Matches in SemOpenAlex for { <https://semopenalex.org/work/W3001816066> ?p ?o ?g. }
- W3001816066 endingPage "59" @default.
- W3001816066 startingPage "41" @default.
- W3001816066 abstract "In neural machine translation (NMT), it has become standard to translate using subword units to allow for an open vocabulary and improve accuracy on infrequent words. Byte-pair encoding (BPE) and its variants are the predominant approach to generating these subwords, as they are unsupervised, resource-free, and empirically effective. However, the granularity of these subword units is a hyperparameter to be tuned for each language and task, using methods such as grid search. Tuning may be done inexhaustively or skipped entirely due to resource constraints, leading to sub-optimal performance. In this paper, we propose a method to automatically tune this parameter using only one training pass. We incrementally introduce new BPE vocabulary online based on the held-out validation loss, beginning with smaller, general subwords and adding larger, more specific units over the course of training. Our method matches the results found with grid search, optimizing segmentation granularity while significantly reducing overall training time. We also show benefits in training efficiency and performance improvements for rare words due to the way embeddings for larger units are incrementally constructed by combining those from smaller units." @default.
- W3001816066 created "2020-01-30" @default.
- W3001816066 creator A5028564314 @default.
- W3001816066 creator A5046084081 @default.
- W3001816066 creator A5068811427 @default.
- W3001816066 creator A5078485689 @default.
- W3001816066 creator A5084103813 @default.
- W3001816066 date "2020-01-24" @default.
- W3001816066 modified "2023-10-03" @default.
- W3001816066 title "Optimizing segmentation granularity for neural machine translation" @default.
- W3001816066 cites W1899794420 @default.
- W3001816066 cites W1978161643 @default.
- W3001816066 cites W2124807415 @default.
- W3001816066 cites W2594229957 @default.
- W3001816066 cites W2756154119 @default.
- W3001816066 cites W2757041753 @default.
- W3001816066 cites W2962784628 @default.
- W3001816066 cites W2963225662 @default.
- W3001816066 cites W2963251942 @default.
- W3001816066 cites W2963736442 @default.
- W3001816066 cites W2963979492 @default.
- W3001816066 cites W2964053711 @default.
- W3001816066 doi "https://doi.org/10.1007/s10590-019-09243-8" @default.
- W3001816066 hasPublicationYear "2020" @default.
- W3001816066 type Work @default.
- W3001816066 sameAs 3001816066 @default.
- W3001816066 citedByCount "21" @default.
- W3001816066 countsByYear W30018160662019 @default.
- W3001816066 countsByYear W30018160662020 @default.
- W3001816066 countsByYear W30018160662021 @default.
- W3001816066 countsByYear W30018160662022 @default.
- W3001816066 countsByYear W30018160662023 @default.
- W3001816066 crossrefType "journal-article" @default.
- W3001816066 hasAuthorship W3001816066A5028564314 @default.
- W3001816066 hasAuthorship W3001816066A5046084081 @default.
- W3001816066 hasAuthorship W3001816066A5068811427 @default.
- W3001816066 hasAuthorship W3001816066A5078485689 @default.
- W3001816066 hasAuthorship W3001816066A5084103813 @default.
- W3001816066 hasBestOaLocation W30018160662 @default.
- W3001816066 hasConcept C104317684 @default.
- W3001816066 hasConcept C105580179 @default.
- W3001816066 hasConcept C111919701 @default.
- W3001816066 hasConcept C119857082 @default.
- W3001816066 hasConcept C138885662 @default.
- W3001816066 hasConcept C144133560 @default.
- W3001816066 hasConcept C149364088 @default.
- W3001816066 hasConcept C154945302 @default.
- W3001816066 hasConcept C162324750 @default.
- W3001816066 hasConcept C162853370 @default.
- W3001816066 hasConcept C177774035 @default.
- W3001816066 hasConcept C185592680 @default.
- W3001816066 hasConcept C187691185 @default.
- W3001816066 hasConcept C187736073 @default.
- W3001816066 hasConcept C203005215 @default.
- W3001816066 hasConcept C204321447 @default.
- W3001816066 hasConcept C2524010 @default.
- W3001816066 hasConcept C2777601683 @default.
- W3001816066 hasConcept C2780451532 @default.
- W3001816066 hasConcept C33923547 @default.
- W3001816066 hasConcept C41008148 @default.
- W3001816066 hasConcept C41895202 @default.
- W3001816066 hasConcept C43364308 @default.
- W3001816066 hasConcept C55493867 @default.
- W3001816066 hasConcept C86251818 @default.
- W3001816066 hasConcept C8642999 @default.
- W3001816066 hasConcept C89600930 @default.
- W3001816066 hasConcept C90805587 @default.
- W3001816066 hasConceptScore W3001816066C104317684 @default.
- W3001816066 hasConceptScore W3001816066C105580179 @default.
- W3001816066 hasConceptScore W3001816066C111919701 @default.
- W3001816066 hasConceptScore W3001816066C119857082 @default.
- W3001816066 hasConceptScore W3001816066C138885662 @default.
- W3001816066 hasConceptScore W3001816066C144133560 @default.
- W3001816066 hasConceptScore W3001816066C149364088 @default.
- W3001816066 hasConceptScore W3001816066C154945302 @default.
- W3001816066 hasConceptScore W3001816066C162324750 @default.
- W3001816066 hasConceptScore W3001816066C162853370 @default.
- W3001816066 hasConceptScore W3001816066C177774035 @default.
- W3001816066 hasConceptScore W3001816066C185592680 @default.
- W3001816066 hasConceptScore W3001816066C187691185 @default.
- W3001816066 hasConceptScore W3001816066C187736073 @default.
- W3001816066 hasConceptScore W3001816066C203005215 @default.
- W3001816066 hasConceptScore W3001816066C204321447 @default.
- W3001816066 hasConceptScore W3001816066C2524010 @default.
- W3001816066 hasConceptScore W3001816066C2777601683 @default.
- W3001816066 hasConceptScore W3001816066C2780451532 @default.
- W3001816066 hasConceptScore W3001816066C33923547 @default.
- W3001816066 hasConceptScore W3001816066C41008148 @default.
- W3001816066 hasConceptScore W3001816066C41895202 @default.
- W3001816066 hasConceptScore W3001816066C43364308 @default.
- W3001816066 hasConceptScore W3001816066C55493867 @default.
- W3001816066 hasConceptScore W3001816066C86251818 @default.
- W3001816066 hasConceptScore W3001816066C8642999 @default.
- W3001816066 hasConceptScore W3001816066C89600930 @default.
- W3001816066 hasConceptScore W3001816066C90805587 @default.
- W3001816066 hasIssue "1" @default.
- W3001816066 hasLocation W30018160661 @default.
- W3001816066 hasLocation W30018160662 @default.