MedCATtrainer icon indicating copy to clipboard operation
MedCATtrainer copied to clipboard

Spell checking functionality

Open MattStammers opened this issue 3 years ago • 2 comments

Currently, it is not possible to correct any typographical errors in the trainer window. Could a text editor function be added to the training window so that incorrect terms do not end up being validated as correct?

MattStammers avatar Jun 17 '22 15:06 MattStammers

Hi @MattStammers - thanks for the suggestion, I've marked as an enhancement. For clarity - the suggestion is that you'd like to be able to edit the text so the model can learn the concept context for the correct spelling rather than an incorrect one right?

In previous annotation projects we've seen that its useful for the underlying model to know that a misspelling is still the intended concept, it's also likely that the concept will appear again spelled correctly so the misspelling is actually helpful.

FYI - MedCAT also includes a spell checker so common misspelling patterns i.e. character pairs being swapped around or single characters missed, will often be picked up the model.

tomolopolis avatar Jun 21 '22 10:06 tomolopolis

Hi, yes this is basically correct. Thanks for replying.

MattStammers avatar Jul 02 '22 12:07 MattStammers