April 30, 2021 –
Automatic translation-quality analysis metrics are indispensable for the fine-tuning of personalized machine translation (MT) fashions in addition to basic pure language processing (NLP) analysis. BLEU–a precision-based metric–stays the most well-liked. However, extra correct metrics, which think about precision and recall along with different elements similar to hLEPOR, have demonstrated higher correlation with human judgments. Previously, among the many elements that prevented the broad use of extra superior hLEPOR was the shortage of public Python implementation.
AI (Artificial Intelligence) builders from Logrus Global, in affiliation with Lifeng Han, the principle writer of the unique metric, have accomplished the Python port of the compound hLEPOR metric, as introduced within the unique article, and made it obtainable to the whole Python improvement neighborhood through PyPi.org.
The hLEPOR is extra exact with respect to the elements of precision, recall, sentence size and variations in phrase positions. Additionally, it permits per-sentence analysis scores in addition to document-level rating (versus BLEU) and is offered freed from cost. The uniform, single-source computerized baseline metrics are simply obtainable to everybody, benefiting practitioners and researchers alike. Further enhancements with the combination of deep studying language mannequin know-how into the metric are on their approach too.
The library is offered at https://pypi.org/project/hLepor/