Posted in:

Taking Care of the Thinking Machine: a Sneak Peek to NMT and AI Translation Maintenance  

© by iStock

Statistics and data are what nourishes Machine Translation. But how do you take care of a high-powered machine that, with incorrect information, could cause problems in translation? This article will focus on Tomedes’ tech-driven initiatives in maintaining their advanced NMT and AI translation.

It will discuss the problems that the translation agency faces in ensuring that the MTs are continuously getting information that will make their translation process and programming more efficient and accurate. It will discuss the relationship between linguists and MTs and some of the issues it’s facing, such as MT’s carbon emissions to AI replacing vendors.

The Present Machine Translation Landscape

Translation technology is relatively new. It’s only been around for the last 50 years and yet during that limited period, the technology has made a lot of progress. Because of the increased interconnectedness of the world, there is an ever-greater need for faster means of translation, and machine translation (MT) is being developed to fill the role. While English remains the business language of the world, other languages are becoming in demand as well.

MT is the technology that can translate text without human intervention. MT is different from Computer Assisted Translation (CAT) in the sense that the latter requires someone to input the text.

There are three types of machine translation that are currently in use. 

Rule-Based Machine Translation (RBMT)

The RBMT operates based on rules that have been inputted into the “machine” and those are mostly grammar and language rules. Those rules can be changed and adjusted by the users.

Statistical Machine translation (SMT)

The next type is the SMT and this technology translates based on a large body of text stored within its system. The more text that the system has been trained on the more accurate the translation will be.

Neural Machine Translation (NMT)

NMT is the most advanced of the three categories of MT. NMT mimics the way that the human brain processes language. NMT is AI-powered and through machine learning, it is able to understand entire sentences and paragraphs instead of just understanding one word at a time. The way that NMT translates text is a lot closer to a human than to a machine.

In theory, an NMT can understand the sentence and the context within which a word is being used. This means it can avoid mistranslations which are common in the other two MT categories.

How a Translation Agency Uses and Maintains NMT Tools

NMT uses artificial neural networks to do the translation work. It utilizes the network to predict the sequence of words. An advantage of NMT is that it only uses a fraction of the memory needed to run SMT. Another thing that separates NMT from the other types of MT is that all of its parts are trained jointly. 

In the translation industry, the use of MT is still being debated. But for Tomedes, a modern translation agency, MT and NMT, in particular, has a definite place in the language industry. It is given as an option to some of the company’s clients. They offer it as part of a package of Machine Translation Post Editing (MTPE). 

According to Rachelle Garcia, the Tech Team Head of Tomedes, tasked with maintaining its NMT and handling MTPE, “We perform MTPE for regular clients for some projects when requiring fast results or at least faster than what human translators can do,” Garcia said.

“For the MTPE process, we sample run the text through different machine translation. Then we have the translator pick the MT. Once we have picked the MT, we will run the entire text. The translator then post-editor examines the file and ensures the quality of the translation,” Garcia explained. 

To maintain NMT, they ensure that they only used clean data when training it. This calls for correct source and translation segments. Inaccurate data used in the training of NMT can cause errors in translation.

Issues Facing Machine Translation

As with other types of technology, machine translation faces some challenges and controversies. Let’s take a look at two of those issues.

MT Vs. Human Translation

One of the issues concerning MT today is the fear that it will soon replace human translation. It’s not surprising for some translators to look at MT with suspicion since other professions are being rendered obsolete by new technology. But the process followed by the translation agency points to the fact that the day that human translators will be completely replaced by machines is still a long way off if that will even become a reality. The general consensus is that machines can help humans in speeding up the translation process but they will never fully replace them.

As of now, the bulk of the translation projects handled by translation agencies is still worked on by human translators. There is an increasing trend to use a mix of the two, in the form of MTPE.

Carbon Emissions and MT

Another issue that has been raised about MT is kind of unrelated to translation. That is the question of carbon emissions. Because climate change has become a major issue today carbon emissions have been included as a possible benchmark for MT. That’s because MT engines have carbon emissions which can impact the environment.

Researchers in India have actually measured the impact on the environment of various language pairs. They found that MTs have varying carbon emissions depending on the language pairs that the engine is working on.

The researchers believe that the language models used in MTs require a great deal of computational power for those to be trained. That computational power can result in significant carbon emissions.

Based on the research, the French>German language was the one with the most significant carbon emission. Although in reality, the carbon emissions of MTs are insignificant when compared to other offenders. While this might be negligible right now, as MT technology continues to develop, it might change in the future.

Where will the future of MT lead? It’s not very clear yet but what’s definite is that it will be used to help make the job of a translation agency a lot easier. As the demand for translation increases like in translating shows and programs from one language to another, MT can play a bigger role.