Reproduce “Evaluating Pixel Language Models on Non-Standardized Languages” 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
Alberto Muñoz-Ortiz, Verena Blaschke, Barbara Plank
The authors test whether PIXEL, which represents rendered text as image patches rather than fixed subword tokens, transfers from Standard German to German dialects better than BERT. It performs better on several dialect evaluations, including Swiss German intent classification, reaching 83.8% accuracy versus 62.4% for cased BERT. Gains vary by task and test set, and results cover German only.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.