Translation
How to translate
LSDE offers powerful tools to allow us to translate into every language in the world.
LLMs ("Large Language Models") are human-designed mathematical architectures that excel at translation, provided they are given the appropriate context.
LS Dialog Editor provides a default context for every request addressed to an LLM, which ensures more relevant results.
Warning: Although the results are satisfactory and can considerably reduce production costs, it remains essential to use a professional studio for final verification.
To translate a dialogue with LSDE, let's start by adding the supported languages to our project via the language manager.
Open the i18n localization configuration window.
Add languages to your project via the indicated button.
Select the desired languages as well as the main project language.
The choices offered are in
The choices offered are in
ISO 639-1 and ISO 3166-1 format.Once the languages are added, make sure to correctly define the main project language.
This step is essential to unlock most of the software's actions.
While the main language should ideally be final, it can be modified at any time.
Next, select a key in the tree structure.
This selection will display all associated dialogues in the main editor window.
This selection will display all associated dialogues in the main editor window.
You can then filter the languages you wish to work on.
A right-click allows you to quickly toggle between displaying all languages or only the main language.
Translate with LLMs
First, click on a dialogue in the editor to activate its editing.
An icon will appear, allowing you to manage the desired task via a menu.
An icon will appear, allowing you to manage the desired task via a menu.
By default, LS Dialog Editor offers pre-configured tasks.
In this specific case, we will use the "Translate" task.
Available tasks can be reconfigured in the global settings if necessary.
Let's select the translation task.
A configuration window opens to adjust the task details:
Ability to enable or disable certain filters before launching.
Adjustment of iteration depth (if the processing concerns a folder).
Showing/hiding ignored keys according to current settings.
Manual activation/deactivation of instances to include if automatic settings are not sufficient.
Adding additional directives to guide the LLM.
Use the up/down arrows on the keyboard to navigate your global history.
Once the settings are adjusted, click "Confirm" (bottom right).
As soon as a task is created, it is transmitted to the automatic processing system.
The latter activates as soon as a project is opened to resume ongoing tasks.
All instances assigned to the task switch to comparison mode, allowing you to visualize the original text on the left and the response on the right.
This button displays the number of letters added compared to the original text.
It also allows you to show or hide added tokens.
This button displays the number of letters removed.
It also allows you to show or hide deleted tokens.
This button displays the final global state (positive or negative character balance).
It also allows you to switch to the final rendering if you cancel changes by clicking on the relevant text portions.
Modification rendering: you can click on each tag to cancel an unwanted change.
You can iterate as many times as necessary by sending additional directives to the LLM.
Warning: depending on the model used, caching is not systematic, which can increase costs exponentially because the entire history is sent back with each iteration.
Telemetry: displays the token costs returned by the LLM provider.
Buttons to accept or refuse changes once refinement is complete.
"Experimental": allows adding automatic corrections according to certain scenarios (e.g., if the LLM does not return the original text during a discussion sequence).
Navigation in the change history to return to a previous version.
Once the result is satisfactory, validate it and manually adjust the final details if necessary.