Machine Translation combined with Channel and Market Expertise
Artificial Intelligence has been receiving a significant amount of attention in the media lately as it finds more applications across a large number of industries. This is especially exciting since recent improvements in Automatic Translation software, referred to as Machine Translation (MT), mean that it’s now possible for numerous language combinations to get translations that are not only intelligible, but actually quite good.
Of course there is still a long way to go for machine linguistic output to be comparable to human, natural language production. Language is not an exact science. Often, there is no ‘perfect’ answer. While much is logical, many elements are harder to explain, untethered as they are to any fixed set of rules. For instance, when is a thought expressed with an indicative vs. a subjunctive mood? When to use polite vs. casual phrasing in languages such as Korean or Japanese? How to articulate an idiomatic expression that doesn’t exist in a target language?
At the same time, online content is booming, with growth in non-English markets fuelling the need for smart approaches to multilingual digital content production that avoid spiralling costs and can provide speed of execution and scalability.
Multilingual content services have seen a huge evolution over the last twenty years, with new methodologies and technologies, an increasing number of languages requiring content, and the need to adjust to the requirements of online platforms.
At Locaria we combine and blend state-of-the-art technology with our channel and market expertise, consultative service and solid human foundations to advise on the optimal methodologies for each type of content, considering intended usage, target, volume of work, budget and KPIs.
Our Machine Translation Post-Editing (MTPE) service is particularly suited to content that is large in volume, less creative in nature, requires fast turnarounds and typically does not affect user experience or final decision-making.
Many language businesses simply offer machine translation at a low rate without much consideration for its usage and desired impact. On the contrary, our MTPE service is based on a blend of outputs from select Neural Machine Translation engines, which is then carefully reviewed by a professional Post-Editor based on agreed KPIs and specific channel or vertical guidelines, before being signed off by our Quality Assurance team.
How Machine Translation Works
Selecting the right type of engine is the crucial first step. No matter how skilled a post editor may be, they will struggle to fix a sub-standard output, especially of high volume. The more dated Statistical Machine Translation (SMT) engines would learn to translate by analyzing large volumes of previously localised content. But the more advanced Neural Machine Translation (NMT) systems learn from observing correlations between the source and target text and modify their output for increased relevance and fluency. While SMT is still used in the market, including by major language businesses, NMT is Locaria’s technology of choice.
But offering the output of only one engine may be a considerable limitation for global clients who require several different language combinations and whose documentation spans many verticals. Our integrated MTPE workflow draws from a pool of select NMT engines which, based on our testing, have been found to be best suited for particular language combinations or verticals. For example, it would be a poor strategy to rely on the same engine to translate e-commerce product descriptions into Russian and also a set of terms and conditions into, say, Korean. We instead select the most suitable engine from our pool of approved ones, which will generate a better starting point for the post editor.
Our main engine is DeepL Translate, which we use for the less technical verticals (such as non creative marketing) and for the main European languages. But Asian languages, for instance, are better served using Asian local, specialised neural engines that are better at capturing the nuances of those markets where syntax and grammar can be so peculiar.
When it comes to the human post-editing element of the workflow, choosing optimal resources is key. Post-editing is a relatively new service for the language industry, which is unlikely to have been covered in many recent university language qualifications. For this reason, we have developed specific training plans for our linguists to become solid Locaria post-editors and ensure their edits focus on what is critical. MTPE will not generate the same quality level as the premium human workflows as this is not what our clients are looking for. Instead, edits must aim not only at removing language errors in the automated output, but also at making the content work for the purpose and the channel it was created for.
We offer two levels of MTPE. ‘Light Post Editing’ involves taking the raw machine translation output and performing as few modifications as possible to the text in order to make the translation understandable, factually accurate, and grammatically correct. ‘Full Post Editing’ is a slower and more in-depth pass which produces more accurate translations that consistently use approved terminology, are stylistically more fluent and are free from any grammatical slips.
After being post-edited, the copy goes through our Quality Assurance process to add a further layer of checks and cath any remaining discrepancies.
Understanding Machine Translation Post Editing rates and pricing
All these considerations will determine what Machine Translation Post Editing rates a company can offer. At Locaria, we believe that a combination of optimal neural engine selection, robust post-editing, channel and market expertise, and in-depth QA make our service reliable and effective.
Locaria’s MTPE workflows are launched from within Locate, our Multilingual Content Intelligence Platform. The copy is fed into the selected engine, machine translated and then distributed to local post-editors without our team having to switch to different software products or platforms, thus ensuring a faster response and higher productivity.
Locate can be connected to most CMS or CRM systems in order to further integrate the process. For example, Locate can be set to ‘watch’ a specific location in the client system, so that whenever new content is placed there, it is automatically extracted and fed into the MTPE workflow at our end. This is the optimal set up for large global clients with recurrent need for MTPE, for example for new product descriptions for an e-commerce website where this methodology has been found to be the optimal solution in terms of value and speed of output.
But while there can be a degree of automation in requests, file preparation and delivery, the overall production process still requires a significant degree of human intervention, driving costs for suppliers.
MT is a valuable tool for language service providers that can deliver benefits to clients, but must be considered as part of the overall arsenal of content creation methodologies and applied appropriately depending on business needs.
This leads onto the most important role that the humans in multilingual marketing and content services can perform for their clients: being the expert consultant that can make sense of the dizzying array of methodologies, linguistic considerations and digital regulations, and come up with the best approach to multilingual content creation for their clients’ business needs.