Skip to content

DiderotEsdFocusGroup

EmilyBender edited this page Aug 30, 2018 · 5 revisions

What were you looking for?

Francis: Enlightenment ... don't remember. Has been a long time. Arg/adjunct distinction? What guidelines you used to make it.

Did you find it?

Francis: No, still can't distinguish arg/adj; but don't blame it on the documentation.

Did you go anywhere else?

Francis: Yes, various pages.

What were you looking for?

Guy: To check whether the ERG makes a certain distinction, and what it is that is made.

Did you find it?

Guy: Don't remember.

Guy: Also read the section about intrinsic arguments. To get the term and for the summary of the arguments for that design decision.

Dan: Did it convince you to have one on neg_rel?

What were you looking for?

Alexandre: Beginner's perspective. Looking for info about the output of the ERG, and kind of structure I can work with after the parser. And for a reference about applications.

Did you find it?

Alexandre: Still learning... it was helpful, got some useful information but still have to read more. Useful but not comprehensive. For begnning it's a lot of different things going on --- lots of variations of the representations. Want to know how things evolve over time. Not clear on the relationship, what's old and has been superceded, what's important to be aware of because they all play a role in the process.

Did you go anywhere else?

Alexandre: Read other papers first.

Olga: Had I remembered that this exists, I would have probably come here in hopes of finding a simpler paraphrase of some of the definitions of Copestake et al 2005. Often go back to that paper when I have to remind myself or be prepared to explain to students. Would be looking for a simpler or more focused presentation.

Things you learned that you were interested in

Something you feel is missing / wanted to find but couldn't

Olga: Suppose I'm a person who wants to use the representations but am lacking in bg, including formal logic. Would be helpful to inform me, just very briefly (without doing the impossible simple tutorial about everything), what kind of background you need to understand these pages. What does "modulo quantifier insertion" mean? Even Copestake et al 2005 doesn't help with that. If we do not assume linguistic background of the readers, it's especially important.

Alexandre: It's better to learn than avoid the technical terms. Help to educate.

Something that surprised you

Berthold: I looked at the quasi-modal infinitivals, but I was a bit surprised that you and they call them relative clauses. There are also predicative uses: "This dog is to guard the tabacco". Do normal relative clauses have predicative uses?

Dan: The observation there is that the purposiveness comes in in both uses, and it has to come from somewhere.

Berthold: Just wondering why you call it a relative clause, given the predictive use...

Emily: We don't actually call them that, but it's hard to see because the title isn't in the page content! (FIX THAT.)

Olga: I have been to these pages before --- the list of phenomena is very helpful. We came here to look at what the ERG does with nominalization and found this summary helpful.

Olga: It's too slow to look in the tdl.

Jan: If someone asks me what do I get in ERS which you don't have in SRL or semantic dependencies, then I would like to be able to use the inventory these are these phenomena and they lead us to have additional predicates that don't correspond directly to surface forms. Some of this is oriented a bit more towards people who already know what the pheneomena are and want to know how they are expressed in the ERG. Not sure if you can serve them both at the same time, but it would be nice. Have a slightly more visual representation of the examples. ==> Show the full ERS for each example on each page.

Dan: Dependency graph representation (from Oslo workshop) nicely shows just how many links are there on a quick glance. Would want to show in that format and then highlight what is relevant to this phenomenon. That notation was widely understood in the workshop --- comes from AMR. Circles where each one is a predication not a surface token.

Weiwei: I checked the page with two goals: (1) to find the meaning of the ERS (2) we are trying to provide deep semantic annotations for Mandarin and I think the ERS can be a good reference. I checked here to find information about linguistic phenomena. Would help to have more references into the literature. Would also be interested in references to counterparts in other languages.

Dan: Making some attempt at moving beyond the very language and grammar specific representation.

Weiwei: And to other linguistic frameworks...

Guy: Regarding using diagrams ... what about incorporating delph-in viz. Don't konw how much work it would be to interface in the wiki.

Francis: Hyperlinks ... to parsing that thing with delphin viz.

