W3C home > Mailing lists > Public > public-openannotation@w3.org > December 2019

Re: Expressing relations between targets?

From: Karin Verspoor <karin.verspoor@unimelb.edu.au>
Date: Wed, 11 Dec 2019 18:53:09 +0000
To: Simon Rainer <Rainer.Simon@ait.ac.at>, Benjamin Young <byoung@bigbluehat.com>, James Smith <jgsmith@gmail.com>
CC: "public-openannotation@w3.org" <public-openannotation@w3.org>
Message-ID: <B803DB23-056F-4C54-BFE1-A241B566723A@unimelb.edu.au>
I think this is similar to the use cases in the paper I sent earlier.

Here we have syntactic annotation over the text:
[A picture containing text, map  Description automatically generated]

[A picture containing screenshot  Description automatically generated]

[A picture containing text, map  Description automatically generated]



From: Simon Rainer <Rainer.Simon@ait.ac.at>
Date: Tuesday, 10 December 2019 at 6:19 pm
To: Benjamin Young <byoung@bigbluehat.com>, James Smith <jgsmith@gmail.com>
Cc: "public-openannotation@w3.org" <public-openannotation@w3.org>
Subject: AW: Expressing relations between targets?
Resent from: <public-openannotation@w3.org>
Resent date: Tuesday, 10 December 2019 at 6:17 pm

Hi Ben,

thanks for this. I agree, we shouldn't try to force knowledge graph-type relations into Web Annotation. That said, I think I probably chose a really bad/misleading example to begin with. As I said, I'm really looking for a way to express the text annotation aspect. A better example might have been:

- Two highlighted text paragraphs

- An arrow by the annotator, with a note attached that says something like "this paragraph contradicts that one" (or something like that)

The important characteristics would be:

- the highlights do really refer to the text, not whatever real-world matter is being described

- the arrow may have direction, but doesn't need to. (The note might just say something like "similar" or whatever)

- the arrow "label" might be a tag, freetext comment, etc. i.e. anything that can be perfectly captured with a WebAnno body already

A said before that the highlighted text segments, in my case, are already expressed as WebAnno annotations. That, I guess, wouldn't even be strictly necessary. The whole construct might just as well be implemented with a single annotation that has two selectors (the text passages). The fact that I'd rather model it as two separate annotation, and then another "link" annotation on top of that, is mostly because its more convenient in our system and/or more along the lines of how the UI functions. (But I'm happy to hear pros/cons on that approach!)



Von: Benjamin Young <byoung@bigbluehat.com>
Gesendet: Montag, 9. Dezember 2019 18:14
An: Simon Rainer; James Smith
Cc: public-openannotation@w3.org
Betreff: Re: Expressing relations between targets?

Thanks for writing, Simon!

At this point, I'd suggest being careful not to reinvent RDF inside Web Annotation. Something like "A is the father of B" or similar is already better expressed via much simpler RDF (assuming you have identifiers for the things. Mixing that into the annotation model starts to create all kinds of painful indirection. 😕

That said, I'm noting a lack of "directionality" when targeting resources in Web Annotation. I'm not (yet) certain it's Web Annotation's job to record that, just noting that it currently isn't possible. Appendix D has the things that get close: https://www.w3.org/TR/annotation-model/#sets-of-bodies-and-targets But even so, I don't think there's a way to turn "Orestes killed Aegisthus" into an annotation per se.

It might be best to narrow in on the "textual editing part" and explore potential needs in that context. Otherwise, RDF proper would likely do a much cleaner job of expressing the things you note below.

Happy to discuss further!




From: Simon Rainer <Rainer.Simon@ait.ac.at>
Sent: Monday, December 9, 2019 8:28 AM
To: James Smith <jgsmith@gmail.com>
Cc: public-openannotation@w3.org <public-openannotation@w3.org>
Subject: AW: Expressing relations between targets?

Hi James,

I agree - we'd ideally want to identify the people via URIs. And then the assertion "is father of" would be a statement between those two entities.

