Comparing the Similarities and Differences between Two Translations

Comparing the Similarities and Differences between Two Translations
Lucic, Ana, Graduate School of Library and Information Science, University of Illinois at Urbana-Champaign, alucic2@illinois.edu
Blake, Catherine, Graduate School of Library and Information Science, University of Illinois at Urbana-Champaign, clblake@illinois.edu
Abstract
Burton Pike described Rainer Marie Rilke’s style in Rilke’s only prose novel, The Notebooks of Malte Laurids Brigge, as “… arresting, haunting, and beautiful, but it is not smooth. His style is explicit, direct, almost laconic, and it has an edge.” (2008, p. xvii) Pike argues that this “edge” was not sufficiently emphasized in previous English language translations and he thus wrote a new translation. The goal of this research is to explore the degree to which automated text analysis tools can capture the different styles used by Burton Pike and Stephen Mitchell in their respective translations. We are particularly interested in what kinds of similarities and differences can be captured between two renderings of Rilke’s novel and in the implications of these findings on the reviews, reader reception, and critical analysis of the original work in two translations.

Method
Two candidate analysis tools were used to identify similarities and differences between the Pike and Mitchell translations of Rilke’s novel The Notebooks of Malte Laurids Brigge. The first approach used a syntactic representation of the texts which was generated using the Stanford Lexical parser ( (link) ). Before running the parser, each translation was tokenized into sentences using the Natural Language Toolkit tokenizer ( (link) ). The generated lexical dependencies, which capture grammatical relations between words in a sentence, were then uploaded to the oracle database for subsequent analysis.

The second approach used a statistical approach–principal component analysis–to compare the manifestations of the novel. Our experiment reflects McKenna et al.’s study (1999) of similarities and differences between Samuel Beckett’s French and English translation of Molloy. Three matrixes were produced comprising the 99 most common words in each of Pike and Mitchell’s translation and the original text. The rows of the matrix reflect 8 text blocks of 7,500 words. Principal component analysis, which measures the variance of 99 most frequent words in the 8 blocks, was then applied to each matrix and the top eigenvectors produced were mapped into a two dimensional space for visual analysis.

Results
Dependency grammar analysis

Table 1 provides the summary statistics of grammatical relations which showed the largest difference captured in the parser for each translation. Although the two translations show a remarkable degree of similarity in the frequency and type of grammatical relations they employ, the dependencies revealed several differences between the two renderings. The main areas of difference were found in the use of negation modifiers, prepositional modifiers, object of preposition, parataxis, and in the word choices for adjectival and adverbial modifiers.

The dependency grammar results show that Pike used many more negation modifiers than Mitchell, which is confirmed by word frequency analysis. Pike uses the non-contracted verb, which places more emphasis on negation and thus also stays closer to the original text whereas Mitchell rarely uses a non-contracted verb form. The word “not” is not the only word in Pike’s translation which indicates a negation that is used with higher frequency. Pike uses “no” with higher frequency than Mitchell, and also “nothing,” “never,” and “none.”

The frequencies of prepositional modifier and object of preposition relations in two texts indicate a difference in how the prepositions are used in the text. This result shows a higher overall frequency of prepositions in Mitchell’s translation (Table 1). The sentence level analysis of two texts, however, revealed that Pike is more likely to repeat and thus emphasize the same preposition throughout the sentence whereas Mitchell is more likely to leave out the preposition rather than repeat it. This finding suggests that in addition to the overall difference in the frequency of prepositions captured through the parser, there are indicators that the use of prepositions, their placement and distribution throughout the sentence, show differences in two texts.

The comparison of two translations revealed a higher frequency of sentences that use a semicolon to separate sub-clauses in Mitchell’s translation (parataxis). Mitchell frequently arranges independent sentences using semicolons and colons, and in this way follows closely the original text, while Pike occasionally intersperses “but” and “and” in place of semicolons and colons. This finding is supported by the frequency of conjunction “but” in two translations which is found at higher frequency in Pike’s translation, 559, than in Mitchell’s, 506. This difference in sentence structure may provide the “edge” which Pike claims was missing in previous translations.

