Looking for Advice on Developing a Specialised Translation Model for Technical Documents

1

Hello Everyone :hugs:,

Since I’m not too experienced in neural machine translation (NMT), I’ve been investigating several frameworks and tools to aid in the development of bespoke translation models. Because of OpenNMT’s extensive feature set and vibrant community, I have selected it.

I’m translating specialised technical engineering documents as part of a project I’m working on. There is a good deal of domain-specific vocabulary and intricate sentence construction in these publications.

Here are some details regarding my project and the difficulties I’m having:

Preparing the Dataset: Although it is not very vast, I have a bilingual collection of technical documents.

Which techniques work best for adding to this dataset so that the translations are more accurate? :thinking: Are there any helpful pre-processing processes or suggested data augmentation techniques? :thinking:

Model Selection: Should I use the normal transformer model given the technical content of the documents, or are there any particular model modifications or configurations that are ideal for technical text translations? :thinking:

Training Process: Which essential hyperparameters should I concentrate on adjusting to improve technical document translation accuracy? :thinking: Are there any recommendations for setting these hyperparameters’ starting values? :thinking:

Evaluation Metrics: In addition to the BLEU score, were there any additional metrics which are particularly helpful in evaluating how well a translation model performs when applied to technical texts? :thinking:

Post-processing and Fine-Tuning: What are the most effective methods for adjusting the original model with fresh data after it has been trained? :thinking: Are there any afterwards methods that can aid enhance the translated output in addition to this? :thinking:

I followed this :point_right: https://www.atltranslate.com/blog/8-best-practices-technical-documentation-translation-sap-sac

Thank you :pray: in advance.

1 Like
2

Hey Jessy,

I think the keyword you are looking for is called ‘domain adaption’ in Machine Translation. Here is a survey paper on it to get a grasp: A Survey of Domain Adaptation for Neural Machine Translation. From there you should find answers to some of your questions and how to proceed.

Disclaimer: I am currently writing my master’s thesis on a related topic but I am rather inexperienced. I just somewhat have gotten a ‘decent overview’ of the field based on what I had to learn to navigate the topic of my thesis. Probably not enough to answer any of your questions though.

Cheers
Jonny

1 Like