Machine translation, or rather computer translation, is also a written translation, since as a result we receive a written text. However, it is not carried out by a translator, but by a special computer program. Modern computer translation programs are quite advanced, but they still cannot solve the most difficult task of the translation process: the choice of a contextually necessary option, which in each text is determined by many reasons. Currently, the result of this type of translation can be used as a draft version of the future text, which will be edited by the translator, and also as a means to get a general idea of the topic and content of the text in extreme situations where there is no translator.
An even more difficult task is translating spoken text using computer programs, since the problem of recognizing spoken speech is only at the initial stage of its solution. Until now, an insurmountable obstacle is the individual coloring of the sound of a speech segment - in any language such speech is poorly formalized.
Preliminary editing of the syntactic structure may include:
· splitting an extra-long sentence (more than 40 words) into several shorter ones, adding (if necessary) connective elements;
· introduction to English text articles where necessary or grammatically justified;
· repetition of elements in the coordinating connection of phrases in a sentence;
· introducing conjunctions when using non-union connections between sentences;
· eliminating constructions in brackets in the middle of a noun phrase or in the middle of a sentence;
· replacing occasional abbreviations with full names or introducing special characters that do not allow their translation;
· eliminating lexical and logical ellipses, informal constructions and metaphors;
· bringing to a single form constructions or complex words that may appear in the text in continuous, hyphenated and free writing.
The manually edited text is then automatically processed in the MP system.
25. General scheme of machine translation.
All over the world, the use of machine translation systems, despite all their weaknesses, has long been an element of the professional work of a translator, who must be able to use a computer not only as a typewriter. The concept of an automated translator's workstation, including a complex of resident dictionaries, thesauruses, spell-checking systems, systems for accessing information over various data networks, should become common knowledge for a philologist.
A machine translation (MT) system for texts can be used as part of such an automated translator workstation, providing high-quality translation that is strictly focused on a specific subject area, user tasks and type of documentation. In addition, such a system can help a user who does not know foreign language, very quickly and at low cost, obtain an approximate (rough) translation of texts in the field of knowledge of interest, a translation sufficient to understand the information conveyed by text in a foreign language.
General requirements for practical systems
machine translation (MT)
· System stability. The MP system should produce a result that can be used even in the case of defects in the source material and incomplete vocabulary.
· Replication of the system. The system should have fairly simple software and linguistic tools to expand the scope of its application.
· System adaptability. The MP system must have the means to customize it to the needs of specific users and the characteristics of the documents being processed.
· Optimal timing parameters. The speed of text translation must correspond either to the volume of information received per unit of time, or to the user’s work standards.
· User comfort. The service facilities of the system must ensure user convenience in all operating modes possible in the system.
When working with a specific machine translation system, you need to remember that translation is carried out at several subordinate levels of the system implementation.
In general, these levels include:
· level of automatic text pre-editing;
· level of lexical-morphological analysis;
· level of contextual and group analysis;
· level of analysis of functional segments;
· level of proposal analysis;
· level of synthesis of the output text;
· level of automatic post-editing.
Introduction
Machine translate- the process of translating texts (written, and ideally oral) from one natural language to another using a special computer program. This is also the name of the direction of scientific research related to the construction of such systems.
Instead of "machine" the word automatic is sometimes used, which does not affect the meaning. However, you should not confuse machine translation with automated translation; it has a completely different meaning - with it, the program simply helps a person translate texts.
Thought of using electronic computing machines(computer) for translation was proposed in 1947 in the USA, immediately after the appearance of the first computers. The first public demonstration of machine translation took place in 1954. Despite the primitiveness of the system, this experiment received wide resonance.
By the mid-1960s, two Russian-English translation systems were provided for practical use in the United States:
- MARK
- GAT
However, the ALPAC commission, created to evaluate such systems, came to the conclusion that due to the low quality of machine-translated texts, this activity is unprofitable in the United States. Although the commission recommended continuing and deepening theoretical developments, in general its conclusions led to an increase in pessimism, a decrease in funding, and often to a complete cessation of work on this topic.
However, research continued in a number of countries, aided by continued progress. computer technology. A particularly significant factor was the emergence of mini- and personal computers, and with them increasingly complex vocabulary, search engines oriented towards working with natural language data. The need for translation itself also grew due to the growth of international relations. All this led to a new rise in this area. The time has come for widespread practical use of translation systems, and a market for commercial developments on this topic has emerged.
However, high-quality translation of texts on a wide range of topics is still unattainable. However, there is no doubt that the translator’s work will be faster when using machine translation systems.
1. Main part
Machine translation systems fall into three categories:
- -systems based on grammatical rules(Rule-Based Machine Translation, RBMT),
- -statistical systems(Statistical Machine Translation, SMT)
- -hybrid systems, combining the advantages of both (are the most promising)
Rule-based machine translation is a general term that refers to machine translation systems based on linguistic information about the source and target languages. They consist of bilingual dictionaries and grammars covering the basic semantic, morphological, syntactic patterns of each language. This approach to machine translation is also called classic. Based on this data, the source text is sequentially, sentence by sentence, converted into the target text. The operating principle of such systems is the connection between the structure of the input and output sentences. computer translation
RBMT systems are divided into three groups:
- · word-by-word translation systems;
- · transfer systems (Transfer) - transform the structures of the input language into grammatical structures of the output language;
- · interlinguistic systems (Interlingua) - an intermediate language for describing meaning.
The main advantage of transfer-based systems is the high completeness of text coverage with an acceptable level of translation quality, as well as the low level of costs for initial development and modernization.
Components of a typical RBMT:
- · Linguistic databases: - bilingual dictionaries; - files of names, transliteration; - morphological tables.
- · Translation module: - grammar rules; - translation algorithms.
- · Advantages of RBMT systems:
- - syntactic and morphological accuracy;
- - stability and predictability of the result;
- - ability to customize to the subject area.
- · Disadvantages of RBMT systems:
- - labor intensity and duration of development;
- -the need to maintain and update linguistic databases;
- - "machine accent" when translating.
Statistical machine translation- a type of machine translation, where the translation is generated on the basis of statistical models, the parameters of which are derived from the analysis of bilingual text corpora (text corpora).
Statistical machine translation is contrasted with rule-based machine translation systems, Rule-Based Machine Translation (RBMT), and example-based machine translation, Example-Based MT (EBMT).
The first ideas for statistical machine translation were published by Warren Weaver in 1949. "Second wave" - early 1990s, IBM. "The Third Wave" - Google, Microsoft, Language Weaver, Yandex.
Statistical translation models:
- · according to words (Word-based translation - WBT)
- · by phrases (Phrase-based translation - PBT)
- · by syntax (Syntax-based translation - SBT)
- · by hierarchical phrases (Hierarchical phrase-based translation - HPBT)
Advantages of SMT:
- · Quick setup
- ·Easy to add new translation directions
- · Smoothness of translation
Disadvantages of SMT:
- · "Shortage" of parallel buildings
- · Numerous grammatical errors
- · Instability of translation
To improve quality, developers of machine translation systems introduce some “end-to-end” rules, thereby turning purely statistical systems into Hybrid machine translation. Adding some rules, that is, creating hybrid systems, somewhat improves the quality of translations, especially if the amount of input data used to build the machine translator index is insufficient.
Hybrid machine translation- integration of different machine translation approaches from possible options MP:
- · Rule-based machine translation (RBMT) - Machine translation based on rules.
- · Corpus-based machine translation (CBMT) - Machine translation using text corpora.
- · Example-based machine translation (EBMT) Machine translation using examples.
- · Statistical machine translation (SMT) - Statistical machine translation.
It is expected that with the help of a hybrid architecture it will be possible to combine the advantages of these approaches.
Hybrid translation technology involves the use of statistical methods to construct dictionary databases automatically based on parallel corpora, the formation of several possible translations, both at the lexical level and at the level of the syntactic structure of the target language sentence, the use of post-editing in automatic mode and the selection of the best (most likely) possible translations based on a language model built from a specific corpus of the target language.