Guy: Would be better to have the visualization in a box so that it stays in sync.

Berthold: If you are looking for what the current state is of pre-modified quantifiers... (nearly every). Hard to tell if it's documented or not. Would be nice to have an index of simpler keywords.

Dan: Ned Letcher and I talked about using typediff to write in a short sentence in English and we have a notion of what's normal background noise of a sentence for English, and Ned's machinery can say which thing stands out as interesting (low frequency) ... and then link into the documentation from the types that stand out. Tried an experiment with H&P Ch 12 on relative clauses.

BK: My general problem with this wiki is the lack of a table of contents, especially missing in this subdomain of pages that seems to be more structured than the rest.

Berthold: Is there a distinction between diamond coverage, gold coverage, silver, ... that I remember from earlier discussion. Is there some indication of how stable some semantic analysis is?

Emily: To a certain extent the open questions section reflects that.

Dan: We typically do have opinions and we ought to express those opinions overtly.

Berthold: You might be in a language with different structural pressure and more easily go for something that is better rather than keeping compatibility just for the sake of it.

Emily: One open question is which representation to foreground. Choices on the table: (1) Simple MRS (like in over page), which wears scope underspecification on its sleeve (2) Cheap scope scoped version (3) Graphical version that downplays scope all together

Dan: How about (2) and (3)?

Weiwei: Sometimes I read about things like caused-motion construction and I'd like to try with the ERG to analyze this phenomenon.

Dan: More than just parsing it in the demo?

Weiwei: I'd like to get information about the way you design the grammar to get it.

Emily: That's out of scope for this project.

Alexandre: The logical form ... that's what goes in the semantic search interface?

Dan: That's yet one more notation. Could we provide a syntactic sugar on top of that so someone could write a logical form and then highlight a bit and generate a search.

Emily: Or a GUI based one, like I think Tigre had?

Francis: More breadcrumbs, please. The keywords are very specific. Would be nicer to have an example for each one.

Dan: Yes, I'd even promised to have that done by now.

Emily: unknown et al aren't interesting...

Francis: Au contraire, some people might be looking at one of those and wonder what they are? [Emily: Index for that?]

Guy: On the essence page you talk about various diathesis alternations (without using that term) but that isn't on the list of phenomena. That's what I usually use to show people that we want semantics rather than syntactic dependencies.

Emily: There's no semantic reflex for the dative alternation.

Guy: In that case, the term "semantic phenomenon" needs to be carefully explained on the inventory page. Dative alternation is a semantic phemonenon in the sense that it's a linguistic phenonenon where we care about the semantics (being shared in this case). If it's not on the inventory page, should still be listed somewhere.

Dan: Maybe a list of systematic paraphrase (and close paraphrase) alternations. Something similar comes up in John's book or that book of John's. Also passive, with its ICONS. Highlight normalization benefit.

Emily: Francis, you didn't tell us to put in hyperlinks to LTDB.

Francis: I think it goes without saying. May also be nice to have it go the other direction (from LTDB to the ESD pages). For many of these phenomena I'd be happy to extend the LTDB so you can give it a list of types as a URL if people want to jump through and find examples and frequency. For now you can search substrings of types. An actual list would be better.

Francis: I like the way it has the gramamr version there, becasue one problem with this kind of documentation is that it does get out of sync a little bit. If it changes is the plan to keep the old verion so people can see the evolution of thought (that's my suggestion).

Emily: At the moment you can see the edit history of the wiki.

Francis: So that people can easily see.

Emily: How about a hyperlink to a specific stable older version in the wiki when there's a major change.

John: Presentational comment. I find lots of abbreviations difficult to distinguish: ESD, EDS, ERG, ERS. Could maybe get rid of "ESD" which Dan couldn't even remember what it stands for.

Jan: What about using colors to show shared handles?

John: Link to the demo before linking to ErgProcessing.

Berthold: It would be nice to link both ways, from demo to ESD pages and from ESD pages to demos.

John/Dan: Or just a link to the ESD pages at the top of the demo.

Clone this wiki locally