Feature suggestion - beat translated versions

Causality Story Sequencer
1

Would be good if beats would have:

  1. Versions for different languages (At least a 2nd one)

  2. A way to see side by side two versions that would sync vertical position, when we scroll, such as for writing our own translation and be able to compare

  3. A flag for showing if it is considered translated (Finish)

  4. A way to click next or previous untranslated beat
    or anyway the next or previous beat (Without regard if it is translation complete or not)

  5. For not inflating file size too much), there could be only 2 language versions, but an option to select which one we delete in an automated saved version (So fail safe, for not overwriting the previous file): Ex: An English-French version, saved in an English-Empty version.
    From there, we could create an English-Spanish version.

  6. If possible, AI translate with an OFFLINE local Python code (Such as with the Argos Translate library), with the appropriate training file for from-to actual langage transition, such as English to French.

My use case:

I write the main version in English, but aim at submitting it both in English and French to potential production/financing partners or actors, for which English is not the first language, for making their first reading more enjoyable.

In my case, the main lead actress mostly understands Brazilian Portuguese, although the script is meant to be shut in English.

Here is a fast search on Python offline translation libraries:

Argos Translate (best simple offline solution)

The most common fully offline Python translation library is Argos Translate.

  • Open-source
  • Works 100% offline
  • Uses neural machine translation models
  • Supports 40+ languages, including English ↔ French
  • Python API available
  • Models must be downloaded (~100-300MB each)

Hugging Face Transformers (offline but heavier)

You can also run transformer translation models locally using Hugging Face Transformers.

Typical models:

  • Helsinki-NLP OPUS-MT
  • MarianMT
  • Higher translation quality
  • Large variety of language pairs

Limitations

  • Larger models (300MB–1GB)
  • Requires PyTorch

OR A MORE FIRST HELPER WORD TO WORD TRANSLATION:

word2word (very lightweight dictionary translation)

A minimal Python library based on bilingual lexicons.

  • Very small
  • Works offline
  • Word-level translation only (not sentences)

It provides translations for thousands of language pairs extracted from parallel corpora.

2

The feature of parallel script expressions has been in the pipeline forever, but we couldn’t do it until now.

The feature is that you have multiple expressions shown side by side, e.g. an English version next to a French version, next to a Storyboard version, next to a Novel version. The underlying story is the same, it’s just expressed different ways.

The idea then also it to be able to track changes made across expressions, so that if the English version is updated, the French version is also prompted to be updated.

I’ve always believed in the this feature, and I’d like to see it soon. The only thing that has held it back is that the entire code base refers to “the script”, so in a thousand places, it must now be specified which expression. That’s some amount of work and some amount of risk of subtle bugs, so it can’t be done casually.

And sure, you could also auto-translate versions. We don’t want to do local models, because the installation hurdle is beyond almost all users, usually requires CUDA hardware or separately installed ONNX inference engine installed to not run very slowly, and is not a thing on mobile or Chromebooks. These features will used hosted models. There’s no need to host one’s own models for something this basic. Our only real use case for hosting our own model is to have an uncensored LLM, so that the provider’s community standards don’t interfere with screenwriting.