One of the modern technologies that allows solving the problem of changing the paradigm of education is modular education, because it is based on the positions of an active, active, flexible approach to building the pedagogical process.

In the late 80s - early 90s of the XX century, a new term from the field of technical sciences "bursts" into pedagogical science, namely "module". Module (from lat. modulus-"small measure") - an integral part, separable or at least mentally distinguished from the general. Modular usually called a thing consisting of clearly defined parts, which can often be removed or added without destroying the thing as a whole.

Much has been written and talked about the benefits of modular learning in the education system. Modular learning - a method of organizing the educational process based on the block-modular presentation of educational information.

The essence of modular training is that the content of training is structured into autonomous organizational and methodological blocks - modules, the content and volume of which may vary depending on the didactic goals, profile and level differentiation of students, the desires of students to choose an individual trajectory of movement along the training course. Modules can be mandatory and elective. The modules themselves are formed: as a structural unit of the curriculum in the specialty; as an organizational and methodological interdisciplinary structure, in the form of a set of sections from different disciplines, united by thematic basis; or as an organizational and methodological structural unit within the framework of an academic discipline. A necessary element of modular training is usually a rating system for assessing knowledge, which involves a scoring of students' progress based on the results of studying each module.

In pedagogical science, the module is considered as an important part of the whole system, without the knowledge of which the didactic system does not work. In terms of its content, it is a complete, logically completed block.

Modular education, partially used in schools in England and Sweden, is built according to the rules of modularity, when the design of the educational material ensures that each student achieves the set didactic tasks, has the completeness of the material in the module and the integration of different types and forms of education. The positive effect achieved as a result of such training is associated with its dynamism, which consists in the variability of the content of elements and modules. The goals of this training are formulated in terms of methods of activity and modes of action, divided into cycles of cognition and cycles of other types of activity. Modular learning is distinguished by a problematic approach, a creative attitude of the student to learning. Its flexibility is associated with the differentiation and individualization of learning on the basis of repeatedly repeated diagnostics in order to determine the level of knowledge, needs, and the individual pace of the learner's learning activity.

The guiding principles of modular learning include:

• 1) principles of modularity;
• 2) structuring the content of training into separate elements;
• 3) dynamism;
• 4) activities;
• 5) flexibility;
• 6) conscious perspective;
• 7) versatility of methodological advice and parity.

The principle of modularity presupposes wholeness and completeness, completeness and consistency of building units of educational material in the form of blocks-modules, within which the educational material is structured in the form of a system of educational elements. From blocks-modules, as from elements, a training course on the subject is constructed. The elements inside the block-module are interchangeable and movable. The development of educational material occurs in the process of a completed cycle of educational activities.

It is the module that can act as a training program, individualized in content, teaching methods, the level of independence, the pace of the student's educational and cognitive activity. In the essential characteristics of modular learning lies its difference from other learning systems.

Firstly, the content of training is presented in complete independent complexes (information blocks), the assimilation of which is carried out in accordance with the goal. The didactic goal is formed for the student and contains not only an indication of the volume of the studied content, but also the level of its assimilation. In addition, each student receives written advice from the teacher on how to act more rationally, where to find the necessary educational material.

Secondly, the form of communication between teacher and student is changing. It is carried out through modules and plus personal individual communication. It is the modules that make it possible to transfer learning to the subject - the subjective basis. Relationships become parity, equal between teacher and student.

Thirdly, the student works maximum time independently, learns goal-setting, self-planning, self-organization, self-control and self-esteem.

Yu.B. Kuzmenkova defines a module as a self-contained mini-course and assumes that this course (like any other) is designed in accordance with the given goals, methods of their implementation and verification of the results achieved. According to Yu.B. Kuzmenkova needs to set a limited number of specific tasks that are realistically feasible in a limited period of study time. To do this, it is convenient to compose several short-term and rotating modules in accordance with the marked core (for any program) components - thematic orientation and target orientation - and build them sequentially and / or in parallel within the framework of the general course. One of the features of the proposed approach is the possibility of differentiation, which makes it possible to clearly distinguish between the selected target settings; At the same time, as practice shows, it is quite convenient to subdivide modules into language and speech modules. This differentiation is based on the principle of focusing attention, according to which the development of educational material, especially unfamiliar ones, becomes more effective when differentiating difficulties and removing them one by one. (11, 21-28)

In accordance with this, when compiling language modules, it is supposed to focus on mastering a specific topic (or a specific sequence of topics) related to the study of an aspect of the language - for example, from the corresponding section of phonetics, grammar or vocabulary, while compiling speech modules assumes as the main task is to master any skills and abilities within the framework of one type of RD (or two related ones - receptive and reproductive).

