Parallel'nye korpusa xudo estvennyx tekstov: principy sostavlenija i vozmo nosti primenenija v lingvisticheskix i perevodovedcheskix issledovanijax

Abstract

Väitöskirjatutkimuksessa käsitellään rinnakkaistekstikorpuksien keräämistä ja käyttömahdollisuuksia. Tähän tarkoitukseen on koottu venäjä-suomi-rinnakkaiskorpus ParRus, jossa on venäjänkielistä kaunokirjallisuutta ja sen suomennoksia. Korpuksen koko on noin 2,2 miljoonaa sanaa kummallakin kielellä ja sitä varten on kehitetty oma ohjelmapaketti. Erikielisten tekstien kohdentamiseksi (aligning) sekä venäjän- että suomenkielisten sanalistojen lemmatisoimiseksi on jouduttu kehittämään työkalut. Korpuksen tekstien valinnassa hyödynnettiin tekeillä olevaa venäjästä suomennetun kirjallisuuden bibliografiaa, jonka tiedoista muodostettiin tietokanta. Tutkimalla bibliografian sisältämää tietoa tehtiin tarpeelliset luokitukset ja määriteltiin valintakriteerit.

Tutkimuksessa esitellään ParRus-korpuksen ohjelmapakettiin kehitettyjä työkaluja, joiden avulla muodostetaan frekvenssisanalistoja, kollokaatiota, konkordansseja jne. Työssä käsitellään erilaisia tietokonelingvistisiä ongelmia kuten lemmatisointia, tekstien kohdentamista, sanavastineiden hakua ym. ja niiden mahdollisia ja tässä tutkimuksessa toteutettuja ratkaisuja.

ParRus-korpuksen aineistolla on tutkimuksessa myös esimerkinomaisesti suoritettu joitakin kieli- ja käännöstieteellisesti kiinnostavia analyyseja. Käännöskielen ominaispiirteitä haettiin analysoimalla välimerkkien käyttöä, sanastoa ja kielioppia. Anaalyysi osoitti, että lähdetekstin vaikutus näkyy tuloksissa ja käännöskieli poikkeaa siten "standardikielestä". Tutkimuksessa on käsitelty myös ns. kulttuurisesti merkittävien sanojen kääntämisen ongelmaa. Rinnakkaiskorpuksen käännökset osoittavat ainakin sen, että maailmankuvan ongelmat eivät estä käännösprosessia. Korpuksesta saatujen käännösekvivalenttien ja sanakirjoista poimittujen sanavastineiden vertailu taas osoittaa sanakirjan vaikuttavan kääntäjään.

The aim of the dissertation was to compile a Russian-Finnish parallel corpus of literary texts and to study possible applications of its data in linguistics and translation studies. The ParRus corpus includes classical as well as modern Russian literary texts and their translations into Finnish. The size of the Corpus is about 2.2 million running words in each language. Full texts rather than samples were collected, which is important for the use of the Corpus in translation studies and humanities. Different authors, translators, genres, and chronological periods are represented. To some extent, the standard principles of linguistic text corpora compilation were revised.

A software package COCOS-P was developed to study the texts of the Corpus. It includes a paragraph-level aligner, Russian and Finnish lemmatizers, a parallel concordancer, a collocation list builder, etc. Furthermore, a special tool for searching possible translation equivalents was designed. The Corpus is stored in a Microsoft Access database and the software is written in Microsoft Visual Basic programming language. A number of case studies of ParRus texts was carried out. It was checked whether translations really tend to be longer than source texts, as suggested by Eugene Nida, applying methods of the theory of information. The research shows that such a tendency really does exist.

Another study topic was whether the language of translations differs from that of texts originally written in the same language. The language of Finnish translations from Russian was compared to the language of the texts of the subcorpus of original literary Finnish texts from The Savonlinna Corpus of Translated Finnish. Differences in lexicon, use of certain grammatical constructions and punctuation were studied, using statistical methods. In most cases, the differences found can be explained by the influence of the source language, which implies that the language of translations from different source languages is different from that of original texts.

Finally, Finnish translation equivalents for some Russian words were studied. The data from both parallel texts and bilingual dictionaries was considered. Equivalents used by translators were often different from those suggested by dictionaries. However, certain influence of bilingual dictionaries on translators can also be traced. Sometimes translators use old-fashioned and even erroneous equivalents; e.g. ylioppilas high school graduate is used as an equivalent for student university student more often than opiskelija student , which would serve the purpose much better.

Parallel corpora of literary texts seem to be a valuable source of data for studying culturally-bound words. Some Russian words, which are believed to be important for the Russian culture and Russian way of thinking, were studied on the Corpus material. It was discovered that, although such words sometimes are difficult to translate, it is still the text and not a single word that is translated, and, in most cases, the translator can find a solution on the context level. The core of the problem is whether the translator understands the text correctly.

The results of the research presented in the dissertation are by no means final. The work on COCOS-P software will continue. There are certain search and query functions, which would be useful to develop; e.g. part-of-speech and grammar-form tagging would be very important for various research tasks, and sentence-level aligning of the texts would have produced more accurate results. Development of the software will help to carry out a more consistent analysis of the Corpus. Nevertheless, most of the results achieved in this first stage of research would have been impossible or very difficult to obtain using traditional methods.