"link":{"text":"Developers","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"https://developers.languageweaver.com/","text":"Developer portal","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"https://docs.rws.com/791593/865454/language-weaver/welcome-to-the-------------language-weaver-documentation","text":"Cloud documentation","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"https://docs.rws.com/833475/886777/language-weaver-edge-documentation/welcome-to-the-------------language-weaver-edge-------------8-5-documentation","text":"Edge documentation","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}}]}]},{"link":{"text":"Resources","isExternal":false,"isVideo":false},"subLinks":[{"link":{"text":"Recommended ","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/language-weaver/events/language-weaver-webinar-series/","internalLink":{},"text":"Webinar Series","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/resources/us-army/","internalLink":{},"text":"U.S. Army case study","isExternal":false,"isVideo":false}},{"link":{"url":"https://webassets.rws.com/images/Language-Weaver-CAS-Case-Study_tcm228-191164.pdf","internalLink":{},"text":"Chemical Abstracts Service case study","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"/language-weaver/resources/machine-translation-cloud/","internalLink":{},"text":"Cloud Data Sheet","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/resources/machine-translation-edge/","internalLink":{},"text":"Edge Data Sheet","isExternal":false,"isVideo":false}}]},{"link":{"text":"Collateral","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=case-study-long","text":"Case Studies","isExternal":true,"isVideo":false}},{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=white-paper","text":"White Papers","isExternal":true,"isVideo":false}},{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=ebooks","text":"eBooks","isExternal":true,"isVideo":false}},{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=webinar-recording","text":"Recorded Webinars","isExternal":true,"isVideo":false}},{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=Brochure","text":"Brochures","isExternal":true,"isVideo":false}},{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=data-sheet","text":"Data Sheets","isExternal":true,"isVideo":false}},{"link":{"url":"/language-weaver/resources/","internalLink":{},"text":"View all","isExternal":false,"isVideo":false}}]},{"link":{"text":"Learn more","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/language-weaver/our-history/","internalLink":{},"text":"History","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/neural-machine-translation/","internalLink":{},"text":"Neural MT","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/the-neural-mt-weekly/","internalLink":{},"text":"Neural MT Weekly","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/supported-languages/","internalLink":{},"text":"Supported Languages ","isExternal":false,"isVideo":false}}]}]},{"link":{"url":"/language-weaver/research/","internalLink":{},"text":"Research","isExternal":false,"isVideo":false}},{"link":{"url":"https://portal.languageweaver.com/","text":"Get Started","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"/language-weaver/contact/","internalLink":{},"text":"Contact","isExternal":false,"isVideo":false}}],"mainLogo":{"image":{"title":"SDL - Part of the RWS Group","isSVG":false,"mediaType":"Image","friendlyFileSize":"2 KB","url":"https://webassets.rws.com/images/Artboard-1_tcm228-187294.svg","fileSize":1113,"mimeType":"image/svg+xml"},"mobileImage":{"title":"SDL - Part of the RWS Group","isSVG":false,"mediaType":"Image","friendlyFileSize":"2 KB","url":"https://webassets.rws.com/images/Artboard-1_tcm228-187294.svg","fileSize":1113,"mimeType":"image/svg+xml"},"alternateText":"SDL - Part of the RWS Group","hasMobileImage":false},"homeLink":{"url":"/","internalLink":{},"isExternal":false,"isVideo":false},"sectionHomeLink":{"url":"/language-weaver/","internalLink":{},"isExternal":false,"isVideo":false},"utilityNav":{"icon":{"desktopImage":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"375 B","url":"https://webassets.rws.com/images/search-1_tcm228-167958.svg","fileSize":375,"mimeType":"image/svg+xml"},"desktopImagePortrait":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"375 B","url":"https://webassets.rws.com/images/search-1_tcm228-167958.svg","fileSize":375,"mimeType":"image/svg+xml"},"mobileImage":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"375 B","url":"https://webassets.rws.com/images/search-1_tcm228-167958.svg","fileSize":375,"mimeType":"image/svg+xml"},"isVideo":false,"hasMobileImage":false},"links":[{"url":"/search/","internalLink":{},"isExternal":false,"isVideo":false}]},"languageSelector":{"icon":{"desktopImage":{"title":"globe","isSVG":false,"mediaType":"Image","friendlyFileSize":"703 B","url":"https://webassets.rws.com/images/globe-1_tcm228-153333.svg","fileSize":703,"mimeType":"image/svg+xml"},"desktopImagePortrait":{"title":"globe","isSVG":false,"mediaType":"Image","friendlyFileSize":"703 B","url":"https://webassets.rws.com/images/globe-1_tcm228-153333.svg","fileSize":703,"mimeType":"image/svg+xml"},"mobileImageValue":{"title":"globe","isSVG":false,"mediaType":"Image","friendlyFileSize":"703 B","url":"https://webassets.rws.com/images/globe-1_tcm228-153333.svg","fileSize":703,"mimeType":"image/svg+xml"},"mobileImage":{"title":"globe","isSVG":false,"mediaType":"Image","friendlyFileSize":"703 B","url":"https://webassets.rws.com/images/globe-1_tcm228-153333.svg","fileSize":703,"mimeType":"image/svg+xml"},"isVideo":false,"hasMobileImage":true},"links":[{"lang":"en","text":"English","messages":[{"message":{"title":"Unfortunately this page is not available in English","body":"

The page is however available in $available_languages_list.

You can also return to your previous page or go to the English language homepage.

","callToAction":{"url":"/","text":"English homepage","isExternal":false,"isVideo":false}},"variant":"variant 1"},{"message":{"title":"Unfortunately this page is not available in English","body":"

You can return to your previous page or go to the English language homepage.

","callToAction":{"url":"/","text":"English homepage","isExternal":false,"isVideo":false}},"variant":"variant 2"}],"isExternal":false,"isVideo":false},{"lang":"de","text":"Deutsch","messages":[{"message":{"title":"Diese Seite ist leider nicht auf Deutsch verfügbar.","body":"

Es gibt die Seite jedoch auf $available_languages_list.

Sie können auch zu Ihrer vorherigen Seite zurückkehren oder zur deutschen Homepage gehen.