Modern studies of the modular principle of organizing the educational process distinguish three main blocks in the structure of the module: a block containing the material to be studied, practical and control blocks.

L.N. and M.E. The Kuznetsovs propose the following module structure:

target block (didactic goal of studying the module, personality-oriented tasks that realize the goal).

Update block (a list of basic knowledge and methods of action necessary to study the topic, exercises, tests and other independent work for updating

basic knowledge).

Problem block (problem situations that are personally significant for students, emotionally rich material to the topic).

Block funds (equipment and didactic materials (constantly updated), methodological finds).

Theoretical block (a list of subject knowledge and methods of action, systematizing ideas, principles, patterns, methods, generalizing methods of action, presentation of the content and structure of the topic in the form of a support (orientation in the topic).

Application block (a system of multi-level tasks for variable repetition and consolidation).

Generalization block (didactic material for concise generalization, systematization of educational material, reflection).

recess block (training material of increased complexity).

Block "exit" (didactic material for tests, tests, reports at educational conferences, for homework).

Table 2. Possible construction of the language module

The objectives of the training may include the complex development of skills and abilities, and when working on the chosen topic, you can use various types of reading / listening, role-playing games, presentations, writing reports, writing essays, etc. - but without fail with an emphasis on the predominant development of lexical skills. Vocabulary (in the volume regulated by means of a minimum dictionary) is well absorbed in the course of developing communication skills in the chosen field - in conditional communicative situations of various types.

If, however, students do not have enough time devoted in the basic course to studying specific sections of phonetics or grammar, then it is possible to similarly build a module focused on mastering the necessary skills within a narrow topic. For example, "Correction of the pronunciation of German vowels" or "Correction of the use of tense verb forms" (in the latter case, a good example is the transfer of emphasis from training exercises to studying grammar in context). At the same time, it is possible to comprehensively develop speech skills based on already known lexical material. The priority, however, will be focused work with material on grammar or phonetics.

It is possible to compose various narrowly thematic modules, choosing target settings based on table 3 as follows.

Table 3. Development of the necessary knowledge on aspects of the language

It should be noted that the duration and frequency of mini-courses are variables and the module is not necessarily a whole lesson, it can be half a lesson, or some part of it, or a series of thematically organized lessons of a longer duration. When compiling the program, they can be varied and used in different combinations: two different language modules within the program of elective courses can be stretched over the entire first quarter, and lined up in parallel within one lesson (10 minutes - phonetic, the rest of the time grammatical), and the lexical module may start from the second quarter, or otherwise. The choice and sequence of introducing modules is determined by the specific needs of students. It is important that from the point of view of planning, the main requirement is a systematic approach to the organization of classes (their cyclicity, continuity, repetition and, accordingly, regular reporting), otherwise all these modular innovations will turn out to be chaotic.

It is most convenient to develop modules for teaching grammar, because it is grammar that causes the greatest difficulties in learning, it is because of ignorance of grammar that students make numerous mistakes in oral and written speech, so we will dwell on this aspect of the language in more detail.

Grammar is a collection of rules about changing words and combining words in sentences. It makes it possible to clothe human thoughts in a material language shell. Grammar cannot be separated from speech; without grammar, mastery of any form of speech is not conceivable, since it, along with vocabulary and sound composition, is the material basis of speech. A characteristic feature of grammar is that, abstracting from the particular and the concrete, it takes the general that underlies the change of words and their combinations in sentences and builds grammatical rules from it.

From the above it follows that grammar is of great practical importance. Therefore, in the ongoing experiment, a module for teaching grammar will be considered.

