Smart Resource and Bible Translation Rubric
No AI was hurt in the process of writing this article.
This project is motivated by two problems that are related. In fact, I see them as two sides of a coin.
Heads
Motivation
Translators on the field suffer from the problem of too much information for drafting and checking. And still, the available information may not always be accessible (e.g. at their reading level). Resources can be narrowly focused on specific details that may not be relevant for the target language or for the immediate task of drafting or checking. Finally, all these resources are static and impossible to adapt for specific field realities.
The local church believer-translator is given a prerequisite verses/day goal and is faced with the dilemma of reading multiple resources to piece together the important information needed to create a faithful draft or check for specific issues. It would not be surprising if the translator chooses to not read through multiple pages before drafting or checking.
Goal
Create a resource such that we can assert that if the translator only read the one relevant portion from this resource, they would have the prerequisite information to produce a faithful draft.
Implementation
- Based on the notes and input of multiple recognized consultants.
- This is an important point- ultimately what would make this resource relevant is if it is backed by multiple recognized consultants in the BT space.
- Grounded-LLM synthesized. Spot checked.
- Generated using an AI harness.
- Granularity Dial: The resource is available at verse, pericope, chapter and the book level.
- Persona Dial: The resource highlights different aspects anticipating the perspective of the translation team member role (e.g. MTT, QC, Community member);in a later release.
Tails
“That which is measured, improves”.
Conversely,
"If you can not measure it, you can not improve it."
Motivation
Quality checking is acknowledged as a bottle neck in the Bible translation process. The underlying assumption for achieving a quality translation is by following a known process. This includes having the translation vetted by a Consultant. The Consultant represents special knowledge that may be amassed only through special training and decades of experience. Despite this, it is remarkable that there is no obvious semblance in the steps that different consultants take to check translations. This hints at a fundamental issue with the work of Bible translation today: there is not a standard rubric to compare a translation draft against. It may be true that it is impossible to develop such a rubric that covers all translation issues. It is, however, feasible to develop an opinionated rubric that when satisfied produces a translation with ‘baseline’ quality, i.e, it sets the floor and not the ceiling of quality. What matters is that the opinions reflected in the rubric are of esteemable Consultants that have shared their knowledge and are in agreement.
Hypothetically, if such a rubric existed and was acknowledged by the authorities/agencies, it has the potential to be the biggest accelerator to both quality and pace of BT. Expressed differently, if a translation satisfies this rubric then it would be true to say that the translation is as good as Consultant checked, i.e., meets baseline quality. Moreover:
- It supports generating targeted resources that guide both drafting and checking stages to ensure the rubric criterias are satisfied early on (as opposed to later stage intervention) in the process.
- It enables AI a clear target to pursue and opens the door to a whole new class of automated quality checks that are specific.
- Makes quality an overt attribute of the product itself and unsubstantiates judgements based only on the training and experience of the translation team (think CBBT teams).
Goal
Based on the opinions of an esteemable committee of BT Consultants, develop a verse/pericope level rubric which when satisfied demonstrably produces a faithful Bible translation that meets ‘baseline’ quality.
Implementation
- Put together big name consultants.
- Share notes, gather information and formulate the concept of ‘baseline’ quality.
- Using the collected data as context, use an LLM to develop a rubric at either a verse/pericope level.
- Along with the original languages, use major LWC Bibles as the basis for both developing and validating the rubric.
- Pilot the rubric on 2-3 target language projects and incorporate feedback.
- Publish the first version of the rubric for more general consumption.
- Positioning it appropriately is key. Showing how it works well with LWC translations and a handful minority languages should inspire confidence.