Matches in SemOpenAlex for { <https://semopenalex.org/work/W4320719765> ?p ?o ?g. }
Showing items 1 to 69 of
69
with 100 items per page.
- W4320719765 abstract "Language tasks involving character-level manipulations (e.g., spelling correction, many word games) are challenging for models based in subword tokenization. To address this, we adapt the interchange intervention training method of Geiger et al. (2021) to operate on type-level variables over characters. This allows us to encode robust, position-independent character-level information in the internal representations of subword-based models. We additionally introduce a suite of character-level tasks that systematically vary in their dependence on meaning and sequence-level context. While simple character-level tokenization approaches still perform best on purely form-based tasks like string reversal, our method is superior for more complex tasks that blend form, meaning, and context, such as spelling correction in context and word search games. Our approach also leads to subword-based models with human-intepretable internal representations of characters." @default.
- W4320719765 created "2023-02-15" @default.
- W4320719765 creator A5018694855 @default.
- W4320719765 creator A5039468724 @default.
- W4320719765 creator A5042601761 @default.
- W4320719765 creator A5077549848 @default.
- W4320719765 date "2022-12-19" @default.
- W4320719765 modified "2023-09-23" @default.
- W4320719765 title "Inducing Character-level Structure in Subword-based Language Models with Type-level Interchange Intervention Training" @default.
- W4320719765 doi "https://doi.org/10.48550/arxiv.2212.09897" @default.
- W4320719765 hasPublicationYear "2022" @default.
- W4320719765 type Work @default.
- W4320719765 citedByCount "0" @default.
- W4320719765 crossrefType "posted-content" @default.
- W4320719765 hasAuthorship W4320719765A5018694855 @default.
- W4320719765 hasAuthorship W4320719765A5039468724 @default.
- W4320719765 hasAuthorship W4320719765A5042601761 @default.
- W4320719765 hasAuthorship W4320719765A5077549848 @default.
- W4320719765 hasBestOaLocation W43207197651 @default.
- W4320719765 hasConcept C137293760 @default.
- W4320719765 hasConcept C138885662 @default.
- W4320719765 hasConcept C151730666 @default.
- W4320719765 hasConcept C154945302 @default.
- W4320719765 hasConcept C157486923 @default.
- W4320719765 hasConcept C204321447 @default.
- W4320719765 hasConcept C2524010 @default.
- W4320719765 hasConcept C2777801307 @default.
- W4320719765 hasConcept C2779343474 @default.
- W4320719765 hasConcept C2780861071 @default.
- W4320719765 hasConcept C28490314 @default.
- W4320719765 hasConcept C33923547 @default.
- W4320719765 hasConcept C37914503 @default.
- W4320719765 hasConcept C41008148 @default.
- W4320719765 hasConcept C41895202 @default.
- W4320719765 hasConcept C86803240 @default.
- W4320719765 hasConcept C90805587 @default.
- W4320719765 hasConceptScore W4320719765C137293760 @default.
- W4320719765 hasConceptScore W4320719765C138885662 @default.
- W4320719765 hasConceptScore W4320719765C151730666 @default.
- W4320719765 hasConceptScore W4320719765C154945302 @default.
- W4320719765 hasConceptScore W4320719765C157486923 @default.
- W4320719765 hasConceptScore W4320719765C204321447 @default.
- W4320719765 hasConceptScore W4320719765C2524010 @default.
- W4320719765 hasConceptScore W4320719765C2777801307 @default.
- W4320719765 hasConceptScore W4320719765C2779343474 @default.
- W4320719765 hasConceptScore W4320719765C2780861071 @default.
- W4320719765 hasConceptScore W4320719765C28490314 @default.
- W4320719765 hasConceptScore W4320719765C33923547 @default.
- W4320719765 hasConceptScore W4320719765C37914503 @default.
- W4320719765 hasConceptScore W4320719765C41008148 @default.
- W4320719765 hasConceptScore W4320719765C41895202 @default.
- W4320719765 hasConceptScore W4320719765C86803240 @default.
- W4320719765 hasConceptScore W4320719765C90805587 @default.
- W4320719765 hasLocation W43207197651 @default.
- W4320719765 hasOpenAccess W4320719765 @default.
- W4320719765 hasPrimaryLocation W43207197651 @default.
- W4320719765 hasRelatedWork W1491795195 @default.
- W4320719765 hasRelatedWork W2482622595 @default.
- W4320719765 hasRelatedWork W2596494451 @default.
- W4320719765 hasRelatedWork W2806021948 @default.
- W4320719765 hasRelatedWork W2900203667 @default.
- W4320719765 hasRelatedWork W2945392649 @default.
- W4320719765 hasRelatedWork W2963461183 @default.
- W4320719765 hasRelatedWork W3107474891 @default.
- W4320719765 hasRelatedWork W4301317594 @default.
- W4320719765 hasRelatedWork W615929232 @default.
- W4320719765 isParatext "false" @default.
- W4320719765 isRetracted "false" @default.
- W4320719765 workType "article" @default.