Using the Three-Step Reflection Translation Method

Using AI translation (such as ChatGPT, Gemini, etc.) for subtitle translation can significantly improve translation quality by leveraging contextual information, although minor issues such as occasional blank lines may occur.

Video translation v3.12 introduces the "Three-Step Reflection Translation Method" to further optimize AI translation results. This method requires the AI to first perform a literal translation, then reflect on the shortcomings of the literal translation and propose modifications, and finally perform a free translation of the original text based on the suggestions. This three-step iteration can significantly improve translation accuracy.

However, the "Three-Step Reflection Translation Method" also has some limitations: first, it consumes more tokens; second, it requires a high level of intelligence from the AI model, and the output of low-intelligence models is often chaotic. Therefore, this feature only supports online large language models such as OpenAI ChatGPT and Gemini, and does not support local small models.

How to Enable:

Step 1: Enable Advanced Options

In the menu bar, click "Tools" -> "Options" -> "Advanced Options" in sequence, and then check the two options shown in the figure below.

Step 2: Select an AI Translation Channel

In the "Translation Channel" option, select an AI translation channel such as "OpenAI ChatGPT/Gemini/Claude AI/302.AI".

Step 3: Select an Appropriate Model

When selecting a model, for the ChatGPT series, be sure to select the "gpt-4" or "gpt-4-mini" model and avoid using "gpt-3.5". This is because the higher the intelligence of the model, the better the effect of the "Three-Step Reflection Translation Method"; otherwise, meaningless results may be obtained.

In conclusion, combining the "Three-Step Reflection Translation Method" with an appropriate AI model can greatly improve the quality and efficiency of subtitle translation. Please be sure to select the appropriate model for best results.