","callToAction":{"url":"/de/","text":"Deutsche Homepage","isExternal":false,"isVideo":false}},"variant":"variant 1"},{"message":{"title":"Diese Seite ist leider nicht auf Deutsch verfügbar.","body":"

Sie stattdessen auch zu Ihrer vorherigen Seite zurückkehren oder zur deutschen Homepage gehen.

","callToAction":{"url":"/de/","text":"Deutsche Homepage","isExternal":false,"isVideo":false}},"variant":"variant 2"}],"isExternal":false,"isVideo":false},{"lang":"fr","text":"Français","messages":[{"message":{"title":"Malheureusement, cette page n'est pas disponible en français.","body":"

La page est toutefois disponible en $available_languages_list.

Vous pouvez également revenir à la page précédente ou accéder à la page d'accueil en français.

","callToAction":{"url":"/fr/","text":"Français homepage","isExternal":false,"isVideo":false}},"variant":"variant 1"},{"message":{"title":"Malheureusement, cette page n'est pas disponible en français.","body":"

Vous pouvez également revenir à la page précédente ou accéder à la page d'accueil en français.

","callToAction":{"url":"/fr/","text":"Français homepage","isExternal":false,"isVideo":false}},"variant":"variant 2"}],"isExternal":false,"isVideo":false},{"lang":"es","text":"Español","messages":[{"message":{"title":"Lamentablemente, esta página no está disponible en español.","body":"

Sin embargo, la página está disponible en $available_languages_list.

También puedes volver a la página anterior o ir a la página de inicio en español.

","callToAction":{"url":"/es/","text":"Español homepage","isExternal":false,"isVideo":false}},"variant":"variant 1"},{"message":{"title":"Lamentablemente, esta página no está disponible en español.","body":"

También puedes volver a la página anterior o ir a la página de inicio en español.

","callToAction":{"url":"/es/","text":"Español homepage","isExternal":false,"isVideo":false}},"variant":"variant 2"}],"isExternal":false,"isVideo":false},{"lang":"it","text":"Italiano","messages":[{"message":{"title":"Siamo spiacenti, questa pagina non è disponibile in italiano.","body":"

Tuttavia, la pagina è disponibile in $available_languages_list.

È possibile tornare alla pagina precedenteo andare nella homepage in lingua italiana.

","callToAction":{"url":"/it/","text":"Italiana homepage","isExternal":false,"isVideo":false}},"variant":"variant 1"},{"message":{"title":"Siamo spiacenti, questa pagina non è disponibile in italiano.","body":"

È possibile tornare alla pagina precedenteo andare nella homepage in lingua italiana.

\n\t","callToAction":{"url":"/it/","text":"Italiana homepage","isExternal":false,"isVideo":false}},"variant":"variant 2"}],"isExternal":false,"isVideo":false},{"lang":"cn","text":"简体中文","messages":[{"message":{"title":"Unfortunately this page is not available in Chinese.","body":"

The page is however available in $available_languages_list.

You can also return to your previous page or go to the Chinese language homepage.

","callToAction":{"url":"/cn/","text":"Chinese homepage","isExternal":false,"isVideo":false}},"variant":"variant 1"},{"message":{"title":"Unfortunately this page is not available in Chinese","body":"

You can return to your previous page or go to the Chinese language homepage.

\n\n","callToAction":{"url":"/cn/","text":"Chinese homepage","isExternal":false,"isVideo":false}},"variant":"variant 2"}],"isExternal":false,"isVideo":false},{"lang":"jp","text":"日本語","messages":[{"message":{"title":"Unfortunately this page is not available in Japanese","body":"

The page is however available in $available_languages_list.

You can also return to your previous page or go to the Japanese language homepage.

","callToAction":{"url":"/jp/","text":"Japanese homepage","isExternal":false,"isVideo":false}},"variant":"variant 1"},{"message":{"title":"Unfortunately this page is not available in Japanese","body":"

You can return to your previous page or go to the Japanese language homepage.

","callToAction":{"url":"/jp/","text":"Japanese homepage","isExternal":false,"isVideo":false}},"variant":"variant 2"}],"isExternal":false,"isVideo":false},{"lang":"kr","text":"한국어","messages":[{"message":{"title":"Unfortunately this page is not available in Korean","body":"

The page is however available in $available_languages_list.

You can also return to your previous page or go to the Korean language homepage.

","callToAction":{"url":"/kr/","text":"Korean homepage","isExternal":false,"isVideo":false}},"variant":"variant 1"},{"message":{"title":"Unfortunately this page is not available in Korean","body":"

You can return to your previous page or go to the Korean language homepage.

