Cleaning up of answers from the ML

[Michael Cochez]

Dear organizers,

We notice that when we get answers from the LM we are using that they does sometimes deviate from the expected outputs in the dev set because of minor issues.

For example, lower vs upper case, use of a different name for the same company, use of an abbreviation instead of the full name, etc. 

Is it acceptable that we include a component in our system to clean up those name based on a dictionary lookup? This would not itself be a language model.

Kind regards,

Michael

[LM-KBC]

Hi Michael,

Yes, basic string cleaning operations and conversion of abbreviations are allowed on top of LM-generated output.

Best,
Sneha