Little Lost Translator: the Good, the Bad, and the Ugly of AI in Translation and Localization

by

When we think of AI, what is the first thing that comes to mind? Do we think about Hal-9000 of 2001: A Space Odyssey, too focused on the mission than the lives and integrity of his crew? Do we think of J.A.R.V.I.S, the helpful AI to Tony Stark in the Marvel Cinematic Universe? Or perhaps to we think of CHAPPIE from naive, moldable AI who is neither wholly good nor wholly bad from the 2015 film of the same name? In any case, these portrayals were made before we had the technology and in most cases, over-estimated just how powerful the technology would be. But now that we have the technology, what does that mean for translators? Are we once again doomed to be replaced by machine translation? Will AI revolutionize our processes and make us more efficient or will it simply be a new flashy technology that ultimately becomes obsolete? As someone with experience in the tech and language realm, let me break down what I’ve experienced into 3 simple categories: the good, the bad, and the ugly.

The Good

Artificial Intelligence has rapidly improved over the last few years with the induction of Open-AI’s ChatGPT and several translation management systems implementing an AI plug-in or assistant to aid the translation process. From clarifying ambiguities to providing a time-efficient option for crucial and immediate translations, AI solidifies its position as another tool in the toolbox of the language industry. So, if that’s the case, what can we categorize as “the good” of AI?

First, we have AI’s ability to look through thousands and sometimes millions of datasets. This gives translators not only the ability to find common translations of a concept but also to see the context of that translation to ensure accuracy and efficiency.  This can aid not only translators by reducing the workload but also benefit LSPs and the industry by providing a cost-efficient and timely option. It can also help translators in their communication with native-speakers by filling in any gaps they might have missed. In my own experience, I’ve used this in my own translations that a concept is conveyed even if it’s only by a machine.

[insert video here]

Then there’s the ability of AI to provide aid in critical-need areas. In low-income or low-literacy areas, AI helps create translations and legible information where translators are scarce or not available. In times like the so-called Infodemic, where access to important information is less and less available, this is especially important for healthcare and other public goods and services.

The Bad

“With great power comes great responsibility” and with AI, it is especially important to remember this fact. Artificial Intelligence is built off data, and that data needs to be good for the machine to produce something good. Without said data, the machine can only work with the information it has. For further clarification, see a brief understanding from the graphic below:

Source: https://www.geospatialworld.net/blogs/difference-between-ai%EF%BB%BF-machine-learning-and-deep-learning/

In an article published by Harvard’s CoVisualize initiative, Manushi Siriwardana discusses the challenges faced by AI during the COVID Pandemic. Among those challenges were the lack of longitudinal datasets (datasets that had information longer than 2 years) and the need for suitable features to prevent unreliable or uncredible data. These features can make the AI costly not only for time but also for money. Therefore, not all information produced from AI can be considered perfect from the get-go.

Another issue with AI is how it can generalize data. According to the article, “Lost in AI translation: growing reliance on language apps jeopardizes some asylum applications”, Johana Bhuiyan discusses Carlos, a Brazillian immigrant who tried to seek asylum in the United States. He was held in a detention center using AI-powered voice-translation to interpret for him as the staff only spoke in Spanish and English. Unfortunately, the AI could not understand his regional accent nor his dialect and produced subpar translations that affected his and others understanding. Other AI-powered translation applications discussed in the article resulted in rejected asylum applications, FAQs with garbage data, and a lack of translations for lesser spoken dialects.

From my understanding of AI that I learned as a CS major, the algorithm follows trends and probabilities that come from generalizations in data. When it overgeneralizes, nuances and subtleties can be omitted or changed which completely changes the meaning of the translation. Additionally, lesser spoken languages will not have sufficient data for the AI to build upon and are often unsupported by these apps. This is especially driven home by a quote from Uma Mirkhail in the article “Afghan languages are not highly resourced in terms of technology, in particular local dialects…It’s almost impossible for a machine to convey the same message that a professional interpreter with awareness about the country of origin can do, including cultural context.”

The Ugly

With that in mind, the ugly part of AI is two-fold: lack of supervision and regulation.

Unsupervised use of AI is the crux of the problem mentioned in the previous section. Without supervision, there is no way to predict what AI will produce, if that information is helpful, harmful, or even usable at all. This will require additional time and money for companies to look at the data it is producing and then edit as necessary. So, wipe the sweat off your brow: your job is not yet in danger.

Secondly, unregulated AI means that any data shared with the program can be shared with anyone without knowledge or acceptance by the provider. No regulations on data privacy and security with AI makes it harder and harder for certain industries to use AI tools especially in companies contacted by the government or other agencies with high security clearances.

Conclusion

AI is not a tool we should fear, but rather a tool we should use with caution. Translators should not see AI as a replacement (it is far from replacing them) and should instead look to AI as a tool to aid them. In the coming years, with more data provided to these algorithms, perhaps the reliability and credibility of the machine will change. But for now, it’s safe to say that we should learn to stop fearing and start understanding the machine.

Sources:

Leave a Comment

css.php