","callToAction":{"url":"/kr/","text":"Korean homepage","isExternal":false,"isVideo":false}},"variant":"variant 2"}],"isExternal":false,"isVideo":false}]},"id":"201944"},"excludeCustomNavigation":false,"excludeMainNavigation":true}},"regions":[{"regions":[],"fragments":[{"content":{"title":"Language Weaver","type":"Default","logo":{"image":{"friendlyFileSize":"412 B","title":"RWS Icon - Large","isSVG":false,"mediaType":"Image","url":"https://webassets.rws.com/images/icon-large-1_tcm228-194189.png","fileSize":412,"mimeType":"image/png"},"mobileImage":{"title":"RWS Icon - Large","isSVG":false,"mediaType":"Image","friendlyFileSize":"412 B","url":"https://webassets.rws.com/images/icon-large-1_tcm228-194189.png","fileSize":412,"mimeType":"image/png"},"alternateText":"RWS Icon - Large","hasMobileImage":false},"links":[{"link":{"text":"Find a solution","isExternal":false,"isVideo":false},"subLinks":[{"link":{"text":"I'm interested in…","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/language-weaver/neural-machine-translation/","internalLink":{},"text":"Neural machine translation","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/government-solutions/","internalLink":{},"text":"Multilingual content intelligence","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/global-enterprise/","internalLink":{},"text":"Making customer support multilingual","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/legal-and-compliance/","internalLink":{},"text":"Enabling Multilingual e-Discovery","isExternal":false,"isVideo":false}}]},{"link":{"text":"By products","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/language-weaver/cloud/","internalLink":{},"text":"Cloud","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/edge/","internalLink":{},"text":"Edge","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/connectors/","internalLink":{},"text":"Connector Hub","isExternal":false,"isVideo":false}}]},{"link":{"text":"By use case","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/language-weaver/global-enterprise/","internalLink":{},"text":"Enterprise","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/legal-and-compliance/","internalLink":{},"text":"Legal","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/government-solutions/","internalLink":{},"text":"Government","isExternal":false,"isVideo":false}}]}]},{"link":{"text":"Service and Support","isExternal":false,"isVideo":false},"subLinks":[{"link":{"text":"Expert Services","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/partners/","internalLink":{},"text":"Partners","isExternal":false,"isVideo":false}},{"link":{"url":"/partners/list/","internalLink":{},"text":"Partner Directory","isExternal":false,"isVideo":false}},{"link":{"url":"/support/professional-services/","internalLink":{},"text":"Professional Services","isExternal":false,"isVideo":false}},{"link":{"url":"https://www.rws.com/support/training/?productSolution=LanguageWeaver","text":"Training ","isExternal":true,"isVideo":false}}]},{"link":{"text":"Support","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"https://gateway.sdl.com/CommunityHome","text":"Support Gateway","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"https://community.rws.com/product-groups/linguistic-ai/cloud/","text":"Community ","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"/support/newsletter-signup/","internalLink":{},"text":"Sign up for support updates","isExternal":false,"isVideo":false}}]},{"link":{"text":"Developers","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"https://developers.languageweaver.com/","text":"Developer portal","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"https://docs.rws.com/791593/865454/language-weaver/welcome-to-the-------------language-weaver-documentation","text":"Cloud documentation","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"https://docs.rws.com/833475/886777/language-weaver-edge-documentation/welcome-to-the-------------language-weaver-edge-------------8-5-documentation","text":"Edge documentation","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}}]}]},{"link":{"text":"Resources","isExternal":false,"isVideo":false},"subLinks":[{"link":{"text":"Recommended ","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/language-weaver/events/language-weaver-webinar-series/","internalLink":{},"text":"Webinar Series","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/resources/us-army/","internalLink":{},"text":"U.S. Army case study","isExternal":false,"isVideo":false}},{"link":{"url":"https://webassets.rws.com/images/Language-Weaver-CAS-Case-Study_tcm228-191164.pdf","internalLink":{},"text":"Chemical Abstracts Service case study","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"/language-weaver/resources/machine-translation-cloud/","internalLink":{},"text":"Cloud Data Sheet","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/resources/machine-translation-edge/","internalLink":{},"text":"Edge Data Sheet","isExternal":false,"isVideo":false}}]},{"link":{"text":"Collateral","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=case-study-long","text":"Case Studies","isExternal":true,"isVideo":false}},{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=white-paper","text":"White Papers","isExternal":true,"isVideo":false}},{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=ebooks","text":"eBooks","isExternal":true,"isVideo":false}},{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=webinar-recording","text":"Recorded Webinars","isExternal":true,"isVideo":false}},{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=Brochure","text":"Brochures","isExternal":true,"isVideo":false}},{"link":{"url":"https://www.rws.com/language-weaver/resources/?resourceType=data-sheet","text":"Data Sheets","isExternal":true,"isVideo":false}},{"link":{"url":"/language-weaver/resources/","internalLink":{},"text":"View all","isExternal":false,"isVideo":false}}]},{"link":{"text":"Learn more","isExternal":false,"isVideo":false},"subLinks":[{"link":{"url":"/language-weaver/our-history/","internalLink":{},"text":"History","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/neural-machine-translation/","internalLink":{},"text":"Neural MT","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/the-neural-mt-weekly/","internalLink":{},"text":"Neural MT Weekly","isExternal":false,"isVideo":false}},{"link":{"url":"/language-weaver/supported-languages/","internalLink":{},"text":"Supported Languages ","isExternal":false,"isVideo":false}}]}]},{"link":{"url":"/language-weaver/research/","internalLink":{},"text":"Research","isExternal":false,"isVideo":false}},{"link":{"url":"https://portal.languageweaver.com/","text":"Get Started","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}},{"link":{"url":"/language-weaver/contact/","internalLink":{},"text":"Contact","isExternal":false,"isVideo":false}}],"mainLogo":{"image":{"title":"SDL - Part of the RWS Group","isSVG":false,"mediaType":"Image","friendlyFileSize":"2 KB","url":"https://webassets.rws.com/images/Artboard-1_tcm228-187294.svg","fileSize":1113,"mimeType":"image/svg+xml"},"mobileImage":{"title":"SDL - Part of the RWS Group","isSVG":false,"mediaType":"Image","friendlyFileSize":"2 KB","url":"https://webassets.rws.com/images/Artboard-1_tcm228-187294.svg","fileSize":1113,"mimeType":"image/svg+xml"},"alternateText":"SDL - Part of the RWS Group","hasMobileImage":false},"homeLink":{"url":"/","internalLink":{},"isExternal":false,"isVideo":false},"sectionHomeLink":{"url":"/language-weaver/","internalLink":{},"isExternal":false,"isVideo":false},"utilityNav":{"icon":{"desktopImage":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"375 B","url":"https://webassets.rws.com/images/search-1_tcm228-167958.svg","fileSize":375,"mimeType":"image/svg+xml"},"desktopImagePortrait":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"375 B","url":"https://webassets.rws.com/images/search-1_tcm228-167958.svg","fileSize":375,"mimeType":"image/svg+xml"},"mobileImage":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"375 B","url":"https://webassets.rws.com/images/search-1_tcm228-167958.svg","fileSize":375,"mimeType":"image/svg+xml"},"isVideo":false,"hasMobileImage":false},"links":[{"url":"/search/","internalLink":{},"isExternal":false,"isVideo":false}]},"languageSelector":{"icon":{"desktopImage":{"title":"globe","isSVG":false,"mediaType":"Image","friendlyFileSize":"703 B","url":"https://webassets.rws.com/images/globe-1_tcm228-153333.svg","fileSize":703,"mimeType":"image/svg+xml"},"desktopImagePortrait":{"title":"globe","isSVG":false,"mediaType":"Image","friendlyFileSize":"703 B","url":"https://webassets.rws.com/images/globe-1_tcm228-153333.svg","fileSize":703,"mimeType":"image/svg+xml"},"mobileImageValue":{"title":"globe","isSVG":false,"mediaType":"Image","friendlyFileSize":"703 B","url":"https://webassets.rws.com/images/globe-1_tcm228-153333.svg","fileSize":703,"mimeType":"image/svg+xml"},"mobileImage":{"title":"globe","isSVG":false,"mediaType":"Image","friendlyFileSize":"703 B","url":"https://webassets.rws.com/images/globe-1_tcm228-153333.svg","fileSize":703,"mimeType":"image/svg+xml"},"isVideo":false,"hasMobileImage":true},"links":[{"url":"/language-weaver/the-neural-mt-weekly/issue-126-learning-feature-weights-for-denoising-parallel-corpora/","text":"English","isExternal":true,"selected":true,"lang":"en","isAvailable":true,"noContentMessage":{"title":"Unfortunately this page is not available in English","body":"

You can return to your previous page or go to the English language homepage.

