Translation Candidates

[Helena S Chapman <hchapman@us.ibm.com>]

I have a need to do "pre-machine
translation" analytics to gauge the quality of MT matches before sending
them to post editors and compare that to post editor productivity over
time. After re-reading the translation candidate module, and I am not sure
if this is a question limited to IBM. 

The attributes "similarity",
"matchQuality" and "matchSuitability"  has value
assigned to each attribute is dependent on the tool adding that information.
Another tool has no idea what method was used to set that value. Since
these attributes can only be a numeric value, 0.0 - 100.0, we can assume
that the larger the number, the better match it is. But if our tool was
adding more matches to a file, you do not know how to set those attributes
to be compatible with the previously added <match> elements.

David pointed out to me that
internally, we would probably have to use the <mda:meta> element
to add our unique information to each <match>. 

For example:
<mda:metadata>
<metagroup category="otm">
<meta type="segmentnumber">23</mtc:meta>
<meta type="documentname">test1.txt</mtc:meta>
<meta type="markup">OTMANSI</mtc:meta>
<meta type="MT-ServiceID">someMTEngine</mtc:meta>
<meta type="MT-MetricName">mtmetric_BLEU</mtc:meta>
<meta type="MT-MetricValue">90</mtc:meta>
<meta type="PE-EditDistanceChars">11</mtc:meta>
<meta type="PE_EditDistanceChars">5</mtc:meta>
<meta type="TM-DomainClass">primdomclass</mtc:meta>
<meta type="TM-DomainSubClass">subdomclass</mtc:meta>
</metagroup>
</mda:metadata>

If this is only limited to IBM, we can
just use metadata to capture that. 

Best regards, 

Helena Shih Chapman
Globalization Technologies and Architecture
+1-720-396-6323 or T/L 938-6323
Waltham, Massachusetts