> I am working on adapting machine translation metrics for the
> purpose of evaluating student-written translations from source text
> (both in terms of generating an overall "score" for accuracy and
> identifying the nature/source of errors). My current application
> is for students learning ancient languages, translating texts from
> the L2 into their L1, but I can imagine other applications as well.
What you describe reminds me of the following conference paper in which the efforts of trainee translators (English/Japanese) was compared to professional translators. Nakamua, S. (2007). Comparison of Features of Texts Translated by Professional and Learner Translators. Proc. Corpus Ling. Conf. (CL2007) University of Birmingham, UK. 27—30 July, 2007. Davies, M., P. Rayson, S Hunston, P Danielsson (eds). Online. URL: http:// ucrel.lancs.ac.uk/publications/CL2007/paper/35_Paper.pdf (13 February 2008) You might also profitably look at the recently published book Augustín Llach, M P. (2011). Lexical Errors and Accuracy in Foreign Language Writing. Second Language Acquisition Series. Multilingual Matters. Although Augustin Llach's personal work (evaluating L2 acquistion in young learners) she does provide what appears to be an extensive literature survey of various L1/L2 errors.
> What I need for the next step in this project is a large collection
> of translations written by students. I know that translating text
> is something done more with ancient/dead languages than modern
> languages, and my local colleagues can't really help, but I was
> hoping someone in this community might do exercises with his or her
> students that might fit the bill.
I'm not sure that the subscribers to the LANTRA-L (Language Translation) list would agree with your assertion. The success of such in vogue authors as Steig Larson, Jo Nesbo, and other writing in their native Scandinavian languages would also suggest you are wrong. And as a sign language interpreter myself daily translating between sign language and English I definitely do not agree with it.
<>< Re: deemed!
-------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: text/html Size: 4680 bytes Desc: not available URL: <https://mailman.uib.no/public/corpora/attachments/20120111/be395a58/attachment.txt>