Translator++ Ver. 8.2.25 - Automatic Retry and Refinement for AI Translation

Hello fellow translators,

This month has been quite busy for me. As it’s the beginning of the year, I’ve been taking care of tax reports, visa paperwork, and various administrative matters that always seem to arrive all at once.

Still, I haven’t forgotten our regular Wednesday update 🙂

✨ Highlight of the Month: AI Translation Improvements

The main focus of this update is a significant improvement to the AI Translation in general.

🔁 Retry and Refinement

One common issue with AI batch translation is formatting inconsistency, such as wrong row counts (yea, I know some of you are annoyed by this 🫣), broken line structures, or outputs that don’t pass validation.

With this update, when a translation result fails validation, Translator++ can now automatically ask the model to retry the translation with additional guide and hints on what needs to be fixed.

What this means for you:

✅ Correct number of rows in batch translations

✅ Proper line formatting (matching line count per row)

✅ Fewer manual corrections

✅ More reliable bulk translation

This new option is available for the OpenAI and LiteLLM add‑ons. This means you can simply use it with any LLM service provider.

This is just the beginning. I’ve laid the groundwork for this retry-and-refinement system, which means I can expand and improve it further in future updates. Expect more intelligent handling, smarter validation, and better automation down the line.

🛡️ New Payload Strategy

Another major improvement is the introduction of JSONOBjectCloaking.

You can read the technical documentation here:https://dreamsavior.net/docs/translator/code-escaping-algorithm/

What does it do?

Instead of sending bulk text as a plain list, Translator++ now wraps the payload into an explicitly numbered JSON object like this:

Each row is assigned a clear numeric key.

Why this matters

Large language models sometimes:

Merge rows

Skip rows

Change ordering

Return the wrong number of lines

By forcing the structure into an explicitly indexed JSON object:

The AI clearly sees the required structure

Each entry has an anchor (0, 1, 2, 3, …)

The output is much harder to misalign

As a result, this dramatically reduces the chance of receiving misaligned or incorrect row counts in bulk translation.

Combined with the Retry & Refinement system, this creates a much more robust AI translation pipeline.

You can update your Translator++ to version 8.2.25 starting now.

Thank you for your continued support, it truly makes ongoing improvements like this possible.

Happy translating,

Dreamsavior