Translation and Delta Revisited: When We Read Translations, Is It the Author or the Translator that We Really Read?

One of the first success stories of Burrows’s Delta,
that of the ‘Englishing of Juvenal,’ strongly points
out that some translators, such as Johnson, translate
other people’s writing as if they were writing their own
work, while others, such as Dryden, are ‘able to conceal
their hand’ (Burrows 2002, 2002a). Previous stylometric
studies of patterns in multivariate diagrams of correlation
matrices derived from relative frequencies of most
frequent words in character idiolects (Burrows 1987) in
a number of originals and translations (Rybicki 2006,
2006a, 2007, 2008) have shown that while some stylometric
tools can show the difference between translator
and translator, the relationship between such patterns
between original and translation is at best unclear – as
unclear, one could say, as is the relationship between the
lists of most-frequent-words in two (or more) languages,
where one-on-one correspondences are rare if not nonexistent.
In the first-mentioned pioneering study, Burrows compared
translations by a variety of authors to their own
writing – a very natural material for translations of poetry,
itself very much the domain of poets. But translations
of prose are usually done by men and women who
often are not novelists in their own right; while there is
a reasonable guarantee that a Dryden or a Miłosz translation,
however ‘unfaithful’ or ‘free’, can still be good
poetry, can one really trust that a novel translated into
another language by someone who has never invented
a plot line of his/her own can be at least remotely connected
to the original in terms of such trifling features
as… literary style? Or, in other words, whose style is it
that we see in the translation: some foreign version of the
original author’s, or simply that of the much less talented
translator? For, obviously, Kurt Vonnegut’s requirement
for a perfect translator (‘that he or she be a more gifted
writer than I am, and in at least two languages, one of
them mine,’ Vonnegut, 1991) is rarely met, and probably
never in translations of prose. In fact, traditional (i.e.
non-computer- and/or non-statistically-assisted) translation
studies speak of ‘inherent intersubjective processes,’
which result in an ‘unpredictability of the target style’
(Wilss, 1996).
The above also signifies that any study venturing into
these uncharted waters is deprived of exactly the ideal
comparative material Burrows could use in his ‘Juvenal;’
and that the best one can count on are other translations
by the same translator, or other translations of the same
author by other translators; if in luck, one can sometimes
use different translations of the same novel. In this context,
the main question of the title boils down to whether
different works of an author translated by the same
translator are going to be stylometrically more similar to
each other than to translations by that translator of other
authors; and whether translations of one author by many
translators are going to resemble each other rather than
translations of other individual authors by these translators.
Two discrete material sets have been used in this study.
The first is a collection of 21 Polish translations of English-
language fiction by Rybicki, made between 1991
and the present, with two authors appearing more than
once: John le Carré (5 different novels) and Douglas
Coupland (3). The second consists of 10 Polish translations
of Jane Austen’s novels by 4 different translators,
produced between 1956 and 2006, and compared to 5
other translations by the same translators (at least 1 by
each). Both sets have been subjected to testing by three different
tools: Principal Component Analysis, Burrows’s
Delta (including Hoover’s and Argamon’s significant
modifications such as DeltaOz, resulting in the powerful
set of Delta spreadsheets, Hoover 2004, 2004a, 2007,
2007a, Argamon, 2008), and the recent black box software,
JGAAP 3.3 (Juola et al., 2006, 2008), still in demo
version. For the first two tools, various combinations of
‘culling,’ wordlist lengths and primary and secondary
test groups have been used; in the third, which presents a variety of ‘events’ and statistical method combinations,
those deemed the most reliable by the makers of JGAAP
were used.
In the first set, all three methods clearly highlighted a
great similarity of the five Rybicki translations of le Carré.
Multivariate Analysis graphs (Cluster Analysis, Factor
Analysis and Multidimensional Scaling) always contained
a visible (yet not exclusive) cluster of this author
(as exemplified in Fig. 1). A huge majority of various
combinations of wordlist size, pronoun deletion, culling
percentage, primary and secondary text selection, and
Delta variety, correctly identified translations of le Carré
from among all other Rybicki translations (the infrequent
yet rule-proving exception being that of Barris, whose
Confessions of a Dangerous Mind can be seen as a parody
of the spy thriller genre, and who twice preceded le
Carré, himself ranked second, as the most likely author
of the original). In contrast, the three Coupland books
exhibited much less stylistic consistency – a possible
consequence of the significant time that had elapsed between
the three translations. On the other hand, given the
choice of only le Carré and Coupland as primary authors,
Delta invariably identified all of their translated novels
correctly and refused to mistake them for any of the secondary
authors – and that at a very wide range of mostfrequent-
word lists. A very similar pattern emerged from
some of the more robust statistical procedures available
in JGAAP (word analyses with Kolmogorov-Smirnov
Distance and Manhattan Distance).
Fig. 2 Multidimensional Scaling Analysis of translations of
Austen (uppercase initials) and of other original writers by
the same translators (lowercase)
In the second set, multivariate graphs contained a visible
cluster of all translations of Austen novels by all
four translators, while their other translations hovered
away from the cluster in other parts of the graph (Fig. 2).