","callToAction":{"url":"/","text":"English homepage","isExternal":false,"isVideo":false}}},{"url":"","text":"Deutsch","isExternal":true,"selected":false,"lang":"de","isAvailable":false,"noContentMessage":{"title":"Diese Seite ist leider nicht auf Deutsch verfügbar.","body":"

Sie stattdessen auch zu Ihrer vorherigen Seite zurückkehren oder zur deutschen Homepage gehen.

","callToAction":{"url":"/de/","text":"Deutsche Homepage","isExternal":false,"isVideo":false}}},{"url":"","text":"Français","isExternal":true,"selected":false,"lang":"fr","isAvailable":false,"noContentMessage":{"title":"Malheureusement, cette page n'est pas disponible en français.","body":"

Vous pouvez également revenir à la page précédente ou accéder à la page d'accueil en français.

","callToAction":{"url":"/fr/","text":"Français homepage","isExternal":false,"isVideo":false}}},{"url":"","text":"Español","isExternal":true,"selected":false,"lang":"es","isAvailable":false,"noContentMessage":{"title":"Lamentablemente, esta página no está disponible en español.","body":"

También puedes volver a la página anterior o ir a la página de inicio en español.

","callToAction":{"url":"/es/","text":"Español homepage","isExternal":false,"isVideo":false}}},{"url":"","text":"Italiano","isExternal":true,"selected":false,"lang":"it","isAvailable":false,"noContentMessage":{"title":"Siamo spiacenti, questa pagina non è disponibile in italiano.","body":"

È possibile tornare alla pagina precedenteo andare nella homepage in lingua italiana.

\n\t","callToAction":{"url":"/it/","text":"Italiana homepage","isExternal":false,"isVideo":false}}},{"url":"","text":"简体中文","isExternal":true,"selected":false,"lang":"cn","isAvailable":false,"noContentMessage":{"title":"Unfortunately this page is not available in Chinese","body":"

You can return to your previous page or go to the Chinese language homepage.

\n\n","callToAction":{"url":"/cn/","text":"Chinese homepage","isExternal":false,"isVideo":false}}},{"url":"","text":"日本語","isExternal":true,"selected":false,"lang":"jp","isAvailable":false,"noContentMessage":{"title":"Unfortunately this page is not available in Japanese","body":"

You can return to your previous page or go to the Japanese language homepage.

","callToAction":{"url":"/jp/","text":"Japanese homepage","isExternal":false,"isVideo":false}}},{"url":"","text":"한국어","isExternal":true,"selected":false,"lang":"kr","isAvailable":false,"noContentMessage":{"title":"Unfortunately this page is not available in Korean","body":"

You can return to your previous page or go to the Korean language homepage.

","callToAction":{"url":"/kr/","text":"Korean homepage","isExternal":false,"isVideo":false}}}]},"id":"201944"},"template":{"content":{},"id":"CustomNavigation","extendedProperties":{}},"id":"1000","extendedProperties":{}},{"content":{"links":[{"url":"/","text":"Home","alternateText":"Home","isExternal":false},{"url":"/language-weaver/","text":"Language Weaver","alternateText":"Language Weaver","isExternal":false},{"url":"/language-weaver/the-neural-mt-weekly/","text":"Neural MT Weekly","alternateText":"Neural MT Weekly","isExternal":false},{"url":"/language-weaver/the-neural-mt-weekly/issue-126-learning-feature-weights-for-denoising-parallel-corpora/","text":"Article","alternateText":"Article","isExternal":false}]},"template":{"content":{},"id":"Breadcrumb","extendedProperties":{}},"id":"153341","extendedProperties":{}},{"content":{"id":"191905"},"template":{"content":{},"id":"NonRenderingComponentLink","extendedProperties":{}},"id":"191905","extendedProperties":{}},{"content":{"title":"Issue #126 - Learning Feature Weights for Denoising Parallel Corpora","date":"04/15/2021","image":{"image":{"title":"Data creation","isSVG":false,"mediaType":"Image","friendlyFileSize":"70 KB","url":"https://webassets.rws.com/images/data-creation-1200x610_tcm228-195130.jpg","fileSize":71678,"mimeType":"image/jpeg"},"alternateTextValue":"The topic of this blog post is data creation.","mobileImage":{"title":"Data creation","isSVG":false,"mediaType":"Image","friendlyFileSize":"70 KB","url":"https://webassets.rws.com/images/data-creation-1200x610_tcm228-195130.jpg","fileSize":71678,"mimeType":"image/jpeg"},"alternateText":"The topic of this blog post is data creation.","hasMobileImage":false},"thumbnailImage":{"image":{"title":"Data creation","isSVG":false,"mediaType":"Image","friendlyFileSize":"70 KB","url":"https://webassets.rws.com/images/data-creation-1200x610_tcm228-195130.jpg","fileSize":71678,"mimeType":"image/jpeg"},"alternateTextValue":"The topic of this blog post is data creation.","mobileImage":{"title":"Data creation","isSVG":false,"mediaType":"Image","friendlyFileSize":"70 KB","url":"https://webassets.rws.com/images/data-creation-1200x610_tcm228-195130.jpg","fileSize":71678,"mimeType":"image/jpeg"},"alternateText":"The topic of this blog post is data creation.","hasMobileImage":false},"author":{"fullName":"Dr. Patrik Lambert","position":" Senior Machine Translation Scientist","description":"

Patrik conducts research on and builds high-quality customized machine translation engines, proposes and develops improved approaches to the company's machine translation software, and provides support to other team members.

