fake tab top


Customer:  Army Research Laboratories (ARL) Computational & Communication Sciences Directorate (CISD)

In integral collaboration with ARL Researchers, ArtisTech designed, implemented, tested, and continues to maintain an on-line experimental system to apply a range of recognized and experimental automated translation metrics to multiple Machine Translation (MT) outputs.  In this system, the researcher inputs a pair of documents, the original Foreign Language (FL) text, and human-expert translation(s) of this same text.  The on-line server-based system allows users to direct the FL text to be translated by any or all of a set of online MT systems in selected languages.  The multiple MT outputs are then run through a user-selected set of automated metrics such as:

BLEU: http://researchweb.watson.ibm.com/people/k/kishore/RC22176.pdf

Meteor: http://www.cs.cmu.edu/~alavie/papers/BanerjeeLavie2005-final.pdf

TER: (TER, Snover et al., 2005)

The results of these runs are then displayed to the user.    

The TIE (Translation Integrated with Evaluation) system supports ARL’s interest in the pertinence of different automated MT metrics for evaluating MT performance for Army requirements.  As these metrics improve and evolve within the linguistic community, ARL needs a research capability like TIE to efficiently and effectively compare, contrast, and characterize the contribution each metric can make to advance the Army’s understanding of improving MT performance.