NIST is a metric used to measure the goodness of translation.

In the paper, Doddington (2002) introduce the notion of "Information weights"

Information weights were computed using N-gram counts over the set of reference translations, according to the following equation:

Which Owczarzak (2008) describes as

Here the precision is modified by the information weights Info(w1 . . . wn), computed on the basis of the frequency of a given n-gram in relation to the frequency of a ‘parent’ (n-1)-gram, in accordance with the equation (2.6)

There's an implementation of the NIST score on https://github.com/moses-smt/mosesdecoder/blob/master/scripts/generic/mteval-v13a.pl

But we can't seem to figure out how should the information weights be computed on https://github.com/nltk/nltk/pull/1840

Does anyone know the details of how this information weights work?

I've tried looking at the mteval-v13a.pl script but I'm not understanding how it works and relates to the formula above.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.