Select Page

Our Processes

Neuro Machine-Translation process

(Neuro MT & Human post-edition)

When it comes to translating, the translation process fully based on human activity is still by far the most requested nowadays, meaning that the content will first be translated by a professional human translator and revised by a second independent professional human linguist. This process fully based on human activity is indispensable for content coming from heavily regulated industries, such as the legal, financial or pharmaceutical sectors, however recent trends and advancements in AI technologies have revived the interest around machine-translated content and AI-generated content.

Transladiem, through our neuro machine translation process (NMT + human post-edition), has been successfully providing machine translation services coupled with human post-editing services since the last decade. Neuro machine translation, or NMT for short, is an advanced machine translation solution developed to use neural network models to learn a statistical model for machine translation. This is indeed a state-of-the-art machine translation approach that uses neural network techniques to predict the likelihood of a set of words in sequence, which allows the machine to check for the segment context, meaning the text coming before and after the target segment so to check that environments match before inputting the translation.
Translating via neuro machine translation engines is in essence different to getting AI-generated content. Unlike the current AI-generated translations, the neuro machine-translation process is more stable in terms of performance and delivers quality nearing the translation process fully based on human output on a number of language combinations. At Transladiem we have this expertise and can advise our clients whether neuro-machine translation is recommended for some rare languages and complex content type or not due to quality risks. Having said that, this process has proved efficient and qualitative on a variety of content types and language combinations which have allowed our clients to save time and money on their projects. Don’t hesitate to reach out to our team to check whether your project could be a good candidate for our neuro machine translation process.

Please see below a step-by-step guide to our Neuro Machine Translation process:

File analysis

Upon receiving the source files from our client, we identify the content type and check that the files received are all editable and open properly. We also analyse them against our translation memories to see the final word count.

Quoting and Planning

At this stage we provide you with a quotation in which we will itemise the services requested and break down the word count for translation. All our quotations are free of charge. This is also the stage where we discuss the costs and timeline associated with the different project steps. Once we are all set and get your go-ahead, we can start the localisation process.

File prep, Post-editor selection & Glossary creation

While our team is organising the material received and making files localisation friendly, we are also lining up the selected resources for the project, in this case, professional post-editors who have developed an acute sense of the typical oversights left by the machine to be fixed. We take great care in selecting the right post-editors for the material you send through. All our linguists have areas of specialisation and native languages which need to match your project to be considered. It is our job to find the best suitable linguists based on the specifics of the project. Depending on the client’s requirements, our team will now build a glossary as well to define the key terms, technical words and highly repeated words on which we need to consult our client’s SMEs in order to obtain approval before kicking off the actual neuro-machine translation stage.

Preparation of neuro engines, Neuro Machine translation & Verification

The approved glossary, together with the potential translation memories in place for our client, will be merged into a neuro machine translation engine prepped specifically for this project and that we will run on the content to perform the neuro-machine translation step. The translation obtained by the machine is then QA’ed by our project team through our internal verification step (adherence to translation memories and project glossary, truncation, punctuation, numeral equivalence, missing works, double space, etc).

Post-edition & Validation

Once the neuro machine translation step is completed and that the output has been verified by our team, we send the translated content to a professional human post-editor, who is an experienced linguist acquainted with the classic machine translation-related oversights and will then work on the content to rework unnatural phrasings, fix awkward wordings, smooth sentence connections and of course compare the source text against the target text to make sure the translation is accurate and true to the source text. The post-edited translation is then QA’ed by our project team through our internal validation step.

Client review & implementation of client review

Before typesetting the translated content back into the source environment, Transladiem always offers our clients the opportunity to review the localised content in a friendly format (a bilingual table with source and target content presented in a word document or via our online review platform) and to add any potential changes they wish to make at this moment. This step exposes our work directly to the client in order to get amendment requests or sign-off on the translation before proceeding to the next steps of the localisation process. This step is highly recommended but remains optional as not all companies have the capacity to run through this. If our stakeholders insert edits into the bilingual tables containing the translations, we will check these changes to make sure there are no typos, grammar errors, double-spaces or any kind or wrong insertions before implementing all amendments.

Formatting or Desktop Publishing & QA

When the localised content is signed off internally or via the client review step, we will typeset the translated content into the original source file environment to mirror the original documents we received. Our team will then QA the work of our formatters, typesetters or desktop publishing operators by skimming through the document to make sure all the elements are well displayed. Here truncation, alignment, punctuation, layout and display issues are fixed to make sure the file is ready to be used or printed.

Post-localisation testing & Final Eye

Through our post-localisation service, Transladiem commissions native testers to run the final linguistic QA before delivering to our clients. Indeed, populating the translation back into the source file will often shift elements and mess up the layout of the file, this is precisely what our testers need to flag to our Desktop publishing operators so they can reorganise the elements on the translated files as per the source file. Before delivering the final localised file, our project managers will go once more through the deliverables and compare them against the source file to make sure they look alike and that all parts have been localised and show in the right places.

Final delivery

Your dedicated account manager will deliver the final files to you via the delivery method agreed upon (email, FTP, secured link) and expect feedback.