English (pdf)
Article in xml format
Article references
Automatic translation
Send this article by e-mail
Cited by SciELO 
Cited by Google 
Similars in
SciELO 
Similars in Google 
Permalink
Introduction
In the changing landscape of global literature, the concept of Weltliteratur serves as a guiding light, embodying the timeless Romantic vision of a literary world that transcends borders and cultures. Weltliteratur has come to refer to a vast panorama of literary forms united by a singular thrust: providing a perspective that embraces multilingualism and multiculturalism within literature and sheds light on cultural diversity in and through literature (Damrosch, 2003; D’haen et al., 2004; Giusti & Robinson, 2021). At the heart of this transformative vision, translation assumes an indisputably indispensable role as it acts as the bridge that enables diverse voices to resonate across linguistic and cultural boundaries, fostering an interconnected and inclusive literary realm.
As a theoretical construct and as an archetype of transformative cultural power, translation plays a valuable role (D’haen, 2012). However, in reality, translation only partially harnesses its enriching emancipatory potential. Beneath the surface of an apparently harmonious cultural exchange through translation, lies a nuanced reality shaped by sociological forces. When taking a closer look at the forces at play, it becomes clear that cultural flows are governed by a three-pronged dynamic where culture, politics and economics are vying for priority. In recent years, sociological investigations have illuminated the stark inequalities inherent to intercultural exchange (Franssen, 2015; Heilbron et al., 1995; Heilbron, 1999). The perpetual dance of regulated cultural exchange amplifies privileged narratives, reinforcing hierarchical structures and oligoculturalism in global literary discourse.
Language technology seems to have emerged as both a disruptor and a potential solution, presenting the promise of redressing the balance in cultural exchange. Techno-optimists believe that the time has come for language technology to capitalise on its potential: machine translation (MT) is envisioned-and increasingly so-as a means to break down linguistic barriers, enhance accessibility, and open up avenues for diverse voices and narratives to reach wider audiences (UNESCO, 2019; NLLB Team, 2022).
Despite literature being long regarded as impervious to technological interference and even characterised as the “last bastion of human translation” (Toral & Way, 2014, p. 174), the impact of technology on the literary landscape is gradually increasing (Hadley et al., 2023; Rothwell et al., 2024). With the recent surges in artificial intelligence (AI), particularly in the form of generative AI, there is a growing recognition that automation may play a role in creative domains, like literary translation (Lyu et al., 2023). In other words, there is reason to consider language technology, particularly machine translation (MT), as a potential driver of the Romantic ideal of Weltliteratur. However, caution is warranted: “literary translation . . . demands a delicate balance between leveraging technological possibilities and preserving the richness of cultural nuances, artistic expression, and the human touch that defines the essence of literature” (Declercq & Van Egdom, 2023, p. 59).
This article constitutes a critical interrogation, positioning language technology within the framework of Weltliteratur. The objective is to delineate the uses, usefulness, and uselessness of language technology in enhancing linguistic and cultural diversity within literature. This exploration encompasses a historical overview of the concept of Weltliteratur. Additionally, a sociological perspective is adopted, examining translation as a constant driving force in the pursuit of Weltliteratur. The critical interrogation culminates in an examination of the merits and drawbacks associated with MT in the literary field. Drawing on relevant literature from translation studies, computational linguistics, cultural and literary studies, this article presents a number of arguments in favour of leveraging MT effectively in support of Weltliteratur, but, more importantly, it also issues a cautionary note against the uncritical adoption of MT. By performing a critical assessment of MT, this article aims to contribute to an informed dialogue on its potential to support the ideals of Weltliteratur.
Weltliteratur and Translation
The conceptual underpinning of Weltliteratur originates in the fertile grounds of the Romantic era’s philosophical milieu. The term was coined by Johann Wolfgang von Goethe, who, beyond his reputation as a writer, earned acclaim as a profound thinker. Ever since the first articulation of the ideal, Weltliteratur stood as a nebulous yet influential term, synonymous with an idealised “global” literature that encompasses diverse cultural and linguistic forms of expression (D’haen et al., 2004; Damrosch, 2003). To this day, the notion compels scholars to engage with it on a conceptual basis as well as with its theoretical implications and its imperatives for society. One of the main reasons why the concept continues to endure scrutiny is because it has elicited a plethora of definitions (laden with ideology and idiosyncrasy).
The responsibility for the proliferation of definitions lies with Goethe himself: his articulations of Weltliteratur are characterised by a lack of a precise definition (Strich, 1957). Throughout his career, the thinker oscillated between promoting Weltliteratur to enrich specific cultural systems, notably the German, and advocating for universal cultural and linguistic enrichment. Regardless of one’s perspective, translation takes centre stage as an indispensable element in realizing Weltliteratur.
