EjaghamISO 639-3 etuJohn 11:17
Nji Jisos a síi ágo, a yûk sê á kən Lasaros á ka mfín énî.
Read it in EnglishHide the English
“When Jesus arrived, he found that Lazarus had already been in the tomb for four days.”
John 11:17, from the Ejagham New Testament. The coloured letters carry tone marks: in Ejagham, tone carries meaning.
A research project by Ekaba Bisong, SiliconBlast
Machine translation between Ejagham and English
We are building translation in both directions from the Ejagham New Testament, and testing it honestly: on books the models never saw, and with people who read and write the language.
01
The language
A language written with tone
Ejagham (ISO 639-3 etu) is an Ekoid language spoken on both sides of the Nigeria–Cameroon border: in Cross River State, Nigeria, and in the South-West Region of Cameroon.
It is written with the letters ə, ʉ and ŋ alongside the Latin alphabet, and with tone marks.
Tone is not decoration. It carries grammatical meaning such as aspect and mood, and it is what tells the third-person singular pronoun from the plural: he or she from they. A translation that loses a tone mark can say something else.
That is why every tone mark counts, in the text we learn from and in every translation we judge.
- əschwa
- ʉbarred u
- ŋeng
- átone mark
Nigeria
Cross River State
Cameroon
South-West Region
02
What and why
Translation from the New Testament alone
For Ejagham, the only parallel text, the same writing in Ejagham and in English, is the New Testament. Its 27 books were translated into Ejagham by a team led by Dr John R. Watters (SIL).
From the New Testament alone, we built translation models in both directions: from Ejagham into English, and from English into Ejagham.
Translation systems usually learn from very large collections of text in both languages. For most of the world’s languages, such collections do not exist. If useful translation can be built from the New Testament alone, the same path is open to many languages in the same position.
The aim is practical: a free translation app for the people who speak Ejagham, and a research paper that shows exactly how it was made and how well it works.
03
How we evaluate
Measured honestly
A translation system is only as good as the test it passes. Ours is built so that it cannot flatter the models.
01
Whole books held out
The models are tested on whole books of the New Testament that they never saw during training, not on verses scattered through books they had already read.
02
Each direction on its own
Ejagham to English and English to Ejagham are different tasks. Each is reported separately, never averaged into one number.
03
A strong point of comparison
Our models are compared with a frontier AI model on exactly the same held-out books.
04
Judged by native speakers
People who read and write Ejagham are now judging the translations, without being told which system wrote which.
Results come after the evaluation
The people judging the translations may read this page, so we publish no results while their work is under way. The results and the technical report will appear here once the study closes.
04
Where we are
Three phases done, one under way
Phase 1, Done
Data
The Ejagham and English New Testaments, prepared and paired verse by verse.
Phase 2, Done
Models, tested on held-out books
Translation models in both directions, tested on whole books they never saw during training.
Phase 3, Done
Comparison with a frontier AI model
The same held-out books, translated by a frontier AI model for comparison.
Phase 4, Now
Native-speaker evaluation
People who read and write Ejagham are judging the translations.
Phase 5, Next
Translation app
A free app for translating between Ejagham and English.
Phase 6, Later
Paper and release
The research paper and the full results.
05
Get involved
Two ways to help
The project depends on the people who speak Ejagham. If that is you, or someone you know, we would like to hear from you.
You read and write Ejagham
Help judge machine translations between Ejagham and English, alongside other native speakers. You are paid for your time, and credited by name if you wish.
You have Ejagham text
Letters, stories, school books, songs, recordings with transcripts: any Ejagham writing beyond the Bible helps the models learn the language as it is used every day.
06
Acknowledgements
With thanks
Dr John R. Watters
SIL, translation team lead
Led the team that translated the Ejagham New Testament, the text this work is built on.
Nta Chris Anom
Access to the text
Facilitated the project’s access to the Ejagham New Testament.
The evaluators
Native speakers of Ejagham
Who are judging the translations now. Names will be added with their permission.