Thus, using various modules, it is possible to successfully implement intra-subject and inter-subject communications, integrate educational content, forming it in the logic of the content of the leading academic subject with the need to differentiate educational content.

Liked our site? Your reposts and ratings are the best praise for us!

Medium garrison building that allows you to exchange crafting reagents for garrison resources and vice versa. In addition, it opens access to the factions (Alliance) and (Horde). Allows you to use the auction.

We recommend using the addon. Master plan- this is the best addon for managing the garrison at the moment. The addon can be found and downloaded. For convenient mission management, we recommend using the addon Garrison Mission Manager. You can find and download it.

1. Overview

2. How to open access to the trading shop?

In this section, we will tell you how to get access to the level 1, 2 and 3 shop.

2.1. Trading Post Level 1

The shop of the first level will be available immediately after the construction of the town hall.

2.2. Garrison Blueprint: Trading Post Level 2

Garrison Blueprint: Trading Post, Level 2 can be purchased at level 98 or after building an outpost in Spiers of Arak. It costs 1000g or exchanges for Outpost Establishment Notes. Notes can be obtained twice: the first time - during the quests in Gorgrond, the second - during the quests in the Spiers of Arak, thus saving some gold. Below is the location of merchants for Horde and Alliance.

2.3. Garrison Blueprint: Trading Post Level 3

Garrison Blueprint: Trading Post Level 3 is a reward from the Savage Friends (Alliance) or Savage Friends (Horde) achievement, exalted with 3 out of 5 Draenor factions: (Alliance), (Horde), (Alliance), (Horde) , (Alliance), (Horde). The achievement is shared by all characters on the account.

After completing the achievement, you should not hope that the blueprint will magically appear in your bag. Head to Spartz Boltwhack (Alliance) or Rezlak (Horde) and buy it for 1000g.

3. Exchange

When visiting the shop for the first time, you will receive a quest (Alliance) or A Little Tricks (Horde). After completing the quest, you will gain access to the merchant. He sells Draenor crafting reagents: ore, grass, meat, fish fillet, fur, leather, and enchanting dust. Price for 5 units Reagent ranges from 20 to 50x Garrison Resources (prices are always different for Horde and Alliance).

4. Orders

Daily orders in the Trading Post allow you to exchange reagents for Garrison Resources. Exchange rate - 5 reagents for 20 resources.

5. Auction

When visiting the second level shop for the first time, you can take the quest from the ancient trading mechanism: (Alliance) or (Horde). You will receive 5 parts, each of which must be combined with different items:

• The Arcane Crystal module requires 4 reagents, which can be obtained from mobs and bosses in raids or garrison attacks:
• The auction management module requires 3 reagents, which can be obtained from mobs and bosses in the dungeons:

ETAP-3 is a multifunctional natural language text processing system that has been developed since the 1980s. a group of Russian linguists, mathematicians and programmers at the Institute for Information Transmission Problems of the Russian Academy of Sciences. The ETAP-3 system is based on the theory “Meaning Û Text”, developed by I.A. Melchuk, and the integral theory of language developed by Yu.D. Apresyan.

STAGE-3 is not a commercial development aimed at achieving a specific application goal. Our main task is linguistic modeling of natural language and computer implementation of such models. This explains our desire to build models that are as adequate as possible from a linguistic point of view. Often, extensive linguistic information is entered into the system, regardless of whether it is necessary to improve the efficiency of computer text processing or not. In particular, we strive to obtain linguistically correct syntactic structures for each sentence, not because otherwise the sentence would not be, for example, correctly translated into another language, but simply because this is required by the task of modeling natural language syntax. However, we are convinced that, in the final analysis, the theoretical adequacy and completeness of linguistic information pay off from a purely practical point of view.

All STAGE-3 applications use the original 3-valued logic system and the detailed formal linguistic description language FORET (see Apresyan et al. 1992a, Apresjan et al. 1992b).

2. Stage-3: modules, properties, architecture, implementation

2.1.Modules

The ETAP-3 system contains the following main modules:

• High Quality Machine Translation System
• Module for generating Russian texts based on the Universal Network Language (UNL)
• Natural language interface for databases
• System of synonymic paraphrasing of sentences
• Syntax error corrector
• Computer-assisted language learning system
• Workplace for syntactic markup of text corpus.

