A recurring challenge when studying history of translation is interpreting catalogue metadata. On one hand such interpretation is limited by the fact that data present in catalogue records is tabular and nominative, and not quantitative. On the other hand, such research is guided by tacit knowledge of scholars in the humanities, and thus it could be challenging to reproduce its results. We take inspiration from phenomenography, a discipline within educational research that examines how students perceive the phenomena being learned. We adopt the view that scienti昀椀c inquiry is a collective form of learning. By doing this, we turn to the phenomenographic theory that variation is necessary to understand the phenomena being studied, and is achieved through three distinct patterns of variance: contrast, generalisation and fusion. We propose an approach to visualise the combination of nominal data and tacit knowledge by subjecting it to these three patterns. We illustrate our approach with two case studies from literary translations between Italy and the UK in the post-war 20th century. Our claim is that on one hand this guides scholars on how to analytically approach their research questions, on the other it drives them to externalise and validate hidden assumptions. Our approach o昀ers a way of doing reproducible science not only when conducting literature research with bibliographic data. It is also applicable in the wider cases within the humanities when tabular data are available.
A recurring challenge in translation history is the interpretation of bibliographic metadata, a
task readily available for analysis within the digital humaniti1e7s, [22]. Yet, most of the data
in bibliography are nominal, which strongly reduces the applicability of commonly used
techniques from comparative analysis, semantic language modelling or quantitative visualisations.
Moreover, complex historical and critical analysis typical for research in the humanities
remains out of the scope of numerical approaches16[]. This is partly due to the fact that in
research projects with a humanistic focus scholars tend to formulate hypotheses within their
discipline relying heavily on previously acquired – and therefore tacit – knowledge shared by
other members of their scienti昀椀c community. This has posed a challenge to scholars’
collaboration in the digital humanities, most relevantly in the creation of a shared research environment
where researchers coming from di昀erent backgrounds can elaborate research questions
relevant at a real interdisciplinary level, i.e. not relying on tacit knowledge. John Biggs places
tacit knowledge within an extensive taxonomy of knowledge types. Beyond other commonly
discussed forms such as theoretical and procedural knowledge, he also includes tacit, intuitive,
metatheoretical and conditional knowledge. Some of these categories include forms of
understanding that are provisional or even counterfactual, and scienti昀椀c rigour would require the
validation of such knowledge through critical reasoning or triangulation with other knowledge
or evidence [
To respond to this challenge, we have turned to phenomenography - i.e. the discipline that
investigates how learners perceive the phenomena they study14[
RQ: How the phenomenographic three patterns of variance could shape a reproducible method to perform exploratory visual analysis of bibliographic data?
We focus on translations of Italian 20th century 昀椀ction into English (UK market) and of British
20th century 昀椀ction into Italian in the period 1945-19701. We created two symmetrical datasets
of bibliographic data: one collecting Italian titles translated in the UK, the other containing
titles in English translated into Italian. For the 昀椀rst dataset, we sourced partial data from
Healey [
We focus on a subset of the collected data, consisting of the bibliographic information of the 昀椀rst edition in the source language, and the 昀椀rst translated edition in the target language. For each pair original-translation, we consider author and title. Furthermore, on both ends we consider publisher and year of publicati4o. nA clear limitation of this approach emerges for translations published several years a昀琀er the original publication, which may imply a change of publisher. Whether the title was acquired from the 昀椀rst or the second publisher is a piece of information that our datasets do not contain. This is only one of the arguments for the need to treat 昀椀ndings of our computational analysis as a provisional result that needs to be veri昀椀ed and explained, as opposed to de昀椀nitive 昀椀ndings. In this sense we propose to combine quantitative and qualitative analysis: since the latter implies a good deal of tacit knowledge for a hypothesis to be formulated and tested, the challenge is to 昀椀nd a method to obtain traceable provisional results by automatising most operations that do not rely on tacit knowledge. As a consequence, this also supports the discernment of where tacit knowledge might be involved.
When working with metadata and without a relevant text corpus, recent research in digital
humanities has turned towards network analysis17[
Educational research has traditionally had strong focus in the communication of knowledge
and di昀erent possible perceptions. In the 1980s the discipline of phenomenography emerged
as the study of the di昀erent ways in which phenomena are perceived. Within this discipline,
a theory of variation was elaborated, for which successful learning followed three distinct
patterns of variance: contrast, generalisation and fusio1n3[]. Contrast posits that a dimension of
the phenomenon under consideration cannot be discerned unless its di昀erent values are
experienced. Akin to the approach of scienti昀椀c controlled experiments, the focus on contrast enables
the understanding of what changes with the variance of the dimension of interest. When it
comes to generalisation, changes in other dimensions of the phenomenon need to be perceived
while keeping the dimension of interest invariant. This enables both further understanding of
3British National Bibliography, accessible from the British Library’s main catalohgtutep:s://explore.bl.uk
4For each dataset, this results into six-columns tables: two columns are numerical, providing years of publication,
and four nominal, containing author, title and the two publishers.
