One of the hottest topics across the language services industry over the past 10-15 years has been the rapid development and implementation of machine translation (MT) solutions. While the use of Computer-Assisted Translation (CAT) and Translation Memory (TM) tools have become nearly standard across the industry, Machine Translation has yet to have achieved the same results in terms of widespread adoption by language service providers. Therefore, we will be discussing in greater detail some machine translation developments and challenges.
What is Machine Translation?
Simply put, Machine Translation (MT) involves processing the translation of one language into another through a specially “trained” computer program. The most commonly used and familiar machine translation tool for most people is Google Translate, and most machine translation tools function roughly in this way; you input your source text, and the “machine” returns the translated content to you.
Google Translate is just one example, though, and if you have ever tried translating anything in Google Translate, you may have noticed that it has gotten exponentially better over the last few years. With certain high-density languages, such as Spanish, the system has benefited from an enormous corpus of bilingual content to “learn” from. The program can be quite useful when used in situations where one wants to “assimilate” content versus “disseminate” content.
When it comes to the latter, where not only technical linguistic accuracy is paramount, but also must capture the subtle nuances of your branding message in each target market. Therefore, there will likely never be a complete replacement for professional human translators and the standard Translation-Editing-Proofreading (TEP) process. However, we will likely continue to see an evolution when it comes to how human translators interact with and apply these technologies in the future.
However, there are many other types of machine translation tools (also called “engines”) available, as well as different types of machine translation. The four basic examples of machine translation systems available today include: Statistical Machine Translation (SMT), Rule-Based Machine Translation (RBMT), Hybrid Machine Translation (HMT), and Neural Machine Translation (NMT).
1) Statistical Machine Translation (SMT) – This type of MT “learns” to translate from the source into target language through the analysis of large amounts of bilingual text corpora. In other words, as more bilingual content is fed into the system, the more the system learns from generating more statistical models. The system then analyzes the content and applies the most statistically likely correct translation.
2) Rule-Based Machine Translation (RBMT) – RBMT involves creating a set of linguistic “rules” that govern the translation of one language (source) into another (target). The rules are created by human language experts and computer programmers who map the rules between the two languages, as well as manually creating lexicons/dictionaries.
3) Hybrid Machine Translation (HMT) – HMT is simply the combination of different types of MT engines into a single engine when a single engine (or type of engine) is unable to achieve the level of acceptable quality. Therefore, examples of HMT could be statistics guided by rules or a rules-based approach that is post-processed by a statistical approach.
4) Neural Machine Translation (NMT) – This is one of the most exciting possibilities for the future when it comes to MT technology. NMT uses a vast neural network that can predict the likelihood of a particular sequence of words. Google is one of the leaders in terms of developing and implementing this technology, with Artificial Intelligence (AI), and has begun rolling it out in its popular Google Translate app.
What are some key developments with Machine Translation?
There is no question that machine translation has changed the way we communicate. While certainly not perfect, Google Translate can help us communicate more easily with people who speak different languages. There are also some exciting possibilities related to neural machine translation and Artificial Intelligence (AI) that are being developed, and that will continue to have an impact on human communication.
We can now talk with people all over the globe, send and receive emails between speakers of different languages, read websites in another language, and much more. Of course, this all comes with the understanding that the quality will not be perfect, and in many cases, far from perfect.
One of the most significant benefits of machine translation is that you can translate large amounts of content very quickly and at a much lower cost than human translation. However, to take advantage of this, the content must be in a format that can be “read” by the program, meaning hard-copy documents or scanned PDFs will not work since the text is not “live” or “editable.”
In some cases, Optical Character Recognition (OCR) technology can handle that, depending on the quality of the text. However, it could still involve a significant amount of manual labor to convert and re-format such documents to where the machine translation engine can effectively read them. Sometimes, it can be much more costly to prepare the documents for the MT engine than only having human translators complete the translations.
What are the key challenges with Machine Translation?
While MT technology has come a long way, it still has several critical issues. With RBMT, for example, it can be an effective solution for a highly specific niche translation (e.g., Japanese chemical patents). Still, because of the manual effort involved, research and development costs for a customized rule-based MT engine can be high. It also tends to produce more stiff and “machine-like” translations (although that may be sufficient for some types of translations depending on their use).
Furthermore, many clients may be under the assumption that MT should be less expensive than human translation. In many cases, that is true. However, it can be more costly, mainly if there is a lot of manual work needed to prepare the files so they can be “read” by the MT engine. Or, if a client wants to custom-build an MT engine to handle their unique content, significant cost savings will likely be realized in the future, but the up-front costs for research and development can be prohibitively expensive.
Conclusion
In summary, there are many things to get excited about when it comes to machine translation developments and challenges that we still face in improving the technology. However, it is still far from perfect, particularly for more “formal” or “official” needs. For example, you would not want your legal translations or medical records translated using only MT technology. Nor would you want diplomats to conduct their sensitive negotiations using automated translations or close a business deal with contracts that a machine has translated, no matter how good.
Even with the advances in AI technology, developers still have not been able to crack the nut when it comes to learning many of the subtle nuances found in language that only the human brain can correctly process and understand.
However, there are still ways to take advantage of this developing technology to benefit our clients when it comes to faster turnaround times, lower costs, greater consistency, all the while maintaining the high level of quality that our clients expect. Translation Source offers several hybrid solutions that include Machine Translation (MT), Computer-Assisted Translation (CAT) tools, and human translators and editors, all working together to provide better language solutions for our valued clients.
