Overcoming Vocabulary Constraints with Pixel-level Fallback
Overcoming Vocabulary Constraints with Pixel-level Fallback
Subword tokenization requires balancing computational efficiency and vocabulary coverage, which often leads to suboptimal performance on languages and scripts not prioritized during training. We propose to augment pretrained language models with a vocabulary-free encoder that generates input embeddings from text rendered as pixels. Through experiments on English-centric language models, we demonstrate that our approach substantially improves machine translation performance and facilitates effective cross-lingual transfer, outperforming tokenizer-based methods. Furthermore, we find that pixel-based representations outperform byte-level approaches and standard vocabulary expansion. Our approach enhances the multilingual capabilities of monolingual language models without extensive retraining and reduces decoding latency via input compression.
Jonas F. Lotz、Hendra Setiawan、Stephan Peitz、Yova Kementchedjhieva
语言学
Jonas F. Lotz,Hendra Setiawan,Stephan Peitz,Yova Kementchedjhieva.Overcoming Vocabulary Constraints with Pixel-level Fallback[EB/OL].(2025-04-02)[2025-05-13].https://arxiv.org/abs/2504.02122.点此复制
评论