However, within the annotation environment, the scenario is that this would be like a note, made a human editor, on the text. So I'm inclined to say that, we are indeed talking about a statement about two strings of text - at least at this point in the workflow.

I guess the example is also not ideal here. The annotations might just as well be highlighting two different text paragraphs, and the annotator would drag an arrow between them saying "the author is repeating him/herself here", or whatever. I.e. irrespective of what ever "real world meaning" might be behind the arrow eventually, I'm primarily interested in using WebAnno/Open Annotation to model the textual editing part.



Von: James Smith <jgsmith@gmail.com>
Gesendet: Montag, 9. Dezember 2019 14:18
An: Simon Rainer
Cc: public-openannotation@w3.org
Betreff: Re: Expressing relations between targets?

How do we know who the text "Aegisthus" refers to? It's a name, so it's identifying something, but who/what is the identity we associate with the string "Aegisthus"?

While we might all agree on whom we link the text to in our minds, it might be helpful to add some context for the computer. I'd add another property on the body that points to a dbpedia entry or other unique URI that is useful in asserting the identity of the person referenced in the text as "Aegisthus". Let's say it's <dbpedia:Aegisthus>. The same could be done for "Orestes" with a link (for the purposes of discussion) of <dbpedia:Orestes>.

Once this is done, then it's a matter of asserting the <dbpedia:Aegisthus> is related to <dbpedia:Orestes>. It's not about the string "Aegisthus" having a familial relationship with the string "Orestes", or that one annotation has a familial relationship with another annotation, but about the person <dbpedia:Aegisthus> having such a relationship with the person <dbpedia:Orestes>. They just happen to be referenced as "Aegisthus" and "Orestes" in this particular text.

If we did make the relationship about the strings in the text, then that relationship wouldn't be true for any other instances of the strings "Aegisthus" and "Orestes" in this or any other text. It would be about the two instances already highlighted.

-- Jim

On Mon, Dec 9, 2019 at 8:03 AM Simon Rainer <Rainer.Simon@ait.ac.at<mailto:Rainer.Simon@ait.ac.at>> wrote:

Dear list,

I have a question regarding the possible use of the WebAnno model for expressing a relationship between two targets.

My scenario is the following:

*) I have two text annotations that identify people. Each annotation has a single target (the person name in a TextQuoteSelector, and character offset in a TextPositionSelector); and a single body (with purpose "identifying").

*) I now want to create a third annotation that expresses a relation between person A and B. (E.g. "A is the father of B" or similar.)


My approach would be to model this third annotation like the sample below. I.e. with:

*) two targets, each holding the ID of one person annotation

*) a body with the relation tag ("is the father of")

*) a motivation of "linking"

According to the definition for "linking", that's not correct though. "Linking" is supposed to express a link between body and all targets, rather than a link between the targets. In addition, there's also no way to express directionality.

Does anyone have recommendations on how to tackle such a use case with WebAnno? I realize that (some of) this may actually be out of scope for the spec as such. In this case, I'd appreciate any thoughts, opinions, and possible recommendations on a custom extension pattern, if needed.

Cheers & thanks in advance,



  "@context": "http://www.w3.org/ns/anno.jsonld",
  "id": "#9ba844a7-e8ec-4127-ad12-1f7f16a240c6",
  "type": "Annotation",
  "motivation": "linking",
  "body": [{
    "type": "TextualBody",
    "value": "isRelatedTo"
  "target": [{
    "id": "#ce0ed291-766b-4763-8e91-90ce1d04e706"
  }, {
    "id": "#447d4bea-08dc-4bd0-ae51-31f5ed7a95a0"

(image/jpeg attachment: image001.jpg)

(image/jpeg attachment: image002.jpg)

(image/jpeg attachment: image003.jpg)

(image/png attachment: image004.png)

Received on Wednesday, 11 December 2019 18:53:30 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 16:38:32 UTC