Label Wording Augmentation
A method for enhancing the diversity of label surface forms in typed decision models. By presenting labels in multiple paraphrased forms, it improves the model's robustness to variations in wording.
Timeline (last 90 days)
The study compares two versions of the multilingual cross-encoder Ru-Jev, of which the newer version introduces a label-wording enhancement method.
On held-out vocabularies composed of ordinary words or loanwords, the enhanced version shows higher accuracy under all settings, with improvements of up to 0.24.
In the abbreviation scenario, the enhanced version performs worse than the baseline on the clean slice.
For opaque code, once the labels related to the training options are removed, both versions remain at or below chance level.
When the number of candidates increases to eighteen, confidence errors in the term and slang scenarios rise somewhat.
The paper points out four evaluation hazards that arise when comparing the two versions.
Label-wording enhancement is markedly effective in handling unfamiliar vocabulary and everyday language variants, but it does not solve the problems of abbreviations, code, or labels lacking semantic transparency.
All Facts (7)
The study compares two versions of the multilingual cross-encoder Ru-Jev, of which the newer version introduces a label-wording enhancement method.
50%UnverifiedOn held-out vocabularies composed of ordinary words or loanwords, the enhanced version shows higher accuracy under all settings, with improvements of up to 0.24.
50%UnverifiedIn the abbreviation scenario, the enhanced version performs worse than the baseline on the clean slice.
50%UnverifiedFor opaque code, once the labels related to the training options are removed, both versions remain at or below chance level.
50%UnverifiedWhen the number of candidates increases to eighteen, confidence errors in the term and slang scenarios rise somewhat.
50%UnverifiedThe paper points out four evaluation hazards that arise when comparing the two versions.
50%UnverifiedLabel-wording enhancement is markedly effective in handling unfamiliar vocabulary and everyday language variants, but it does not solve the problems of abbreviations, code, or labels lacking semantic transparency.
50%