He received a master in Physics from McGill University. Then he worked for several years as technical translator and as software developer. He completed in 2008 a PhD in Artificial Intelligence at the Polytechnic University of Catalonia (UPC, Spain). He then worked as research associate on machine translation and cross-lingual sentiment analysis.

","image":{"image":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"9 MB","url":"https://webassets.rws.com/images/patriklmbert_tcm228-193200.jpg","fileSize":9153345,"mimeType":"image/jpeg"},"alternateTextValue":"Dr. Patrik Lambert","mobileImage":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"9 MB","url":"https://webassets.rws.com/images/patriklmbert_tcm228-193200.jpg","fileSize":9153345,"mimeType":"image/jpeg"},"alternateText":"Dr. Patrik Lambert","hasMobileImage":false},"url":"/language-weaver/the-neural-mt-weekly/author/drpatriklambert"},"topic":{"translatedTitle":"Neural MT Weekly","title":"Neural MT Weekly","description":"Neural MT Weekly","key":"Neural MT Weekly"},"tags":[{"translatedTitle":"Data creation","title":"Data creation","description":"Data creation","key":"data-creation"},{"translatedTitle":"Model improvement","title":"Model improvement","description":"Model improvement","key":"model-improvement"}],"displayDate":"15 Apr 2021","resources":{"topicsLabel":"Category","tagsLabel":"Tags"}},"template":{"content":{},"id":"BlogHero","extendedProperties":{}},"id":"191903","extendedProperties":{}},{"content":{"description":"

Introduction

Large web-crawled parallel corpora constitute a very useful source of data to improve neural machine translation (NMT) engines. However, their effectiveness is reduced by the large amount of noise they usually contain. As early as in issue #2 of this series, we pointed out that NMT is particularly sensitive to noise in the training data. In issue #21, we presented the dual cross-entropy method to filter out noise in parallel corpora (Junczys-Dowmunt, 2018). In this post, we take a look at a paper by Kumar et al. (2021), which goes a step further and proposes a denoising method combining several sentence-level features with weights learned using reward modeling.

Method

The method focuses on learning weights for combining the following sentence-level features:

IBM Model 1 (basic word alignment model),
source language model,
target language model,
dual cross-entropy and sentence length ratio.

The aim of training is to evaluate a reward function R(w|Φ), where w are the weights and Φ is the sentence-level feature vector averaged over a batch of sentences. The optimum weight vector is the one maximizing the reward.

The reward is chosen in a way allowing a direct relation between the value of w and translation performance. This is carried out in an NMT training loop. To sample each batch, the features are calculated for a number of sentence pairs, and the best scoring ones according to a random weight vector are selected. The reward is related to the perplexity of a validation set following a neural network parameter update with the selected batch. The random weight vector, the average feature vector over the batch and the reward calculated in this parameter update constitute one training sample to learn the weight vector. These steps are repeated for a number of batches and the whole procedure is repeated with different candidate NMT runs.

After having collected enough (weight vector, feature vector, reward) training samples, a simple feed-forward neural network is learned. This neural network maps the input weights (possibly modulated by the batch quality, represented by the feature vector) to the observed reward. The variant with the raw weight vectors as input is referred to as “Weight based”, while the variant in which the weight vectors are scaled by the sum of the corresponding feature vector is referred to as “Feature based”. Once this neural network is learned, a grid search over the weights is performed to find the best weight vector.

The final step consists of calculating the sentence-level features for each sentence-pair in the corpus, and using the best weight vector to score it. The corpus can then be filtered based on this score.

Experiments

The experiments are performed on the Estonian–English Paracrawl corpus. The Estonian and English language models were trained on their respective NewsCrawl data sets. To compute the dual cross-entropy scores, machine translation models were trained on clean data, namely the Europarl-v8 set. The best weight vector obtained was: dual cross-entropy: 0.81, IBM Model 1: 0.07, Sentence length ratio: 0.07, source language model: 0.03, and target language model: 0.02. These weights were used to score each sentence pair, and three sets, with the best 10, 15 and 20 million (M) English words, were selected. The results are shown in the following table. $\"NMT$ The table first shows the BLEU score corresponding to the sets obtained by filtering the corpus with a number of baseline methods. Zipporah, BiCleaner and LASER are three filtering toolkits. The best one in terms of BLEU score is LASER, which is based on pre-trained multilingual embeddings (it could actually be interesting to include LASER as a feature). The rest of the baselines are the individual features. The best scoring one is the dual cross-entropy, which performs similarly as LASER. The very simple sentence length ratio performs surprisingly well, and the monolingual features (language models) perform the worst. The combination of features with uniform weights is worse than the dual cross-entropy. However, with tuned weights, the combination performs between 0.7 to 1 BLEU point better than the best baseline. Scaling the input weights by the sum of the corresponding feature vector (Feature based variant) gives a small improvement. The number in parentheses (14 and 15) are the number of candidate NMT runs for which the best result was achieved.

Finally, the authors try to transfer the best weight vector to another language pair (Maltese-English paracrawl). However, in that case the combination of features is not better than the dual cross-entropy. This suggests that learning the weight vector is necessary.

In summary

\nKumar et al. (2021) propose a parallel corpus filtering (denoising) method which combines the dual cross-entropy feature with four other sentence-level features. The feature weights in the combination are learned based on reward modeling. The training samples are generated in an NMT training loop, with an NMT quality objective related to the perplexity on a validation set. Filtering the corpus based on the feature combination yields slightly better BLEU scores than based on the best baselines, which are the LASER and dual cross-entropy methods. Unfortunately, the best weights learned in one language pair could not be transferred to a different language pair. This method achieves top results, but the calculation of some of the features is slow, as is the process to collect training samples to tune the weights."},"template":{"content":{},"id":"TextBlockEx","extendedProperties":{}},"id":"191904","extendedProperties":{}},{"content":{"topic":{"translatedTitle":"Neural MT Weekly","title":"Neural MT Weekly","description":"Neural MT Weekly","key":"Neural MT Weekly","count":0},"tags":[{"translatedTitle":"Data creation","title":"Data creation","description":"Data creation","key":"data-creation","count":0},{"translatedTitle":"Model improvement","title":"Model improvement","description":"Model improvement","key":"model-improvement","count":0}],"blogTopicUrl":"/blog/categories","blogTagUrl":"/blog/tags","resources":{"topicsLabel":"Category","tagsLabel":"Tags"}},"template":{"content":{},"id":"BlogTags","extendedProperties":{}},"id":"1802200285","extendedProperties":{}},{"content":{"fullName":"Dr. Patrik Lambert","position":" Senior Machine Translation Scientist","description":"

