SLTev currently assigns a token the timestamp of its first occurence even though later occurences of the same token may not have been generated at that time. When translating longer segments that may consist of multiple sentences, the likelihood of encountering tokens that were previously seen increases. In such cases, all of these tokens would be assigned the timestamp of their initial occurrence. Hence, this tends to favour longer segments in the translation process.
This issue has been fixed to take into consideration all the timestamps of the repeated tokens.
obo
commented
1 year ago
Fix for delay computation
SLTev currently assigns a token the timestamp of its first occurence even though later occurences of the same token may not have been generated at that time. When translating longer segments that may consist of multiple sentences, the likelihood of encountering tokens that were previously seen increases. In such cases, all of these tokens would be assigned the timestamp of their initial occurrence. Hence, this tends to favour longer segments in the translation process. This issue has been fixed to take into consideration all the timestamps of the repeated tokens.