Matches in SemOpenAlex for { <https://semopenalex.org/work/W3170702943> ?p ?o ?g. }
- W3170702943 abstract "We propose a new training objective named order-agnostic cross entropy (OaXE) for fully non-autoregressive translation (NAT) models. OaXE improves the standard cross-entropy loss to ameliorate the effect of word reordering, which is a common source of the critical multimodality problem in NAT. Concretely, OaXE removes the penalty for word order errors, and computes the cross entropy loss based on the best possible alignment between model predictions and target tokens. Since the log loss is very sensitive to invalid references, we leverage cross entropy initialization and loss truncation to ensure the model focuses on a good part of the search space. Extensive experiments on major WMT benchmarks show that OaXE substantially improves translation performance, setting new state of the art for fully NAT models. Further analyses show that OaXE alleviates the multimodality problem by reducing token repetitions and increasing prediction confidence. Our code, data, and trained models are available at https://github.com/tencent-ailab/ICML21_OAXE." @default.
- W3170702943 created "2021-06-22" @default.
- W3170702943 creator A5016804474 @default.
- W3170702943 creator A5024040521 @default.
- W3170702943 creator A5059251959 @default.
- W3170702943 date "2021-06-09" @default.
- W3170702943 modified "2023-10-14" @default.
- W3170702943 title "Order-Agnostic Cross Entropy for Non-Autoregressive Machine Translation" @default.
- W3170702943 cites W1617088490 @default.
- W3170702943 cites W2101105183 @default.
- W3170702943 cites W2222512263 @default.
- W3170702943 cites W2892213699 @default.
- W3170702943 cites W2939049558 @default.
- W3170702943 cites W2962784628 @default.
- W3170702943 cites W2962926939 @default.
- W3170702943 cites W2963246629 @default.
- W3170702943 cites W2963248296 @default.
- W3170702943 cites W2963341956 @default.
- W3170702943 cites W2963366552 @default.
- W3170702943 cites W2963403868 @default.
- W3170702943 cites W2963463964 @default.
- W3170702943 cites W2963532001 @default.
- W3170702943 cites W2963736842 @default.
- W3170702943 cites W2964089333 @default.
- W3170702943 cites W2964121744 @default.
- W3170702943 cites W2970832665 @default.
- W3170702943 cites W2976965654 @default.
- W3170702943 cites W2988975212 @default.
- W3170702943 cites W2990372437 @default.
- W3170702943 cites W2990389671 @default.
- W3170702943 cites W2995999067 @default.
- W3170702943 cites W2996843693 @default.
- W3170702943 cites W3034363136 @default.
- W3170702943 cites W3034892578 @default.
- W3170702943 cites W3035416964 @default.
- W3170702943 cites W3035725000 @default.
- W3170702943 cites W3100753857 @default.
- W3170702943 cites W3127901106 @default.
- W3170702943 cites W3174255604 @default.
- W3170702943 hasPublicationYear "2021" @default.
- W3170702943 type Work @default.
- W3170702943 sameAs 3170702943 @default.
- W3170702943 citedByCount "0" @default.
- W3170702943 crossrefType "posted-content" @default.
- W3170702943 hasAuthorship W3170702943A5016804474 @default.
- W3170702943 hasAuthorship W3170702943A5024040521 @default.
- W3170702943 hasAuthorship W3170702943A5059251959 @default.
- W3170702943 hasBestOaLocation W31707029431 @default.
- W3170702943 hasConcept C105795698 @default.
- W3170702943 hasConcept C106301342 @default.
- W3170702943 hasConcept C11413529 @default.
- W3170702943 hasConcept C114466953 @default.
- W3170702943 hasConcept C119857082 @default.
- W3170702943 hasConcept C121332964 @default.
- W3170702943 hasConcept C153083717 @default.
- W3170702943 hasConcept C154945302 @default.
- W3170702943 hasConcept C159877910 @default.
- W3170702943 hasConcept C167981619 @default.
- W3170702943 hasConcept C199360897 @default.
- W3170702943 hasConcept C203005215 @default.
- W3170702943 hasConcept C33923547 @default.
- W3170702943 hasConcept C38652104 @default.
- W3170702943 hasConcept C41008148 @default.
- W3170702943 hasConcept C43126263 @default.
- W3170702943 hasConcept C48145219 @default.
- W3170702943 hasConcept C52692508 @default.
- W3170702943 hasConcept C62520636 @default.
- W3170702943 hasConcept C75782508 @default.
- W3170702943 hasConcept C9679016 @default.
- W3170702943 hasConcept C98036226 @default.
- W3170702943 hasConceptScore W3170702943C105795698 @default.
- W3170702943 hasConceptScore W3170702943C106301342 @default.
- W3170702943 hasConceptScore W3170702943C11413529 @default.
- W3170702943 hasConceptScore W3170702943C114466953 @default.
- W3170702943 hasConceptScore W3170702943C119857082 @default.
- W3170702943 hasConceptScore W3170702943C121332964 @default.
- W3170702943 hasConceptScore W3170702943C153083717 @default.
- W3170702943 hasConceptScore W3170702943C154945302 @default.
- W3170702943 hasConceptScore W3170702943C159877910 @default.
- W3170702943 hasConceptScore W3170702943C167981619 @default.
- W3170702943 hasConceptScore W3170702943C199360897 @default.
- W3170702943 hasConceptScore W3170702943C203005215 @default.
- W3170702943 hasConceptScore W3170702943C33923547 @default.
- W3170702943 hasConceptScore W3170702943C38652104 @default.
- W3170702943 hasConceptScore W3170702943C41008148 @default.
- W3170702943 hasConceptScore W3170702943C43126263 @default.
- W3170702943 hasConceptScore W3170702943C48145219 @default.
- W3170702943 hasConceptScore W3170702943C52692508 @default.
- W3170702943 hasConceptScore W3170702943C62520636 @default.
- W3170702943 hasConceptScore W3170702943C75782508 @default.
- W3170702943 hasConceptScore W3170702943C9679016 @default.
- W3170702943 hasConceptScore W3170702943C98036226 @default.
- W3170702943 hasLocation W31707029431 @default.
- W3170702943 hasOpenAccess W3170702943 @default.
- W3170702943 hasPrimaryLocation W31707029431 @default.
- W3170702943 hasRelatedWork W10104832 @default.
- W3170702943 hasRelatedWork W10491538 @default.
- W3170702943 hasRelatedWork W11023528 @default.
- W3170702943 hasRelatedWork W11297145 @default.
- W3170702943 hasRelatedWork W11792228 @default.