“What is 802.11?”

[progval]

parsed as (?,definition,802.11)

http://askplatyp.us/?lang=en&q=What+is+802.11%3F

[Ezibenroc]

The problem come from the Stanford parser. This is even worse for “What is P=NP?”.
We could try to train it on such cases.
An other solution would be to do some preprocessing: if there is a part of the sentence with numbers or symbols like =, then handle it as if it was within quotations.

[yhamoudi]

(Partially) Fixed in https://github.com/ProjetPP/PPP-QuestionParsing-Grammatical/tree/reverse_predicates (but the problem comes from the stanford parser)

Possible solution for such cases: apply the algorithm used with quotations:

• identify "strange" entities (an entity = a sequence of letters without space, "strange" = contain a strange symbol : ., =, ...). Ex: 802.11, P=NP,...
• replace them by a random string
• parse with the stanford parser
• replace the random string by the initial word

If someone wants to implement this, please do it in branch reverse_predicates in file preprocessingMerge.py.

[yhamoudi]

See #118

[Ezibenroc]

+1 for this solution. But this does not concern this branch. You should do one pull request per problem to avoid huge modifications.

[yhamoudi]

But this does not concern this branch. You should do one pull request per

I will remove the advice once the branch will be merged. It's just to avoid that someone implement the solution in master before, otherwise there will be a big conflict.