Below we will briefly characterize all these modules, and we will dwell on one of them - the UNL module - in more detail.

2.1.1. ETAP-3 machine translation system

The main module of ETAP-3 is a machine translation (MT) system serving five pairs of languages. Translation systems are available for (1) English to Russian, (2) Russian to English, (3) Russian to Korean, (4) Russian to French, and (5) Russian to German.

To date, the first two systems have been developed in the most detail. The translation system from English into Russian and from Russian into English, which can be considered as a single bidirectional module, is designed to translate real texts, mainly scientific and technical topics. The best results were obtained for texts on computer science, electrical engineering, economics, and politics, since the combinatorial dictionaries of the working languages ​​of the system (each containing about 50,000 entries) are mainly focused on the vocabulary of these subject areas. However, ETAP-3 also copes with texts on everyday topics, since recently the dictionaries have been significantly replenished with everyday vocabulary. For each lexeme in the combinatorial dictionary, its syntactic, word-forming, semantic and word-forming features, its control model, as well as information about set phrases with this lexeme are given.

In addition, there is a Russian morphological dictionary (100,000 dictionary entries), which, in addition to purely morphological information, contains basic syntactic information about the lexeme and its approximate translation equivalent. The English morphological dictionary has the same structure (60,000 entries). The system is based on comprehensive grammatical descriptions of the English and Russian languages, compiled by the developers of STAGE-3.

For other pairs of languages, translation systems exist at the level of prototypes.

If STAGE-3 receives a homonymous sentence and the system cannot resolve this homonymy, then several translation options are offered at the output. In all other cases, the system produces one most plausible syntactic structure and one most probable translation. If the user of the system wants to get all possible translations, he can select the appropriate option, and the system will “remember” all cases of unresolved homonymy and give out all possible syntactic structures of the sentence with lexical content acceptable for them. Let's consider one real example. Sentence They made a general remark that... with the option “all translation options” selected, it was translated into Russian in two ways, which differ both in syntactic structures and in the choice of vocabulary: (a) They made a general remark that... and (b) They forced the general to point out that…

2.1.2. Natural language interface for databases

This module of the ETAP-3 system translates queries given in free form in natural language (English or Russian) into SQL query language expressions. The module also translates from SQL to natural language. The module is based on a semantic component developed specifically for this purpose, which translates the deep syntactic structure into a formal semantic representation, from which one can easily switch to a representation in the SQL language.

2.1.3. System of synonymic paraphrasing

This module is designed to conduct linguistic experiments to obtain a variety of synonymous and quasi-synonymous paraphrases of Russian and English sentences. The system is based on the apparatus of lexical functions, one of the most important innovations of the "Meaning Û Text" theory. The result of the synonymous paraphrasing module can be illustrated by the following example:

(1) The director ordered John to write a report – The director gave John an order to write a report – John was ordered by the director to write a report – John received an order from the director to write a report.

This area of ​​linguistic research seems to be very promising, as it can have a wide variety of applications, for example, in teaching native and foreign languages, in authoring systems and text planning systems.

2.1.4. Syntax error corrector

This module is designed for processing texts in Russian. Its purpose is to find and correct various kinds of errors in grammatical agreement, as well as in case management.

2.1.5 Computer language teaching system

This module is a stand-alone software application, namely, a computer game in the form of a dialogue. This program can be used when teaching Russian, English and German as a foreign language. The game is intended for those who have already mastered the language well, but would like to expand their vocabulary, primarily through set phrases and paraphrasing. The system is based on the apparatus of lexical functions. The program can also be successfully used by native speakers of the above languages ​​who want to enrich their vocabulary (for example, journalists, teachers and even politicians).

2.1.6. Workplace for syntactic markup of text corpus.

This newly developed module uses the ETAP-3 dictionaries, as well as the system's morphological and parsing analyzers, to build the first syntactically marked corpus of Russian texts. This is a mixed application: the tree structure resulting from automatic analysis is then edited by a human using convenient graphical tools.

