Blending machine translation and human translation

Machine translation (MT) is the automatic translation of text by software, or by a learning algorithm based on a neural network. At the core of MT is the translation memory, which has been trained in a language, based on specialised terms and reference materials.

With machine translation services, it is now possible to produce translations in many languages quickly, economically and to an acceptable quality. However, no MT system is currently capable of producing error-free translations or translations at the level of human translations. This is where machine translation post-editing (MTPE) comes in.

MTPE is a hybrid approach that combines the quality and accuracy of human translation with the speed of machine translation. It’s a relatively new service that’s grown rapidly with the rise of AI and its integration into daily life. Our philosophy is that, instead of clinging to the status quo, savvy language service companies and linguists must adapt to advances in technology if they wish to remain relevant. For us, savvy linguists are those who embrace these changes by working with machines as opposed to against them. We see MTPE as a symbiotic relationship between man and machine.

MTPE at the intersection of translation and editing

As far as language services go, the MTPE service is an interesting offering as it blends two previously distinct language services: translation and editing. This hybrid approach means that, while the final product is essentially an edited product, the human translator or team of translators assigned to the project, in addition to being adept professional editors, must also be seasoned translators. This is because MTPE involves comparing two versions of a text: the original document (known as the source text) and the version that has been translated by a machine (known as the target text), in order to compare the two and determine which changes and corrections need to be applied.

Types of MTPE

Light post-editing is recommended when you need to resolve major errors and unreadable content. This approach is recommended for limited timeframe projects or documents intended for internal use (internal communication).

Full post-editing involves a more substantive edit and more thorough and rigorous quality checks. Full post-editing is a more suitable choice for larger-scale projects designed for public posting or dissemination (external communication).

When to use MTPE

MTPE services are typically best suited to high-volume projects that are extremely time sensitive, or internal communications.

MTPE is a good choice if you need to translate a large volume of non-critical text quickly and cost effectively. If you need to translate sensitive content, creative content, or content that requires specialised knowledge, then human translation is a better choice. This is especially true for sensitive or confidential content, as your material will be fed into a machine and potentially be used to train future language models. This poses some obvious risks relating to privacy and security.

It cannot be disputed that human translation still trumps machine translation in terms of accuracy, quality, context, nuance, and idiomatic language. This is especially true when translating between the Romance languages and English. Take French, for example. The French language has fewer words than English: the Larousse French Dictionary lists around 130 000 French words, whereas the Oxford English Dictionary lists 200 000 English words. By implication, the same word in French, spelled the same way, can have several different meanings. Most automated translation software programmes lack the ability to translate context, so they may churn out a mistranslation, or a translation that has a completely different meaning that which was originally intended. Human translators also possess one quality that machine translation lacks: judgement.

With the MTPE projects we handle, the client’s source text is fed into a machine translator. One of our post-editors then carefully reviews the text, following the protocol described above, for accuracy and meaning. They then deliver a target text that has been covered by both phases in the MTPE process: translation and editing.

Advantages of MTPE

It can be faster than 100% human translation. By leveraging the speed of machine translation, you can save time by limiting the amount of time required for a human linguist to refine and polish your text. With one important caveat: the speed of an MTPE task varies depending on the quality of the MT software you use. When the software is rudimentary, it can take human post-editors longer to edit the translated text than they would have had they started translating from scratch.

Potential cost-saving benefits. This is because the “heavy lifting” (the initial translation phase) has already been undertaken by MT in rough draft format, freeing up time for a human linguist to edit and refine the target text without necessarily needing to translate from scratch. Again, though, this varies widely depending on the quality of the MT software.

It’s scalable. MTPE is particularly well suited to very high-volume projects and projects that are extremely time sensitive, where manual translation has the potential to slow down the process.

Pricing and Process

For maximum transparency, we will always quote you per source word for MTPE. We’ll also give you an estimated turnaround time at the time of quoting, based on the word count and the level of complexity of your project. This way, you will know upfront exactly what the job will cost you and you can budget accordingly. In the unlikely event we run into delays, we will communicate with you immediately. No hidden fees. No mixed messages or unkept promises. Total transparency.

At The Savvy Linguist, all our language services come with two free reverts. We follow a rigorous quality assurance process for all written content. We use the Track Changes function in Word, Excel or Google Docs to mark up additional edits, so you can see exactly what we’ve changed or what fixes we’ve suggested, and why, allowing you to maintain control over your product at every step of the process.

For MTPE projects, we prefer using Excel as it allows our post-editors to compare the source text with the machine-translated text in adjacent columns and record the edits alongside both the source and target texts.  If, after two rounds of review, you still want to make changes or additions to the text, we will quote you an hourly fee from that point onward, which we will, of course, communicate with you upfront.