"date of birth of Barack Obama"

[Tpt]

queries like: http://ppp.pony.ovh/?lang=en&q=date+of+birth+of+Barack+Obama

[yhamoudi]

This kind of queries is really difficult because we have to perform multiwords expressions (MWE) recognition if we want to parse it correctly. (from the grammatical point of view, barack obama and birth have exactly the same role).

We didn't find efficient + easy to use tools that perform MWE recognition. However, it seems now that conceptnet could be used to perform this task. Indeed, if a MWE exists then it is a concept and it can be in conceptnet (first president is not in conceptnet, but prime minister or date of birth are \o/). I think we could investigate this a little more. If conceptnet is powerful enough, here is what we can do to deal with MWE:

• use conceptnet to find all the MWE in the input sentence
• concatenate each MWE into a single word (same thing than with quotations). Be sure that it has no effect on the stanford parsing
• parse with the stanford parser, run the usual algorithm

[yhamoudi]

The parsing of date of birth of Barack Obama is not good (same problem of "distributivity" than the example here: #90).

However, let's assume the parsing was good, we will be able to produce ((Barack Obama, birth,?),date,?). Then we could use the same collapsing trick than here: #85

(this solution is less powerful than detecting the multiwords expressions and merging them before using the stanford parser, but it is more likely to succeed)

[yhamoudi]

See #123