Machine translation doesn’t spell the end of the professional language services industry era – instead it marks the beginning of a new era, in which our work becomes more challenging, yet more efficient and more fulfilling.
Machine translation (MT) is a technology that translates text from one language to another in a matter of seconds, and has many uses in multilingual content creation and management. Often assumed to be a fully automated process without any human involvement, the building and training of customised MT engines is actually an undertaking that requires persistent dedication and expert knowledge.
A translation expert will typically use MT as a productivity tool that speeds up the process of producing human-quality translation. This requires full integration of MT into the human translation process and plug-in connectivity with your chosen CAT environment. If your goal in deploying MT is to improve the productivity of your translators, you must be able to measure the quality of your MT output and convert the results into a discount model.
Content owner companies can also choose to use unedited or ‘raw’ MT output in their communications portfolio. Raw output can work as a low-budget translation solution for short-lived content or internal information exchange in a company, but the nuances of crafting elegant prose will certainly be lost. If you are looking to invest in your own, custom-built MT engines that process specialised texts in a very narrow field of expertise, you should take into account how much bilingual data you need to build and train such engines.
MT engines need to be created, hosted, trained, used and maintained. Depending on your budget, time constraints and expertise, you could perform some or all of these tasks yourself. It is more common, however, to buy the services from MT vendors and pay for the use of their engines.
Every MT vendor has their own set of tools for managing the engines. Some tools have limited user-configurable options, whilst others allow users to fine-tune the training data. There are vendors who do not allow any user configuration at all, instead requiring that all changes be implemented by their own technicians. You have to pay for this hands-on service, but the option of influencing the engine’s output is only relevant to you if you have staff who are capable of using it.
The introduction of machine translation as a service and technology raises many questions in the minds of those working in multilingual content management. In our free machine translation e-book, we answer these questions and share our experience of working with MT. Download your copy now.
(but were afraid to ask)
Sandberg was the first company in the UK to achieve both ISO 17100 and ISO 18587 certification. ISO 18587 certification is awarded to language service providers who satisfy official requirements for the preparation and processing of full human post-editing of machine-translated (MT) material.