Using AI And ML For Translation Solutions
Let's take a look at using Artificial Intelligence and Machine Learning for translation solutions.
Join the DZone community and get the full member experience.Join For Free
Natural Language Processing; it’s Artificial Intelligence that learns words and patterns of words so that it can respond to human searches and questions. Siri and Alexa are examples of this technology.
And this technology is continually improving. As more and more conversations are held with these machines, they continue to learn and respond more accurately.
Machines are also in use for translations. The concept is the same. Over time, machines will continue to learn words and patterns of both the original and the target languages and become more proficient in translation. Right now, however, machine translations have not been exclusively used, because they have not been able to provide the accuracy and all of the idiomatic expressions and grammar rule exceptions that all languages have. The learning curve, indeed, is deep.
What’s on the Horizon
In 2017, two research papers were published with the results of Machine Learning methods for language translation.
One of the methods reported on was something called “back translating.” First, dictionaries were built using common connections among words in both languages — shoes and sock, for example. With this method, the machine translated a sentence from one language to another. Then it translated back again, from the target to the original language. If there was not a match back to the original language, the machine attempted again, tweaking the next translation, until it gets much closer. The researchers believe that this is a first but major step in moving toward accurate translations, without human intervention.
Google and Facebook have also both been working on AI methods to improve their current adequate, but also quite imperfect, machine translation functions.
Developments in 2018
This year, researchers from Microsoft announced the creation of an AI/Machine Learning system that can be compared in quality to human translation. The test cases were news article translated from Chinese to English.
Once the translations were completed by the machine, they were also translated by bilingual humans. Then, an evaluator compared the two, finding them remarkably close. According to the technical team working on natural language and Machine Learning for Microsoft, this is a pretty major breakthrough in removing the barriers that have long held machine translation back. They caution, however, that there still a way to go. The news stories that were translated, for example, included more common vocabulary. Moving into more complex translations, particularly in scientific and technical areas, will require more complex learning, and the Microsoft team is not there yet.
Machine Translation Is Not New
For years, translation services have used a form of machine translations, but these have been at the very rudimentary level. When they have long-term clients, they have been able to develop systems that will automatically translate recurring vocabulary and terms, as well as simple phrasing and more common vocabulary. Over time, they have also been able to “train” machines to recognize and translate within broader contexts. But still, the human element cannot be eliminated, and the largest translation companies use a dual approach. Bilingual translators must do the heavy lifting, by going through each machine translation and correcting the errors and issues.
The Microsoft team used several training methods in their work — methods that have implications beyond just the translation industry.
- Dual learning. This is similar to the research reported earlier in this article. The news report was first translated from Chinese to English and then back again. The process was repeated over and over again until a natural, accurate translation was achieved. All along, the machine was learning.
- Deliberation networks. This process involved translating the same sentence over and over again, refining that translation each time, and thus improving the final product gradually.
- Agreement regulation. This process involved having the machine read from right to left as it translated a sentence and then left to right until the two translations reached a “match.” The goal is to reach a consensus from both directions, and that consensus proves to be more accurate.
All of these methods depend on the basic concept of Machine Learning. The more a machine is asked to perform a task, the better it gets at doing so. And when new language patterns are added to the mix, machines will “absorb” those patterns and put them in their memories, being able to use them from that point forward. This is what makes Siri and Google voice searches continue to become more efficient and accurate.
Disrupting the Translation Industry – How Far Will It Go?
We have already experienced major disruptions to many industries because of data science and AI. Individual investors are no longer relying on brokers for advice and counsel. They have access to the same real-time data that those brokers have and have chosen, in many instances, to make their own decisions and conduct their own trades. The same is true for the insurance and mortgage industries. Consumers have access to data and are taking control. As a result, there are professions that are dying.
In the translation industry, there is still a heavy reliance on humans. And reputable translation agencies still rely on bilingual pros — pros who have not only mastery of the two languages in question but also the cultural background to ensure that translations and localization of content are appropriate and not offensive to the populations of the target language. This is a huge factor when companies are attempting to do business in foreign countries.
The question for machine translation is this: Can machines “learn” the cultural nuances that have to be taken into account in the translation process? Can a machine learn, for example, that references to owls mean wisdom in one culture and evil in another? Perhaps.
It seems that the advances in AI and Machine Learning will ultimately disrupt the translation industry more than they already have, and even result in less demand for human translators. But it also appears that the human element cannot be completely eliminated. Because, in the end, translating for a foreign audience is still about forming relationships. And who wants a relationship with a machine?
Opinions expressed by DZone contributors are their own.