Comment
Author: Admin | 2025-04-28
A fellow localizer, who works at an organization with a very mature localization model and an insane scale, told me recently: “I hardly ever work with human beings these days. It’s all MT+AI.”This may be an extreme example, but Machine Translation (MT) is pretty commonplace these days. Any organization that needs to scale translation of content, is using MT in some form. These organizations are not jeopardizing quality of the translations but are able to achieve acceptable levels of quality using MT.So the answer to the question, “to MT or not to MT” is … Yes! What are you waiting for? Why isn’t everybody using it?The reality is that MT is an incredible technology and has progressed immensely because of AI. However, ultimately it is one of many tools in a tool box. A very productive one, but usually not a complete solution and best when not used in isolation. Like every tool, for it to be effective you need to understand it and apply it properly.MT Pros and ConsSome tasks are perfect for machines. Anything that involves analyzing large amounts of data, crushing through complicated calculations, repetitive (let’s face it, boring!) tasks, are all activities that benefit from the processing power of machines. In these cases, machines are light years faster and more accurate than humans — and for a fraction of the cost.On the opposite spectrum, machines are famously bad at creative work, especially interpreting creative … well, anything! There are advanced AI applications that are able to “recognize” (detect elements in the image) and categorize images, but the buck pretty much stops there. Any interpretation, optimization, and emotional connection to the audience still relies heavily or solely on human talent … thank goodness!In the MT world, the same criteria hold true. MT engines, in particular Neural Machine Translation (NMT) engines, require huge amounts of training data. In translation, data usually refers to the number of words. By processing massive volumes of content, NMT identifies patterns and trends and learns how to translate one language into another. On the flip side, NMT as well as any MT system perform best on larger “chunks” of content, which provide more “information” for the machine to process. This is actually true for human translators as well: translating a tagline is harder than translating a whole sentence. Finally, higher volumes of content benefit from greater economies of scale on the cost of MT. This is of course a gross over-simplification of MT, but it’s good enough for the purposes of this blog.At a very high level we can identify the two main benefits of MT:· Scale/Speed· Cost SavingsHow does MT perform with creative, highly nuanced copy?You guessed it! Not brilliantly. Creative copy is quite unique and served in small doses. Copy is used in conjunction with visuals to connect emotionally with the user. Marketing content in particular is created with a thorough analysis of the target audience, messaging to the user, and objectives of what the copy is trying to achieve. In
Add Comment