{"id":1698,"date":"2022-07-01T17:46:52","date_gmt":"2022-07-01T15:46:52","guid":{"rendered":"https:\/\/www-staging.systransoft.com\/papers\/alias-et-quia-sint-velit-sit\/"},"modified":"2023-11-16T18:21:44","modified_gmt":"2023-11-16T17:21:44","slug":"alias-et-quia-sint-velit-sit","status":"publish","type":"paper","link":"https:\/\/www-staging.systransoft.com\/de\/ressourcen\/papers-and-publications\/alias-et-quia-sint-velit-sit\/","title":{"rendered":"Latent Group Dropout f\u00fcr maschinelle \u00dcbersetzung mit mehreren Sprachen und Dom\u00e4nen"},"content":{"rendered":"<p><em>Multidomain and multilingual machine translation often rely on parameter sharing strategies, where large portions of the network are meant to capture the commonalities of the tasks at hand, while smaller parts are reserved to model the peculiarities of a language or a domain. In adapter-based approaches, these strategies are hardcoded in the network architecture, independent of the similarities between tasks. In this work, we propose a new method to better take advantage of these similarities, using a latent-variable model. We also develop new techniques to train this model end-to-end and report experimental results showing that the learned patterns are both meaningful and yield improved translation performance without any increase of the model size.<\/em><\/p>","protected":false},"excerpt":{"rendered":"<p>Mehrdom\u00e4nen- und mehrsprachige maschinelle \u00dcbersetzungen st\u00fctzen sich h\u00e4ufig auf Strategien zum Teilen von Parametern, bei denen gro\u00dfe Teile des Netzwerks die Gemeinsamkeiten der anstehenden Aufgaben erfassen sollen, w\u00e4hrend kleinere Teile reserviert sind, um die Besonderheiten einer Sprache oder einer Dom\u00e4ne zu modellieren. Bei adapterbasierten Ans\u00e4tzen sind diese Strategien in der Netzwerkarchitektur fest kodiert, unabh\u00e4ngig ... <a href=\"https:\/\/www-staging.systransoft.com\/de\/ressourcen\/papers-and-publications\/alias-et-quia-sint-velit-sit\/\">anhaltend<\/a><\/p>","protected":false},"featured_media":0,"template":"","class_list":["post-1698","paper","type-paper","status-publish","hentry"],"acf":[],"_links":{"self":[{"href":"https:\/\/www-staging.systransoft.com\/de\/wp-json\/wp\/v2\/paper\/1698","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www-staging.systransoft.com\/de\/wp-json\/wp\/v2\/paper"}],"about":[{"href":"https:\/\/www-staging.systransoft.com\/de\/wp-json\/wp\/v2\/types\/paper"}],"wp:attachment":[{"href":"https:\/\/www-staging.systransoft.com\/de\/wp-json\/wp\/v2\/media?parent=1698"}],"curies":[{"name":"WP","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}