Deep Learning Essentials

METEOR

Metric for Evaluation of Translation with Explicit ORdering (METEOR) was originally used for measuring machine translation. It combines recall and precision as weighted score components by looking only at the unigram precision, and recalling and aligning output with each reference individually and taking the score of best pairing (instead of BLEU’s brevity penalty). It takes into account translation variability via word inflection variations, synonymy and paraphrasing matches, which enables the match between semantic equivalents. Also, it addresses fluency via a direct penalty for word order: how fragmented is the matching of the MT output with the reference? METEOR has significantly better correlation with human judgments compared ...