This research focuses on the interaction between human and computer in processing text of fiction. It is related to operations as applied to overall managing and creating (p)arallel (t)ranslation (c)orpus (PTC) in regard to English as well as Ukrainian language pair. Corpus linguistics comprehensive tools are used both for processing parallel translation corpus results obtained and for the afterwards analysis. Preliminary findings can be exemplary for better understanding of how corpora helps in the scrutiny of the individual style of an author dominants and in what way it can stimulate more qualitative and faithful rendition in translation. The paper provokes a number of issues on how quantitative parameterization can be useful for translation studies analysis. Also, we elucidate the possibilities of NPL manipulation in regard with tagging, hence, researching emotional dislocation as verbalized in fiction.
ST – (S)ource (T)ext.
TT – (T)arget (T)ext.
PTC – (P)arallel (T)ranslation (C)orpus.
XML – (E)xtensible (M)arkup (L)anguage.
HTML – (H)yper (T)ext (M)arkup (L)anguage.
NLP – (N)atural (L)anguage (P)rocessing.
EDA – (E)motional (D)islocation and (A)lienation (tag used for PTC in a current research project). 2.2.
A case study of the novel The Goldfinch written by the Pulitzer winner novelist Donna Tartt is chosen. The reason is that it has not yet been researched in English-Ukrainian parallel, especially from statistical angle.
In addition, The Goldfinch was awarded with Andrew Carnegie Medal for Excellence in Fiction and was film adapted in 2019.
Typologically, the analysed corpus bares the following characteristics; it is annotated; bilingual; illustrative; literary with (a) a certain author (Donna Tartt), (b) a single author type of a text ( Donna Tartt’s novelThe Goldfinch), (c) a single author particular text (corpus of Donna Tartt’s noTvheel Goldfinch), (d) a distinct genre of text (novels), (e) a separate period (twenty-first century American novel), (f) a peculiar group (Modern American mysterious and psychological thriller), (g) a definite theme (emotional register: dislocation, emotions centered around loss, alienation).
Considering the theoretical, practical and applicable issues in regard to author’s idiolect and taking into account the overlap with translator’s individu–alprosftyoluendly discussed and studied from humanitarian perspective, – the particular difficulty of such approach, among other challenges, is to find a golden middle between philological profiling and multifaceted involvement of mathematical linguistics and applied linguistics tools within human language operations.
That such cases are common is of little surprise since there is often a conflicting need to objectively justify the results and support research elaborations with statistics.
In the method and analysis, modern philological researches heavily rely on computer-centered
approaches [
Thus, translation studies analysis incorporates corpora-oriented and descriptive approaches along
with some statistical methods for probability measurement [
In the introductory book [
Authors believe corpus linguistics to be among fundamental domains of applied linguistics [4, p. 88 – 89].
Such processing techniques as word listing and concordancing make it possible to have same data viewed and approached from a variety of angles.
It allows stimulating multifaceted analyses and incorporating researchers to constantly rethink the positions of theirs [19].
Language corpora can empower translation studies textual analysis.
As for the advances, in translation studies it should be mentioned that it stems from the fact that they consist of texts in electronic form, and can thus be stored, shared, and processed in ways that enhance their usefulness and comfortable portable processing as compared hard copy corpora [7, p. 69].
Corpus-based translation studies, to recall, has the potential to be a decentering, dynamic force in translation studies on the whole. In this respect, many scholarly publications have already proven its effectiveness and privileges for the sake of their potential benefit.
J. Munday has it that this is a potentially powerful tool to help analyze translation shifts [12, p. 7].
However, little research is available if to mention quantitative analyses in translated text assessment
within English-Ukrainian parallel. Therefore, in this article we opt for describing practical application
of NPL (Natural Language Processing) methods [
We made sure that tags are correct and error-free, since if the tag is missing, the XML file will not work. At this stage, tagging, based on the context, was carried out.
We compared text samples from the source and the target languages.
In focus – word phrases defining emotional dislocation and alienation. <p> </p> – paragraph mark <s> </s> – sentence tag <loss> </loss> - EDA tag for loss (of a beloved relative) <addiction> </addiction> – EDA tag for addiction (drug, alcohol, cigarettes) <anxiety> </anxiety> – EDA tag for anxiety (fear of losing the friend) <punishment> </punishment> – EDA tag for punishment (fear of accusation and arrest by police for stealing the painting) <rambling> </rambling> – EDA tag for rambling (orphanage/homelessness) <guilt> </guilt> – EDA tag for guilt (a burden of guilt for stealing the painting and selling out the fake paintings) <suicide> </suicide> – EDA tag for suicide (suicidal thoughts/depression) <abandonment> </abandonment> – EDA tag for abandonment (unrequited love) <disappointment> </disappointment> – EDA tag for disappointment (father’s indifference disappointment in him) and <obsession> </obsession> – EDA tag for obsession (obsession with the painting and impossibility to watch it constantly) <failure> </failure> – EDA tag for failure (father’s unrealized carre)e <eda n=1 > </eda n=1 > – verbalized sample of EDA in the text (n=1; n – numbering, 1 – sequence number) 3.2.