Universal translatability is considered a tenet of early Romantic thought and lies at the heart of Weltliteratur (for a detailed discussion of universal translatability, see Berman, 1984). Paradoxically, the pursuit of translatability commenced with the neglect of translation sensu stricto: as the Jena Romantics would have it, translation, as a broadly construed concept, was considered the fulcrum for interdisciplinary dialogue, interweaving literature, music, and science, and ultimately contributing to the formation of a national culture. The overarching goal envisaged by Novalis and Schlegel-this interdisciplinary confluence wherein literature assumes significance as the embodiment of comprehensive potential-underscores the profoundness of the concept of translation.
Goethe was the first to place translation sensu stricto at the forefront of Romantic considerations (Berman, 1984, pp. 87-110). His interest in translation was a direct consequence of his own “trials of the foreign”: not only did Goethe try his hand at translating literature, but he was also inspired by the encounter with numerous translations of his own works. Both experiences imbued him with a heightened awareness of the transformative potential of translation. In a socio-cultural milieu characterised by youthfulness and a fervent quest for means to sculpt cultural landscapes, Goethe was the first to touch upon the role of interlingual translation, positing that translation alters the dynamics between languages and cultures. He positioned Weltliteratur as a programmatic concept conducive to cultural and linguistic enrichment.
As said, Goethe’s engagement with the term is marked by conceptual ambivalence. Fritz Strich noted that Goethe shied away from a succinct articulation of the term, creating a conceptual terrain fraught with divergent trajectories (Strich, 1957, p. 15). The deliberate vagueness in Goethe’s conceptualisation of Weltliteratur serves as a testament to the complexity and multifaceted nature of his intellectual pursuits. He left room for a spectrum of interpretations.
On the far end of the spectrum, Weltliteratur is presented as a nationalist concept (Scherr, 1869). When viewed through this nationalist lens, Goethe’s conceptualisation of Weltliteratur takes on a strategic dimension. At the time, German culture was positioned as young and peripheral, endeavouring to attain international recognition amidst fierce competition with well-established (or, to put it in polysystemic terms, “central”) juggernauts such as the French. In this milieu, Goethe’s vision of Weltliteratur becomes not merely a cultural aspiration but a deliberate endeavour to facilitate the assimilation of German culture on the global stage. Here, the translation of international works serves a conduit for assimilating external influences, particularly from developed cultures, thereby expanding the national aesthetic heritage. In illuminating discussions with Johann Peter Ackermann, Goethe heralds “die Epoche der Weltliteratur”, articulating the epoch where the imperative of “Aneignung” (appropriation) takes centre stage (1835). These discussions demonstrate the urgency with which Goethe envisions the German literature appropriating all that is beneficial from external sources. As Berman put it: Goethe designates the German language and culture as the vehicle of Weltliteratur (“le médium privilégié de la littérature mondiale”, 1984, p. 92). Within this framework, the concept emerges as the shortest, most pragmatic route to an elevated status of German language and culture.
At the other end of the spectrum, a “mondialist” trajectory can be discerned (Veit, 1834). The mondialist notion of Weltliteratur can be considered a decentralised force fostering cultural cross-fertilisation. The notion of mutual enrichment among cultures is articulated in Goethe’s letter to the historian Sulpiz Boisserée. The emancipatory objective of an “allgemeine Weltliteratur” is characterised by Goethe as facilitating nations to acquire mutual benefits (in Boisserée, 1831/1862, p. 565). In his letter, Goethe aspired toward the creation of a literary space where the constraints of national borders would dissolve, and the human experience could be conveyed through the exchange of ideas (Ideenverkehr -Boisserée, 1831/1862). Berman underscores the necessity of this enrichment through “intertraduction généralisée”, promoting the translation of all significant literature into all languages-transforming “national languages” into langues-de-traduction (literally: translation languages) (1984, p. 94). This ideal of a Weltliteratur can be characterised as more befitting of the current Zeitgeist. It departs from the neo-imperialist tendencies inherent to the concept forwarded during conversations with Ackermann, and aligns closely with the contemporary conviction that literature should be inclusive, providing space for diverse voices and promoting cultural and linguistic representation. Viewed through this particular lens, Weltliteratur is seen as the epitome of a literature that transcends boundaries and celebrates diversity.
Translation and the World Book Market
In the contemporary global landscape, the promotion of Weltliteratur through translation seems almost intuitive. Over the past two centuries, technological advancements, globalisation, and cultural exchanges have significantly eased the breaking down of barriers between diverse national literatures. Despite these advancements, the realisation of a literary realm characterised by boundlessness, multilingualism, and multiculturalism proves to be a far more intricate task when confronted with the complexities of reality.