","image":{"image":{"url":"https://webassets.rws.com/images/patriklmbert_tcm228-193200.jpg","fileSize":9153345,"mimeType":"image/jpeg","mediaType":"Image","friendlyFileSize":"9 MB"},"mobileImage":{"url":"https://webassets.rws.com/images/patriklmbert_tcm228-193200.jpg","fileSize":9153345,"mimeType":"image/jpeg","mediaType":"Image","friendlyFileSize":"9 MB"},"alternateText":"Dr. Patrik Lambert"},"url":null,"link":{"title":"All from Dr. Patrik Lambert","url":"/language-weaver/the-neural-mt-weekly/author/drpatriklambert"}},"template":{"content":{},"id":"AuthorBlock","extendedProperties":{}},"id":"1802200285","extendedProperties":{}},{"content":{"marketoForm":{"formId":"6ddd9314-e2d7-40c2-ac8c-fa27435acd3d","formProvider":"HubSpot"},"title":"Subscribe","portalId":"147991"},"template":{"content":{},"id":"BlogSubscription","extendedProperties":{}},"id":"1802200285","extendedProperties":{}},{"content":{"dynamicResources":{"totalItems":183,"take":6,"currentPage":0,"numberOfPages":31,"results":[{"articleTitle":"Issue #182 - DYLE: Dynamic Latent Extraction for Abstractive Long-Input Summarization ","subTitle":"Kyle looks at ACL 2022 paper by Mao et al. (2022) “DYLE: Dynamic Latent Extraction for Abstractive Long-Input Summarization”, where they present a technique for abstractive summarization of long documents based on the “extract-then-generate” framework. ","date":"05/26/2022","readTime":"7 minute read","image":{"image":{"url":"https://webassets.rws.com/images/other-1200x610_tcm228-195137.jpg","fileSize":0,"id":"195137"},"mobileImage":{"url":"https://webassets.rws.com/images/other-1200x610_tcm228-195137.jpg","fileSize":0,"id":"195137"}},"author":"Kyle Shaffer","topic":"Neural MT Weekly","tags":["other-miscellaneous"],"tagTranslations":["Other / Miscellaneous"],"dateMonth":"5","dateYear":"2022","displayDate":"26 May 2022","id":"tcm:228-205106-64","linkUrl":{"url":"/language-weaver/the-neural-mt-weekly/issue-182-DYLE/","isExternal":false,"isVideo":false},"publicationDate":"2022-06-30T16:05:56.001Z","title":"Issue #182 - DYLE: Dynamic Latent Extraction for Abstractive Long-Input Summarization","publicationId":"228","itemType":"blog","itemTypeTranslation":"Blog"},{"articleTitle":"Issue #181 - Incorporating Source text in Automatic Post-Editing ","subTitle":"Karen looks at work by Pal et al., 2019, which incorporates the source text in an Automatic Post-Editing (APE) architecture, and also looks at how Wang et al., 2021 apply it to train a context-aware APE system.","date":"05/19/2022","readTime":"5 minute read","image":{"image":{"url":"https://webassets.rws.com/images/other-1200x610_tcm228-195137.jpg","fileSize":0,"id":"195137"},"mobileImage":{"url":"https://webassets.rws.com/images/other-1200x610_tcm228-195137.jpg","fileSize":0,"id":"195137"}},"author":"Dr. Karin Sim","topic":"Neural MT Weekly","tags":["other-miscellaneous"],"tagTranslations":["Other / Miscellaneous"],"dateMonth":"5","dateYear":"2022","displayDate":"19 May 2022","id":"tcm:228-204729-64","linkUrl":{"url":"/language-weaver/the-neural-mt-weekly/issue-181-incorporating-source-text-in-APE/","isExternal":false,"isVideo":false},"publicationDate":"2022-06-30T16:05:55.488Z","title":"Issue #181 - Incorporating Source text in Automatic Post-Editing","publicationId":"228","itemType":"blog","itemTypeTranslation":"Blog"},{"articleTitle":"Issue #180 - Detecting Over- and Under-translations with Contrastive Conditioning ","subTitle":"This week, Danielle looks at work by Vamvas & Sennrich (2022) who explore two ways to evaluate MT for over- and under-translation (where a translation has either added or omitted information). ","date":"05/12/2022","readTime":"7 minute read","image":{"image":{"url":"https://webassets.rws.com/images/evaluation-1200x610_tcm228-195132.jpg","fileSize":0,"id":"195132"},"mobileImage":{"url":"https://webassets.rws.com/images/evaluation-1200x610_tcm228-195132.jpg","fileSize":0,"id":"195132"}},"author":"Dr. Danielle Saunders","topic":"Neural MT Weekly","tags":["evaluation"],"tagTranslations":["Evaluation"],"dateMonth":"5","dateYear":"2022","displayDate":"12 May 