Statistical MT seeks to use linguistic data, while systems with a “classical” rule-based approach use statistical methods. Adding some “end-to-end” rules, that is, creating hybrid systems, somewhat improves the quality of translations, especially when the amount of input data used to build index files for storing linguistic information of a machine translator based on N-grams is insufficient.
Architecture of Hybrid Technology "SMT and RBMT"[
The RBMT system is supplemented with two components: a statistical post-editing module and a language models module. Statistical post-editing allows you to smooth out the RB translation, bringing it closer to natural language while maintaining a clear structure of the synthesized text. Language models are used to evaluate smoothness and grammatical correctness translation options generated by the hybrid system.
Typical HMT architecture:
- · Parallel body;
- · Education;
- · Language model;
- · Data for post-editing;
- · Synthesis rules;
- · Dictionary of terminology.
Advantages of hybrid translation:
- · Fast automatic configuration based on the customer's Translation Memories;
- · Terminological accuracy of translation, as well as unity of style;
- · Obtaining additional useful data - a bilingual terminological dictionary.
Conclusion
The main advantage of machine translation is that it allows you to quickly cope with very large volumes of text and therefore sometimes turns out to be more cost-effective than manual translation. It should be remembered that the quality of machine translation will always be inferior to human translation. Therefore, it is advisable to use it only in certain cases.
Many types of materials are not intended for machine translation in principle. Thus, you cannot trust a machine with texts where inaccurate translation may jeopardize human health, the performance of a complex device, or a large contract - the time saved here does not justify the risk. Any documents that imply legal liability require human control. Machine translation is not suitable for marketing materials, where the text is essentially re-interpreted in a new cultural context and created anew.
Acceptable quality can be expected when translating strictly formalized technical texts, while literary and advertising texts cannot be translated by machine.
When resorting to machine translation, it is important not only to clearly understand the desired result and understand the limitations of this method, but also to take into account one more factor. MP systems usually require complex individual configuration and modification, including “training” on a specific topic - without this they show much worse results. In this regard, it makes sense to use machine translation only if you have to translate huge volumes of similar texts. In this case, it will be economically feasible to spend some time training the system, then apply machine translation and obtain the output text suitable for post-editing. If we are talking about several dozen pages, trying to implement machine translation is pointless and simply unprofitable.
Thus, machine translation with post-editing can be really beneficial if the right type of texts are translated in very large volumes. Since large volumes of translations go through translation companies, which often specialize in specific subject areas, the introduction of fairly effective, but expensive machine translation systems of the latest generation is economically justified in such companies: neither content providers, even large ones, nor individual translators can effectively independently use machine translation.
Used Books
- 1. http://www.logrus.ru
- 2. http://www.moluch.ru/
- 3. https://www.academia.edu
- 4. http://study-english.info/
58. Retsker Ya.I. On regular correspondences when translating into native language// Theory and methodology of educational translation. – M., 1950. S.
59. Retsker Ya.I. Translation theory and translation practice. – M., 1974
60. Semenova M.Yu. Basics of text translation. – M., 2009
61. Strelkovsky G.M. Theory and practice of military translation. – M., 1979
62. Ter-Minasova S. G. Language and intercultural communication. – M.. 2000
63. Tyulenev S.V. Translation theory. - M.: Gardariki, 2004
64. Fedorov A.V. Basics general theory translation. – St. Petersburg, 2002
65. Florin S., Vlahov S. Untranslatable in translation. – M., 1980
66. Khaleeva I.I. Fundamentals of the theory of learning to understand foreign language speech (training of translators). – M., 1989
67. Chomsky N. Language and thinking. – M., 1972
68. Chernov G.V. Theory and practice of simultaneous translation. – M., 1978
69. Chuzhakin A., Palazhchenko P. The world of translation, or the Eternal search for mutual understanding. M., 1997 - 2004
70. Chukovsky K.I. High art. – M., 1968.
71. Schweitzer A.D. Translation theory (status, problems, aspects). M., 1998
72. Schweitzer A. D. Translation and linguistics. – M., 1973.
73. Shiryaev A.F. Translation as an object of complex scientific study // Linguistic problems of translation. – M., 1981.