Below we provide the screenshots of the marked fragments obtained. To specify, the tagging procedure was conducted for the whole scope of the original and translated documents.
See below the screenshots of the sample fragments of the created corpus of the ST with focus on EDA
Screenshot of the sample fragment 1
The same results were obtained for all the EDA in the texts (original and translated).
Having conducted text mark-up (as depicted in subsection above), there was a spreadsheet created that depicted all the EDA samples from the ST in the chosen subtexts.
The Figure 5 provides the visual illustration of the corpus obtained from ST data. 4.2.
TT EDA in PTC
Following the ST, there was a spreadsheet created that depicted all the EDA samples from the TT in the chosen subtexts.
The Figure 6 provides the visual illustration of the corpus obtained from TT data.
Next, we created a new spreadsheet that contained calculations of words and word usages in chosen subtext. 4.3.
In total, 294619 words were used in the original of the novel, and 268409 in the translation of the work. EDA sample parallel translation corpus counts 2248 ST words and 2031 TT words.
Explicit EDA and its grammatical variants occur 13 times in ST and 16 times in TT. Implicit EDA are used 45 times in ST and 59 times in TT.
Total in the In chosen text EDA subtexts
EDA verbalization
Implicit EDA
Explicit EDA
TT ST ST TT
The number of contextually explicit EDA in ST and TT coincides and totals in 107, which comprises a holistic embodiment of the EDA in The Goldfinch novel.
Most frequent is EDA verbalized via abandonment and has 23 examples.
The least common is EDA verbalized via failure, it has 2 examples.
Among others there are: loss - 15 examples, addiction - 14 examples, anxiety - 12 examples, punishment - 15 examples, rambling - 7 examples, guilt - 6 examples, suicide - 4 examples, disappointment - 3 examples, obsession - 6 examples.
Below are the percentages of EDA verbalized via abandonment, addiction, anxiety, disappointment, failure, guilt, loss, obsession, punishment, rambling, suicide, when EDA by default remains 100%. Now, let us present the visualization of EDA persentage interrelation. abandonment anxiety failure loss punishment addiction disappointment guilt obsession rambling
For this data we made use of the available on opVenICTAacNceAss platform se(e:
http://victana.lviv.ua/nlp/linhvometriia). We focus on contrastinlgexical diversity: syntactic
complexity; speech coherence coefficient; exclusiveness index; concentration index as defined in [
№
Coefficient of lexical diversity : Kl = W / N Coefficient of syntactic complexity : Ks = 1 - P / W
Exclusivity index : Iwt = W1 / W
W = 431 Z = 83 S = 69
P = 29 W1 = 330
Iwt = 0.76566125290023
Concentration index : Ikt = W10 / W
№ Coefficient
Now, let us consider the results in comparison.
All in all, the findings have proven the coefficient of speech coherence (Kz) is most distinctive; compare: ST Kz equals to 1.7471264367816 while TT Kz totals to 2.2307692307692 and exceeds the ST Kz for the amount of 0,4836427939874.
Let us consider the dominant EDA for future Translation Studies analysis of the original and the translation from the viewpoint of pragmatic interference.
The loss of a loved relative is one of the main themes of Donna Tartt's novel The Goldfinch. There are numerous examples of this EDA in the first chapter of the novel. In one of them, the main character of the novel is having a dream in which he remembers his mother. The chain of EDA makes it possible to understand the psycho-emotional state of the main character of the novel. Excerpts below accurately trace the author's accentuation on the emotional and psychological state of the protagonist, who tries to convince himself that his mother survived the horrific terrorist attack. In these examples, the climax of EDA is manifested in two disclosures: explicitly and implicitly. Below is the screenshot to make it easier to visualize.
Next, the key word generator can help to heighten the awareness of meaningful choices.
Block 1 is preconditioned by the norms of the English language of possessive pronouns placement and usage as well as the presence of definite article the, the allomorphic feature for Ukrainian.
Thus, Block 1 proves objective necessity.
In focus – Block 3. These are the lexemes that capture, so to say, EDA in a most illustrative way. This serves kind of a “d-ochuebclek” for the translator at th-eediptionsgt and proof-reading stage.
