Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Those numbers are very bad, given that proper phonemic orthographies can give you a 90+% confidence with far fewer rules.

There's a simple and consistent way to compare languages in this way too, too: train a neural net to map spelling to pronunciation on one half of the dictionary, then test it on the other half. The more complicated and less consistent the orthography is, the more mistakes it'll make. People have in fact done this exact experiment, and English scores extremely poorly in it; for spelling, closer to Chinese, in fact, than many other European languages: https://aclanthology.org/2021.sigtyp-1.1/



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: