Babeling Tongues: Literary Translation in the Technological Era

According to the Canada Council for the Arts, “It all starts with a good book. Then a translator, writer, or publisher is inspired to see it translated.” Indeed, in the present moment, translation is becoming ever important to both the globalizing world generally and the publishing industry specifically. Despite the increased role translations must undoubtedly take in the world market today, Three Percent, a resource for international literature at the University of Rochester, reports that “Unfortunately, only about 3% of all books published in the United States are works in translation… And that 3% figure includes all books in translation—in terms of literary fiction and poetry, the number is actually closer to 0.7%.” This paper justifies the need for increasing the number of translations available in the market, and explores the problems and possibilities doing so.

This essay is divided into three sections. It begins by examining the role of language in literature. It will use the political importance of a wider canon and the mass appeal of World Literature to establish the importance of works in translation. It will then explore the different processes by which professional translators and machine translation softwares approach the translation of texts. In this section, I will demonstrate how machine translations differ from human translations in their conception and execution. The final section of this essay will discuss the limits and possibilities applicable to both types of translation. In particular, it will suggest that machine based translations are, at present, largely capable of translating only literally, while literary translations require translations that go beyond simply literal forms, relying as they do on cadence, metaphor, connotations, and a detailed knowledge of context. Finally, I conclude by showing how work in this regard remains nonetheless open as different groups are attempting to perfect machines equipped with Artificial Intelligence that can deal with more complex types of decision-making required for such translation.

On Translation and World Literature

In order to realize the importance of translation today, we must first recognize that we are at present in the present 21st century, in which the world has become incredibly globalized. In this globalized world we have, for the first time in history, so many individuals from so many different cultures interacting with each other on a daily basis. Consequently, such individuals speak to each other on a daily basis, and also take interest in the literatures of each other. This having been said, the communication predominantly takes place in English, which has, because of the British empire, historically been a very important language at the global scale. Thus, though individuals may speak several languages at once, the dominant language of communication across cultures is English. Indeed, one would be hard-pressed to find someone who does not find that English is the global language of the present world.

The position of English deserves more thought, particularly as, according to the Kenyan scholar Ngugi wa Thiong’o, languages are not politically neutral. They are not inert objects used simply for communication, but rather, every language is “both a means of communication and a carrier of culture [emphasis added]” (Ngugi 13). By calling language a carrier of culture, Ngugi informs us that each language, in its very form and vocabulary, carries the experience of a certain people. Effectively, languages carry “the entire body of values by which we come to perceive ourselves and our place in the world” (Ngugi 390).

Though insightful in itself, Ngugi’s provocation proves particularly pertinent when we consider the historic progenitors of English language: usually first-world, native speakers, who are almost always white. And while, in the past, this fact may not have been an issue as the language circulated in only this same sphere of people, it is problematic in the modern world where the literature published has failed to match the diversity of its leadership. Indeed, as The Guardian’s recent list of the 100 Best Novels attests, the majority of what is considered ‘great literature in English’ is still considered to come from the people mentioned above. Despite surveying novels across five centuries, the Guardian list acknowledges less than 10 novels by authors of colour. Political repercussions notwithstanding, this difference reflects present literary trends even outside of novels published originally in English. At present, we find ourselves in a world whose supposedly ‘global’  literature does not represent the diversity of the people who read it. Given the role of editors and publishers in shaping the literary landscape, such individuals and groups must strive to ensure that the available literature reflects the experiences of those who are to read it. An integral step in this process is to make more English translations available for the current, global readership of the language.

Apart from the responsibility they hold, publishers need to increase the translations they publish because, quite frankly, it makes economic sense. Literature in translation is a growing market, particularly in diasporic communities whose second and third generation readers cannot read their original languages, yet still desire to reconnect with their roots. Moreover, as more and more people are have become aware of the limited perspectives inherent in ‘purely English’ literature, they have also recognized the importance of books in translation. As a result, they have created a huge demand for literature in these languages that reflects their respective cultures. Moreover, such books are also of interest to native English readers as such individuals are curious to know what people from other countries are writing about. For this reason, many authors have gained worldwide appeal despite never having written in English: Gabriel Garcia Marquez, Milan Kundera, Yukio Mishima, and Jorges Luis Borges– to name just a few.

