Theatres of the Mind: What modelling and visualizing translation processes tell us about the power of comparison

Have you ever been wondering what actually happens when you translate a phrase from a foreign language into your own? It seems like a pretty straight-forward process, doesn’t it? You substitute the foreign words by their respective equivalent in your mother tongue and – voilà – you have achieved your goal. Of course, it is not that simple; the process of translation just described, i.e. a word-for-word or verbatim translation, is merely one of the many different manifestations of translation processes there are. Yet, all of these processes share a common feature across disciplines: due to its complex nature, translation is a topic met with reluctance. The inclination to vindicate the field and demonstrate its effectiveness is what Dr Civiliene pursues in her research at King’s College London. In her talk “Digital Humanities and the politics of comparison: translated, lived, lost(?)” she points out the significance of researching and understanding translation in context.

Dr Civiliene starts out her talk by criticizing the dismissiveness with which the method of comparison, lying at the heart of translation studies and many other scientific and humanistic disciplines, is treated in scientific discourse. Since disregarding a core method of research is detrimental to the field of translation studies per se, translation is denied its potential to serve as a fruitful source of scientific endeavor. Good comparison is difficult to achieve as the relation between two or more things encompasses diverse aspects beyond the scope of differences and similarities. In its early stages, translation studies hinged on this understanding of comparison, as the discipline followed a prescriptive approach by establishing how translation should be done rather than observing the socio-cultural determinants informing translation processes. By including aspects of socio-cultural significance, the field broadened its horizon and scope. This is an important step because biases in comparison, and hence translation, may seem harmless at first, but a coherent narrative spun by perpetuating these biases prefigures our understanding of things and thereby the world. Perpetuating certain biases by featuring some aspects more prominently or focusing on salient features while being oblivious of other seemingly less significant aspects can produce stereotypes, as well as epistemic and social inequalities. Obviously, this critique of comparison is not totally unfounded or unsound and comparison as a tool and method faces constraints and limitations. One of these constraints derives from what Dr Civiliene coined hypercollective epistemic goods, denoting those practices and topics in research and knowledge acquisition assigned a higher value than others. Thus, monocultural perspectives are reproduced which leads to ignorance towards smaller topics in the case of translation studies and languages. In stressing one topic over another, the method of comparison may lead to overgeneralization hence losing track of detail. This may ultimately lead to the manifestation of stereotypes as for example linguistic hegemonies. In this vein, translation studies have been liable to the universalist promise of identifying universal features allegedly shared by everybody.

Pursuing the premise that translations are much more complex and fuzzier than this universalist approach tends to allow for, Dr Civiliene introduced the “Distant Reading across Languages” (DRaL) project. The project launched in 2019 at King’s College and aims to demonstrate the complexity of translation processes. The questions motivating the project included: how can repetition be computed so as to avoid building a classical corpus1? How can repetition be operationalized so that it can be used as a heuristic tool in multilingual alignment? And how can the translator’s perspective on importance be captured and the process itself be elucidated and visualized appropriately? To answer these questions, the project seeks to demonstrate the complexity involved in translation by means of computing the word frequencies in William Faulkner’s “The Sound and the Fury” as repetition strings which are linked with lexical choices in the translations into Western and Eastern European languages. The choices are modelled and visualized as patterns of relative omissions and variation in the translation of repetition. Currently, the website features six translations of the first chapter, also called the Benjy chapter, and three of the last chapter of Faulkner’s book. The website enables the visualization of their results but does not feature the modelling as of now. One challenge that remains pervasive is the linguistic complexity underpinning these translation processes, a complexity which resists conventional methods of representation. As the team is keen on avoiding any compression of data to make it statistically more tangible, the common practice of network analysis to represent significant features as nodes and their relation in terms of the proximity between these nodes, proves problematic. Distance, or proximity, is a relative concept and any statistical abstraction is bound to lose sight of details deriving from context and relations outside the binary poles of original and translation.

Figure 1: Example of the visualization of relative omissions in two Russian translations of William Faulkner’s “The Sound and the Fury”. Note the juxtaposition of the noun woman (highlighted) when comparing the number of omissions of the word. (Link)

In the above figure, the visualizations of relative omissions as they can be browsed on the project’s website is exemplified by the juxtaposition of two Russian translations. The bars represent the number of omitted words, the empty spaces in between signify that no omission occurred. The two Russian translations focused on here were produced roughly around the same time during the Soviet regime, yet the upper representation (1973) was published during the Soviet reign whereas the lower representation (2001) was published afterwards. The contrast in these translations, according to Civiliene, does not stem from systemic linguistic differences, a common argument put forth in comparative studies, but, to her, were clearly motivated by choice. This point can be elucidated when considering more context. In this regard, the website offers a tabular, yet truncated, view of each word with the corresponding choices of the translator (see figure below).

