Skip to content

NH TODO Discuss

Jump to bottom
jonathanPoitz edited this page Jun 6, 2016 · 107 revisions

TODO

  • highlight last label in an SDRS
  • show SDRS box using outscopings using discourse graph of SDRS. Discuss: Should we display crdRels side by side and subRels below each other? Also, what SDRSs should this method fail with? E.g. two root nodes (one node not attached)
  • make underspecified show, replacing drss with variables
  • DRS alpha-equivalence
  • export DRS.LambdaCalculus.renameSubDRS
  • export DRS.Variables.increase
  • change @d@ in line 28 Binding.hs to @r@

Discuss

if -- target node is on RF
then 
  if -- if the target node is arg in a relation
  then
    if relType rel == Crd
    then 
      if not $ isTopic rel
      then -- if the crdRel is not topic, simply add it to the current level with an And
      else -- if it's a topic relation, more updates need to be done since the outscoping variables change
        if -- relation pointing to target node is subordinating and entails \Downarrow (i.e., )
        then -- put new rel under new outscope and update old references to left arg with new outscoping label
        else -- not sure, we might have to make an explicit \Downarrow relation
    else -- simply add it to the current level with an And
  else -- target node is not an arg in a relation -> this can only be the root node of the sdrs.
    if relType rel == Crd
    then -- add it to the root node, put this relation under new outscoping label. (no references to be updated, b/c root still outscopes rest)
    else -- also add to root node, put this relation under new outscoping label and update old references (b/c new DRS will now outscopes rest)
else error = "Cannot add to node that's not on the RF"

  • adding a relation to the SDRS that already exists, add it nonetheless or skip it? -> add it

  • Would this discourse graph be right for the following example discourse:

    • Peter had a bad evening
    • His dinner was bad.
    • The next morning he got stomach cramps.
    • He went to the doctor.

  • Order of relation adding when adding multiple relations at once

Thoughts on sensible isomorphism constraints

  1. structural isomorphism: Tree structure and relations must be identical, DU labels can be different, DRS contents can be different: I do not currently see a reason for implementing this. An application like authorship attribution would not profit from something like this since we only get a boolean as return. One would need to Implement a function calculating the similarity of two discourse structures in terms of their structural isomorphism which is out of scope for this thesis.
  • full isomorphism: Tree structure and relations must be identical, DU labels can differ, DRSs must be alpha-equivalent. Depends on the implementation of an alpha-equivalence method of two DRSs.
  • sdrs isomorphism: Tree structure and relations must be identical, DU labels can differ, DRSs must be identical. The stricter version of above full isomorphism, two structures are only isomorphic if the DRSs are equal.

I would prefer version 2 since it comes with no apparent drawbacks compared to version 3, but offers more flexibility for (very possible differences in DRSRef naming)

  • Say we add a DRS to an SDRS that has an overlapping DRSRef "x1", we thus alpha convert the DRSRef in the new DRS. For this process we only take accessible DRSs and their DRSRefs into account. If there is a DRS with DRSRef "x2" already in the SDRS that however is not accessible to the newly attached DRS, then this doesn't matter right?
  • DRT Merge. Did you define it the way you implemented it or did previous work do that? (re: variable binding)
  • Do we need another meeting with Prof Pinkal?
  • Decisions on open Asher questions -> Segments in recursive SDRSFormulae
  • sdrsDRSRefAlphaConvert -> do we need this? What usecase would there be? I find alphaConvDRS much more usesful currently.
  • Basil example, pi1 also accessible?
  • edu/cdu implementation, use it? -> book example for background/narration where topic is union of ind. drss?
  • topic, what about the following example: "Peter had a great meal. He then went to see a theatre play. He enjoyed his evening a lot." There 3 would be the topic for Narration(1,2), so Topic(3,1_2). But this is problematic. Up to now, relations always went from an existing node to a new node. But the topic relation is defined as ⇓(a,b) meaning a is a topic for b. In the above example, the topic occurs after their subparts (otherwise it would be easy elaboration/narration). So how do we deal with this? :(
  • negation in composition. how do we implement it, if we implement it?
  • unresolved structures
  • relations
    • sdrt in theory allow for n-ary relations, but then in the text say "From now on we'll restrict disocurse relation symbols to being binary", but in la07 they use the binary definition
    • sdrt doesnt disallow self-referencing relations, does it? "If R is an (2-place) discourse relation symbol and pi1 and p2 are labels, then r(pi1, pi2) is an sdrs-formula.
  • I've identified structural constraints on the discourse imporsed by Narration and Background (but what about Continuation btw?)
  • since sdrs merge doesn't seem to be part of the formalism, do we have to properly define it?
  • section 4:
    • do i need to more properly explain how the discourse structure comes about? (Succ, outscopes etc)
    • wellformedness of sdrs, implementation? I'd like some high level function that combines some checks and maybe a section within section 4 to talk about it
  • in tikz graphs, highlight last label? al didnt do that, but i'd like to
  • what do they mean?

  • check whether inferLast works as desired
  • haskell: when should i import an entire module, when just a couple functions
Clone this wiki locally