Traditionally, translation flows are influenced by three factors: culture, politics, and economics. According to Heilbron and Sapiro, these factors fundamentally affect the selection of foreign book titles for translation (Heilbron & Sapiro, 2007, p. 95). Culture plays a paramount role, as it encompasses the literary values, norms, and aesthetics of both the source and target languages. The cultural alignment of a work determines its resonance with potential target audience’s tastes and preferences. However, politics and economics have always assumed a more dominant role in the configuration of the global book market (Heilbron & Sapiro, 2007, p. 97; Sapiro, 2003). Political conditions, including censorship laws, cultural policies, and international relations, can facilitate or restrict the movement of literary works across borders (Heilbron & Sapiro, 2007; Heilbron & Sapiro, 2018). The promotion of certain literature types can reflect a government’s ideological stance or diplomatic objectives.
A historical example that illustrates the role of politics in the selection or boycott of foreign literature is the Cold War era, particularly the relationship between the former Soviet Union and the United States. During this period, the ideological conflict between the communist Soviet Union and the capitalist United States was not only a matter of geopolitical tensions but also extended into cultural domains. In the Soviet Union, Western literature was only selectively translated, based on its alignment with communist ideology. Furthermore, the economic aspect is increasingly dominant in today’s globalised world, where market forces dictate the feasibility of publishing foreign works (Heilbron & Sapiro, 2007, p. 98). As the decision to translate is primarily driven by the marketability of a work and its potential financial return, economic considerations often outweigh cultural and political factors. The interplay of culture, politics, and economics can open doors for diverse literature to be accessed globally, but research in translation sociology has shown that it also contributes to the reinforcement of literary hierarchies.
The threefold distinction between culture, politics, and economics is further substantiated through bibliometric research, which provides concrete support for these historical observations and conceptualisations (Franssen, 2015; Heilbron et al., 1995, 1999; also see Grbić, 2013). This research strand has also provided some added depth to the debate, as it has also managed to reveal a threefold law that governs translation choices within national culture: the law of hierarchy, proximity, and kinship (Declercq & Van Egdom, forthcoming).
Stemming from polysystem theory, which positions translation (as a cultural phenomenon) in a centre-periphery structure (Even-Zohar, 1990), the law of hierarchy asserts that works from central languages are translated more often than works from other languages (Casanova & Jones, 2013; Heilbron & Sapiro, 2007, pp. 95-96). English occupies a uniquely central position in the international book market: it accounts for approximately half of all global book translations, German and French follow suit, each contributing 10 to 12% of the world’s translation market. These three languages are considered “central.” Translations from peripheral languages are clearly rarer, predominantly because these languages are endowed with lower status. Peripheral languages usually have one percent (or even less) of the global market share. Still, these percentages, however small, suggest that translations do occur from these languages. Paradoxically, the extent to which the law of hierarchy holds sway over the international book market becomes apparent when examining concrete translation flows between peripheral languages (Heilbron, 1999, p. 8; Van Es & Heilbron, 2015). It is often observed that interest in a translation from a peripheral language emerges only after a central language has endorsed it, typically through translation into that particular central language. In other words, translation flows between peripheral languages frequently occur via a central language. This phenomenon is exemplified by Marieke Lucas Rijneveld’s De avond is ongemak [The Discomfort of Evening], which was awarded the International Booker Prize. The English translation and its acclaim in the Anglo-American context spurred initiatives to translate it into peripheral languages (Janssen, 2022).
The second law, the law of proximity, suggests that cultural exchange is traditionally fostered by geographical closeness. The translation trends in the Low Countries, for instance, are significantly shaped by their geographical and historical ties, especially those with German- and French-language literature. Interestingly, while the Low Countries often translate from German and French due to their proximity, there are subtle differences in the nature of said ties. In Flanders, the Dutch-speaking region of Belgium, both the historical proximity to France and the influence of Walloon-French significantly shapes the translation landscape, leading to a higher frequency of French literary translations and of Flemish translators having French as their working language. Conversely, the Netherlands shows a stronger gravitation towards Germany. These variations exemplify how geography and history uniquely influence translation choices in different regions.
The third law is the law of kinship. Sometimes, the selection of translation is not driven by hierarchy or proximity, but by cultural-aesthetic and ideological kinship. In the 19th century, the German literary world emphasised Greek literature, as a deliberate move to counteract Roman influences found in French literature. This preference for Greek over Roman literary elements symbolised a broader cultural debate, contrasting “culture”, associated with creativity epitomised by Greek aesthetics, with “civilisation”, linked to regulation and evident in Roman-influenced French literature (Sagnol, 2003).
Bibliometric research has certainly provided a firmer grasp on translation flows, offering a more concrete view of cultural exchange. Still, it tends to aim for generalizability of knowledge, often abstracting from concrete dynamics at play. Encouragingly, growing attention is paid to actors within the literary field. Sociological studies reveal that agency is ever more important: as a result of the liberalisation of the book market, new actors, like scouts and translator-ambassadors, have entered the scene and helped shape the international cultural landscape (Heilbron & Sapiro, 2002; Kalinowski, 2002). In other words, the prospect of realizing a Weltliteratur that truly embraces multiple languages and cultures should not be viewed with undue pessimism. There are actors who actively contribute to localised diversification of cultural landscapes. Their impact is observed in the available bibliometric data (Franssen, 2015; Van Voorst, 1997): while it is true that these data suggest that the liberalised market favours financially viable titles, often from the English-speaking world, they also reveal an increasing variety of source languages, partly due to small-scale initiatives taken by individual agents.
Machine Translation as a Gateway to Weltliteratur
In the pursuit of achieving Weltliteratur, the previous section highlighted how political and commercial influences simultaneously facilitate and impede cross-cultural accessibility. This situation presents a significant challenge to the actualisation of Weltliteratur, envisioned as an emancipatory aspiration for a multilingual and multicultural literary sphere. The focus now shifts towards MT: its dual role as both a perpetuator of disparities and a potential equaliser is examined. This section critically interrogates the promises as well as the pitfalls of incorporating MT into literary translation, examining whether the utopian ideals of a Weltliteratur can be advanced through this technological medium.
In recent years, the utilisation of MT in literary translation has evolved into a realm of investigation. In their seminal paper “Is Machine Translation Ready for Literature?”, Antonio Toral and Andy Way referred to literature as the “last bastion of human translation” (2014, p. 14). They argued that the maturation of MT, coupled with its increasing adoption within the so-called “language industry”, was likely to spark a profound investigation into its applicability in the literary realm. The discussion surrounding the applicability of MT to literature had primarily been in conceptual terms. However, Toral and Way proficiently crafted a compelling case for the use of empirical methodologies as an essential step in evaluating MT’s suitability for literary translation.
Until recently, scepticism prevailed within computational linguistics and translation studies. During the era when statistical MT was the state-of-the-art technology, research outcomes offered little promise (Hoover & Sommer, 2010; Jones & Irvine, 2013). The growing optimism in recent years can be attributed to advancements in neural MT, which have been greatly enhanced through machine learning techniques and transformer architectures (Toral & Way, 2014, 2018; Voigt & Jurafski, 2012). These improvements not only result in higher-quality translations, but also open up opportunities for advanced customisation (Guerberof-Arenas & Toral, 2022). Recent research has demonstrated the value of feeding MT engines with tailored content (literary texts), optimizing their performance. Furthermore, in the domain of literary MT, exciting experiments are underway to explore the translation of textual characteristics, including rhyme, meter, and metaphor, showcasing the more nuanced and artistic approach to MT (Genzel et al., 2010; Van de Cruys, 2022). Nonetheless, optimism has been somewhat curbed, as research has indicated that the creativity of this cutting-edge technology is still lagging behind (Guerberof-Arenas & Toral, 2022). Therefore, it is exceptionally interesting to note that generative AI has recently been added as an option to the translation toolbox. While initial research suggests that generative AI applications may not yet produce outstanding translations, they are regarded for the creativity they exhibit (Lyu et al., 2023). These developments indicate that the time appears ripe for a contemplation of MT as a means to enrich literary landscapes.
Machine Translation’s Promises
First, the various possibilities that MT offers will be examined. A distinction will be made between the advantages intrinsic to the technology itself and the social and cultural benefits of using MT in a literary context.
One of the key advantages intrinsic to MT is its cost-effectiveness. MT’s ability to operate beyond the confines of economic viability, intimated by Besacier as early as 2014, is transformative. Lowering translation costs can be crucial in a literary market where the financial risks of translating works are often prohibitive. Countries with relatively favourable cultural infrastructures are usually governed by the laws of marketability, so translation often gravitates towards works that promise commercial success, leaving a vast array of literature untranslated (European Commission, 2022a). Lowering the economic threshold for translation makes it financially feasible and more attractive to publish literature from less commonly spoken languages and lesser-known genres. Furthermore, in many parts of the world, infrastructures are simply lacking, and translation is an even more precarious endeavour. MT can make a difference in these settings, not only by affordably increasing the volume of translations but also by providing a general cultural boost to local cultures. In this way, MT can directly contribute to the cultural edification or Bildung within (hyper)peripheral cultures with weaker structures (Rivière, 2017). By lowering the economic threshold, MT opens the door to literature from diverse cultures that might otherwise never find a place in the global literary market. This inclusivity is essential for achieving the ideals of Weltliteratur, ensuring that diverse cultural narratives are accessible and appreciated worldwide.
Another intrinsic benefit of the use of MT within a literary context is its speed. Compared to traditional methods, involving human translation, MT allows for quicker translation and higher rates of turnaround, which can be crucial in maintaining the topical relevance of works in translation. For publishers, a speedy translation is often particularly advantageous for titles with contemporary, historical, and socio-cultural relevant themes (Trentacosti & Pilcher, 2021). Additionally, MT can expedite the production of translations of best-selling authors, enabling near-simultaneous multilingual releases, which is particularly valuable in maintaining the relevance and marketability of these works.
Viewed from a socio-cultural vantage point, MT can also serve as a cultural barometer for publishers. By monitoring the uptake of MT titles, publishers can gain insights into emerging genres and cultures, allowing them to adjust their marketing strategies accordingly. This information can be used to commission high-quality retranslations of MT titles or produce follow-up titles through skilled literary translators. Such market-response strategies would reduce financial risks associated with translated titles while maintaining the publisher’s role as a cultural gatekeeper-thereby nourishing (rather than devouring) the translation profession.
Another important socio-cultural factor is the availability of resources. As mentioned, the translation of literary works is traditionally constrained by political, and economic factors. A factor that has not been touched upon, but that can be said to play a paramount role in the selection of translation is human resources. Available human resources are limited, as literary translators tend to be more adept in a select number of working languages. Since political and economic dynamics create a demand for specific working languages, diversity is usually restricted, as training programs focus on languages that are of political or economic interest, thus perpetuating the status quo. MT can assume a pivotal role in the democratisation of literature, primarily because it offers a broader palette of working languages.
Over the past few years, initiatives have been rolled out to ensure that MT systems cover less commonly taught or translated languages (lesser-known or less-resourced languages). The No Language Left Behind (NLLB Team, 2022) initiative is a prime example of how MT supports underserved linguistic communities and preserves cultural heritage. NLLB is focused on prioritizing the needs of low-resource language communities, redressing power relations and creating a more equitable digital space. NLLB recently reported on advancements in improving MT for low-resource and multidialectal translation: their NLLB-200 model, which covers 202 languages, is believed to outperform baseline models, particularly in non-English translation directions, demonstrating potential for quality MT across numerous languages. Their approach is said to help in saving languages from extinction (so-called “endangered languages”), and in enriching the global linguistic landscape. Initiatives like NLLB show that MT holds potential to be a powerful tool in broadening linguistic and cultural horizons. It can make literature more inclusive and accessible, irrespective of human resources or the economic or political weight it carries in the global arena.
A final argument in favour of MT use in literary translation is its potential to reshape and enrich the socio-professional landscape. As mentioned, traditionally speaking, literary translation has been the domain of a select group of literary translators working with a limited range of languages. Interestingly, MT has given rise to new roles that have contributed to the dissemination of often underrepresented literature (including in genres publishers consider less economically attractive). Fan translation has long existed (O’Hagan, 2009), but it only recently started to influence the literary world (Zhang, 2023). Traditionally, user-generated translation, a trend that has been significantly amplified by globalisation and the digitisation of media, has been dominated by a multilingual fanbase. In recent years, however, improvements in the quality of MT output have led to the emergence of the “monolingual translator” (Zhang, 2023), marking a shift in the literary landscape. While there are valid questions about the methods and skills of fan translators, it is clear that translation is gaining a broader base of support through an increase in resources and a diversification of voices (Rivière, 2017). Minako O’Hagan argues that fan translation often involves subcultural knowledge and high engagement (2009, 2020). Fan translators, whether multilingual and producing translations “from scratch” or monolingual and relying on MT in target text production, are creating a space for a greater diversity of voices and multiple perspectives in culture. This development enhances the multi-voiced and multiperspective nature of culture, opening up literature to a wider range of interpretations and understandings-owing to the broadening reach of MT. Viewed as such, MT emerges as an important instrument in democratizing access to literature, facilitating a more inclusive literary world where previously marginalised voices and narratives gain visibility.
Machine Translation’s Pitfalls
Next, the significant challenges and limitations that come with the use of MT will be explored. A distinction will be made between intrinsic limitations related to the technology itself and extrinsic social, cultural, and ethical concerns. A discussion of challenges and limitations is crucial to temper optimism with regard to MT.
First of all, it seems wise to reconsider the concept of “accessibility”, if we are to make great claims about MT’s potential to democratise access. “Accessibility” pertains firstly to the physical realm, encompassing access to locations, services, and technologies (European Commission, 2021). This aspect of accessibility ensures that every individual can fully engage with various aspects of society. Secondly, accessibility involves facilitating the availability of diverse forms of information, including textual, auditive, and visual media (Neves, 2022). Human rights bodies have recognised the fundamental and legal significance of the right to freedom of information, underscoring the necessity of effective legislation to uphold this right, and highlighting its role in promoting an informed and empowered society. This legal recognition extends the scope of accessibility to encompass equitable access to cultural content (Rivas Carmona & Ávila Ramírez, 2023).
In literary translation, accessibility acquires a somewhat unique dimension. Language barriers of all sorts often impede access, making (interlingual, intralingual, and intersemiotic) translation a critical means of achieving accessibility. In the realm of literature, the nature of what one seeks access to differs from most forms of translation. Unlike straightforward pragmatic texts, literature tends to involve a deeper engagement with the text: what readers seek in literature is not just surface-level content, but an immersive experience, capturing its style and emotional depth (Van Egdom et al., 2023, p. 55). This distinction draws our attention to the specific nature of MT output and highlights the challenges of providing access to the full literary experience. Assessing accessibility in the literary realm demands a nuanced understanding of the intricate relationship between form, content, and text function.
Research has shown that even if accessible literature were conceived primarily as accurately translated literature (in terms of meaning), MT reveals its limitations in translating foreign works (Vanmassenhove et al., 2021; Van Egdom et al., 2023). Despite advancements, MT struggles to render source material flawlessly. This is all the more true in a literary context: the complexity of languages (e.g. complex syntax) and the nuanced meaning (e.g. polysemy) embedded within literary works can elude the grasp of sophisticated MT systems. These shortcomings may hinder the accurate conveyance of the original text’s meaning and compromise accessibility when defined merely as content accuracy.
However, the stylistic or formal elements of literature are believed to form the crux of its appeal and depth. Literature’s richness, often emanating from intricate word choices, metaphors, and symbolism, is deeply rooted in the source language’s context. Therefore, in assessing the democratisation of access through MT, it is only fair and comprehensive to consider how well MT handles these aspects. While MT has made strides in neural processing, it seems clear that it continues to struggle in accurately replicating these stylistic subtleties: the resultant translations frequently lack the original work’s stylistic depth, leading to a distortion of the reader-experience (Ruffo, 2022; Ruthven, 2023; Van Egdom et al., 2023). This invites a dialogue on how to enhance MT ‘s capabilities to preserve the artistic integrity of literary works to facilitate literature accessibility.
The intertwining of literature with its cultural context further complicates access. Literary works are laden with cultural references and idiomatic expressions unique to their source culture (Lai & Nissim, 2022; Van Egdom et al., 2023). MT systems’ documented struggles to interpret these elements often leads to cultural misrepresentation. In efforts to bridge linguistic divides proposing MT as a solution, automated translations are likely to present readers with texts lacking the cultural depth and authenticity vital for comprehending the original work’s splendour. Thus, MT ‘s contribution to foreign literature accessibility may come at the cost of diluting its cultural richness. Consequently, at this point in time, MT might serve as an introductory tool to foreign literature, rather than a medium offering an immersive reading experience.
But even if MT were capable of rendering the meaning and cultural attributes of source material and of displaying some degree of literariness in automated output, concerns remain about the architecture and the algorithmic properties of MT and their potential to create a skewed representation of foreign literature. Over-reliance on MT can lead to a homogenisation of literary styles. This homogenisation can erase the (specific) “voice” and style of an author, the unicity of a literary work of art, resulting in a literary landscape devoid of diversity and original expression. Kenny and Winters (2020) highlight the risk of using an MT engine for translating literature: they observed a clear uniformity in style in MT output. This issue persists even with post-editing (PE) by qualified translators, as multiple studies have shown that the MT engine creates a “priming effect”, allowing the “mechanical voice” of MT to permeate these PE texts (Castilho & Resende, 2022; Declercq & Van Egdom, 2023, p. 52; Kenny & Winters, 2020). The situation is likely to exacerbate when MT outputs are handled by non-professional bilingual or even monolingual post-editors: due to their often-inadequate grasp of the literariness of a foreign text, it is expected that the literary PE text will derive its literariness from an all too simplistic view of literature. This homogenisation effect was described as “ennoblissement” (ennoblement) by Antoine Berman (1984).
The potential for stylistic homogenisation is not the only concern when it comes to the architecture and the algorithmic properties of MT. Linguistic data imperialism also presents a profound challenge (Choudhury, 2023; Wang et al., 2023; see also Demont-Heinrich, 2011). This is exemplified by the English language’s dominance in language models. The Anglocentric bias in the foundation of these models has led to a disparity in the quality of automated translation tools for less-resourced languages. Studies focusing on initiatives to enhance models for low-resource languages have underscored the limitations and biases inherent in English-dominated systems. A pressing concern in MT is the phenomenon of English-language interference-the so-called Anglocentric bias-: the over-representation of English linguistic structures, idioms, and cultural references in training data leads to a “shine-through effect” where translations are unduly influenced by English norms. Reliance on English-centric models can be expected to impede the process of capturing the nuances and complexities of less-represented languages and literary works. These works can lose their unique idiomatic expressions, stylistic nuances, and cultural references in translations that are biased towards English. Eventually, this Anglocentric bias is likely to contribute to diminished linguistic diversity (Kranich, 2014). This can result in a homogenisation not just of style, but of world literature as a whole, where diverse literary voices and perspectives are increasingly overshadowed by dominant languages and cultures.
Linguistic imperialism can be considered as just one facet of the broad issue of algorithmic bias. Time and again, research shows that these technologies are plagued by the risk of inheriting and amplifying biases from training data. For example, a growing body of research highlights how language models often exacerbate linguistic biases. This leads to an artificially impoverished language, aptly termed “Machine Translationese” (Castilho & Resende, 2022; Vanmassenhove et al., 2019; Vanmassenhove et al., 2021). Such bias occurs because machine learning models are predisposed to reinforce prevalent patterns in their training data, often at the expense of unique expressions. Consequently, translations may favour literalism, and are found lacking in preserving lexical and syntactic diversity.
Beyond linguistic bias, MT systems also tend to exhibit cultural and societal prejudices (Vanmassenhove et al., 2018). For instance, studies on gender bias in translation from gender-neutral languages to gendered languages reveal a marked tendency towards male-oriented defaults (Savoldi et al., 2021). This trend illustrates how algorithmic biases can be said to perpetuate stereotypes that lead to misrepresentations. Moreover, in scenarios involving translations between languages with varying degrees of formality or hierarchy, MT systems tend to default to a formal tone or inadvertently reflect damaging stereotypes, particularly when source content is ambiguous (Lee et al., 2023; Niu et al., 2017). This trend shows that MT systems are susceptible to socio-cultural bias. Based on these findings, one can reach the tentative conclusion that such MT output is culturally insensitive, and prone to misinterpretation and cultural distortion. Evidently, the prevalence of bias in MT algorithms mirrors a larger issue of human bias. AI systems are, after all, a product of their developers’ unconscious biases and, more importantly, of the data they are trained on.
When development teams lack diversity, training data does not adequately represent different cultures and “toxic” content is badly monitored, these disadvantages will be mirrored in AI output. This may lead to the further perpetuation of unfair standardised characterisation. These ethical implications are especially critical in literary translation, as literature is tied to cultural and linguistic roots, and biased or inaccurate translations can misrepresent entire cultures. Additionally, there is the risk of cultural appropriation, where unique cultural elements in literature are translated without proper understanding, leading to distortion or commodification. Addressing these challenges necessitates a more inclusive and equitable approach to the development of language technologies. It is imperative to acknowledge and cater to the diversity of global languages and cultures to mitigate these risks and ensure fair representation.
But even if we are able to enhance MT output in literary translation, fine-tuning algorithms to do justice to the unique voices of authors, and adequately address biases through data cleaning and algorithmic refinement, ethical (and other) objections to MT in literary translation would still arise. Therefore, it is essential to broaden the perspective to encompass concerns extrinsic to technology itself. Excessive reliance on MT introduces intricate intellectual property issues, particularly when translations occur without the consent of rights holders (Moorkens & Lewis, 2019). It can be safely maintained that this trend encroaches upon the concept of “data sovereignty”, which underscores the rights of individuals or communities over their cultural and intellectual data (Chander & Sun, 2023). When MT systems are trained on copyrighted material without proper authorisation, developers potentially undermine the control and rights of the original creators, leading to violations of copyright laws. Moreover, the intellectual ownership of translated materials becomes a particularly contentious issue when PE is added to the equation. Relying on PE for literary translation is fraught with risks due to ambiguity surrounding the artistic origins of the target expression.
Legal frameworks are currently being put in place to address these concerns (European Commission, 2022b), but even favourable legislation may inadvertently undermine the position of translators utilizing MT. MT adoption is likely to lead to a shift in societal perception, where literary translation is increasingly viewed as less artistic. This change in perception can dilute the recognition of the translator’s creative contribution impacting the valuation of their expertise in cultural and economic terms. In other words, addressing these concerns is not just a legal imperative but also a moral one, particularly in preserving the integrity and authenticity of literary works across different languages and the integrity of the profession (European Commission 2022a).
To get an estimate of the extent to which an industry can be affected by technologisation, one only needs to glance at an adjacent domain: the Language Industry (Dam & Zethsen, 2008; Vieira, 2018). The ascendance of MT has had profound implications for language professionals across the industry, raising concerns about job displacement and the devaluation of expertise. Initially heralded for its promise of efficiency and enhanced job satisfaction, the language industry has instead been caught in what has been dubbed “a race to the bottom”. The focus has shifted toward productivity and the acceptance of “good-enough translations”, often at the expense of translation quality and true creativity. This has not only resulted in a decline in rates but also in an increasing pressure from clients for lower costs. The translation industry’s marketplace dynamics, fuelled by rising demand and advancements like MT, have exerted a downward pressure on translation rates. This competitive environment, driven by technological disruption, poses a serious challenge to the sustainability and perceived value of human translation. It is not difficult to discern a potential parallel with the cultural sector. Significant effort is required to ensure that this situation is addressed justly in the literary field. Efforts must be focused on creating balanced frameworks that uphold the rights and contributions of human translators while integrating technological advancements responsibly.