74. Shiryaev A.F. A guide to simultaneous translation. – M., 1982.
75. Shiryaev A.F. Translation theory. Status. Problems, aspects. M., 1988
76. Shadrin V.I. Theory soon in modern translation studies // Materials of the XXVІІІ interuniversity scientific and methodological conference of teachers and graduate students. Vol. 5. Actual problems theory and practice of translation. March 15-22, 1999 St. Petersburg, 1999.
Alys shetelde shykkan adebietter:
1. Bassnett-McGuire S. Translation Studies. - Methuen, L. and N.-Y., 1980.
2. Brislin R.W. (ed). Translation. Application and Research. - N.-Y., 1976.
3. Brower R. (ed). On Translation. - Cambridge (Mass.), 1959.
5. Catford J. A Linguistic Theory of Translation. - L., 1965.
6. Coseriu E. Teoria del lenguaje y lingliistica general. Cinco estudias. - Madrid, 1973.
7. Dolet E. De la manie’re de bien traduire d’une langue en l’autre. – P., 1540
8. Delisle J. L’analyse du discourse comme methode de traduction. – Ottawa, 1984
9. Dryden J. Ovid's Episteles // W.P.Ker (ed.) Essays of John Dryden. - Oxford, 1926.
10. Firth J.R. Linguistic Analysis and Translation // For Roman Jakobson. - The Hague, 1956.
11. Goethe J.W. Drei Stucke vom Übersetzen // H.J.Storig (Hrsg.). Das Problem des Übersetzen. - Stuttgart, 1963.
12. Gutt E.-A. Translation and Relevance. Cognition and Context. - Cambridge (Mass.), 1991.
13. Güttinger F. Zielsprache. Theorie und Technik des Übersetzens. - Zürich, 1963.
14. Halliday M.A.K. Comparison and Translation // M.A.K. Halliday, A. McClntosh, P. Strevens. The Linguistic Sciences and Language Teaching. - L., 1964.
15. Halliday M.A.K. The comparison of languages // A.Mclntosh, M.A.K.
16. Halliday. Patterns of Language. L., 1966.
17. Holz-Mänttäri J. Transtatorisches Handeln. Theorie und Methode. - Helsinki, 1964.
18. Humboldt, W. Einleitung zu “Agamemnon” // H.J. Störig (Hrsg.). Das Problem des Übersetzens. - Stuttgart, 1963.
19. Jäger G. Translation und Translationslinguistik. - Halle (Saale), 1975.
20. Jäger G., Müller D. Kommunikative und maximale Äquivalenz von Texten
// Äquivalenz bei der Translation. - Leipzig, 1982.
21. Jäger G. Die sprachlichen Bedeutungen - das zentrale Problem bei der Translation und ihrer wissenschaftlichen Beschreibung // Bedeuting und Translation. Leipzig, 1986.
22. Jakobson R. On Linguistic Aspects of Translation // R. Brower (ed.). On Translation. - Cambridge (Mass.), 1959.
23. Kade O. Zufall und Gezetzmassigkeit in der Übersetzung. - Leipzig, 1968.
24. Kade O. Die Sprachmittling als gesellschaftliche Erscheinung und Gegenstand wissenschaftlicher Untersuchung. - Leipzig, 1980.
25. Kelletat A.F. Die Ruckschritte der Übersetzungstheorie. - Vaasa, 1986.
26. Koller W. Einfürung in die Übersetzungswissenschaft // 6. Auflage. Wiebelsheim, 2001
27. Krings H.P. Was in den Kopfen von Übersetzern vorgeht. - Tübingen, 1986.
28. Lederer, M. La traduction simultanée. Experiance et théorie. - Paris, 1981.
29. Luther M. Sendbrief vom Dolmetschen // H.J.Störig (Hrsg.). Das Problem des Übersetzens. - Stuttgart, 1963.
30. Lyudskanov A. Prevezhdat chovekt i mashinata. - Sofia, 1957.
31. Mounin S. Les Problèmes théoriques de la traduction. - Paris, 1963.
32. Mounin G. Teoria et storia, della traduzione. - Turin, 1965.
33. Mounin G. Linguistique et traduction. - Bruxelles, 1976.
34. Neubert A. Pragmatische Aspekte der Übersetzung // A. Neubert (Hrsg.).
Grundfragen der Übersetzungswissenschaft. - Leipzig, 1968.
