Are humans better than machines at translation?
By Lee Jun-ho
.jpg?w=728)
To explore whether machine translation can or cannot replace human translators, it is logical to shed light on what humans can do easily but machines cannot or vice versa. It is well known that machines are much better than humans at processing information, but humans are a lot better at strategic planning. That is why NASA suggested a human-to-machine collaboration model where humans plan and control how machines should process information.
Then, is the model applicable to the work of translators? Translation has been wrongfully regarded as a mere replication of the original text into another language form. Therefore, it is no wonder people believe there is no room for a translator to do strategic planning or make a decision.
However, translated text is another form of communication just like many other texts. Every text is written with a certain purpose and translated texts are not the exception. To be more specific, the inherent communicative purpose will have an impact on the way a text should be written.
For example, a text written in a newspaper may need to be translated into a speech transcript. Then, the text should be “re-written” in the process of translating so the translated text can be read as a transcript. In that sense, translators have played active roles in rewriting the text, considering the purpose of texts, target audience, target culture and many other factors.
This strategic planning has been a part of translation work process and one of the important factors used to distinguish “good” translators from “bad” ones. However, most of the current, if not future, Neural Machine Translation services were designed to process information sentence by sentence, without taking the whole text into account.
Consequently, it is highly likely that machine translation ends up with an outcome that needs a lot of human engagement just to re-configure the right tone and manner.
To narrow the discussion from the text to sentence level, logical links between sentences are poor in machine translation. To put it in simple terms, machines cannot read between the lines. To make matters worse, same terminology is sometimes translated differently in two sentences because a machine cannot recognize that the two sentences belong to the same text.
This kind of inconsistency is hard to imagine in human translation unless a human translator was negligent in the reviewing process. That means, again, human engagement is necessary just to connect the dots among the machine-translated sentences.
Mistakes in translation, of course, cannot be fully avoided in human or machine translation. However, one big difference between humans and machines is that human translators can correct errors on their own during the review process, while the machine translation output will be always the same unless the input text or processing algorithm is modified.
A fair analysis would be: 1) humans are better than machine translation in making a strategic plan, understanding context, and correcting mistakes and 2) machines are much better at processing information.
The best solution always comes from the sum of goods from multiple sources. To make the machine translation useful, it is high time to think how machine translation can use human capability to overcome its shortcomings.
Lee Jun-ho is a professional conference interpreter specializing in information technology. He has been teaching translation at a graduate school and other universities for seven years. He has completed PhD coursework, and studies intensively the feasibility of machine translation. He can be reached at cuefit@gmail.com.