[all] Is disambiguation doing more harm than good sometimes for incorrect sentences?

Today I tried to write an English rule that matches the sentence ‘Its great!’ and would be supposed to suggest ‘It’s’ instead of ‘Its’ for the first token.

The obvious way to do it would be a rule that matches ‘Its’, followed by an adjective and SENT_END. But the text analysis shows that in the sentence ‘Its great!’ the adjective reading is removed from the token ‘great’ by the disambiguator.

That would be a great feature if we could assume that all sentences are correct, but this is an assumption we must not make in a grammar checker.

Maybe we should add a feature that allows access to the tags that were there before the disambiguation process.

Related post by Michael: [de] Improvement of POS analysis

I think you found a case of wrong disambiguation rule. Just remove it.

Its + adj/noun” should not disambiguate by itself, maybe it just needs another token or an exception.

Also notice that sometimes some constructions are syntactically correct even though semantically weird. It depends on your policy whether you raise an alarm on them. To me as a writer it is annoying to be warned about some constructions just because they aren’t common.