35. Neubert A. Text and Translation. - Leipzig, 1985.
37. Nida E. Principles of Translation as Exemplified by Bible Translating. // R. Broper (ed.) On Translation. - Cambridge (Mass.), 1959.
38. Nida E. Toward a science of translation. - Leiden, 1964
39. Nida E., Taber C.R. The Theory and Practice of Translation. - Leiden, 1964.
40. Nida E, Reyburn W.D. Meaning Across Cultures. - NY., 1976.
41. Quine W. Meaning and Translation. // R. Brower (ed.) On Translation. - Cambridge (Mass.), 1959.
42. Reiss K. Möglichkeiten und Grenzen der Übersetzungskritik. - Munich, 1971.
43. Reiss K.. Vermeer H.J. Grundlegung einer allgermeinen Translationstheorie. Tubingen, 1984.
44. Rose M.D. (ed.). Translation Spectrum. Essays in Theory and Practice. - Albany, 1981.
45. Ross C.D. Translation and Similarity // M.D.Rose (ed.) Translation Spectrum. - Albany, 1981.
46. Savory T. The Art of Translation. - L., 1952.
47. Schleiermacher F. Methoden des Übersetzens // H.J.Störig. Das Problem des Übersetzens. - Stuttgart, 1963.
48. Seleskovich D., Lederer M. Interpreter pour traduire. - Paris., 1987.
49. Snell-Hornby M. Translation Studies. An Integrated Approach. - Amsterdam, Philadelphia, 1988.
50. Sperber D., Wilson D. Relevance: Communication and cognition. - Oxford, 1986.
51. Tirkkonen-Condit S. Textual Criteria in Translation Quality Assessment, - Jyvaskyla, 1982.
52. Toury G. In Search of a Theory of Translation. - TelAviv, 1980.
53. Tytler A.F. Essay on the principles of translation. - L., 1791.
54. Vehmas-Lehto I. Quasi-Correctness. - Helsinki, 1989.
55. Vinay J.-P., Darbelnet J. Stilistique comparée du français et de 1"anglais. - Paris, 1968.
56. Voegelin C.F. Multiple Stage Translation // IJAL. - Vol.20, No.4, 1954.
57. Wilss W. Übersetzungswissenschaft: Probleme und Methoden. - Stuttgart, 1978.
58. Wilss W. Kognition und Übersetzen: Zu Theorie und Praxis der menschlichen und der maschinellen Übersetzung. - Tüsbingen, 1988.
59. Shuttleworth M., Cowie M. Dictionary of Translation Studies. – Manchester, UK: St. Jerome Publishing, 1997. – 234 p.
Paydalangan sozdikter, resmi kuzhattar, okulyktar, audarylgan enbekter:
