Try training a language model for correcting minor issues remaining if ulf2english

[genelkim]

We can start by simply training a typo-correcting model (take a corpus, add typos, delete words, etc.) and train model to recover it. Then try running it on ulf2english with mistakes. Then we might also extend it to include the idea of ULF operators and syntax which should not appear in the output (e.g. k operator, .suffix, etc.).