RE: issue-41 (mtconfidence), issue-42 (mtConfidence, textAnalysisAnnotation, quality)

Thanks for the explanations Felix.

-- As far as I can tell (but I may have missed one) we don't have any example in 1.0 or the test suite that demonstrates the overriding is for all information of a multi-information data category, even when the local information is undefined. And indeed we don't have an example demonstrating the contrary either.

-- The case of defining 'termInfoRef' as optional along with 'term' for the local markup doesn't demonstrate anything IMO, it says it's optional, but doesn't say if its absence also overrides a previous rule. In general absence of an attribute means it has no effect. 

-- The modified example you send in the other email simply shows that when the non-complete overriding semantics are used you have to override all information explicitly. One cannot guess an intention from an example. If the intention is to have alerts for the three notes, the example is fine.

-- Even your implementation, Sebastian's (or mine which happen to also to do complete overriding) isn't a formal proof that the intention of 1.0 was to do complete overriding: they are not tested against anything that would give a different output with the non-complete overriding semantics.


...But, there is the outputType attribute in the 1.0 test suite result format.
Since it is defined at the node level rather than in each information it does indeed prove to me the intent to have a complete overriding in 1.0. Sorry I missed that.

Then, you are right obviously: we must stay backward compatible.

But we probably need something more explicit than the current sentence: "Override semantics are always complete, that is all information that is specified in one rule element is overridden by the next one" in 2.0 to make the intent very clear.
Or we should change the example so it shows a local undefined information overriding a global one.

Cheers,
-yves

Received on Wednesday, 19 September 2012 11:54:42 UTC