Machine Translation Quality Statistics

This page contains comparisons of translation quality of different machine translation services. Although this is not an universal ranking of best machine translators for all situations, this type of scoring provides an indicative state of play.

Machine translation quality is a highly controversial topic, with a lot of debate around the usefulness of automatic translations. The discussion has contained a plenty of personal views and experiences, what has made the topic a matter of opinions rather than facts. We hope to add a little more actual data into the conversation with these statistics from our tests.

Results: English – Spanish

es-machine-translation-quality-scores-2015-04

 

Results: English – Portuguese

pt-machine-translation-quality-scores-2015-04

Test Conditions

The analysis was made using MT-Qualifier, a technology that estimates machine translation quality automatically and without reference translations. MT-Qualifier evaluates the quality of several machine translation services automatically. The quality calculations are made separately for each segment and the overall quality score is the average of these numbers. The analysis contains up to six publicly available machine translation services: Apertium, Bing/Microsoft Translator, Google Translate, LinguaTec, Promt and Systran. The number of services may vary due to the available language options.

The comparison and analysis was made using the data from WMT14 – ACL 2014 Ninth Workshop on Statistical Machine Translation (For more information about the conference and material visit the official website). The material contains a representative collection of many types of texts. The topics include politics and economics. The analysis is based on translations from English to Spanish or Portuguese.

The results between languages may not be comparable because the test conditions (e.g. number of services or material) may vary.

Disclaimer: Machine translation quality is always context-sensitive. The machine translation services included in this analysis may perform very differently with other material or between other languages. Also any possible changes that are made in the technologies may affect the quality. These changes are mostly silent and released without an announcement.

If you have any questions or if you are interested in testing MT-Qualifier, please send us email: mt-quality.multilizer.com

[Updated on April 28th, 2015.]

 

 

Want to know more?
Write your email if you are interested in hearing more about MT-Qualifier.

Thank you!

©Multilizer