By Niina Bailey and Lucy Clark
Artificial Intelligence has been a big talking point in recent months, especially with the release of ChatGPT by OpenAI in November 2022. ChatGPT is a chatbot to which the public can ask questions and it will provide them with detailed answers. ChatGPT is currently the most popular and well-known chatbot, but its release has spurred competition from rival companies, with Google releasing its own chatbot Bard earlier than initially planned. While some have embraced this AI technology and have used ChatGPT to help them write essays and articles, others have expressed concern over the usage of it and what it could mean for a number of industries. The publishing industry is one such industry that could be significantly affected by using AI technology, especially in translating literature. For this issue, we wanted to take a look at how translation in the industry could be affected by AI and whether translation AI already exists.
One of the main concerns a lot of people have over the use of AI is that it will produce inaccurate work or errors. For example, ChatGPT will sometimes provide factually incorrect information and because it does so confidently, people are more likely to believe it. This is a significant problem with using AI for translation as AI struggles with the complexity and nuance of language. Google Translate, the most well-known translation AI, will often translate things wrong because it translates text literally and word-for-word, failing to take into account the context of the words. Translating texts wrong is even more likely in regards to literary translation as the language is often more complex than in regular speech.
Another major concern that people have is that AI will replace people and perform their jobs in their stead. Why would you need a person to write an article when ChatGPT can do it in a fraction of the time? Within the publishing industry, AI could replace translators, at least partly. Publishers could use AI to initially translate a book and then only hire a translator to fix the mistakes the AI makes. You might think that that is still a long time away from happening, but the technology is improving and translation AI is learning, which makes the publishing industry more likely to use it.
Understanding how different translation systems work is essential to seeing how AI will affect the translation industry. There are a number of different types of systems, the most successful of which is called Neural Machine Translation.
Neural Machine Translation (NMT)
Launched in 2016, the success of NMT has been due to its speed of translation as well as its translation accuracy as it boasts an impressive 60% reduction in translation errors compared to Statistical Machine Translation. This state-of-the-art machine translation approach utilises neural network techniques to predict the likelihood of a set of words in sequence, whether that’s a fragment of text or an entire document. What makes this system so effective and unique is the utilisation of a 'common language' made up of numbers. Individual words are encoded into numbers; these numbers enter the neural network. Then, based on the learnt language rules, a new set of numbers are produced which correspond to the most appropriate words in the target language. Essentially, this system approaches translation like the human brain: it takes the presented information, translates them into its own language, and then 'thinks' about the best way to make a comprehensible sentence.
Although this very simplified outline of the workings of the system explains the process of translating individual words, the question remains: how does it produce accurate meaning? The key to this is NMT’s contextual approach to translation; it looks at the words that appear before and after rather than treating each word as an individual translation. To go one step further, this system actually learns over time, allowing the system to continuously improve.
AI and the future of publishing
These are, of course, impressive technological advances; however, it does bring into question what will happen to human translators. Since literary translation is so much more than content as it requires the translator to take into account literary style, will AI be able to achieve a full translation of texts? Perhaps the result will be a hybrid approach to literary translation where AI is used to create a first draft of the text, which the translator will then be able to use to produce a final translation. However, a writer’s voice is incredibly nuanced, and some literary styles are more obvious, while others are more subtle. No matter the style, the effect is that the reader experiences a text in a certain way, so in literary translation, the content and style need to be translated with care to achieve this. Despite the advancements in translation technology, it is clear there will remain, for some time, the need for human translators. The flexibility and natural ambiguity of the human language means literary translation will remain a 'human job' for the time being as machines are not yet able to untangle its complexities.