Ask yourself these ten questions when considering an MT solution:
1. Language combination? Despite the quantum leap in quality achieved since the advent of Neural Machine Translation (NMT), some combinations still perform much better than others. Languages for which there are few available resources, such as minority, regional, endangered, or heritage languages, are poor candidates for MT.
2. Literal or creative content? If the content is literal, MT is much more likely to perform well. Legal or manufacturing content, such as instructions for use, are typical scopes that are highly suitable for MT. Magical realism in a novel, on the other hand, is likely a recipe for disaster.
3. Scanned or readily editable? Be sure to weigh the advantages of MT against the costs of preparing the input in order for the MT engine to generate a decent output. Cost and time spent on file preparation may outweigh the benefits of MT in some cases.
4. How’s the grammar and style? Care must be taken with structure and punctuation when developing global content to be localized later with the help of MT. Certain trades in particular tend not to concern themselves with the quality of their written materials, and poor-quality input inexorably leads to poor-quality output from any machine; translation engines are no exception.
5. Simple or complex formatting? Some MT solutions may disrupt formatting. However, this drawback seems to be rapidly improving, and will most probably be dropped from this checklist in the near future.
6. Plain language or highly technical? Simple, well-written grammar will improve MT performance. Terminology has been shown to be a challenge, particularly when discussing the more powerful NMT engines.
7. A baseline or customized engine? Please refer to our section on types of engines and the differences between generic and specific ones.
8. Translation Memory to leverage? MT is naturally a great option when generating large volumes of new content. However, when content must be updated, leverage will mostly come from a well-maintained Translation Memory.
9. MT already integrated in your workflows? Nowadays, most CAT tools are integrated with either baseline or customized engines via API. GTT (Google’s useful translation toolkit) is gone for good.
10. Capitalization in your content? One stumbling block for MT that has not yet been overcome is the use of capital letters. For example, a title in English showing capitalization will be better translated to a Romance language if you manage to lower all uppercase and, if needed, restore that in the target. Needless to say, content that makes liberal use of capitalization, such as legal documents, may pose a serious challenge.
