The Duluth Word Alignment System

The Duluth Word Alignment System participated in the 2003 HLT-NAACL Workshop on Parallel Text shared task on word alignment for both English--French and Romanian--English. It is a Perl implementation of IBM Model 2. We used approximately 50,000 aligned sentences as training data for each language pair, and found the results for Romanian--English to be somewhat better. We also varied the Model 2 distortion parameters among the values 2, 4, and 6, but did not observe any significant differences in performance as a result.