Evaluation of parallel text alignment systems
alignment tools are of primary importance in many domains
such as translation, terminology or lexicography, where
they help the construction of multilingual lexical
resources. There has been recently a growing body of
research in the area and various results have been
reported in the literature. However, these results are
difficult to compare, and to our knowledge, there was
previously neither a reference corpus nor a methodology
for evaluation of parallel text alignment.
The goal of the ARCADE exercise, financed by AUPELF-UREF, is to promote research in the field of multilingual alignment. The exercise is built around an ARPA-like competition between systems. It is expected that the evaluation and comparison of various systems and algorithms will result in a better understanding and an improvement of alignment technology.
Two two-years campaigns are planned. Six systems participated in the first campaign (1996-1997), and competed on the problem of sentence-to-sentence alignment. The second campaign (1998-1999) is underway, in collaboration with the SENSEVAL/ROMANSEVAL exercise on word sense dismabiguation. It involves new teams and tackles the problem of word and term alignment.
exercise will be concluded by a conference in Spring 2000
where progress and results will be reported.
First campaign (1996-1997)
Second campaign (1998-1999)
Final conference (Spring 2000)
Laboratoire Parole et Langage
tel : (+33) 4 42
95 31 37