apptek / SubER

SubER - Subtitle Edit Rate
Apache License 2.0
21 stars 3 forks source link

Added length_ratio as metric #8

Closed patrick-wilken closed 3 months ago

patrick-wilken commented 1 year ago

Auxiliary metric to check whether the tested system has a general tendency to produce too long / too short hypotheses.

Should be pretty uncontroversial, only design decision is the choice of tokenization. (Even character ratio would be plausible.)