W3C home > Mailing lists > Public > public-multilingualweb-lt@w3.org > October 2012

RE: [ISSUE-22] Provenance and Agents

From: Yves Savourel <ysavourel@enlaso.com>
Date: Mon, 15 Oct 2012 06:03:21 -0700
To: "'Felix Sasaki'" <fsasaki@w3.org>, "'Dave Lewis'" <dave.lewis@cs.tcd.ie>
CC: <public-multilingualweb-lt@w3.org>
Message-ID: <assp.0635f38981.assp.0635276fdb.006601cdaad5$7463dcb0$5d2b9610$@com>
Hi Felix, Dave, all,


Felix: I think there is a difference in the way you use transProvRef and the way locQualityIssuesRef is currently defined. You use a list of URIs for transProvRef while locQualityIssuesRef defines a single URI that points to a set of issues.


To have both data categories be similar, you would have to have transProvref to point to a translationProvenanceRecords with one or more records. So in your example, two translationProvenanceRecords elements (one for each of the transProvRef).


But I agree that a similar stand-off structure could be used for both.






From: Felix Sasaki [mailto:fsasaki@w3.org] 
Sent: Sunday, October 14, 2012 11:22 AM
To: Dave Lewis
Cc: public-multilingualweb-lt@w3.org
Subject: Re: [ISSUE-22] Provenance and Agents


Hi Dave, all,


I added the translation provenance agent to


with a big warning that this is in an early stage. I changed a few things from your draft:


- XPath expressions in pointer attributes in the example:  these were quite general; e.g. //dc:creator selects all "dc:creator" elements in the document. Esp. given the discussion we just have here 


this seems to be too general


- XPath expression in the selector, e.g. "selector="/html/body/legalnotice"" > "selector="/text/body/legalnotice""

I changed "/html/body/par" to "/text/body/par[1]", so that here only the first "par" element is selected. I realized here again that we haven't resolved the "tool many global rules" issue. Dave, can you take up this thread


Because depending on the outcome both provenance and many other data categories might change a lot


- I removed local XPath expressions, e.g. transToolPointer or transToolRefPointer attributes. We don't have local XPath - that has been discussed several times. If needed I can dig up the threads again, but it would save a lot of time if we could just agree on this. 


- I changed the local example. What you tried in the local example was a combination of global and local provenance information. But that doesn't work: we said now several times that overriding is always complete. So you cannot "through a local selection overriding part of the global rule.". You will override the complete rule. It doesn't matter whether the local attributes are in HTML5 or in XML, that doesn't change overriding.


In general I'm quite frustrated about the data category. The issue is not the pieces of information itself; what you specify (person, organization, tools) makes a lot of sense. The issue is that obviously the specification is not implementation driven, as can be seen by the non tested XPath expressions and the overriding that wouldn't work, even with a conformance only processor.


The other frustration comes from the speed and continuation of progress: to wrap this up we need a continuous discussion. So my main question is: will you and Phil have time to engage in this by the end of November, that is within the last call period? Or: can we engage somebody else interested in implementing this?


Now, about the data category in general ...


I think what you are trying to achieve is:

conveying several pieces of provenance information for agents:

initial revision = translation agent provenance;

subsequent revision = translation revision agent provenance;

complex revision information: standoff provenance.


We may have a similar picture like with quality issue: the complexity of this information might be better dealt with a standoff approach. I am not talking about the standoff approach in your example, Dave, but something like this:



<text xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:its="http://www.w3.org/2005/11/its"
        <dc:creator>John Doe</dc:creator>
        <title>Translation Revision Provenance Agent: Global Test in XML</title>
            <its:translationProvenanceRecord xml:id="tp1"
                transToolRef="http://www.onlinemtex.com/2012/7/25/wsdl/" transOrg="acme-CAT-v2.3"/>
            <its:translationProvenanceRecord xml:id="tp2" transPerson="John Doe"
            <its:translationProvenanceRecord xml:id="tp3" transPerson="Carl Meyer"
            <its:translationProvenanceRecord xml:id="tp4" provRef="http://www.examplemtservice.com/prov/e76547"/>
        <par its:transProvRef="#tp1"> This paragraph was translated from the machine.</par>
        <legalnotice postediting-by="http://www.vistatec.com/" its:transProvRef="#tp2 #tp3 #tp4">This text was
            translated directly by a person.</legalnotice>



The interaction between "its:translationProvenanceRecords" and the local its:transProvRef attribute is identical to "its:locQualityIssues" and "its:locQualityIssuesRef" attribute.


In its:translationProvenanceRecords you have a list of "its:translationProvenanceRecord" elements. Each element has an "xml:id" attribute. We could say that the order of "its:translationProvenanceRecord" specifies whether this is translation agent provenance or revision agent provenance information. Or we could say that this is specified by the order of the values in "its:transProfRev". ”Your" standoff data category could be accommodated by <its:translationProvenanceRecord xml:id="tp4" provRef="http://www.examplemtservice.com/prov/e76547"/>.


You seem to have the use case of attaching several pieces of provenance information to the same node. With the ITS overriding that is not possible. But with the above approach tools can still do that, locally:

- first tool creates

<legalnotice postediting-by="http://www.vistatec.com/" its:transProvRef="#tp2">This text was
            translated directly by a person.</legalnotice>

- second tool creates

<legalnotice postediting-by="http://www.vistatec.com/" its:transProvRef="#tp2 #tp3">This text was
            translated directly by a person.</legalnotice>

- third tool creates

<legalnotice postediting-by="http://www.vistatec.com/" its:transProvRef="#tp2 #tp3 #tp4">This text was
            translated directly by a person.</legalnotice>


This all works without global "adding" rules (but keeping the pointer attributes in global rules). We just need guidance for the tool developers how to attach such complex pieces of information.


Also, for the simple local case we could still have 

<legalnotice postediting-by="http://www.vistatec.com/" its:transPerson="John Doe"
                its:transOrgRef="http://www.legaltrans-ex.com/" its:provRef="http://www.examplemtservice.com/prov/e76547">This text was translated directly by a person.</legalnotice>


But would say that you either have local markup or the external record, not both.


So in summary, above proposal would mean

- have only one provenance data category

- realize the need of specifying initial translation provenance, revision and standoff provenance at the same time like this: having lq issue like standoff elements

- realize the need of providing several pieces of information via several references to provenance records, e.g. its:transProvRef="#tp2 #tp3"

- have global rules only for pointing, see the other thread.






2012/10/12 Dave Lewis <dave.lewis@cs.tcd.ie>

Hi All,
Please find attached updates to the provenance related data categories ready to be included in the draft. Many thanks to Phil for reviewing these in detail.

There are three separate data categories:
- Translation Agent Provenance: which record machines, people and organsiations responsible for translating the selected text

- Translation Agent Provenance: which records machines, people and organsiations responsible for revising the translation the selected text (e.g. from posteding or linguistic review)

- Standoff Provenance: which provides a link to standoff provenance record using the W3C PROV standard.

Comments welcome.




Felix Sasaki

DFKI / W3C Fellow

Received on Monday, 15 October 2012 13:03:56 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 16:31:55 UTC