Reproduce “Robust Open-Vocabulary Translation from Visual Text Representations” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Elizabeth Salesky, David Etter, Matt Post
The visual-text translation system represents source text as overlapping image slices, embeds them with a convolutional network, and passes them through a Transformer, avoiding fixed subword segmentation. On synthetic German-to-English tests where every eligible source token receives an adjacent-character swap, it scores 25.9 BLEU versus 1.9 for a subword baseline. Clean-text results are near parity, and natural-noise tests are domain-confounded.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.