Matches in SemOpenAlex for { <https://semopenalex.org/work/W2892095314> ?p ?o ?g. }
Showing items 1 to 99 of
99
with 100 items per page.
- W2892095314 abstract "Neural Machine Translation (NMT) typically leverages monolingual data in training through backtranslation. We investigate an alternative simple method to use monolingual data for NMT training: We combine the scores of a pre-trained and fixed language model (LM) with the scores of a translation model (TM) while the TM is trained from scratch. To achieve that, we train the translation model to predict the residual probability of the training data added to the prediction of the LM. This enables the TM to focus its capacity on modeling the source sentence since it can rely on the LM for fluency. We show that our method outperforms previous approaches to integrate LMs into NMT while the architecture is simpler as it does not require gating networks to balance TM and LM. We observe gains of between +0.24 and +2.36 BLEU on all four test sets (English-Turkish, Turkish-English, Estonian-English, Xhosa-English) on top of ensembles without LM. We compare our method with alternative ways to utilize monolingual data such as backtranslation, shallow fusion, and cold fusion." @default.
- W2892095314 created "2018-09-27" @default.
- W2892095314 creator A5061001839 @default.
- W2892095314 creator A5083524596 @default.
- W2892095314 creator A5091317839 @default.
- W2892095314 date "2018-09-01" @default.
- W2892095314 modified "2023-10-01" @default.
- W2892095314 title "Simple Fusion: Return of the Language Model" @default.
- W2892095314 cites W179875071 @default.
- W2892095314 cites W1902237438 @default.
- W2892095314 cites W1915251500 @default.
- W2892095314 cites W1916559533 @default.
- W2892095314 cites W2006969979 @default.
- W2892095314 cites W2101456909 @default.
- W2892095314 cites W2109664771 @default.
- W2892095314 cites W2130942839 @default.
- W2892095314 cites W2132339004 @default.
- W2892095314 cites W2252272516 @default.
- W2892095314 cites W2552838200 @default.
- W2892095314 cites W2555428947 @default.
- W2892095314 cites W2561274697 @default.
- W2892095314 cites W2597891111 @default.
- W2892095314 cites W2748679025 @default.
- W2892095314 cites W2756566411 @default.
- W2892095314 cites W2765961751 @default.
- W2892095314 cites W2766182427 @default.
- W2892095314 cites W2797913374 @default.
- W2892095314 cites W2798362442 @default.
- W2892095314 cites W2886540570 @default.
- W2892095314 cites W2903012348 @default.
- W2892095314 cites W2962784628 @default.
- W2892095314 cites W2963216553 @default.
- W2892095314 cites W2964308564 @default.
- W2892095314 hasPublicationYear "2018" @default.
- W2892095314 type Work @default.
- W2892095314 sameAs 2892095314 @default.
- W2892095314 citedByCount "5" @default.
- W2892095314 countsByYear W28920953142019 @default.
- W2892095314 countsByYear W28920953142020 @default.
- W2892095314 countsByYear W28920953142022 @default.
- W2892095314 crossrefType "posted-content" @default.
- W2892095314 hasAuthorship W2892095314A5061001839 @default.
- W2892095314 hasAuthorship W2892095314A5083524596 @default.
- W2892095314 hasAuthorship W2892095314A5091317839 @default.
- W2892095314 hasConcept C111472728 @default.
- W2892095314 hasConcept C137293760 @default.
- W2892095314 hasConcept C138885662 @default.
- W2892095314 hasConcept C154945302 @default.
- W2892095314 hasConcept C16910744 @default.
- W2892095314 hasConcept C199360897 @default.
- W2892095314 hasConcept C203005215 @default.
- W2892095314 hasConcept C204321447 @default.
- W2892095314 hasConcept C2777413886 @default.
- W2892095314 hasConcept C2777530160 @default.
- W2892095314 hasConcept C2780586882 @default.
- W2892095314 hasConcept C28490314 @default.
- W2892095314 hasConcept C41008148 @default.
- W2892095314 hasConcept C41895202 @default.
- W2892095314 hasConceptScore W2892095314C111472728 @default.
- W2892095314 hasConceptScore W2892095314C137293760 @default.
- W2892095314 hasConceptScore W2892095314C138885662 @default.
- W2892095314 hasConceptScore W2892095314C154945302 @default.
- W2892095314 hasConceptScore W2892095314C16910744 @default.
- W2892095314 hasConceptScore W2892095314C199360897 @default.
- W2892095314 hasConceptScore W2892095314C203005215 @default.
- W2892095314 hasConceptScore W2892095314C204321447 @default.
- W2892095314 hasConceptScore W2892095314C2777413886 @default.
- W2892095314 hasConceptScore W2892095314C2777530160 @default.
- W2892095314 hasConceptScore W2892095314C2780586882 @default.
- W2892095314 hasConceptScore W2892095314C28490314 @default.
- W2892095314 hasConceptScore W2892095314C41008148 @default.
- W2892095314 hasConceptScore W2892095314C41895202 @default.
- W2892095314 hasLocation W28920953141 @default.
- W2892095314 hasOpenAccess W2892095314 @default.
- W2892095314 hasPrimaryLocation W28920953141 @default.
- W2892095314 hasRelatedWork W2121870595 @default.
- W2892095314 hasRelatedWork W2122270629 @default.
- W2892095314 hasRelatedWork W2125823897 @default.
- W2892095314 hasRelatedWork W2284660317 @default.
- W2892095314 hasRelatedWork W2396575863 @default.
- W2892095314 hasRelatedWork W2898793635 @default.
- W2892095314 hasRelatedWork W2902128069 @default.
- W2892095314 hasRelatedWork W2911689843 @default.
- W2892095314 hasRelatedWork W2963109507 @default.
- W2892095314 hasRelatedWork W2963174344 @default.
- W2892095314 hasRelatedWork W2963403868 @default.
- W2892095314 hasRelatedWork W2988121211 @default.
- W2892095314 hasRelatedWork W2988335601 @default.
- W2892095314 hasRelatedWork W2996854111 @default.
- W2892095314 hasRelatedWork W3002887707 @default.
- W2892095314 hasRelatedWork W3021357296 @default.
- W2892095314 hasRelatedWork W3091540052 @default.
- W2892095314 hasRelatedWork W3127593070 @default.
- W2892095314 hasRelatedWork W3208011254 @default.
- W2892095314 hasRelatedWork W3211447745 @default.
- W2892095314 isParatext "false" @default.
- W2892095314 isRetracted "false" @default.
- W2892095314 magId "2892095314" @default.
- W2892095314 workType "article" @default.