A reference based subtitle quality assessment model is a model developed to assess the quality of machine made subtitles by cross referencing with human made subtitles of the same subject. The models return a score, usually between 0 and 1 (with 2 decimals, e.g 0.50), where 1 is the highest of quality and 0 is the lowest.

The different models use different metrics in order to assess quality and are therefor not comparable by the score returned by the models. Some are also made for a specific use case, e.g interlingual or live subtitles.

List of assessment models

  • SubER1
  • FAR2
  • WER
  • NER
  • BLEU
  • METEOR
  • ROUGE

Footnotes

  1. Wilken, P. (2022). SubER: A Metric for Automatic Evaluation of Subtitle Quality. Retrieved from https://arxiv.org/abs/2205.05805

  2. Pedersen, J. (2017). The FAR model : assessing quality in interlingual subtitling. JoSTrans, (28), 210–229. Retrieved from https://urn.kb.se/resolve?urn=urn:nbn:se:su:diva-146053