the impact of each value of the dimension of interest and the understanding o昀琀he
surrounding context of this dimension. Finally, fusion of di昀erent dimensions is crucial to understand
the interplay between them. This means having a way to experience the independent
combinations between the di昀erent dimensions under scrutiny. Visualisation plays an important
role in communication of knowledge and in fact, these patterns of variance are also familiar in
data visualisation. For example, Bach and colleagues present the patterns of comparison and
repetition [
Bibliographic records are represented as a combination of names (title, author, publisher, etc.), dates (e.g. publication date), identi昀椀ers (e.g. ISBN, URL) and possibly other data. In computer science terms, these data are the attributes of each bibliographic record and all attributes for the same data across all records form a dimension. From an analytical perspective, these dimensions are mostly nominal. There are only few formal operations that can be performed on nominal attributes. Without additional taxonomic knowledge of the underlying domain of the attribute (e.g. publishers), all that can be said about two di昀erent instances is whether they are equal or not8[]. Such an equality operation allows for the possibility to group and count equal instances. However, further operations typical for quantitative analysis, like ranking any two values or measuring the distance between them, are not well de昀椀ned8][. This limitation precludes further analytical operations like ordering or clustering the data.
Partially due to the above reasons, exploratory visualisations have been traditionally used for quantitative [18, 19] and only rarely for nominative dat5a][. Moreover, research in data visualisation has explored the dimensionality of visualisations, in which projections of data along a single line is considered to be a visualisation of one dimension, mapping of two data dimensions one against the other is considered to be of two dimensions and so o1n9][. Examples of one-dimensional visualisations that could be used for the visualisation of nominal data and its counts are bar charts, line charts and pie charts, of two-dimensional - heatmaps (Figure2), and of two or more dimensions - diagrams that represent dimension parallelly like alluvial diagrams (Figure3sand 5) and slope chart diagrams (Figur6e).
We operationalise variation by applying the three patterns to the de昀椀ning characteristics we have identi昀椀ed for our research domain: the combination of nominal data and tacit knowledge of contributing scholars. In phenomenographic terms the de昀椀nition of dimensions and values may be 昀氀exible – one aspect could be a dimension in one case and a value in another13[], e.g. a speci昀椀c publisher can be seen as both one of many, and the dimension in which a speci昀椀c book series are de昀椀ned. However, the consideration of such 昀氀exibility remains beyond the scope of this paper and subject to future research. Rather, we take a naive approach by considering the dimensions formed by attributes of bibliographic records as phenomenographic dimensions and the corresponding nominal data as the phenomenographic values. Here is how the three patterns of variance could be represented in visualisations: Contrast. A 昀椀rst step is focusing on a single dimension of interest. This could be performed not only as considering the possible values of this dimension, but also aggregating the other dimensions by counting the distinct values there that correspond to each of the values of the dimension of interest. Even though multiple other dimensions could be aggregated, any resulting charts representing counts are unidimensional.
Generalisation. The second way to experience variation is to 昀椀x a dimension to a particular value and visualise a chart juxtaposing other dimensions under this condition. In this case it is bene昀椀cial to take advantage of multidimensional representations, visualising two or more complementary dimensions.
Fusion. The 昀椀nal step in experiencing variation is to create a visualisation of the big picture, possibly a type of map that includes all possible data, risking to obscure the possible details of interest. To examine the interplay of several dimensions, they need to be plotted on a graph with high dimensionality.
When it comes to externalising tacit knowledge, we propose that be done by explaining the why and how the patterns of variation were applied. Clearly, this is not exclusive to tacit knowledge, but also could lead to eliciting intuitive, metatheoretical and conditional knowledge.
We have tested our approach by visualisin5gtwo case studies from our datasets of literary transfer. The 昀椀rst one focuses on Heinemann as a sort of wholesale importer to Italy. The second on Vasco Pratolini as the only Italian author in the period who had two distinct publishers translating more than one literary work of his. Heinemann’s case was chosen because of the high number of partnerships with Italian publishers, and especially with Mondadori. Besides being Italy’s major publisher, Mondadori is the one with most accessible data, which provided an apt test-case for our study. Pratolini’s case was chosen because of the author’s presence across several publishers’ catalogues, which in merely visual terms, is outstanding and therefore elicits questions as to possible interpretations on the qualitative level.
