SaarlandIcons

Discussion moderated by SanghounSong, scribed by MichaelGoodman

Apologies for the many ..., where the scribe's fingers were slower than the speakers' lips.

Sanghoun: ICONS processed by ACE, other processors (LKB, PET, AGREE)

Sanghoun: ICONS can be used for anaphora resolution,

• also reflexives, apposition, non-restrictive relative clauses, etc. (planned)

Glenn: We need to know what you need for ICONS, otherwise it seems straightforward as a bag

Woodley: Similar to HCONS

Emily: ICONS are present in TDL, with trigger rules, etc. using them

Ann: Is there any description of ICONS?

Woodley: somewhere

Emily: For a start I can distribute my documentation, which helped Woodley

Glenn: Are there test parameters?

Emily: We can link them from the Wiki page

Glenn: So there's 5 trivial grammars for testing?

Emily: yes

Stephan: It's a good example of the process for revising formalisms, in this case MRS. It's very much a collaborative process; Ann and Dan and others seem excited about ICONS, and Sanghoun formalized it to some degree, Woodley implemented it, etc.. I don't see anything wrong with this process.

Ann: It's easier to add code when you know how it is supposed to behave. I'm glad someone else has done the experimentation.

Stephan: Now we need volunteers to test them.

Woodley: We think we know the right thing to do for generation with ICONS

Glenn: This is like the issue discussed the other day; MRS as a logic used within our group

Ann: Well no, ICONS is not something that is interpretable in the MRS logic, so I would expect no effects at all in things like Utool, and if you're converting to a first-order representation, etc., then the ICONS will just be lost

Emily: Right, this is a separate layer..

Glenn: So this is neither a description nor formal logic...

Ann: I think you can use the term Meaning Representation, now we just need an S

others: System, Simple, Stuff

Ann: If the ICONS constraints about anaphora, etc. then you can't create an MRS that will scope in the desired way, so you can think of it as going toward discourse representation. Somewhere there is a proposed DTD of MRS including ICONS (Ed: perhaps SofiaMrsRfc ?), but the differences are things like the identification of structure, or sentence structures... Once you're thinking about anaphora across sentences you need a way to identify those sentences..

Woodley:...

Dan: Appostion...

Ann: Well there is now some problem with the appos_rel, ... you can't equate those things and get a scope out...

Dan: Can I get a hypernormally connected graph from Utool?

Stephan: Bec, do you remember the rule?

Glenn: he's suggesting we make this more accessible to lay people

Dan: Perhaps we should take this offline

Ann: I don't remember the issue with appos..

Stephan: If UW is pushing ICONS in wider use in DELPH-IN, it may lead to other analyses of info struct, so a formal proposal is needed. I asked Emily in the Spring, do ICONS affect truth conditions, do they affect scope... personally I would like to see more of the proposed formalization and implementation documented on a wiki page.

Woodley: For info struct, they don't affect truth conditions, maybe scope

Stephan: Once you have a wiki page, I'll follow up with my questions

David: It would be interesting to know the types of individuals that can be related... i.e. what's the set of indivs that can be related with an MRS versus those with ICONS

Ann: The issue is with entities that can be events or instances ('i'; "individual").. The relationships we're talking about are those that don't arise naturally from what we call semantics... they have something to do with linguistic structure, but they're not in what we traditionally call compositional semantics. We need a way to process them, dealing with various labels they can take, etc.. as a convenience rather than a formal interpretation

---

As requested, here is a start of a wiki page on IconsSpecs --Emily