CS159 NLP Final Project

We compared the performances of IBM Models 1 and 2 (in the NLTK package) by evaluating the quality of the alignments obtained after the training process on 3 corpora (French, English and Vietnamese) with 4 metrics (precision, recall, F1, and AER (Alignment Error Rate)), with our own hand alignments as the “golden alignments”. The results indicated that IBM Model 2 performed marginally better than IBM Model 1, and both models performed worse on longer sentences.