Applying AI to Technical Document Translation

This is what Acolad did for a leading aerospace and defense company, which has worked with us for more than 20 years and continues to place its trust in us.

The client works with a team at Acolad that is specialized in aerospace and defense and is aware of the sector’s growing challenges. Having worked on various technical translation projects, this team presented the client with a range of innovative technological solutions, including neural machine translation.

A few months later, the opportunity arose for the client to use artificial intelligence to translate urgent, very lengthy documents that only required FIPO (For Information Purposes Only) quality, rather than the quality required for publication, and could be produced on a limited budget.

The Challenge: Overcoming Biases Against Machine Translation

The project involved translating 525 pages of technical specifications from Italian to French. They had to be ready within 24 hours for a team of French engineers who were traveling to Italy.

Initially, both the client and Acolad encountered a range of difficulties that could have impacted the project. One was the client’s lack of familiarity with more innovative translation solutions, which led them to doubt the quality of technical document translation performed by Neural Machine Translation (NMT). To reassure the client, our team carried out a demonstration and conducted tests, showing them the quality of work our technologies can deliver.

Since this was the first time we had offered an NMT service to this client, we also faced the challenge of outlining a clear and effective internal process that would facilitate the implementation of the new solution. However, the team had perfected their methods, drawing on many years of experience, and was able to convince the client to follow it.

A third challenge was the quality of the translation engine. Because the Italian-French engine was less developed than other engines we use more frequently, our proposed solution couldn’t be implemented immediately. What’s more, the source files requiring translation were in PDF format, which the client had converted into Word files using an OCR tool. This resulted in poor sentence segmentation, which was incompatible with neural translation and made syntax recognition by the engine impossible.

Despite these challenges, the client and Acolad were able to overcome them by maintaining regular communication throughout the project.

The Solution: Customized NMT and a Tailored Process

After examining these difficulties, we offered our client a solution that included the following steps:

Retrieve source files compatible with NMT (Word files at the very least).
Carry out an initial test of the raw NMT translation using the engine in its current state (i.e. not tailored to the client’s industry) in order to assess the relevance of the solution and submit it to the client.
Correct the sentence segmentation in the source files.
Improve the engine by feeding it with client terminology gleaned from translation memories from past projects.
Pre-translate using raw NMT, without any proofreading.
Check that all relevant information and required data is present and complete before delivery to the client.
Add a note to every page specifying that the content was generated by the Acolad neural translation engine to ensure that the client knew they were reading an optimized automatic translation, not one performed by a human translator.

Innovations Tailored to the Client’s Strategic Projects

The client reiterated its desire to continue working with Acolad because of our team’s ability to anticipate its potential needs and identify content and formats that could benefit from NMT.

While the economic benefits and speed of execution were undeniable arguments in favor of our solution, what really made the difference for the client was the engine’s reliability and the security of the data it contained, which is essential when operating in an environment where confidentiality is key.

The advantage of our solution was twofold: it allowed us to process documents within 24 hours, and the client received a special price.

The Results

Continued Trust and a Stronger Partnership

During our most recent steering committee meeting with the client, their team said they were “satisfied with the maturity of the engine”. In addition, our contact from the department responsible for translation requests would like to incorporate our NMT offering into its internal platform for translation requests.

It is important to take an educational approach with the client and to explain to them both the advantages and the limitations of NMT.