Dear Kristina:
You write:
1) I'm not sure how the files with the source sentences are obtained. Are they
hand-crafted / auto-generated? Or are they taken from the same texts used in
creating the dictionaries?
Not necessarily. They should be sentences taken from texts of the style
that is supposed to be processed by the particular machine translation
system.
2) Does "reference translation" mean sentence translated by human, and is it in
target language?
Yes, and yes. The easiest way to do this would be to have a parallel set
of sentences in the source and target language, so that there is no
specific preparation needed.
3) Will all the possible combinations of information be compared? i.e.
just the reference translation / source sentence + reference translation /
machine translation + reference translation / all three ?
I think tests should contain questions of all types:
1. No hints (just fill in the holes using your imagination=
2. Source sentence as a hint
3. Target-language output of MT as a hint
4. Both source and MT output.
And, optionally,
1. [if available] an additional reference translation.
4) I was thinking of using Python for this, is that alright? I will be happy to
use (and learn if needed) other languages as well.
Python could be a choice, yes. As you are planning to have a user
interface, you should consider the possibility of using some web
scripting language for the web interface so that the test can be embedded.
Try to have a proposal draft as early as possible so that you can get
feedback from mentors.
I am also worried that I have not participated in open source projects before,
so any feedback/advice is greatly appreciated.
If there is anything specific that you would like to ask, just email me
as I am one of the mentors in that task. But the best is to hang around
in the #apertium IRC channel at freenode.net and ask the developers there.
Thank you very much for your time.
Good luck
Mikel L. Forcada
Kris.
Mikel L. Forcada
------------------------------------------------------------------------------
Better than sec? Nothing is better than sec when it comes to
monitoring Big Data applications. Try Boundary one-second
resolution app monitoring today. Free.
http://p.sf.net/sfu/Boundary-dev2dev
_______________________________________________
Apertium-stuff mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/apertium-stuff