Finally, objections to MT use can also stem from concerns about the ecological footprint of AI-driven technology (Moniz & Escartin, 2023). Language models demand considerable computational resources for their training and operation. This processing requirement translates into high energy consumption, leading to ecological repercussions such as increased carbon emissions. For instance, a study revealed that to achieve marginal improvements in output quality, a standard neural MT system used up to 7MWh of energy in three days (Waites, 2019). This level of energy consumption is on par with the annual usage of a large, inefficient household. The environmental impact goes beyond energy use: the carbon footprint resulting from training an MT model can amount to 284 tonnes of carbon dioxide-five times the lifetime emissions of an average car. Initiatives are being deployed to develop more sustainable practices in machine learning, but Big Tech clearly needs to step up their efforts to balance the benefits of advanced MT technology with the need to reduce their carbon footprint.
Discussion
When considering the use of MT as a tool to foster Weltliteratur, it is essential to weigh all pros and cons, acknowledging the constant need for renegotiation. It has been argued above that the potential benefits of MT in fostering a diverse literary landscape appear to be significant. The arguments adduced under “Machine Translation’s Promises” lend credence to the belief that MT can democratise access to literature, by breaking down linguistic barriers, enabling literature from lesser-known languages to reach a wide audience, and tapping into new forms of intercultural agency. This is crucial in a world where traditional literary translation is constrained by economic and political factors, as well as by the availability of translators, limiting the range of voices that can be heard on the global stage. Initiatives like the NLLB project (NLLB Team, 2022) illustrate MT ‘s capability to support underserved linguistic communities and enhance cultural exchange, and to do so in a quick and cost-effective manner. Moreover, MT can provide a boost for publishers who are seeking ways to diversify their offering but are currently deterred by the economic viability of new projects. Finally, MT can facilitate the translation and dissemination of more diverse literary works by allowing new agents, such as monocultural translators, to make their voices heard.
Despite the potential for increased access, this critical interrogation also highlighted several challenges, which must be addressed in order for MT to realise its full benefits. At the moment, MT ‘s ability to accurately convey the nuances, style, and cultural depth of literary works seems rather limited. The stylistic richness and complex word choices that characterise literature often get lost in MT, leading to a dilution of literary style and distortion of the voices of authors. Additionally, MT systems are prone to inherent biases that can lead to a homogenisation of style and language, the perpetuation of socio-cultural stereotypes and flagrant misrepresentation of (marginalised or Indigenous) cultures, thus compromising the ideal of a Weltliteratur worthy of the name.
Beyond technological limitations, substantial socio-cultural, ethical, and ecological concerns have been raised. There are intellectual property concerns, particularly when copyrighted material is used without authorisation to train MT systems. Moreover, further language automation may also devalue the expertise of human translators, potentially leading to poorer working conditions and reduced appreciation for literary translation as an art form. Lastly, the ecological footprint of MT, due to the use of vast datasets and powerful computational resources, will lead to increased carbon emissions and energy consumption, a factor that cannot be overlooked in an era of sustainability. In short, MT may be said to hold the promise of enhancing access to a more diverse range of literary works, but it is imperative that its limitations and impacts be considered.
Conclusion
The critical interrogation in this paper underscores the nuanced potential of MT in nurturing a multilingual and multicultural literary landscape, in line with the ideals of Weltliteratur. Close contact with different linguistic and cultural backgrounds, as emphasised by Pettigrew (1998) and Visbal (2009), is crucial for the formation of intercultural relations, the erosion of stereotypes, and the establishment of empathy and solidarity. However, in reality, contact is often hindered by numerous factors, restricting readers’ exposure to cultural perspectives. Recently, MT seems to have emerged as a beacon of hope in this scenario: it stands as a promising tool for democratizing the global book market and bridging linguistic and cultural divides. Advancements in language technology seem to go hand in hand with the digitalisation of the book market, which further enhances the feasibility of publishing works from underrepresented languages, potentially leading to the further emancipation of Weltliteratur.
Nevertheless, the deployment of MT in literature, as a driver of linguistic and cultural diversity, is fraught with challenges that must be addressed. Issues like text quality, representation, algorithmic bias, and legal and ecological concerns cannot be overlooked. Publishers and policymakers must make balanced decisions when considering language technology, a task complicated by the complexity of the issue and the relative lack of concrete empirical data on the impact of MT in the literary realm. This task is further complicated by the unpredictable nature of AI advancements. Rapid developments in AI underscore the need for ongoing dialogue and adaptive strategies across all cultural domains.
In conclusion, as we move forward, it remains imperative to ensure that technological evolution in the field of literature respects and augments the concept of Weltliteratur. Evidently, this will require continuous research, an open dialogue, and a flexible approach to technology application, always keeping in mind the overarching goal of creating a truly multilingual and multicultural global literary landscape without disregard for fellow human beings (and the expertise they bring to the table) and the environment. In this context, a balanced approach is essential: one that weighs both the benefits and the challenges of MT which could foster or hinder the development of a diverse and inclusive literary world. In essence, the journey towards a Weltliteratur worthy of the name is ongoing, rich with promise, but also requires careful consideration to realise its full potential.
