Re: [Issue-41][Action-195] Produce next draft of mtConfidence

Hi Dave,

I'm also not clear on the override semantics, maybe Felix can confirm
whether in fact that by specifying a local value for mtConfidence also
requires to specify locally the value for both mtEngine and mtProducer.

Referring to point 2 below, the issue is that mtConfidence on it's own is
rather meaningless and difficult to interpret, as what it's value
represents is entirely dependant on being able to tie it to the engine that
produced it seeing as there's no "industry standard" of how mtConfidence is
calculated.

Intuitively it makes more sense to expect mtProducer and mtEngine to be
specified at a global level in the majority of cases and mtConfidence at a
local level (actually mtConfidence should always be specified at a local
level in reality, unless the entire content is translated in one huge chunk
which is highly unlikely, as even then an MT system generally operates on a
sentence or segment level).

(Attached is the current verion of the mtConfidence doc, taking into
account Yves and Felix suggested ammendments and fixing the </span> tag
error as pointed out by Dave)

Declan

On 18 September 2012 15:54, Dave Lewis <dave.lewis@cs.tcd.ie> wrote:

>  Hi Declan, David,
> i have a question about the confidence score data category and how it uses
> inheritance and overriding. You have example 2:
>
> Example 2:  Global usage of mtConfidenceRule, mtProducer, and mtEngine
> (specified with a sample privately structured string) along with local
> usage of mtConfidenceScore.****
>
> <text>
>     <its:rules
>         xmlns:its="http://www.w3.org/2005/11/its"<http://www.w3.org/2005/11/its>
> version="2.0">
>         <its:mtConfidenceRule selector="/text/body/p/"
>         its:mtProducer=”vanilla Moses”
>         its:mtEngine="medical:EN-ES_LA” />****
>
>     </its:rules>****
>
>   <body>
>     <p><span its:mtConfidenceScore=”0.9876543”> Lavar y secar bien las
> manos es fundamental para prevenir la propagación de gérmenes.</p>
>    </body>
> </text>****
> But isn't it the case that overriding of a data category is _complete_,
> i.e. the span element in the example (missing a </span> by the way) will
> complete overwrite the global values for the <p>. So it will have the value
> for mtconfidenceScore, but not for mtProduce or mtEngine which i don't
> think is the intended outcome?
>
> i know you and Yves touched on this in:
>
> http://lists.w3.org/Archives/Public/public-multilingualweb-lt/2012Aug/0136.html
>
> ... but you were talking about defaults addressing this in the general ITS
> overide case, but this isn't helpful here since the values are open ended,
> so no meaningful default (apart from 'undefined' perhaps) is possible right?
>
> I'm not entirely sure i'm interpreting this completely correctly, since
> the statement about completeness in
> http://www.w3.org/TR/2012/WD-its20-20120829/#selection-precedence
> says:
> "Override semantics are always complete, that is all information that is
> specified in one rule element is overridden by the next one.", i.e. it
> refers to inter-rule overrides rather than local-global override
>
> But from the very useful if-then-else description of overriding felix
> provided in:
>
> http://lists.w3.org/Archives/Public/public-multilingualweb-lt/2012Sep/0011.html
>
> I take it that local value override global values in the same way later
> rules override earlier rules - right Felix?
>
> if so then:
> 1) we should clarify local-global override is complete in that section
> 2) we may need to revisit the idea of specifying engine and producer in a
> separate data category
>

> cheers,
> Dave
>
>
> On 31/08/2012 13:25, Dr. David Filip wrote:
>
> All,
>
>  please find attached the next draft.
>
>  I tried to resolve most of the editorial comments Yves and others had re
> the first draft.
>
>  I now refer to general inheritance and override behavior.
> I consolidated to <0;1> giving a hint to implementers that they can
> interpret this as a percentage etc..
>
>  After a brief discussion with Dave on provenance and translationAgent,
> we agreed that covering mtConfidence needs in translationAgent would make
> the translation agent unnecessarily convoluted (this might change if WG
> decides that ITS needs more complex translation agent reporting and
> recording). mtConfidence information is extremely perishable and the
> information needed for identifying an MT engine virtually does not overlap
> with the translation agent in its present shape.
>
>  BTW I miss a processing requirements section in the template..
> Where should I say that
> For self-reporting MT confidence of produced MT quality, MT producers MUST
> use mtConfidence.
> Any agent replacing or modifying raw MT in a segment (text element) MUST
> remove or archive all the MT confidence markup.
>
>  Rgds
> dF
>
>
>  Dr. David Filip
>  =======================
> LRC | CNGL | LT-Web | CSIS
> University of Limerick, Ireland
> telephone: +353-6120-2781
> *cellphone: +353-86-0222-158*
> facsimile: +353-6120-2734
> mailto: david.filip@ul.ie
>
>
>


-- 
Dr. Declan Groves
Research Integration Officer
Centre for Next Generation Localisation (CNGL)
Dublin City University

email: dgroves@computing.dcu.ie <dgroves@computing.dcu.ie>
 phone: +353 (0)1 700 6906

Attachments

Received on Tuesday, 18 September 2012 16:51:36 UTC