Grammar development time/effort
  + are developers more open-minded to automatic (at least non-traditional) methods of grammar dev

Bart: A good grammar takes immense amounts of effort
  + by the time the grammar is developed, perhaps our formalism will be deprecated?
  Antske doesn't think it's an issue. MRS is more important than HPSG (maybe)
  + Also, treebanks that are formalism-dependent similarly have limited value
  + Perhaps we need a less specific "standard" for our treebanks, resources
  Francis: but that results in less interesting information
    + DELPH-IN has not pushed the point that depth-of-coverage is useful (compared to breadth)
    + How do we get to high-depth & high-coverage while producing something also useful at intermediate stages

     d |H         +     H: HPSG
     e |                +: Goal
     p |                C: Cheetah
     t |       C        S: Statistical/Shallow systems
     h |         
       |_________S_
         breadth
  Antske: HPSG treebanks (and MRS) can be converted to other formalisms, so maintains usefulness

Regarding grammar induction
  Yi: more advanced methods can be applied (data-driven)
  Bart: Laurie brought up using robustness rules for error detection, etc. This is data-driven but not necessarily statistical

Discussion of uses of bilingual text:
  Francis: disambiguation (e.g. prepositions, noun-verb in Japanese-English)
        also for improving coverage (e.g. ERG can parse more of a bitext than Jacy, Jacy developer can look at semantics from ERG output, use to inform Jacy development)

Talk about shared resources/standards
 ... about grammar development time being the bottleneck
 ... about the walls of reaching + (in chart above)
 ... about using dependency treebanks to thin trees when treebanking

+ Not always about finding funding for grammar development, but a grammar developer


Plans for future grammars:
  Francis wants to find a student for Malay
  Yi says Mandarin will be developed at Saarbruecken (perhaps starting at S and climbing to +)
  Tania says the slavic languages will be further expanded (Polish, Russian...)

What are 'goals' of a grammar?
  Of course, one is applications
  One could be learning about/describing grammar (as opposed to just parsing)

Last update: 2010-07-04 by MichaelGoodman [edit]