Matches in SemOpenAlex for { <https://semopenalex.org/work/W2897414631> ?p ?o ?g. }
- W2897414631 abstract "In neural machine translation (NMT), it is 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 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 without any additional 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.
- W2897414631 created "2018-10-26" @default.
- W2897414631 creator A5028564314 @default.
- W2897414631 creator A5046084081 @default.
- W2897414631 creator A5068811427 @default.
- W2897414631 creator A5078485689 @default.
- W2897414631 creator A5084103813 @default.
- W2897414631 date "2018-10-19" @default.
- W2897414631 modified "2023-09-23" @default.
- W2897414631 title "Optimizing Segmentation Granularity for Neural Machine Translation" @default.
- W2897414631 cites W1899794420 @default.
- W2897414631 cites W1978161643 @default.
- W2897414631 cites W2107468211 @default.
- W2897414631 cites W2117873652 @default.
- W2897414631 cites W2133564696 @default.
- W2897414631 cites W2144600658 @default.
- W2897414631 cites W2178031510 @default.
- W2897414631 cites W2311921240 @default.
- W2897414631 cites W2339995566 @default.
- W2897414631 cites W2594229957 @default.
- W2897414631 cites W2744654301 @default.
- W2897414631 cites W2756154119 @default.
- W2897414631 cites W2757041753 @default.
- W2897414631 cites W2899771611 @default.
- W2897414631 cites W2949563612 @default.
- W2897414631 cites W2962784628 @default.
- W2897414631 cites W2963979492 @default.
- W2897414631 cites W2964053711 @default.
- W2897414631 cites W46679369 @default.
- W2897414631 hasPublicationYear "2018" @default.
- W2897414631 type Work @default.
- W2897414631 sameAs 2897414631 @default.
- W2897414631 citedByCount "4" @default.
- W2897414631 countsByYear W28974146312019 @default.
- W2897414631 countsByYear W28974146312020 @default.
- W2897414631 crossrefType "posted-content" @default.
- W2897414631 hasAuthorship W2897414631A5028564314 @default.
- W2897414631 hasAuthorship W2897414631A5046084081 @default.
- W2897414631 hasAuthorship W2897414631A5068811427 @default.
- W2897414631 hasAuthorship W2897414631A5078485689 @default.
- W2897414631 hasAuthorship W2897414631A5084103813 @default.
- W2897414631 hasConcept C10485038 @default.
- W2897414631 hasConcept C111919701 @default.
- W2897414631 hasConcept C119857082 @default.
- W2897414631 hasConcept C12267149 @default.
- W2897414631 hasConcept C138885662 @default.
- W2897414631 hasConcept C154945302 @default.
- W2897414631 hasConcept C162324750 @default.
- W2897414631 hasConcept C177774035 @default.
- W2897414631 hasConcept C187691185 @default.
- W2897414631 hasConcept C187736073 @default.
- W2897414631 hasConcept C203005215 @default.
- W2897414631 hasConcept C204321447 @default.
- W2897414631 hasConcept C2524010 @default.
- W2897414631 hasConcept C2777601683 @default.
- W2897414631 hasConcept C2780451532 @default.
- W2897414631 hasConcept C33923547 @default.
- W2897414631 hasConcept C41008148 @default.
- W2897414631 hasConcept C41895202 @default.
- W2897414631 hasConcept C43364308 @default.
- W2897414631 hasConcept C8642999 @default.
- W2897414631 hasConcept C89600930 @default.
- W2897414631 hasConcept C90805587 @default.
- W2897414631 hasConceptScore W2897414631C10485038 @default.
- W2897414631 hasConceptScore W2897414631C111919701 @default.
- W2897414631 hasConceptScore W2897414631C119857082 @default.
- W2897414631 hasConceptScore W2897414631C12267149 @default.
- W2897414631 hasConceptScore W2897414631C138885662 @default.
- W2897414631 hasConceptScore W2897414631C154945302 @default.
- W2897414631 hasConceptScore W2897414631C162324750 @default.
- W2897414631 hasConceptScore W2897414631C177774035 @default.
- W2897414631 hasConceptScore W2897414631C187691185 @default.
- W2897414631 hasConceptScore W2897414631C187736073 @default.
- W2897414631 hasConceptScore W2897414631C203005215 @default.
- W2897414631 hasConceptScore W2897414631C204321447 @default.
- W2897414631 hasConceptScore W2897414631C2524010 @default.
- W2897414631 hasConceptScore W2897414631C2777601683 @default.
- W2897414631 hasConceptScore W2897414631C2780451532 @default.
- W2897414631 hasConceptScore W2897414631C33923547 @default.
- W2897414631 hasConceptScore W2897414631C41008148 @default.
- W2897414631 hasConceptScore W2897414631C41895202 @default.
- W2897414631 hasConceptScore W2897414631C43364308 @default.
- W2897414631 hasConceptScore W2897414631C8642999 @default.
- W2897414631 hasConceptScore W2897414631C89600930 @default.
- W2897414631 hasConceptScore W2897414631C90805587 @default.
- W2897414631 hasOpenAccess W2897414631 @default.
- W2897414631 hasRelatedWork W1501640207 @default.
- W2897414631 hasRelatedWork W2017761627 @default.
- W2897414631 hasRelatedWork W2066159919 @default.
- W2897414631 hasRelatedWork W2143774016 @default.
- W2897414631 hasRelatedWork W2176263492 @default.
- W2897414631 hasRelatedWork W2234895705 @default.
- W2897414631 hasRelatedWork W2738569017 @default.
- W2897414631 hasRelatedWork W2934533289 @default.
- W2897414631 hasRelatedWork W2949849824 @default.
- W2897414631 hasRelatedWork W2963380118 @default.
- W2897414631 hasRelatedWork W2978806756 @default.
- W2897414631 hasRelatedWork W3006698530 @default.
- W2897414631 hasRelatedWork W3091985378 @default.
- W2897414631 hasRelatedWork W3103853042 @default.