Is it possible to have a thread where people can submit examples or testcases where the text normalizer is not doing perfectly?

[huangruizhe]

Thanks for the great work!
Just wondered if we can submit or share ad-hoc test cases to make the text normalizer more robust?

For example, some wrong normalizations are:
2020 Third Quarter => 2023rd quarter
Monroe's financial release => monroe is financial release

[jongwook]

Yes the normalizer definitely has many rough edges. Your first example seems easy to fix, the second is a bit more tricky because it's hard to differentiate between possessive 's and contractions of is or has with the rule-based approach. But would appreciate more examples like this to be aware of its failure modes!