Heuristic Word Alignment with Parallel Phrases

This paper presents a method for word alignment that uses parallel phrases from manually word aligned sentence pairs to align words in new texts. Experiments on an English–Swedish parallel corpus showed that the heuristic phrase-based method produced word alignments with high precision. Furthermore, alignment recall was improved by generalizing phrases with part-of-speech categories. We also compared the phrase-based method to statistical word alignment and found that a combination of phrase-based and statistical word alignments outperformed pure statistical alignment in terms of Alignment Error Rate (AER).