An examination of the unique adverbs and adjectives used in each translation revealed that although Mitchell and Pike may use similar grammatical relations in their respective renderings, their word choices are frequently different. This suggests that a semantic rather than the syntactic analysis may reveal additional differences.

PCA analysis

Figure 1 shows the score plots of the translations and suggests that the distribution of frequent terms is very similar between the two translations. However, the grouping of the 7th block with 4th, 5th, and 6th rather than with 8th block, in Mitchell’s translation, calls for a closer analysis of the 7th block of the novel and its comparison with the 7th block in Rilke’s original text and Pike’s translation. The last part of the novel, which corresponds to the 8th block of text, is visibly different in style and tone from the rest of the novel and this difference is indicated by the location of the 8th block in the far right corner of the plot in each version.

Conclusion
Our results thus far suggest that syntactic grammatical relations reveal differences between the two translations that are not captured when using a bag-of-word approach (word frequencies of function words and 1,000 most frequent words). In contrast, the PCA analysis using matrices of frequent terms in the 8 blocks of text suggests that only small differences exist between the translations, with the exception of the 7th block of text, which warrants further investigation.

The similarity captured between translations using the PCA analysis brings to mind Jan Rybicki’s article “Burrowing into Translation: Character Idiolects in Henryk Sienkiewicz’s Trilogy and its Two English Translations” in which Rybicki analyzes the distinctiveness of character idiolects in Henryk Sienkiewicz’s trilogy and concludes that: “… the patterns of difference and similarity are almost mysteriously preserved in the translations—so well that the above-mentioned linguistic differences might be the sole reason for the small differences between the original and the translation. In the greater picture, characters differ one from another in the translations just as they do in the original.” (Rybicki, 2006, p.102)

The differences between two translations that were established will help create the linkages between these findings and the reader reception, reviews, and critical analysis of two translations. We also hope that they will help trace the contours along which the creation of the new variant and new literary rendering begins to emerge and the differences that speak directly to the explicit, direct, and laconic style of Pike’s translation.

We plan to extend this work to include M. D. Herter Norton’s translation (1949) of The Notebooks of Malte Laurids Brigge and to investigate how well this method generalizes to different translations of different genres.

Table 1 - Dependency distributions with the largest difference between the Pike and Mitchell translations of The Notebooks of Malte Laurids Brigge by Rainer Maria Rilke

Full Size Image

Figure 1 - Score plots

Full Size Image

References:
de Marneffe, M. & Manning, D. C. 2008 Stanford typed dependencies manual., 30 October 2010 (link)

Lucic, A. 2010 “Measuring Similarities and Differences between Two Translations., ” Research Showcase 2010 at the Graduate School of Library and Information Science, University of Illinois at Urbana-Champaign, Champaign, IL, 2010

McKenna, W., Burrows, J., Antonia, A. 1999 “Beckett’s Trilogy: Computational Stylistics and the Nature of Translation., ” RISSH, 35 151-71

Rilke, R.M. 2000 Die Aufzeichnungen des Malte Laurids Brigge. [Plain Text UTF-8], 30 October 2010 (link)

Rilke, R.M. 1949 The Notebooks of Malte Laurids Brigge (Translated by M. D. H. Norton)., W.W. Norton & Company New York

Rilke, R.M. 1982 The Notebooks of Malte Laurids Brigge (Translated by S. Mitchell)., Random House New York

Rilke, R.M. 2008 The Notebooks of Malte Laurids Brigge (Translated by B. Pike)., Dalkey Archive Press Champaign

Rybicki, J. 2006 “Burrowing into translation: Character Idiolects in Henryk Sienkiewicz’s Trilogy and its Two English Translations., ” Literary & Linguistic Computing, 21(1) 477-495