2022","id":"tcm:228-204453-64","linkUrl":{"url":"/language-weaver/the-neural-mt-weekly/issue-180-detecting-over-and-under-translations/","isExternal":false,"isVideo":false},"publicationDate":"2022-06-30T16:05:55.036Z","title":"Issue #180 - Detecting Over- and Under-translations with Contrastive Conditioning","publicationId":"228","itemType":"blog","itemTypeTranslation":"Blog"},{"articleTitle":"Issue #179 - Charformer: Fast Character Transformers via Gradient-based Subwords","subTitle":"This week, Patrik looks at the Charformer model, proposed by Tay et al. (2022), in which subword tokenization is learned as a gradient-based, end-to-end task. ","date":"05/05/2022","readTime":"5 minute read","image":{"image":{"url":"https://webassets.rws.com/images/model-improvement-1200x610_tcm228-195135.jpg","fileSize":0,"id":"195135"},"mobileImage":{"url":"https://webassets.rws.com/images/model-improvement-1200x610_tcm228-195135.jpg","fileSize":0,"id":"195135"}},"author":"Dr. Patrik Lambert","topic":"Neural MT Weekly","tags":["model-improvement","robustness"],"tagTranslations":["Model improvement","Robustness"],"dateMonth":"5","dateYear":"2022","displayDate":"05 May 2022","id":"tcm:228-204336-64","linkUrl":{"url":"/language-weaver/the-neural-mt-weekly/Issue-179-Charformer-model/","isExternal":false,"isVideo":false},"publicationDate":"2022-06-30T16:05:53.436Z","title":"Issue #179 - Charformer: Fast Character Transformers via Gradient-based Subwords","publicationId":"228","itemType":"blog","itemTypeTranslation":"Blog"},{"articleTitle":"Issue #178 - Length-based Overfitting in Transformer Models","subTitle":"This week, Jingyi reviews a paper by Variš and Bojar (2021) which suggests that the reason neural-based MT systems normally perform worse on very long sequences might be due to the mismatch between the length distributions of training data and evaluation set.","date":"04/28/2022","readTime":"7 minute read","image":{"image":{"url":"https://webassets.rws.com/images/model-improvement-1200x610_tcm228-195135.jpg","fileSize":0,"id":"195135"},"mobileImage":{"url":"https://webassets.rws.com/images/model-improvement-1200x610_tcm228-195135.jpg","fileSize":0,"id":"195135"}},"author":"Dr. Jingyi Han","topic":"Neural MT Weekly","tags":["model-improvement"],"tagTranslations":["Model improvement"],"dateMonth":"4","dateYear":"2022","displayDate":"28 Apr 2022","id":"tcm:228-204180-64","linkUrl":{"url":"/language-weaver/the-neural-mt-weekly/Issue-178-Length-based-Overfitting-in-Transformer/","isExternal":false,"isVideo":false},"publicationDate":"2022-06-30T16:05:53.042Z","title":"Issue #178 - Length-based Overfitting in Transformer Models","publicationId":"228","itemType":"blog","itemTypeTranslation":"Blog"},{"articleTitle":"Issue #177 - Parallel Corpus Filtering via Pre-trained Language Models ","subTitle":"In today’s blog post, we will take a look at the work of Zhang et al. (2020) who propose a novel approach for parallel corpus filtering. ","date":"04/21/2022","readTime":"7 minute read","image":{"image":{"url":"https://webassets.rws.com/images/other-1200x610_tcm228-195137.jpg","fileSize":0,"id":"195137"},"mobileImage":{"url":"https://webassets.rws.com/images/other-1200x610_tcm228-195137.jpg","fileSize":0,"id":"195137"}},"author":"Akshai Ramesh","topic":"Neural MT Weekly","tags":["other-miscellaneous"],"tagTranslations":["Other / Miscellaneous"],"dateMonth":"4","dateYear":"2022","displayDate":"21 Apr 2022","id":"tcm:228-203923-64","linkUrl":{"url":"/language-weaver/the-neural-mt-weekly/Issue-177-Parallel-Corpus-Filtering-LMs/","isExternal":false,"isVideo":false},"publicationDate":"2022-06-30T16:05:52.316Z","title":"Issue #177 - Parallel Corpus Filtering via Pre-trained Language Models","publicationId":"228","itemType":"blog","itemTypeTranslation":"Blog"}]},"resources":{"relatedArticles":"Related Articles"},"blogTopicUrl":"/blog/categories","blogTagUrl":"/blog/tags"},"template":{"content":{},"id":"RelatedArticlesBlock","extendedProperties":{}},"id":"1802200285","extendedProperties":{}}],"id":"Main","extendedProperties":{}},{"regions":[{"regions":[],"fragments":[],"id":"Header","extendedProperties":{}}],"fragments":[],"id":"Header","extendedProperties":{}},{"regions":[{"regions":[],"fragments":[{"content":{"linkList":[{"title":"RWS","description":"\n

