Machine Translation

[Kent Karlsson]

Not that I'm necessarily arguing that a language tag extension should be
defined for this (in particular I'm not arguing for using BLEU scores), but
I'd just like to clarify a few things.


Den 2009-09-12 09.06, skrev "Felix Sasaki" <felix.sasaki at fh-potsdam.de>:

> Of course one could define several machine translation related extensions,
> 
I think so far, just one extension related to this has been alluded to,
varyingly using "-t-" or "-m-" in examples as singletons for it (in messages
earlier in this thread).

> but at some point the suitability of language tags is really in question. E.g.
> how would you represent a bleu score as an extension? The point is that proper
> metadata for evaluation of machine translation soon goes beyond a simple "can
> be represented as a closed set of strings" pattern.

The RFC describing the extension can allow for numbers to be expressed in
the extension, as long as the number is expressed as strings (could be
several per number) of ASCII letters and digits of length 2-8. The RFC need
not enumerate them. In particular, subtags that occur in an extension have
no relation to subtags that occur in the LSR.

However, I'm not arguing for actually doing this. So far there is no
consensus that an extension for "translation status" or similar should be
defined.

        /kent k


> Felix
> 
> 2009/9/11 Kent Karlsson <kent.karlsson14 at comhem.se>
>> What you say is correct for a (single) variant subtag, as initially
>> suggested, but extension subtags
>> work differently. See http://tools.ietf.org/search/rfc5646#section-2.2.6.
>> Data like that you refer
>> to can be put in the part that follows the extention "singleton".
>> 
>> Note also that section 2.2.6 starts:
>> 
>> "Extensions provide a mechanism for extending language tags for use in
>>    various applications.  They are intended to identify information that
>>    is commonly used in association with languages or language tags but
>>    that is not part of language identification.
>> "
>> 
>>         /kent k
>> 
>> 
>> 
>> Den 2009-09-11 18.35, skrev "Felix Sasaki" <felix.sasaki at fh-potsdam.de
>> <http://felix.sasaki@fh-potsdam.de> >:
>> 
>>> I would agree with Yves Savourel that for translation tool developers, this
>>> kind of information is better provided via a different field. Other
>>> practical information which one could not pack into a broad data category
>>> "machine translation" easily (to use Peter's terminology), but not easily in
>>> the "language tag" field would be: name of system that generated the
>>> translation (maybe several ones where used ...), quality of the input,
>>> quality rating of the system (e.g. BLEU score). IMO these fine grained
>>> differences are necessary for making use of this kind of metadata, and I
>>> don't see a clear use case for a broad "machine translated" sub tag.
>>> 
>>> Felix
>>> 
>>> 2009/9/11 Kent Karlsson <kent.karlsson14 at comhem.se
>>> <http://kent.karlsson14@comhem.se> >
>>>> 
>>>> Den 2009-09-11 17.32, skrev "Peter Constable" <petercon at microsoft.com
>>>> <http://petercon@microsoft.com> >:
>>>> 
>>>>> > From: ietf-languages-bounces at alvestrand.no
>>>>> <http://ietf-languages-bounces@alvestrand.no>
>>>>> > [mailto:ietf-languages-bounces at alvestrand.no] On Behalf Of Felix Sasaki
>>>>> >
>>>>>> >> There is a difference in the case of XLIFF. If the extension subtag is
>>>>>> just
>>>>>> >> similar,
>>>>>> >> but not identical to MT related information in other technologies like
>>>>>> XLIFF,
>>>>>> >> you
>>>>>> >> will end up with a mess of *values*. This is IMO different from the
>>>>>> script
>>>>>> >> subtag
>>>>>> >> case: Here you have the same values, but different *occurences*
>>>>> >
>>>>> > Expressed with different terminology: you end up with a mess of data
>>>>> > categories; in the script subtag case, you have a single data category
>>>>> with
>>>>> > many values.
>>>> 
>>>> I don't think that should be a major issue. XLIFF, and other formats having
>>>> separate attributes for this, could simply have that attribute take
>>>> priority, even to the extent that "language extensions", in particular one
>>>> that overlaps with an attribute, can be completely ignored in those
>>>> formats.
>>>> 
>>>>         /kent k
>>>> 
>>>> 
>>>> _______________________________________________
>>>> Ietf-languages mailing list
>>>> Ietf-languages at alvestrand.no <http://Ietf-languages@alvestrand.no>
>>>> http://www.alvestrand.no/mailman/listinfo/ietf-languages
>>> 
>>> 
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.alvestrand.no/pipermail/ietf-languages/attachments/20090912/8151967d/attachment.htm