International Workshop on Spoken Language Translation (IWSLT) 2004

Keihanna Science City, Kyoto, Japan
September 30-October 1, 2004

PolyphraZ: A Tool for the Quantitative and Subjective Evaluation of Parallel Corpora

Najeh Hajlaoui, Christian Boitet

GETA, CLIPS, IMAG, Université Joseph Fourier, Grenoble, France

The PolyphraZ tool is under construction in the framework of the TraCorpEx project (Translation of Corpora of Examples), for the management of parallel multilingual corpora (coding, format, correspondence). It is a software platform allowing the preparation and handling of parallel corpora (languages, codings...), parallel presentation, and addition of new languages to existing corpora by calling several MT systems, and letting human translators produce the final reference translations by using a web-based editor. It integrates the computation of some objective evaluation metrics (NIST, BLUE), and enables subjective evaluations thanks to parallel presentations, and formating based on distance computations between sentences (at several levels). In the future, PolyphraZ should also support versioning and provide feedbacks to developers of the MT systems used: unknown words, badly translated words, and comparative presentations of the outputs of the various systems.

Full Paper    Presentation

Bibliographic reference.  Hajlaoui, Najeh / Boitet, Christian (2004): "Polyphraz: a tool for the quantitative and subjective evaluation of parallel corpora", In IWSLT-2004, 123-129.