One important research question within literary transfer studies concerns relationships between publishers, a crucial factor for the establishment of translated authors. For the sake of simplicity, we consider a case study contained only in the UK to Italian dataset. 5We illustrate our case studies with the RAWGraphs visualisation ap1p5][ for two reasons. First, beyond the widerspread chart types for quantitative data, it also features charts that support dimensions containing nominal data. The second useful feature of RAWGraphs is the export of the charts in vector graphics, which allows for easier post-processing.
A 昀椀rst step is identifying the (quantitatively) big players in the process. To employ contrast, we look at all UK publishers involved in literary transfer to Italy. This view shows us that the market of exports to Italy is very segmented with only 4 out of 49 publishers having translated more than 15 books each (Figure1). Yet these 8% of publishers are cumulatively responsible for 43% of studied translations. Focusing on these four key UK publishers, in a further generalisation step we look at authors and Italian publishers of translations by this UK publisher (Figure2). In line with contextual knowledge that Heinemann is close to exclusive publisher of G. Greene, W. Somerset Maugham, and others, here it is visible that these authors form an important part of Heinemann’s exports and correspondingly that there is a strong partnership between Heinemann and Mondadori. Finally, fusion through an alluvial diagram allows us to visualize the role of Heinemann in the big picture (Figu3r)e- as one of the two strongest exporters to Italy thanks to its important role in the UK and strong relationship with one of the big players in Italy. Whereas Gollancz outperforms Heinemann in numbers of translated titles, it appears less persistent with its partnerships, both with authors and Italian publishers.
A further step in this investigation would be to engage in a study of archives to understand whether Heinemann lost its role on the Italian market in 1961 when it was acquired by Tilling and was le昀琀 by Greene and others [20], but this is beyond the scope of this paper.
Another question that helps understand the role of publishers is what authors were successful when not working with the same publisher.
This leads us to consider Italian authors that have been translated by more than one publisher (Figure4), a case of applying contrast on the authors. We consider all 昀椀ve authors that have at least four works translated and at least two publishers, for each of the Italian and UK markets. This leads us to generalise over these authors and compare how their translations were published (Figure5). This reveals that Pratolini is the only author that was published twice by each of two UK publishers. This suggests that Italian publishers were not the ones that de昀椀ned these consecutive relationships. Rather a possible interpretation could be that Vallecchi had a role in opening the door to Pratolini to diverse UK publishers. Finally, to visualise the potential impact of this limitation alongside with the changes of UK publishers, we employ fusion to display the Italian authors that have four or more translated titles showing the publication years by country and the corresponding UK publishers (Fig6u)r. eHere we see that there was a particularly big gap in some translations of Pratolini.
Archival research into the communication between editors reveals that it was not the 昀椀rst publisher (Vallecchi) that managed the translation on the Italian side. This illuminates an issue that still needs to be addressed. Whereas for short periods between 昀椀rst publication in original language and the 昀椀rst translation, it might be realistic to assume that it was the 昀椀rst publisher that managed the translation, when more time passes this becomes more complicated, and we need to also include subsequent Italian reprints and editions in our analysis.
In this paper we proposed a new approach towards visually analysing bibliographic data, addressing two challenges that we identify to be critical: the nominal character of bibliographic data and the tacit nature of complementary knowledge of scholars in translation studies and in the humanities in general. One could argue that the steps of contrast, generalisation and fusion are intuitive and thus already widely applied. Yet, it is the systematic use of the combination of these that makes the approach conform with phenomenography. To validate our proposed approach, we apply it to two case studies from literary translation between Italy and the UK in the post-war period of the 20th century. We approached simplistically the mapping between the conceptual terms of phenomenography and the constructed objects of tabular data representation. This interpretation is overly limiting, e.g. research in cognitive lo1a2d] s[uggests that people are capable of mentally constructing domain-speci昀椀c schemas that can potentially span across multiple descriptive dimensions. This opens to the possibility that for example a phenomenographic value could include a combination of values across bibliographic 昀椀elds, possibly also including aspects of tacit knowledge, e.g. considering the partnership between UK and Italian publishers as a dimension, incorporating corresponding researcher knowledge.
The creation of a research environment where data is sourced and analysed in collaboration between scholars in the humanities and in computer sciences right from the start is for us a priority. To do this, it is crucial to negotiate a set of questions that are relevant on both sides and likely to contribute to both quantitative and qualitative analysis, since neither of these, alone, can impact the way of writing translation history.
This publication is part of the MiGATE (Milan gateway to Europe) project, funded by the Italian National Programme for Research (PNR 2021-2027) and hosted by the Department of Languages, Literatures, Cultures and Mediations of the University of Milan.