FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3150702986> ?p ?o ?g. }

• W3150702986 abstract "Byte-pair encoding (BPE) is a ubiquitous algorithm in the subword tokenization process of language models as it provides multiple benefits. However, this process is solely based on pre-training data statistics, making it hard for the tokenizer to handle infrequent spellings. On the other hand, though robust to misspellings, pure character-level models often lead to unreasonably long sequences and make it harder for the model to learn meaningful words. To alleviate these challenges, we propose a character-based subword module (char2subword) that learns the subword embedding table in pre-trained models like BERT. Our char2subword module builds representations from characters out of the subword vocabulary, and it can be used as a drop-in replacement of the subword embedding table. The module is robust to character-level alterations such as misspellings, word inflection, casing, and punctuation. We integrate it further with BERT through pre-training while keeping BERT transformer parameters fixed–and thus, providing a practical method. Finally, we show that incorporating our module to mBERT significantly improves the performance on the social media linguistic code-switching evaluation (LinCE) benchmark." @default.
• W3150702986 created "2021-04-13" @default.
• W3150702986 creator A5008147177 @default.
• W3150702986 creator A5012052463 @default.
• W3150702986 creator A5027235926 @default.
• W3150702986 creator A5028709336 @default.
• W3150702986 creator A5047351656 @default.
• W3150702986 creator A5052900932 @default.
• W3150702986 date "2021-01-01" @default.
• W3150702986 modified "2023-09-25" @default.
• W3150702986 title "Char2Subword: Extending the Subword Embedding Space Using Robust Character Compositionality" @default.
• W3150702986 cites W1899794420 @default.
• W3150702986 cites W1938755728 @default.
• W3150702986 cites W2048967369 @default.
• W3150702986 cites W2101761627 @default.
• W3150702986 cites W2131571251 @default.
• W3150702986 cites W2153579005 @default.
• W3150702986 cites W2194775991 @default.
• W3150702986 cites W2250539671 @default.
• W3150702986 cites W2250729567 @default.
• W3150702986 cites W2251012068 @default.
• W3150702986 cites W2252209719 @default.
• W3150702986 cites W2462831000 @default.
• W3150702986 cites W2525778437 @default.
• W3150702986 cites W2757041753 @default.
• W3150702986 cites W2798935874 @default.
• W3150702986 cites W2880875857 @default.
• W3150702986 cites W2886146035 @default.
• W3150702986 cites W2887800417 @default.
• W3150702986 cites W2946558277 @default.
• W3150702986 cites W2962732637 @default.
• W3150702986 cites W2962739339 @default.
• W3150702986 cites W2962784628 @default.
• W3150702986 cites W2962818281 @default.
• W3150702986 cites W2963174553 @default.
• W3150702986 cites W2963251942 @default.
• W3150702986 cites W2963324947 @default.
• W3150702986 cites W2963340990 @default.
• W3150702986 cites W2963341956 @default.
• W3150702986 cites W2963403868 @default.
• W3150702986 cites W2963421945 @default.
• W3150702986 cites W2963831310 @default.
• W3150702986 cites W2963887123 @default.
• W3150702986 cites W2963979492 @default.
• W3150702986 cites W2964046515 @default.
• W3150702986 cites W2964090065 @default.
• W3150702986 cites W2965373594 @default.
• W3150702986 cites W2972324944 @default.
• W3150702986 cites W2973049837 @default.
• W3150702986 cites W2975529437 @default.
• W3150702986 cites W3032020872 @default.
• W3150702986 cites W3034238904 @default.
• W3150702986 cites W3034559121 @default.
• W3150702986 cites W3035390927 @default.
• W3150702986 cites W3035441470 @default.
• W3150702986 cites W3101140821 @default.
• W3150702986 cites W3105604018 @default.
• W3150702986 cites W3118485687 @default.
• W3150702986 doi "https://doi.org/10.18653/v1/2021.findings-emnlp.141" @default.
• W3150702986 hasPublicationYear "2021" @default.
• W3150702986 type Work @default.
• W3150702986 sameAs 3150702986 @default.
• W3150702986 citedByCount "4" @default.
• W3150702986 countsByYear W31507029862021 @default.
• W3150702986 crossrefType "proceedings-article" @default.
• W3150702986 hasAuthorship W3150702986A5008147177 @default.
• W3150702986 hasAuthorship W3150702986A5012052463 @default.
• W3150702986 hasAuthorship W3150702986A5027235926 @default.
• W3150702986 hasAuthorship W3150702986A5028709336 @default.
• W3150702986 hasAuthorship W3150702986A5047351656 @default.
• W3150702986 hasAuthorship W3150702986A5052900932 @default.
• W3150702986 hasBestOaLocation W31507029861 @default.
• W3150702986 hasConcept C121332964 @default.
• W3150702986 hasConcept C137293760 @default.
• W3150702986 hasConcept C138885662 @default.
• W3150702986 hasConcept C154945302 @default.
• W3150702986 hasConcept C165801399 @default.
• W3150702986 hasConcept C176982825 @default.
• W3150702986 hasConcept C204321447 @default.
• W3150702986 hasConcept C2524010 @default.
• W3150702986 hasConcept C2780861071 @default.
• W3150702986 hasConcept C28490314 @default.
• W3150702986 hasConcept C33923547 @default.
• W3150702986 hasConcept C41008148 @default.
• W3150702986 hasConcept C41608201 @default.
• W3150702986 hasConcept C41895202 @default.
• W3150702986 hasConcept C62520636 @default.
• W3150702986 hasConcept C66322947 @default.
• W3150702986 hasConcept C90805587 @default.
• W3150702986 hasConceptScore W3150702986C121332964 @default.
• W3150702986 hasConceptScore W3150702986C137293760 @default.
• W3150702986 hasConceptScore W3150702986C138885662 @default.
• W3150702986 hasConceptScore W3150702986C154945302 @default.
• W3150702986 hasConceptScore W3150702986C165801399 @default.
• W3150702986 hasConceptScore W3150702986C176982825 @default.
• W3150702986 hasConceptScore W3150702986C204321447 @default.
• W3150702986 hasConceptScore W3150702986C2524010 @default.
• W3150702986 hasConceptScore W3150702986C2780861071 @default.
• W3150702986 hasConceptScore W3150702986C28490314 @default.
• W3150702986 hasConceptScore W3150702986C33923547 @default.

• next