We pay attention to main parts of speech (verb, noun, and pronoun, adjective) and/or their derivatives.
ST B k her - 8; and - 7; the - 6; was - 6; c lo 1
TT
EDA key words of ST in the chosen subtext: nouns – mother, night, world; verbs – wanted, woke.
EDA key words of TT in the chosen subtext: nouns – матір, серце, очі, ніч (вно,чсів)іт; verbs – виючи.
The discrepancies become obvious; they, however, kindle the avid interest. With a close look, the following translator’s decisions can be noticed:
For ST lexemes wailing and (nocturnal) howl the translator offered one correspondence виючи. Applying componential analysis, it becomes obvious that the English lexeme to wail has a sememe weeping with corresponding ридання, while the ST lexeme howl is rendered with its direct Ukrainian correspondence вити. The ST fragment Sometimes, in the night, i woke up wailing might have easily be transformed with the relevant lexeme Іноді я прокидався вночі, ридаючи. This alternative choice would not have distorted EDA sequence or “communicative dynamics” (in M. Baker’s terms), vice versa, it would havuecedprmodore diverse utterances in the exclusivity coefficient of the TT.
The key word generator does not recognize word phrases such as heart-piercing as separate meaningful units. Therefore, it was omitted from the automatic calculation. The translator’s choice i preconditioned by the allomorphic feature of English and Ukrainian, i.e. the absence of Participle constructions in the Ukrainian language. Thus, to sound fluent and natural, the ST Participial construction a million other heart-piercing sights could not have been translated other way than мільйона інших деталей, що розривали мені серце.Translator’s choice is well justified.
The ST phrase to look at her directly was reproduced by means of подивитися на неї очі в очі (back translation: to look at her eyes to eyes). This choice re-projected EDA verbalization means more expressive in the translation than intended in the original of the chosen subtext. If frequent, such instances with additional translation markedness might serve a fruitful platform for researching translator’s individual style and manner of translation.
In the analyzed excerpt, world is a counterpart of EDA since it is metaphorized via psychoemotional pangs of pain, e.g. I knew I couldn’t turn around, that to look at her directly was to violate the laws of her world and mine (i.e. the worlds of dead and alive) and world where there was nothing so unusual in a nocturnal howl of pain. This is well-preserved, partially accentuated, я знав, що подивитися на неї очі в очі було б порушенням законів її світу і мого світу and ніби прийшов сюди зі світу, в якому прокидатися вночі, виючи від болю, не вважалося чимось незвичним correspondingly.
Despite the small number illustrated, these few samples show that NLP tools are profound resources for Translation Studies analysis. Along with linguistics-oriented approaches, they are useful at different stages of translation process:
(a) pre-translation ST analysis, original author’s statistical profiling, defining the authors idiostyle via linguomenty and key word generators, key lexical and stylistic features digital markup – along with translator’s expertise and experien–cwe,ill, undoubtedly, navigate the professional translator and add up to the successful translation;
(b) translation process: NLP tools can help to identify dynamic and stable elements in a text; decide on strategies such as explicitation or compensation, what is more important, to be able to balance them; it makes lexical choice easier for the translator; mismatches would become a conscious translator’s decision rather than mistake or negligence.
(c) post-translation: from the viewpoint of translation assessment and critical evaluation, NLP tools, as well as PTC, help to detect the holistic picture, to trace objective and subjective shifts of cohesion and coherence in TT. More emphasis might be placed on transitivity structure and ideational function of the text.
The project has been conducted within the scope of scholarly research topic A“pplication of modern technologies in information optimizing of the processes in natural language” Lvaitv Polytechnic National University.
At the initial stage the project underwent the consultancy of Ihor Kulchytskyy, to whom we express our gratitude.
The text was digitally marked-up by Petro Vus (graduate of Lviv Polytechnic National University).
The process was scrupulously supervised by the authors of the current article.
Results and findings support the idea that quantitative indicators have the potential to demarcate principal vectors of the analysis of EDA dominants and their correspondences in translation. This elaboration gives impetus to talk about the quantitative equality and response with minor fluctuation of corresponding units.
The study analysis was conducted on the material of a corpus of texts, digitally marked up, from the viewpoint of computational and corpus linguistics.
Applied was qualitative analysis, dominant in modern linguistic studies. This allowed us to trace a holistic statistics picture of the author and the translator.
The upcoming results help to draw the discrepancies of parallel profiling; some of the unavoidable (due to the difference in pair language systems), and avoidable (translator’s choices).
The perspective is recognized if proceeding with the comparison of multiple translation reproductions of the same source text writing.