Machine Translations: Two Approaches

Human translators quite often wonder why would anyone use machine translations?  Larger companies quite often wonder why can’t machine translations be made better?   As simple as this sounds, we can not under estimate the complexities involved when deciding which option to consider.  To begin with, I think it’s important to define machine translation. It is “the application of computers to the task of translating texts from one natural language to another.”  This concept is much harder than it sounds. There are 2 different approaches that have been used thus far.

  1. Machine Translation (MT)
  2. Machine Translation PLUS Human Reviewer

For certain cases where extremely structured content is the norm it makes sense to use a MT system. These cases are far and few, however. Ford Motor Company uses a MT system to translate some of their content that is continuously and dynamically updated such as vehicle assembly build instructions. In these kinds of situations, the content has to be very structured and detailed; preferably written in a language known as Standard Language. This language is a subset of the English language and is restricted to about 5,000 to 10,000 words. Companies can create content in Standard Language that is specific to their operating environment. Additionally, the source content must be developed and/or converted into a format that a MT system can easily understand. Note that the MT system learns your environment specific vocabulary overtime so it can not be used right out of the box. Lot of engineering time is needed to customize the MT system for your environment.  For some companies this might be a cost prohibitive option.

Machine Translation PLUS Human Reviewer option is viable for situations where there are not lot of dynamic changes and at least some of the content is structured. Such examples include machine / engineering specs that have to be shared between engineers in different countries and the need for the translation is not urgent. In these scenarios, an MT system must be deployed for a company specific operating environment and previous structured text must be fed into the MT system in order to do any translations. The need for engineering time and cost to customize the MT system must be given a serious thought.