Using AI in Bible Translation

Nguyen's MAP article on translating the OT using Artificial Intelligence

Duy Niem Nguyen just posted an interesting essay at MAP, "Using Neural Translation Technology to Translate the Old Testament from the New Testament." His essay includes step-by-step instructions for how to execute the process of using the data from a NT translation to provide a rough draft of the entire OT with the AI tools available by Google (read the whole thing here): 

The purpose of this essay is to raise awareness of using neural translation technology for Bible translation. My idea is using the New Testament vocabulary of existing translation and create the first draft of the Old Testament for that language with the neural translation technology provide by Google. In this essay, I will explain the neural translation technology, its potential, and practical usages for Bible translation with all the tools freely available for public usage. 

First, Google provides The Seedbank (https://research.google.com/seedbank/). It is a depository of the pre-made, ready-to-test programs in Python code for many applications of Artificial Intelligence. In this Seedbank depository, there is a program name Neural Translation with Attention (https://research.google.com/seedbank/seed/neural_translation_with_attention). This program is a demonstration of neural translation technology, and especially it shows the implementation of two well-known algorithms: seq2seq and Attention. This program exposes the inner working of a neural translation engine, and the best thing is: it is open-sourced, ready to modified and free to use for any kind of application. It is the program that I want to modify for the task of Bible translation. Google also provides the tool to test this program: the Google Colab cloud server...(continue reading here)

My brief analysis

The use of such AI, while perhaps helpful in fronting a significant amount of data for a team, should be exercised with caution by a translation team (and/or consultant). Some of my initial concerns are:

     1. Key terms 

The key terms in the OT are not the same in the NT, and the OT contains several key terms that do not occur in the NT. 

     2. Accurate and natural OT discourse

In addition, AI is not able to produce a document that is sensitive to natural TL discourse, let alone conduct good exegesis and analysis of the OT Hebrew/Aramaic, since the data is limited to two input translations--one in the target language and the other in a language of wider communication. Thus, the risk of producing a grand-daughter translation might be rather high, especially if the other translation steps are not practiced in rigor. Translation of individual terms and/or expressions does not equal faithful and natural clauses and discourse.

     3. Actual payoff or net-negative impact?

In the end, I have to wonder if using AI in the way Nguyen suggests is actually time efficient or productive (helpful in producing a draft of the Scriptures that is faithful to the Hebrew/Aramaic). If significant revision of the rough draft would be necessary, or if the discourse and other terms need complete overhaul (for example), how helpful is an AI-produced rough draft? Does it have net-negative benefits? Such analysis is impossible to make at present, but as more exploration occurs in this area, hopefully scholars will conduct careful studies to research these and other questions. 

Despite these shortcomings that seem intrinsic to the current limits of AI, I am nevertheless open to its (limited) use and development, in the hope that AI might become an effective tool for the global Bible translation movement.