This was even more obvious in Delta analyses, where all
Austen novels in Polish, irrespective of their translators,
were almost always correctly identified as Austen novels
(in 9 out of 9 cases at various combinations of parameters,
Table 1). Conversely, standard Delta and DeltaOz
never exceeded a 5/11 correctness ratio when asked to
tell translator from translator (Table 2). Faced with an
Austen text by translator A, JGAAP almost invariably
chose a translation of any Austen novel by translator A or
B, rather than a translation of another author by translator
A, as the more similar of the two. The conclusions from this study are twofold. First, the
results – reliable, because they were obtained using three
tools using a variety of statistical calculations (traditional
multivariate analysis, state-of-the-art Delta, and the
emerging JGAAP black box with its various distances)
– are an interesting addition to Burrows’s study of the Juvenal
translations. The fact that translations of prose tend
to ‘disguise’ the individual styles of the translators and
seem to create individual styles of the original authors
in a foreign language – styles common to a number of
different translators even more consistently than translations
of poetry – can be a consequence of the greater
formal complexities of poetic translation, which usually
calls for the task to be done by literary creators in their
own right – exactly such as the translators studied by
Burrows. Secondly, and perhaps more importantly, the
fact that ‘unpredictable target style’ is in fact predictably
similar between translations of the same original author
done by many people – as measured on the basis of freeliminate individual word-types characteristic for the vocabulary
of individual novels and translators) – suggests
possible and previously unsuspected affinities between
original and translation at very ‘mechanical’ levels of
the text. Indeed, since many of these words – especially
those at higher frequencies – carry grammatical rather
than lexical meaning, there is hope on the horizon that
one day computational stylistics and cognitive linguistics
(and particularly cognitive translation studies) might
be two sides, experimental and theoretical, respectively,
of the same phenomenon – as has already been suggested
by Connors (2006, 2008).
References
Argamon, S. (2008). ‘Interpreting Burrows’s Delta:
Geometric and Probabilistic Foundations,’ Literary and
Linguistic Computing 23(2): 131-147.
Burrows, J. F. (1987). Computation into Criticism:
A Study of Jane Austen’s Novels and an Experiment in
Method, Oxford: Clarendon Press.
Burrows, J. F. (2002). ‘The Englishing of Juvenal:
Computational Stylistics and Translated Texts,’ Style 36:
677-99.
Burrows, J. F. (2002a). ‘”Delta”: A Measure of Stylistic
Difference and a Guide to Likely Authorship,’ Literary
and Linguistic Computing 17: 267-287.
Connors, L. (2006). ‘An Unregulated Woman: A Computational
Stylistic Analysis of Elizabeth Cary’s The
Tragedy of Mariam, The Faire Queene of Jewry,’ Literary
and Linguistic Computing 21(Supplementary Issue):
55-66.
Connors, L. (2006). ‘Function Word Analysis and
Questions of Interpretation in Early Modern Tragedy,’
Oulu: Digital Humanities 2008.
Hoover, D. L. (2004) ‘Testing Burrows’s Delta,’ Literary
and Linguistic Computing 19: 453-475.
Hoover, D. L. (2004a) ‘Delta Prime?’ Literary and Linguistic
Computing 19: 477-495
Hoover, D. L. (2007). ‘Corpus Stylistics, Stylometry,
and the Styles of Henry James,’ Style 41(2): 174-203.
Hoover, D. L. (2007a). ‘Quantitative Analysis and Literary
Studies,’ A Companion to Digital Literary Studies,
Oxford: Blackwell, 517-33.
Rybicki, J. (2000). A Computer-Assisted Comparative
Analysis of Henryk Sienkiewicz’s Trilogy and its Two
English Translations. PhD thesis, Kraków: Akademia
Pedagogiczna.
Rybicki, J. (2006). ‘Burrowing into Translation: Character
Idiolects in Henryk Sienkiewicz’s Trilogy and its
Two English Translations,’ Literary and Linguistic Computing
21(1): 91-103.
Rybicki, J. (2006a). ‘Can I Write like John le Carré?’
Paris: Digital Humanities 2006.
Rybicki, J. (2007). ‘Twelve Hamlets: A Stylometric
Analysis of Major Characters’ Idiolects in Three English
Versions and Nine Translations,’ Urbana-Champaign:
Digital Humanities 2007.
Rybicki, J. (2008). ‘Does Size Matter? A Re-evaluation
of a Time-proven Method,’ Oulu: Digital Humanities
2008.
Vonnegut, K. (1991). Fates Worse than Death, New
York: Berkley Trade, 181.
Wilss, W. (1996). Knowledge and Skills in Translator
Behaviour, Amsterdam: John Benjamins, 34-5.
Juola, P., Noecker, J., Ryan, M., and Zhao, M. (2008).
‘JGAAP3.0 -- Authorship Attribution for the Rest of Us,’
poster, Oulu: Digital Humanities 2008.
Juola, P., Sofko, J. and Brennan, P. (2006). ‘A Prototype
for Authorship Attribution Studies,’ Literary and
Linguistic Computing 21:169-178.