This immense demand for translations is visible in that such books generally claim much more than of the market than the ratio in which they are produced. According to Nielsen,

“‘On average, translated fiction books sell better than books originally written in English, particularly in literary fiction.’ Looking specifically at translated literary fiction, [we can see that] sales rose from 1m copies in 2001 to 1.5m in 2015, with translated literary fiction accounting for just 3.5% of literary fiction titles published, but 7% of the volume of sales in 2015.” (from The Guardian)

Given the above-mentioned statistics, it is obvious that the translation market is an incredibly fruitful avenue for publishers to explore. Moreover, given how small the current production share of translations is, there is still a lot of potential for exploiting the market for translated literature. Rebecca Carter corroborates this point as she notes that “Amazon had identified an extremely strong and under-served community: readers with an interest in books from other countries.”

In light of these findings, it is obvious that we need to increase the number of translations available, and to see what avenues are available for rendering high-quality translations. As we seek to do so, I argue that it becomes prudent that we look into newer methods of translation, particularly machine-based translations (MT), which could possibly prove more efficient and economical than traditional means. As such, it is necessary for us to see how these two processes of translation, by humans and by machines, work, and what the problems and possibilities are of each.

Translating: By Machine and By Hand

At present, the dominant translation methodology is that followed by professional translators. Given that translation is a niche profession, we must examine the motivations of professional translators in order to understand the techniques they use in translating works. Deborah Smith, translator of, The Vegetarian, winner of the 2016 Man Booker International Prize, explains that “[p]art of the reason I became a translator in the first place was because Anglophone or Eurocentric writing often felt quite parochial” (from The Guardian). Smith’s view is very much in tune with the current ascendancy of World Literature and the movement towards a more global canon. Andrew Wilson, author of Translators on Translating, and a translator himself, is “struck by the enjoyment that so many translators seem to get from their work” (Wilson 23). In fact, the various accounts from translators in his book indicate that passion is the driving force behind the profession. Per Dohler is immensely proud of himself and fellow translators because “[w]e come from an incredible wealth of backgrounds and bring this diversity to the incredible wealth of worlds that we translate from and into” (Wilson 29). While many translators, like Dohler, come with a background in literature and linguistics, others, like Smith, are self-taught.

Building off the motivations expressed by these other translators, Andrew Fenner describes a general approach to translation. He points out that, firstly, the translator reads the whole work thoroughly, in order to get a sense of the concepts in the text, the tone of the author, the style of the document and the intended audience. The translator then follows this by translating what they can. They prepare the first draft, leaving unknown words as is. After doing so, the translator leaves the work aside for a day, and allows their subconscious to mull over ambiguous words or phrases. They then return to the work sometime to later to make checks, to correct any errors, and to refine the translation. Lastly, the translator repeats this last process a few more times (Wilson 52-3).

The salient feature of Fenner’s process is that the human translator takes into consideration the work as a whole. They do not imagine the text simply as an object built from the connection of the literal meanings of words. This idea of the work as a whole being reflected in each individual segment will become exponentially important later, when we explore machine-based translations. For now, however, we must only note that this approach ties into Peter Newmark’s diagram of the dynamics of translation:

The Dynamics of Translation

Note: 9 should say “The truth (the facts of the matter) SL = Source language TL = Target Language

As the above diagram shows, the translator must keep in mind both the source and target language’s norms, culture, settings, as well as the literal meaning of the text – a challenging task to say the least– one that involves a complex system of processes and judgements.

Machine translations, in contrast to human translations, use a different series of processes, which generally do not take into account different factors in the same way. For the purposes of this essay, we will look at two machine translation softwares: Duolingo and Google Translate. By its own admission, Duolingo uses a crowdsourcing model to “translate the Internet.” Founder Luis von Ahn strove to build a “fair business model” for language education– where users pay with time, not money, and create value along the way. Duolingo allows users to learn a language on the app, while simultaneously contributing to the translation of the internet.

Ahn introduced the project and the process of crowdsourcing these translations at a Ted Talk:

He claims that Duolingo “combines translations of multiple beginners to get the quality of professional translators” (from video above). In the video, Ahn demonstrates the quality of translations derived from the app. The image below contains translations from German to English by a professional translator, who was paid 20 cents a word (row 2), and by multiple beginners (row 3). Comparing Translations: Professionals versus Duolingo Beginners

As is evident, the two translations seen in the bottom rows are very similar to each other. Using the ‘power of the crowd,’ Ahn estimates that it would take 1 week for him to translate Wikipedia from English to Spanish, a project that is “$50 million worth of value” (from video). From this estimate alone, we can see that machine translation provides the possibility of saving a lot on the cost of translation– a prospect that, in itself, may allow for many more  translations to be produced with the same amount of financial capital.

