Details
-
Bug
-
Resolution: Fixed
-
None
-
Critical
-
Text segmentation and text segmentation search in instant-translate only will be done only when for the current search TM is available or risk-predictor (ModelFront) is enabled.
-
Empty show more show less
Description
Problem
Currently the only validation if the searched text is valid for segmentation is when the segment contains more then 1 sentence.
Example of possible search test and the number of sentences:
| nr of segments = 0 | nr of segments = 1 | nr of segment > 1 |
|---|---|---|
| ' ' | This house is blue. | One house is blue. One is red. |
| . | A house?? | Red? Yes! |
| A house? | ||
| A house ? | Red: one | |
| A house! ! |
The problem occurs when for the current request there are only MT resources configured. And since all MT resources have static match-rate (no risk-prediction available ex: ModelFront), always the first queried resource will be displayed as result. Instant-translate segmentation should be active only when for the current search request there is TM configured and available or when machine translation risk-predictor is enabled (ex: ModelFront)