Reasons for Blank Lines in Translation Results

When using traditional translation channels such as Baidu Translate, Tencent Translate, etc., or when using AI translation but without selecting Send Complete Subtitles, the subtitle text is sent to the translation engine line by line, expecting to return the same number of translated lines.

If the number of lines returned by the translation engine does not match the number of lines sent, blank lines will appear.

During translation, the default trans_thread=15 means that 15 lines of subtitles are sent to the translation engine at the same time. The translation engine may merge two adjacent lines into one, especially when the two lines are semantically one sentence. This results in the number of lines returned not matching the original number of lines. This problem is more serious in traditional translation channels and less intelligent small models (especially locally deployed small models).

Example:

The original subtitles sent to it are 2 lines:

Line 1: You're one of my

Line 2: best friends.

The optimized result returned by the translation engine may be one line, that is, the two lines are automatically merged into one line for translation according to semantics, and the translation result is also one line:

You are one of my best friends.

Instead of the expected 2 lines:

Line 1: You are one of the

Line 2: best friends

Prompt Words Appear in Translation Results

When using AI translation channels, the translation results also output prompt words. This is more common with locally deployed small models, such as 14b, 32b, etc. The root cause is that the model size is too small, and there is insufficient intelligence to strictly follow instructions.

How to completely avoid these two situations?

0. Avoid using local small models, especially 7b, 14b, 32b, etc. If you want to use them, it is recommended to change Number of subtitles to translate simultaneously to trans_thread=1 and uncheck Send Complete Subtitles.
1. Open Menu -- Tools -- Advanced Settings and change Number of subtitles to translate simultaneously to trans_thread=1. But obviously this method slows down the speed and cannot take into account the context, so the effect is not good.
2. Use more intelligent online AI large models, such as gemini/deepseek online api, etc.