Apart from Duolingo, one of the more common translation softwares is Google Translate. Unlike Duolingo, which relies on the input of many users translating the same sentence, Google translate works in an entirely computational manner. It performs a two-step translation, using English as an intermediary language, although it undertakes a longer process for certain other languages (Boitet et al.). As the video below shows us, Google translate relies on pre-existing patterns in a huge corpus of documents, and uses these patterns to determine appropriate translations.

While we grant professional translators the benefit of doubt, in that we do not expect their translations to be ‘perfect,’ it is important to note that we seem to have exalted expectations of work done by machines. Machine translations, with their statistically sound algorithmic models, are assumed to provide accurate and appropriate translations. As we go forward with this essay, especially as we discuss the limitations and possibilities of these approaches to translation, it is important to realize that while machine-based translations may indeed advance the pace and quality of translations, we still cannot always assume their translations to be perfect, or always reliable.

Translating: Problems and Possibilities

In terms of limitations, I have noticed that the primary issue with machine-based translation at present is that seem capable only of doing literal translations. In short, this method of translation is most suitable in translating individual words occurring in simple sequences, one after the other. Now, this limitation proves especially debilitating because many texts, particularly literary texts, do much more than simply convey literal meaning. As Philip Sidney explained in his Defence of Poesie, Literature with the capital L means to both “teach and delight.” Literature, in its attempt to delight and entertain, involves an infinitely complex interaction between words, their sound and cadence, their denotation and connotation. It is not simply an object of beauty, but ascends to the level of metaphor, symbolism, and leitmotif, and, n so doing, becomes an object of beauty. To put it simply, when we talk about Literature, it is not just what is said (what we can see in literal translation) that matters, but also how it is said (which is not easy to execute).

Given this supra-literal quality of literary fiction, we must question the applicability of machine translations to such literary forms. Indeed, because machine translations do not seem capable of accounting for this metaphoric dimension of literary language, they may be better suited to types of writing whose goal is the simple transferral of meaning, or communicative writing. Machine translations are thus more applicable to knowledge-orientated genres of writing such as encyclopaedic articles, newspapers, academic texts, whose main focus is to educate and whose core linguistic operations are literal and not metaphoric. However, though machines seem less apt for translating these more complex forms of writing, I maintain that there is the possibility of having machines perform translations of such texts with the aid of limited human intervention, and Artificial Intelligence.

According to a recent study of Google’s Neural Machine Translation system conducted at MIT, the quality of machine translations could possibly be made to be very similar to translations performed by a professional. Tom Simonite reveals that, “When people fluent in two languages were asked to compare the work of Google’s new system against that of human translators, they sometimes couldn’t see much difference between them.” The inherent challenges of translating literary works lie in that multiple connotations of the same word are often context dependent, and therefore, programming a system that can intelligently select one connotation over the other is no easy feat. Will Knight explains the advancement of artificial intelligence, “In the 1980s, researchers had come up with a clever idea about how to turn language into the type of problem a neural network can tackle. They showed that words can be represented as mathematical vectors, allowing similarities between related words to be calculated…By using two such networks, it is possible to translate between two languages with excellent accuracy.”

The official MIT report concludes: “Using human-rated side-by-side comparison as a metric, we show that our GNMT system approaches the accuracy achieved by average bilingual human translators on some of our test sets.” Now, there is definitely no indication that this model is perfect, as yet, but it is a fascinating possibility for the future of translation.

Similar to how we read and process language in texts, Google’s software “reads and creates text without bothering with the concept of words” (web). Simonite describes that the software, in a manner similar to humans’ processing of language, “works out its own way to break up text into smaller fragments that often look nonsensical and don’t generally correspond to the phonemes of speech.” Much like the professional translators approaching the text in chunks that they feel are appropriate, this software does the same. For publishing, the benefits of machines performing high-quality translations equivalent to that of professional translators are manifold.

