To many of us, the term “machine translation” (MT) brings to mind the times when we passed our Chinese essay through Google Translate after writing it out in English before submission. For the more paranoid, it could even conjure images of Skynet conquering the world in the future.
The former is actually a fairly accurate representation of MT as it involves passing a source text through a machine for it to be translated into a target language. However, I’m sure that many of you would have discovered several issues with the final work after passing it through Google Translate. You might even have tried translating a website with it only to be puzzled by the gibberish that comes out (depending on what languages you’re translating between).
As for translation memories (TM—not to be confused with MT), these are tools that translators can use to aid them in their work. In essence, a TM is a glossary of sorts which makes it more convenient when translating a large number of words that have repetitions. With this, consistency can be achieved and the target text can be understood easily.
How does this fit into MT you might wonder? TMs help MTs just as they would help human translators. With these TMs, machines can learn the correct terms to use for certain situations and texts, allowing them to improve their resulting translations.
In recent years, there has been growing interest and confidence in MTs and TMs, due in part to Google’s efforts at perfecting machine translation, and its prominent presence on the internet. At the same time, the availability of other MT and TM tools has made translations more available for everyone. However, is this the beginning of the end for human translations?
To find that out, we prepared a mini-test in which we ran a sample text through various MT/TM tools to see how each fared against the other, and against a human translator. A sample of the translations is shown below.
[mk_table style=”style1″]
| Source | Human/Official | Machine A | Machine B | Machine C |
|---|---|---|---|---|
| Diseases affecting the brain, nervous system and spinal cord are usually diagnosed with a neurologic examination. | 大脑、神经系统和脊髓方面的疾病诊断通常需要用到神经学检查。 | 影响脑子、神经系统和脊髓的疾病通常诊断与一次神经学考试。 | 影响大脑，神经系统和脊髓的疾病通常被诊断为神经系统检查。 | 疾病影响到神经系统、大脑和脊髓通常被诊断患有一种 neurologic 检查。 |
| XYZ ist ein weltweit tätiges Unternehmen mit dem Fokus auf Elektrifizierung, Automatisierung und Digitalisierung. Als einer der größten Anbieter energieeffizienter, ressourcenschonender Technologien ist XYZ führend bei Systemen für die Energieerzeugung und -übertragung sowie die medizinische Diagnose. | XYZ is a global powerhouse focusing on the areas of electrification, automation and digitalization. One of the world’s largest producers of energy-efficient, resource-saving technologies, XYZ is a leading supplier of systems for power generation and transmission as well as medical diagnosis. | XYZ is a world-wide active enterprise with the focus on electrification, automation and digitization. As one of the largest providers of energy-efficient, resources-protective technologies XYZ is prominently in systems for the energy production and – transmission as well as the medical diagnosis. | XYZ is a global company focused on electrification, automation and digitization. As one of the largest providers of energy-efficient, resource-saving technologies, XYZ is a leader in power generation and transmission systems and medical diagnostics. | XYZ is a global company with a focus on electrification, automation and digitization. As one of the largest providers of energy-efficient, resource-conserving technologies, XYZ is the leader in systems for energy generation and transmission, as well as the medical diagnosis. |
[/mk_table][vc_row][vc_column width=”1/1″]
From the results, it is clear that:
- There are more errors occurring when translating from English to Chinese, than from a European language to English (e.g. German).
Linguistic differences have a significant effect as different sentence structures used in the different languages will affect how a translation will sound.
- Source texts with fixed phrases/expressions tend to have less issues than descriptive text.
Fixed phrases/expressions are easier to translate because they tend to have a fixed phrase counterpart. This is particularly true of terms such as, “Except as otherwise expressly stated herein…” etc.
- Less is more (accuracy) when it comes to sentence length.
Longer sentences have a higher chance of being mistranslated as the machine translates linearly and can translate the sentences out of context.
Of course, the mini-test was conducted using free-to-use services. This means that there was no access to the industry-specific TMs that no doubt would improve translation accuracy, as more appropriate and accurate vocabulary, terms, and phrases could be used.
Even so, it is clear that the gist is carried across in the translation; you would likely be able to understand the general idea behind the text. However, going back to the example of throwing an essay into Google Translate, your teacher would return you your essay marked in red. This is because the structure, form and terms used might not be the most appropriate or natural for the translated language.
Hence, when it comes to understanding the world, MT is no doubt your best friend. However, when it comes to getting your message across, which would you rather choose?