When to Use Machine Translation vs Human Translation?

Human and Machine Translation

Deciding whether to use machine translation (MT) or human translation depends on the text type, complexity of the text, and purpose of the target text, among other factors. But do we really need to choose between one or the other?

The difference between MT and human translation is the use of technology. MT involves software automatically translating content without any human input, while human translation sees a person decipher the source text and manually translate it. 

Human translators may utilize technology, such as CAT tools, but they take the lead and the tech simply assists in producing the final output.

The question of when to use machine or human translation also comes down to the target text quality required. MT quality has witnessed significant leaps forward over recent years and is constantly improving: from Rule-Based Machine Translation (RBMT) and Statistical Machine Translation (SMT) to Neural Machine Translation (NMT) and the quality level achieved today.

Machine Translation: Pros and Cons

The primary advantage of machine translation is the high speed and low price tag. This balance of cost and speed means it can be beneficial for leveraging large volumes of content.

In terms of quality, MT can allow better consistency in wording and style; it can reliably implement the same translation of a term or phrase throughout a text. However, the issue is that the translation itself may not be perfect.

Indeed, MT output is not 100% accurate. MT cannot understand the context in which a word or phrase is used or recognize where there might be a deeper meaning than the surface sense. MT particularly struggles with nuances, connotations, and cultural sensitivities.

Additionally, biases have been found in MT. E.g., gender bias. MT biases are cases of ambiguity whereby a source language expression may have two or more senses. An example of MT gender bias would be translating a source expression that is gender-neutral (e.g., the English doctor ) which has to be gender-specific in the target language (e.g., the German Arzt for a male doctor or Ärztin for a female doctor). If there is no trace of the intended meaning in the text, no AI or machine can just guess what the meaning should be so a human-in-the-loop is needed; we cannot fix MT bias by improving the AI. 

Some are also concerned by MT’s susceptibility to manipulation by attackers that could subsequently produce potentially harmful output.

Human Translation: Pros and Cons

The output of human translation is (generally undisputedly) of higher quality. Unlike MT, human translators can be curious, investigate, use their own experiences, collaborate with other translators, and utilize all the physical / online tools available. They can identify the deeper meaning and nuances to work out the right way to translate a specific source meaning.

Human translators can also take brand style, glossaries, target audience, intended purpose, and visual context into consideration during translation. This is particularly important when it comes to localization and transcreation. Humans are capable of transforming a message and ensuring it resonates with the target language audience in the same way as the original. Utilizing linguistic assets also helps to ensure consistency within and across texts.

However, human translation is more labor-intensive which also contributes to a higher price tag. Sourcing this labor is an additional drawback; unless businesses have their own in-house linguists, they have to outsource this work to translation agencies or find their own freelance translators.

The expense of human translation is also exacerbated by the fact that the process is a lot slower than MT.

Furthermore, human translations may lack consistency if translators are not meticulous. Careful attention to detail is required, however using technologies, like CAT tools and translation memory systems, can improve consistency.

Machine Translation and Human Translation in Use

Machine translation is suitable for content that simply needs to be fit for purpose, rather than perfect. For instance, non-critical content or texts to be used within a company.

Similarly, MT should be used when high volumes of content need to be translated as well as for low-priority content (e.g., user reviews, comments, or social posts). Companies may also opt for MT for translating texts which are part of a new business area, for sections that are not yet bringing in a lot of revenue, or for trying out a new market as using human translators is a greater commitment in terms of time, money, and labor.

On the other hand, human translation is preferred for content demanding extra care and attention; when accuracy and high quality are more important than a quick turnaround time. For example, legal, healthcare, and technical texts as well as more creative content like marketing and literary translations

Human translation should also be used for high-priority content and content hoping to incite sales / conversions as inaccuracies can give the wrong impression to potential buyers / users and cause a lack of trust.

Hybrid Translation

A mash-up combination of machine translation and human translation is becoming increasingly recognized as the perfect solution to balance quality, cost, and speed. In late 2022, The Economist named the human-machine hybrid as the “translator of the future”. Expert-in-the-loop workflows, including human-in-the-loop evaluation and machine translation post-editing (MTPE), could be the ultimate answer. Check out Slator’s Machine Translation Expert-in-the-Loop Report for more.