Machine translation does solve a lot of problems for marketeers and global content managers: it brings speed, it can chew through gigabytes of content, and because of these two reasons it makes translation possible for many content types that would never see the light of day in another language otherwise.
However, MT content is still not on par with human-quality translation in many languages and domains, even with the newest MT kid on the block: neural MT (NMT). Of course, the jury is still out on how to measure MT quality itself, as we have now largely moved from previously used approaches such as statistical and rule-based MT to NMT. Some of the previous parameters of MT quality would analyse in terms of the number of corrections required (edit distance) or how close MT strings are to human translation (BLEU score). In neural MT, however, often there are no grammatical errors in sentence structure, etc., but sometimes the overall text may lack coherence. So, we may need different ways to gauge MT quality.
Not all content types may need the same treatment. There will always be some that can be published without editing. For instance, user-generated content such as reviews or feedback. But if higher quality is required depending on either locale (such as quality-conscious Japan) or sector that is when you’d deploy MTPE. Industries such as legal, travel, IT/software (knowledge bases, user guides, manuals) automobile, education and training are the usual candidates for MTPE.
Ideally, you would not want to edit the MT output heavily, quite simply because that defeats the purpose of using MT in a way. How much post-editing is required will depend on the quality of the MT raw output and the level of quality required. Post-editing is used to fill in the gap between these two factors.
To make post-editing fruitful, sit down with your language service provider (LSP) providing the service so that everyone is on the same page. Your LSP can also provide you structured feedback which will help you understand the types of errors creeping up in the output and go towards improving the engine over time.
In general, there are two levels of post-editing your LSP will offer: light and heavy or full.
Light post-editing is used when the raw output is not of very poor quality or when “good enough” quality will do. In this approach, the purpose is to inform the reader, not create flawless content. Generally, the types of errors that need to be corrected in this approach are lexical and syntactical. Lexical errors refer to wrong word usage, while errors in syntax refer to faulty sentence structure.
In heavy or full post-editing, the goal is to equal human-quality translation. As such, apart from the above mentioned error types – lexical and syntax – style, fluency, and less obvious errors too are corrected.
Below are a few guidelines to help you get the maximum value out of MT post-editing:
First off, write for machine translation. When creating the source content, follow practices such as using the active voice, writing shorter sentences, restricting one idea to a sentence, and avoiding ambiguity. This helps in getting relatively error-free raw MT output and thus bringing down the post-editing effort.
Pre-edit where required. Some tweaks such as fixing spelling errors, formatting the document, and flagging words not to be translated will contribute to a better output.
Invest in trained MT engines. While generic MT engines such as those freely available on the internet are a strict no-no, spend some time and effort in modeling a custom engine suited to your domain and target languages.
Work closely with post-editors. They are experts in their domain and are highly knowledgeable in the target language. Talk to them about what your expectations are from the content and they will help you arrive at the quality that will align with these expectations.
Choose the level of post-editing carefully. It does not always have to a black-and-white choice between light and full post-editing. Don’t go for blanket decisions that will apply to all of your content. Analyse every content project and determine how much editing you require.
To achieve scale and speed in global communication, companies are using MT than ever before and this trend is only set to go up. However, you may still have some concerns about the quality and brand voice that you are delivering to your customers. Post-editing gives the control you seek and lets you smartly leverage MT to the maximum.