What happens to the term “cucumber sandwiches” in Oscar Wilde’s The Importance of Being Earnest when it is translated from English to French and from French to Italian? Literary translation researchers and computer scientists at Trinity College Dublin and Dublin City University are joining forces to help address the age-old issue of important cultural meaning being diluted in translation.
Translation is a ubiquitous feature of life today. In fact, it is so ubiquitous that most often, we do not even notice it is there at all.
Have you read Haruki Murakami’s Killing Commendatore? – Did you learn Japanese first? Have you heard about the political situation in Syria, Venezuela, and North Korea? – Do you know Arabic, Spanish and Korean?
Translation surrounds us and interconnects the whole world. Even if we never come into contact with a translator or their work directly, translations are what make it possible for us to understand other people despite the often radical differences between their lives and ours.
Translation is not as simple as we might think. It is convenient to imagine the translator as something like the classical image of St Jerome, a dishevelled character, poring over a tome, quill in hand, his only companion a skull siting on his desk. But if this image was ever accurate, it certainly is not now.
Today, translators are highly interconnected individuals, using sophisticated technology that synchronises the vocabulary choices they make, maintains the formatting of documents between languages, and offers suggestions based on previous translations. And all this efficiency is needed. As the unseen army that makes interlingual communication possible, translators exist in a niche that requires constant financial, practical and societal justification.
But things get more complicated than that. Translations do not simply pass from one language to another. Companies, governments, cultural organisations and NGOs often need highly technical, highly context-specific, content-rich texts translated from one language to many, or in some cases, from many languages to many more.
That seems simple if the start point, the source language, is a globally well-known language like English, but what about when it is a language learned by fewer people around the world? It is one thing to translate a computer game from English to Chinese, Hindi, French and Spanish. It is quite another to translate a similar computer game from Latvian to Greek, Indonesian, Portuguese and Swahili. There is nothing obscure about any of these languages in global terms. However, the number of translators who are able to work languages combinations like these to the level required is comparatively small.
That is why in such situations, translation goes indirect. Indirect translation is when one layer of translation renders a text from a relatively small language to a relatively large one, and then that translation is used as the basis for all the subsequent translations to be produced. In this scenario, an organisation that operates in five different languages only needs to employ enough translators to work in eight directions: one language serves as the intermediary, and the other four work into and from that. On the other hand, an organisation that uses no intermediary language will need to employ enough translators to work in as many as 20 directions, to allow communication from, to and between every language and every other.
You may be thinking that this would not be an issue now, in the age of Google Translate and other forms of machine translation, where the answer is just a click away. But you would be wrong. In fact, indirect translation is precisely how neural machine translation engines like Google’s get around the lack of interconnecting data between comparatively small languages. Neural machine translation systems essentially learn from translations that already exist in the world. If a text exists in multiple languages, the system can infer which parts in one language correspond to the parts in the other languages, and so, essentially learn how to translate.
But that means a system that has more examples will work better than a system with fewer. So, if there are not many examples of parallel texts in say, Latvian and Indonesian, then the system has little to learn from, so the quality of the translations it produces between these two languages is likely to suffer. That is why such systems capitalise on the fact that certain global languages, especially English, are so widely translated into and from, that the available body, or corpus, of example texts is big enough to ensure the quality of the new translations is high. The upshot is that the quality of a translation going from Latvian to English and from English to Indonesian is likely to be higher than it would be if English did not feature in the equation at all.
So indirect translation solves practical issues by reducing the amount of work required to move texts between languages, and maximising quality. However, it also raises new issues, because every time you translate, you produce a new text, and that new text is underpinned by certain cultural and linguistic assumptions that are particular to how that language in particular works and relates to the people who speak it. If you translate translations, those assumptions can end up piling up, and the cultural elements that set texts apart in the first place can be eroded.
What isn’t known, however, is to what extent this cultural erosion takes place, and whether there is anything that can be done to minimise it. The QuantiQual Project, an Irish Research Council-funded collaboration between literary translation scholars at Trinity Centre for Literary and Cultural Translation, and computer scientists at the ADAPT Centre for Digital Content Technology is setting out to answer these two questions.
The project uses digital humanities methodologies to analyse huge numbers of literary texts translated indirectly during the 18th, 19th and 20th centuries to identify lexical and syntactical features that are statistically unusual in each context. It is also producing new translations using specially created neural machine translation software, in order to test whether the patterns identified are particular to human translators or are shared by machines too.
While the initial focus of the project is the translation of literature because it is so difficult for humans and machines alike, the potential impact of the project goes far beyond this.
The early stages of research will examine what happens to cultural artefacts during the translation process, such as when Oscar Wilde’s The Importance of Being Earnest is translated into French, and when that French text is subsequently used as the source of the Italian translation, what happens to culturally specific words like “cucumber sandwiches”, “Tories”, or even “Earnest”.
The project will involve the development of a corpus of literature translated indirectly into a range of languages during the 18th, 19th and early 20th centuries. This corpus will be used to identify patterns in the translation process, and assess what effect indirect translation has had over the texts and the elements that associate them with their home languages.
However, the ultimate aim of the QuantiQual Project to find ways to train translators and program translation systems to produce translations of the highest possible quality in the many contexts that rely on globally small languages.
These contexts are often some of the world’s most impoverished, and the inability to access information in a readily understandable form constitutes a major hurdle for people struggling to lift themselves out of poverty. In the past, the go-to solution was simply to learn a global, often colonial or postcolonial language, but this can lead to language death, where the local language ceases to be relevant and people stop speaking it. Alternatively, it can lead to a widening of social strata between those who can afford this language education and those who cannot. High quality indirect translation may be a powerful tool that may enable people to thrive in their own languages.
James Hadley is Ussher assistant professor in literary translation at the Trinity Centre for Literary and Cultural Translation