Visual Text Representations

AbstractMachine translation models have discrete vocabularies and commonly use subword segmentation techniques to achieve an ‘open vocabulary.’ This approach relies on consistent and correct underlying unicode sequences, and makes models …...