Machine translation of ELAN tiers

I just saw this on RNLD. It looks like a relatively simple idea that could be a boost to make documentary more available and usable. Has anyone tried it out?

Raphaël Merx has written an online service that takes an Elan file and translates tiers. He wrote this for work I am doing with Bislama (Vanuatu) and it translates from Bislama to English, but you can select to translate to and from a number of languages. It creates a new tier and you then download the .eaf file with the translation. https:// elan-trans-api.rapha.dev/. Feedback welcome.

2 Likes

It looks pretty neat, but I wish there was just a little more information about what’s under the hood

From Raphaël’s website:

Translate ELAN files (used for audio transcription) from over 100 languages using an LLM backend. […] The translation is context-aware, and of good quality for mid-low resource languages (Tetun, Bislama).

Might be related to this work of his: GitHub - raphaelmerx/tulun: Transparent and Adaptable Low-resource Machine Translation, through LLM post-editing

1 Like

ah, ok, so maybe it’s that on a server with an ELAN formatting wrapper?