RE: Tool info specification (Re: action-221 summary of overriding discussion)

Hi Felix, all,

> I think it would be OK to have a data category "ITS Tool information" 
> which is available both locally and globally. Locally,

I can see one implementation complication with this method.
It has to do with overriding (obviously):

Imagine the following file:

<doc>
 <head>
 ...
 </head>
 <body>
 ...
 </body>
</doc>

Tool ABC processes this document and add the tool-ref attribute to <body> to be sure any annotation it does inside <body> is properly associated with its information:
 
<doc>
 <head its:toolRef="#tABC">
 ...
 </head>
 <body>
 ...<span its:someStuff="abc">...
 </body>
</doc>

Then Tool XYZ takes this as input and process it for another data category.
But this time Tool XYZ choose to set the tool reference information on <doc>:

<doc its:toolRef="#tXYZ">
 <head its:toolRef="#tABC">
 ...
 </head>
 <body>
 ...<span its:someStuff="abc">...
 ...<span its:someOtherStuff="xyz">...
 </body>
</doc>

Now the information its:someOtherStuff added by Tool XYZ is seen as being added by Tool ABC (or more exactly by no tool since Tool ABC is not labeled for the data category corresponding to its:someOtherStuff).

Obviously a tool can no solve such issue by searching for toolRef in the whole document and adding a reference to itself when appropriate. But it can be a bit complicated to do. Especially if the tool is stream-based rather than DOM-based. My point is that the mechanism is not necessarily easy to implement correctly.

Another question is whether tools will have tot support this new data category if they support one of the data categories that make use of it. 
My first thought would be no: so far we've keep each data category separated, and identifying the tool that added some information may be seen as optional in many cases. So there are little reasons to make it mandatory.
But this can lead to new problems: if it's optional two tools can process the same document for the same data category e.g. mtConfidence, but if only one provides the tool reference, then all mtConfidence markup is seen as done by the lone tool that provided the tool reference.
So, it seems we must force tools to implement that data category, but it's not one easy to implement...

Anyway, some fruits for thoughts...

Cheers,
-yves

Received on Saturday, 22 September 2012 03:16:16 UTC