Primarily, such form of translations would mean lesser production times per translation, and increased accessibility of the work. In the current system where translations are usually performed only when funding or grant money is available, or when there is an assured demand or number of sales in the target market, quality machine translations would ensure that lack of funds would not hinder the development of a translation project. When professional translators themselves may not be readily available for certain languages, machines could step in to do the work. Of course, the financial and physical accessibility of such software to publishers themselves is another matter of consideration. But these are dreams worth considering, and pursuing.

The question remains, however: how can this machine translation model be perfected? Without delving too much into the technicalities of the matter, one finds that it is evident that one of the best ways to fine-tune translation models such as these is to provide the system as much parallel data as possible. According to Franz Josef Och, the former head of Machine Translation at Google, Google Translate has relied on documentation from the Canadian government (in both English and French), and files from the United Nations database. In a similar manner, we can ask publishers to provide literary texts, either original works or translations, to which they currently hold the copyright. By providing copious amounts of data, and by using processes of machine learning, we may be able to teach computers to increasingly translate better. This, in turn, could lead to very advanced machine translations, capable of even translating highly metaphoric forms of literature. In so doing, we can possibly arrive at a stage where, as in the words of Jo-Anne Elder, the former president of the Literary Translators Association of Canada, “A translated book is not a lesser book.” In pursuit of this goal, our aim must be to not simply give up in recognition of the present hurdles confronting machine-based translations, but, like a literary Usain Bolt, we must strive to ascend above them, and succeed.


Works Cited

About Three Percent.” Three Percent. University of Rochester. Web. 7 Nov. 2016.

Boitet, Christian, et al. “MT on and for the Web.” (2010):10. Web. 24 Nov. 2016.

Carter, Rebecca. “New Ways of Publishing Translations – Publishing Perspectives.Publishing Perspectives. 05 Jan. 2015. Web. 20 Nov. 2016.

Duolingo – The Next Chapter in Human Computation. YouTube, 25 Apr. 2011. Web. 28 Nov. 2016.

English Essays: Sidney to Macaulay. Vol. XXVII. The Harvard Classics. New York: P.F. Collier & Son, 1909–14; Bartleby.com, 2001. Web. 24 Nov. 2016.

Flood, Alison. “Translated Fiction Sells Better in the UK than English Fiction, Research Finds.” The Guardian. Guardian News and Media, 09 May 2016. Web. 10 Nov. 2016.

Google. Inside Google Translate. YouTube, 09 July 2010. Web. 26 Nov. 2016.

Knight, Will. “AI’s Language Problem.MIT Technology Review. MIT Technology Review, 09 Aug. 2016. Web. 25 Nov. 2016.

Literary Translators Association of Canada.Literary Translators Association of Canada. Web. 28 Nov. 2016.

Medley, Mark. “Found in Translation.National Post. National Post, 15 Feb. 2013. Web. 28 Nov. 2016.

McCrum, Robert. “The 100 Best Novels Written in English: The Full List.The 100 Best Novels. Guardian News and Media, 17 Aug. 2015. Web. 26 Nov. 2016.

Och, Franz Josef. “Statistical Machine Translation: Foundations and Recent Advances.” Google Inc. 12 Sept. 2009. Web. 25 Nov. 2016.

Simonite, Tom. “Google’s New Service Translates Languages Almost as Well as Humans Can.MIT Technology Review. MIT Technology Review, 27 Sept. 2016. Web. 28 Nov. 2016.

The Butterfly Effect of Translation.Translation. The Canada Council for the Arts. Web. 13 Nov. 2016.

Thiong’o, Ngugi Wa. Decolonizing the Mind: The Politics of Language in African Literature. London: J. Currey, 1986. Web. 26 Nov. 2016.

Wilson, Andrew. Translators on Translating: Inside the Invisible Art. Vancouver: CCSP, 2009. Print.

Wu, Yonghui, et al. “Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation.” (2016). Web. 25 Nov. 2016.

One Reply to “Babeling Tongues: Literary Translation in the Technological Era”

  1. This essay is thoughtfully organized to discuss the current state of translated works, the current state of machine translations, and the possibilities going forward. It also identifies and describes, rather well, the main limitation with machine and crowd-based translations (their focus on the sentence or paragraph level and on literal meaning). However, despite its clear strengths, the essay somehow does not quite stand together. Each part feels separate from the other, and as a result the main point is sometimes lost. It is not clear, for example, the link between the ant-colonial sentiment expressed in the first section and the limitations of machine translations in the second is never made explicit.

Leave a Reply