1. Ayapova Zh.M., Arynov E.M. Isker Adamny Oryssha-Kazakh Economy Tusіndіrme Sozdіgі. A., 1993
2. Egeubay A. Kisilik kitaby. – Almaty, 1998.
3. Ibatov A. Kutbtyn “Khusrau ua Shirin” poem by Masyn Sozdigi (XIV Gasyr). - Almaty, 1974.
4. Ibraeva A. Zan terminderinіn kazaksha-oryssha zhane oryssha-kazaksha kyskasha tussіndіrme sozdіgі. A., 1996
5. Kazakhstan Republic Syn Constitution. A., 1995
6. Kazakhstan Republic Son's Constitution son's body of creation. A., 1996
7. Kaliev G. Tіlіmі terminderің tүsіndіrme сөздігі. A., 2005.
8. Kaliev G., Bolganbaev A. Kazirgі kazakh tіlіnіn lexicology and phraseology. A., 2006
9. Kozhakhmetova Kh.K., T.B. Kazakh-oryssha phraseologylyқ sozdіk. A., 1988
10. Rakhmatullin H. Bank termderin oryssha-kazaksha kyskasha sozdigi. A., 1992
11. Rustemov L.Z. Arab-Iran kirme sozderin kazaksha-oryssha tusindirme sozdіgi. A., 1989
12. Russian-Kazakh proverbs and sayings / Comp. A. Turekhanov and others - A., 1999
13. Sapargaliev G. Zaң terminderіnѣ tүsіndіrme сөздігі. A., 1995
14. Tazkire-i-Bugra Khan - room Serikbay Kosan. – Almaty: Tolagai, 2007.
Paydalangan derekkozder:
1. Abay. Shyғarmalarynyң eki tomdyk tolyk zhinagy. Ekinshi volume. Olender men audarmalar. Poemalar. Kara sozder. A., 1995
2. Altynsarin Ybyray. Eki tomdyk shiғarmalar zhinagay. A., 1988
4. Auezov M. The Path of Abai / Translated by A. Kim. – A., 2007. Book 1.
5. Auezov M. Zhiyrma tomdyk shygarmalar zhinagay. A., vol. XII, 1983, vol. XІX, 1985.
6. Auezov Mukhtar. Abai Zholy: An epic novel. A., 1989.
7. Dzhambul. Selected works. A., 1981
8. Esenberlin I. Almas kylysh (tarihi novel). A., 1971
Approaches to machine translation
Machine translation systems can use a translation method based on linguistic rules. The most suitable words from the source language are simply replaced with words from the target language.
It is often argued that to successfully solve the problem of machine translation, it is necessary to solve the problem of understanding text in natural language.
Typically, a rule-based translation method uses a symbolic representation (an intermediary) from which the text in the target language is created. And if we take into account the nature of the intermediary, we can talk about interlinguistic machine translation or transfer machine translation. These methods require very large dictionaries with morphological, syntactic and semantic information and a large set of rules.
If a machine translation system has sufficient quantity data, then you can get a translation good quality. The main difficulty lies in generating this data. For example, large text corpuses required for statistical translation methods turn out to be insufficient for grammar-based translation. Moreover, for the latter, an additional grammar task is required.
For translate related languages(Russian, Ukrainian) a simple replacement of words may be sufficient.
Modern machine translation systems are divided into three large groups:
· rules-based;
based on examples;
Rule-based EMS
Rule-based machine translation systems are a general term that refers to machine translation systems based on linguistic information about the source and target languages.
They consist of bilingual dictionaries and grammars covering the basic semantic, morphological, syntactic patterns of each language. This approach to machine translation is also called classical.
Based on this data, the source text is sequentially, sentence by sentence, converted into the target text. Often, such systems are contrasted with machine translation systems that are based on examples.
The operating principle of such systems is the connection between the structure of the input and output sentences. The translation is not of particularly good quality. But on simple examples works.
Translation from English to German would look like:
A girl eats an apple. Ein Madchen isst einen Apfel.
These systems are divided into three groups:
· word-by-word translation systems;
· transfer systems;
· interlinguistic;
Word by word translation
Such systems are now used extremely rarely due to the low quality of translation. The words of the source text are converted (as is) into words of the target text. Often such a transformation occurs without lemmatization and morphological analysis. This is the simplest machine translation method. It is used to translate long lists of words (such as directories). It can also be used to compile a subscript for TM systems.
Transfer systems
How transfer systems, and interlinguistic, have the same general idea. To translate, it is necessary to have an intermediary who carries the meaning of the expression being translated. In interlinguistic systems, the intermediary does not depend on the pair of languages, while in transfer systems it does.
Transfer systems operate on a very simple principle: rules are applied to the input text that match the structures of the source and target languages. The initial stage of work includes morphological, syntactic (and sometimes semantic) analysis of the text to create an internal representation. The translation is generated from this representation using bilingual dictionaries and grammatical rules. Sometimes, based on the primary representation that was obtained from the source text, a more “abstract” internal representation is built. This is done in order to emphasize places that are important for translation, and discard unimportant parts of the text. When constructing a translation text, the transformation of the levels of internal representations occurs in the reverse order.
When using this strategy, it turns out quite high quality translations, with an accuracy of around 90% (although this greatly depends on the language pair). The operation of any transfer transfer system consists of at least five parts:
· morphological analysis;
· lexical transfer;
· structural transfer;
· morphological generation.
