{Machine translation of mathematical text}
{Aditya Ohri, Tanya Schmah}
{We present a machine translation system, the PolyMath Translator, for
\LaTeX\ documents containing mathematical text. The system combines a
\LaTeX\ parser, tokenization of math and labels, a deep learning
Transformer model trained on mathematical and other text, and the Google
Translate \API\ with a custom glossary. Ablation testing shows that math
tokenization and the Transformer model each significantly improve
translation quality, while Google Translate is used as a backup when the
Transformer does not have confidence in its translation. For \LaTeX\
parsing, we have used the pandoc document converter, while our latest
development version instead uses the TexSoup package. We will describe
the system, show examples, and discuss future directions.}