\n ","links":[{"url":"/about/careers/","internalLink":{},"text":"Careers","isExternal":false,"isVideo":false},{"url":"/contact/","internalLink":{},"text":"Contact us","isExternal":false,"isVideo":false},{"url":"/about/offices/","internalLink":{},"text":"Offices","isExternal":false,"isVideo":false},{"url":"/about/","internalLink":{},"text":"About","isExternal":false,"isVideo":false},{"url":"/about/corporate-sustainability/","internalLink":{},"text":"Sustainability","isExternal":false,"isVideo":false}]},{"title":"Investors","links":[{"url":"/about/investors/","internalLink":{},"text":"Overview","isExternal":false,"isVideo":false},{"url":"/about/investors/investment-proposition/","internalLink":{},"text":"Investment proposition","isExternal":false,"isVideo":false},{"url":"/about/investors/directors/","internalLink":{},"text":"Directors","isExternal":false,"isVideo":false},{"url":"/about/executive-team/","internalLink":{},"text":"Executive team","isExternal":false,"isVideo":false},{"url":"/about/investors/rns-announcements/","internalLink":{},"text":"RNS announcements","isExternal":false,"isVideo":false},{"url":"/about/investors/results-and-reports/","internalLink":{},"text":"Results, reports and presentations","isExternal":false,"isVideo":false}]},{"title":"What we do","links":[{"url":"/localization/","internalLink":{},"text":"Localization","isExternal":false,"isVideo":false},{"url":"/content/","internalLink":{},"text":"Content","isExternal":false,"isVideo":false},{"url":"/artificial-intelligence/","internalLink":{},"text":"Artificial Intelligence","isExternal":false,"isVideo":false},{"url":"/intellectual-property-services/","internalLink":{},"text":"IP Services","isExternal":false,"isVideo":false},{"url":"/industries/","internalLink":{},"text":"Industries","isExternal":false,"isVideo":false}]}],"promo":{"title":"Communication preferences","description":"Looking for more information on RWS? Sign up with our preference centre, and tell us your marketing communication preferences.","link":{"url":"/about/preference-centre/","internalLink":{},"text":"Sign up ","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":false,"isVideo":false}},"socialLinks":{"title":"Follow us","links":[{"url":"https://twitter.com/RWSGroup","text":"Twitter","icon":{"desktopImage":{"title":"Twitter icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"499 B","url":"https://webassets.rws.com/images/twitter_tcm228-153337.svg","fileSize":499,"mimeType":"image/svg+xml"},"desktopImagePortrait":{"title":"Twitter icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"499 B","url":"https://webassets.rws.com/images/twitter_tcm228-153337.svg","fileSize":499,"mimeType":"image/svg+xml"},"mobileImageValue":{"title":"Twitter icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"499 B","url":"https://webassets.rws.com/images/twitter_tcm228-153337.svg","fileSize":499,"mimeType":"image/svg+xml"},"mobileImage":{"title":"Twitter icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"499 B","url":"https://webassets.rws.com/images/twitter_tcm228-153337.svg","fileSize":499,"mimeType":"image/svg+xml"},"isVideo":false,"hasMobileImage":true},"target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false},{"url":"https://www.linkedin.com/company/rws-group","text":"LinkedIn","icon":{"desktopImage":{"title":"Linkedin icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"491 B","url":"https://webassets.rws.com/images/linkedin_tcm228-153329.svg","fileSize":491,"mimeType":"image/svg+xml"},"desktopImagePortrait":{"title":"Linkedin icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"491 B","url":"https://webassets.rws.com/images/linkedin_tcm228-153329.svg","fileSize":491,"mimeType":"image/svg+xml"},"mobileImageValue":{"title":"Linkedin icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"491 B","url":"https://webassets.rws.com/images/linkedin_tcm228-153329.svg","fileSize":491,"mimeType":"image/svg+xml"},"mobileImage":{"title":"Linkedin icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"491 B","url":"https://webassets.rws.com/images/linkedin_tcm228-153329.svg","fileSize":491,"mimeType":"image/svg+xml"},"isVideo":false,"hasMobileImage":true},"target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false},{"url":"https://www.facebook.com/TheRWSGroup","text":"Facebook","icon":{"desktopImage":{"title":"facebook icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"357 B","url":"https://webassets.rws.com/images/facebook_tcm228-153331.svg","fileSize":357,"mimeType":"image/svg+xml"},"desktopImagePortrait":{"title":"facebook icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"357 B","url":"https://webassets.rws.com/images/facebook_tcm228-153331.svg","fileSize":357,"mimeType":"image/svg+xml"},"mobileImage":{"title":"facebook icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"357 B","url":"https://webassets.rws.com/images/facebook_tcm228-153331.svg","fileSize":357,"mimeType":"image/svg+xml"},"isVideo":false,"hasMobileImage":false},"target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false},{"url":"https://www.youtube.com/c/RWSGroupPLC","text":"YouTube","icon":{"desktopImage":{"title":"Youtube icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"405 B","url":"https://webassets.rws.com/images/youtube_tcm228-153328.svg","fileSize":405,"mimeType":"image/svg+xml"},"desktopImagePortrait":{"title":"Youtube icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"405 B","url":"https://webassets.rws.com/images/youtube_tcm228-153328.svg","fileSize":405,"mimeType":"image/svg+xml"},"mobileImage":{"title":"Youtube icon","isSVG":false,"mediaType":"Image","friendlyFileSize":"405 B","url":"https://webassets.rws.com/images/youtube_tcm228-153328.svg","fileSize":405,"mimeType":"image/svg+xml"},"isVideo":false,"hasMobileImage":false},"target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false},{"url":"https://www.instagram.com/RWS.group/","icon":{"desktopImage":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"784 B","url":"https://webassets.rws.com/images/instagram_tcm228-197748.svg","fileSize":784,"mimeType":"image/svg+xml"},"desktopImagePortrait":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"784 B","url":"https://webassets.rws.com/images/instagram_tcm228-197748.svg","fileSize":784,"mimeType":"image/svg+xml"},"mobileImage":{"isSVG":false,"mediaType":"Image","friendlyFileSize":"784 B","url":"https://webassets.rws.com/images/instagram_tcm228-197748.svg","fileSize":784,"mimeType":"image/svg+xml"},"isVideo":false,"hasMobileImage":false},"target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":true,"isVideo":false}]},"utilityLinks":{"links":[{"url":"/legal/terms-and-conditions/","internalLink":{},"text":"Our Terms of Use","isExternal":false,"isVideo":false},{"url":"/legal/copyright/","internalLink":{},"text":"Copyright ","isExternal":false,"isVideo":false},{"url":"/legal/privacy/","internalLink":{},"text":"Privacy","isExternal":false,"isVideo":false},{"url":"/legal/security/","internalLink":{},"text":"Security","isExternal":false,"isVideo":false},{"url":"/legal/modern-slavery-statement/","internalLink":{},"text":"Modern Slavery and Human Trafficking Statement","isExternal":false,"isVideo":false},{"url":"/legal/privacy/cookie-notice/","internalLink":{},"text":"Cookie Notice","target":{"title":"New Window","description":"New Window","key":"_blank"},"isExternal":false,"isVideo":false}]},"cookiePreferencesCta":"Cookie Preferences"},"template":{"content":{},"id":"OneTrustSiteFooter","extendedProperties":{}},"id":"153373","extendedProperties":{}}],"id":"Footer","extendedProperties":{}}],"fragments":[],"id":"Footer","extendedProperties":{}},{"regions":[{"regions":[],"fragments":[{"content":{"headScripts":[{"codeText":"","id":"178027"}],"bodyScripts":[{"codeText":"","id":"204655"},{"codeText":"","id":"204698"},{"codeText":" \n","id":"190221"}]},"template":{"content":{},"id":"GlobalScripts","extendedProperties":{}},"id":"178028","extendedProperties":{}}],"id":"Scripts","extendedProperties":{}}],"fragments":[],"id":"Scripts","extendedProperties":{}}],"template":{"content":{"params":null},"id":"BlogDetail","extendedProperties":{}},"id":"191907","extendedProperties":{"resources":{"marketoAccountId":"689-LLU-525","alphavantageId":"A3Q2SUJ7HYI4N3A3","portalId":"147991","gtmTagId":"GTM-WGXQM6"},"rootPath":"","configs":{"textDirection":"ltr","language":"en-US","culture":"English","oneTrustScriptId":"0829f7a9-9cc9-475a-bac1-3acee089735b"}}}