SaarlandMweDiscussion

Discussion of MWEs, inspired by Ann's participation in PARSEME.

We have started a new page on MWEs (MweTop) to which we will link various relevant things.

What do we do now

• things with spaces

  • some interfacing with morphology

• things made into a single predicate

  • look up

• things recognized as larger things with idiom matchin

  • determiner-less PPs in hospital

    • also occurs outside/slightly genericy
    • semi-lexicalized

  • idiom thingies keep tabs on

    • some work at NTU/CSLI on possessive idioms

    + note: not marked as a unit in the MRS output + supported by LKB and ACE

  • different types of idiomaticity detless_pp vs flexible idioms

  • we have paraphrase rules for many of these

    • but not perfect out of your tiny mind

• how does the interface with chart mapping/tokenization

• what about the idiomatic/non-idiomatic distinction

  • we don't enforce it perfectly

• we maybe have more examples of MWEs with structure than anyone else

  • although we don't have as many examples as e.g. in wordnet

• SRG: words with spaces, verb+particle, idioms (take into account)

• Matrix: no idioms (FCB: there is documentation on the wiki)

• NorSource: not yet

• Burger: some types for verb+complement

• Jacy: all kinds, even documentation

  • not so good with things like te-nakareba-narimasen, complex pps

  • (http://moin.delph-in.net/JacyIdiom)

• Hegram: nothing

• MCG: nothing

  • Chengyu (four character idioms)
    • treat them as non-compositional
    • NTU has a list of these with some more information (with help from Mike and Ning)
    • there are also non-Chengyu idioms

• we can have both internal and external modification (for some idioms)

  • the cat kicked all nine buckets (Mike)

• a lot of regional use

• treat proverbial the same as fucking (can go anywhere)

• in general adding MWEs adds ambiguity so we tend not to add them

  • if they help in parse-selection it would be worth putting them in

  • even very common things like Thank you and good morning

Things we don't have an account for:

• institutionalized phrases traffic light/traffic signal

• light verbs/light verby idioms give a rat's arse [about]

• proverbs --- how do we handle these a stitch in time saves nine)

  • interestingly cross-lingually
  • often contains frozen bits of older grammars

• fixed foreign phrases (que sera sera)

  • interesting to see if there are differences
    • in flexibility between old English vs foreign

• NPIs are on the edge of this phenomenon

• things like you may wish to -> you should (post-process)

• If you like currently words-with-space in ERG

Other projects

• MWEs with structure in wordnets
• Lots of work in Japan, e.g. on idiom/literal (Chikara)