Morphological analysis. Words in the source text are classified by parts of speech. They are revealed morphological characteristics. Word lemmas are defined.
Lexical categorizations. In any text, some words may have more than one meaning, causing ambiguity in the analysis. Lexical categorization reveals the context of a word. Various kinds of notes and clarifications are possible.
Lexical transfer. Based on a bilingual dictionary, the lemmas of words are translated. The action is very similar to word-by-word translation.
Structural transfer. The words agree in a sentence.
Morphological generation. Based on the output data of the structural transfer, word forms of the translated text are created.
One of the main features of transphenial machine translation systems is the step during which an intermediate representation of the source language text is “transferred” into an intermediate representation of the target language text. This can work at one of two levels of linguistic analysis, or at both.
1. Superficial (syntactic) transfer. This level is characterized by the transfer of “syntactic structures” between the source and target languages. Suitable for languages in the same family or type, for example in Romance languages, between Italian Spanish, Catalan, French, etc.
2. Deep (semantic) transfer. The level is characterized by a semantic representation. It depends on the original language. This representation may consist of a number of structures that represent meaning. Translation also usually requires a structural transfer. This level is used for translation between more distant languages.
Interlinguistic machine translation
Interlinguistic machine translation is one of the classical approaches to machine translation. Original the text is transformed into an abstract representation that is independent of language (unlike transfer translation). The translated text is created based on this representation. The main advantage of this approach is that it allows you to add a new language to the system. It can be proven mathematically that within the framework of this approach, the creation of each new language interpreter for such a system will reduce its cost, compared, for example, with a transfer translation system. In addition, within this approach it is possible
· implement “text retelling”, paraphrasing the source text within one language;
· relatively simple implementation of translation of very different languages, such as Russian and Arabic.
However, there are still no implementations of this approach that would work correctly for at least two languages. Many experts express doubts about the possibility of such implementation. The biggest challenge for creating such systems is designing an interlingual representation. It must be both abstract and independent of specific languages, but at the same time it must reflect the features of any existing language. On the other hand, within the framework of artificial intelligence, the task of identifying the meaning of a text has not yet been solved.
The interlinguistic approach was first proposed in the 17th century by Descartes and Leibniz, who proposed universal dictionaries using numerical codes. Others such as Cave Beck, Athanasius Kircher and Johann Joachim Becher worked to develop an unambiguous universal language based on the principles of logic and iconography.
In 1668, John Wilkins, in his treatise “An Essay on Genuine Symbolism and Philosophical Language,” spoke about his interlingua.
During the 18th and 19th centuries, many universal languages were developed, including Esperanto. It is known that the idea of a universal language for machine translation did not manifest itself in any way at the initial stages of the development of this technology. Instead, only pairs of languages were considered. However, during the 1950s and 60s, researchers in Cambridge led by Margaret Masterman, in Leningrad led by Nikolai Andreev and in Milan by Silvio Ceccato began work in this area.
In the 1970s and 1980s, some progress was made in this area and a number of machine translation systems were built.
In this translation method, interlingual representation can be seen as a way of describing the analysis of a text in the original language. At the same time, the morphological and syntactic characteristics of the text are preserved in the representation. It is assumed that in this way the “meaning” can be conveyed when creating a translated text.
In this case, two interlingual representations are sometimes used. One of them more reflects the characteristics of the source language. The other is the target language. The translation in this case is carried out in two stages.
In some cases, two or more representations of the same level are used (equally close to both languages), but differing in topic. This is necessary to improve the quality of translation of specific texts.
This approach is not new to linguistics. It is based on the idea of the proximity of languages. To improve the quality of translation, natural language is used as a bridge between two other languages. For example, when translating from Ukrainian to English, Russian is sometimes used.
To use the interlinguistic machine translation system you need:
· dictionaries for analysis and generation of texts;
· description of language grammars;
· knowledge base of concepts (to create an interlingual representation);
· rules of concept projection for languages and representation.
The hardest part about creating this type is the inability to build a base for broad areas of knowledge. And those databases that are created for very specific topics have high computational complexity.