Figure 2: Tabular representation of the use of respective words in context and across translations (here: two Russian versions). (Link)

The project’s findings highlight the significance of repetition as a unit for analysis as it captures the thematic nature of the original text at the same time bringing forth conceptual and lexical shifts in translation. A deep comparison of multilingual data, as is sought after in the DRaL project, neatly aligns with Geertz’s concept of thick description. Geertz’s method of observation, characterized by a back and forth between the whole conceived through the parts informs the project’s objective of bridging the gap between distant and close reading. Along with this view, cross-linguistic comparison is thus alleviated from a mere means to an end to a tool and method of visual representation.

At the heart of every translation lies its genetic relation to the original. This statement, however, should be taken with a grain of salt, since a translation’s manifold sources cannot be known without further investigation. To demonstrate this point, Dr Civiliene emphasized the different translations of “Jimson weed” as a flower in some works and as grass in others. This difference alone does not support the claim that translators’ choices were informed by outside sources, i.e. beyond the original text, but nevertheless highlights the fact that each language has a specific translation for Jimson weed, yet no target language uses it. When cross-referencing an example like this with the many other instances the team found in the data, they strive to contrive a tool enabling cross-linguistic genetic criticism, allowing them to postulate whether the observed phenomenon is mere coincidence or stems from intertextual relations between translations. This train of thought is also important in considering semantic variation across the different translations. When comparing the different choices of translating the verb “say”, the findings suggest that the choices were motivated by design rather than systemic linguistic differences. The visualization, which is not part of the DRaL project but modelled on the same data, also demonstrates the constraints of visually representing this kind of data on a two-dimensional plane. The more data is rendered, the less contrastive the visual representation can be, and thus less powerful in displaying patterns. To counter this constraint, the data can be clustered into different semantic categories, enabling a more nuanced and continuous description of certain passage. When clustering the data in this fashion, another aspect of translations quickly surfaces: their performative and experiential nature. Being influenced by Gadamer’s view that we are in the world by being in language, another problem in the modelling and representation of translation arises; the use of language as an experiential and dynamic process. This process is difficult to model as phenomena like the observer’s effect, i.e. the act of observing changes the object of observation, is taken into consideration. It becomes evident that translations should rather be viewed as theatres of the mind than mere textual surfaces. A translator’s spatial imagination directs their mental imagery of the space and place evoked by the original text and thus the perspective induced by the translation itself. Due to this dynamic and experiential nature of text and language, a visual representation of this process becomes all the more challenging.

Figure 3: Example of immersive visual representation of spatial imagination in the use of the word “barn” and its translations in William Faulkner’s “The sound and the Fury”(Link)

This conceptualization of translation deriving its sense of space and place from multiple sources beyond the original text brought about a second project: the VR Ethnography and Literature across Language (VRELL). In this project, Dr Civieliene and her team try to address two questions: How can the limitations of the flat dynamics of the two-dimensional plane be overcome? And how can one speculate about translation as a playground for ethnographic exploration? The project’s premise holds that spatial imagination is not transferrable between texts and is instead derived from the translator’s spatial perception. This premise prompted the search for ways of demonstrating the differences in perceived spatial makeups and focus of the writer and translator respectively via immersive technologies. As the famous Wittgensteinian distinction between saying and showing implies, the limits of visualization lie in the distinction of description versus demonstrating (see figure above). By rendering this conceptualization as a tangible source of information experiences via the use of immersive technologies, the possibility to recontextualize translations and integrate them in a broader scientific discourse as a source of ethnographic and anthropological significance arises. This opens up the possibilities for modelling, visualizing and further investigating translation in a broader cultural context, leading to a more sophisticated understanding of the many-faceted process of translation.

About the author: Cosima Granderath-Miegel is currently majoring in Informationsverarbeitung at the University of Cologne. She holds a B.A. in Linguistic- and Literary studies where she specialized in Cognitive Linguistics. Currently, she is employed at the Language Archive Cologne as a developer.

1 This is the object of research in corpus linguistics (or studies), which should not be confused with translation studies although the fields share common features. In corpus linguistics, languages are studied via their textual representation, i.e. a corpus, considered to denote the real world use of language (which can, evidently, be contextualized).


Schreibe einen Kommentar

Deine E-Mail-Adresse wird nicht veröffentlicht. Erforderliche Felder sind mit * markiert.

Diese Website verwendet Akismet, um Spam zu reduzieren. Erfahre mehr darüber, wie deine Kommentardaten verarbeitet werden.