2.2. Main properties of the system

Among the main features of the ETAP-3 system as a whole and its individual modules, the following can be noted:

• Using Rules as the Basic Unit of an Algorithm
• Tiered approach
• Translation through the transfer stage
• Using Dependency Syntax Trees
• Lexicalist approach
• Ability to receive translation options
• Possibility of diverse use of linguistic resources

In the current version of ETAP-3, all modules use only rule-based algorithms. However, in a number of recent experiments, the MT module was supplemented with a component based on access to the translation memory (translation memory) , and a statistical component that semi-automatically extracts translation equivalents from bilingual text corpora (see Iomdin & Streiter 1999).

Like many other natural language text processing systems, ETAP-3 is characterized by a layered approach. In the course of processing, each sentence goes through several stages and at each stage it is presented in the form of a certain structure: 1) morphological, 2) syntactic, and 3) normalized (or deep-syntactic). The actual translation (transfer) is carried out at the level of a normalized syntactic structure, i.e. English normalized structures are converted to corresponding Russian normalized structures and vice versa.

What distinguishes ETAP-3 from most similar systems is the use of syntactic dependency trees to represent the structure of a sentence (most natural language text processing systems around the world use structures of direct constituents).

STAGE-3 is characterized by a lexicalist approach in that the information recorded in the dictionary is considered as important as the information recorded in the grammar. Accordingly, ETAP-3 dictionaries contain significantly more information than dictionaries used in other similar systems. The dictionary entry of ETAP-3 contains, in addition to the name of the lexeme, information about the syntactic and semantic features of the lexeme, its management model, the translation equivalent, various rules, as well as the values ​​of lexical functions, the keyword of which is this lexeme. Syntactic signs words characterize his ability or inability to act in certain syntactic constructions. A word can be assigned several syntactic features from a general list containing more than 200 features. Semantic features needed to check semantic agreement between words in a sentence. Management model the word contains information about the surface expression of the actants of the given word (for example, the word can control one or another preposition or conjunction, or one or another case form of a name). The most important component of the dictionary entry are regulations. All rules in STAGE-3 are distributed between grammar and vocabulary. Grammar rules are more general and apply to broad classes of words, while rules referred to in dictionary entries (directly or by reference) apply to small groups of words or even individual words. Such an organization of rules ensures that the system is automatically configured to process each individual proposal. During the translation process, only those rules are activated, the reference to which is explicitly contained in the dictionary entries of the words contained in the sentence.

As an illustration, we present a fragment of the dictionary entry of the English word chance:

SYNT:COUNT,PREDTO,PREDTHAT

DES:"FACT","ABSTRACT"

D1.1: OF,"PERSON"

D2.1: OF,"FACT"

D2.2: TO2

D2.3: THAT1

SYN1: OPPORTUNITY

MAGN: GOOD1/FAIR1/EXCELLENT

ANTIMAGN: SLIGHT/SLIM/POOR/LITTLE1/SMALL

OPER1: HAVE/STAND1

REAL1-M: TAKE

ANTIREAL1-M: MISS1

INCEPOPER1: GET

FINOPER1: LOSE

CAUSFUNC1: GIVE /GIVE

TRANS: CHANCE/ HAPPENING

R:COMPOS/MODIF/POSSES

1.1 DEP-LEXA(X,Z,PREPOS,BY1)

1 ZAMRUZ:Z(PO1)

2 ZAMRUZ:X(RANDOMNESS)

1 ZAMRUZ:Z(RANDOM)

TRAF:RA-EXPANS.16

TRAF:RA-EXPANS.22

When developing the ETAP-3 system, we tried to build its components in such a way that they could be used for a variety of purposes. In particular, the main grammatical and vocabulary resources of the system are used in all its modules. For example, Russian dictionaries are used at the analysis stage when translating from Russian into English and at the synthesis stage when translating from English into Russian; the same dictionaries are used in the MP module, in the paraphrasing system, in the syntactically marked corpus, etc. Moreover, some of the system resources can be "alienated" from it and, after being refined depending on the requirements of the customer, can be used in various processing systems naturally - language texts.

2.3. General architecture of the ETAP-3 system

To give a general idea of ​​the functioning of the ETAP-3 system, we present the general algorithm of the MP module (Scheme 1). All other modules can, with a certain reservation, be considered as derivatives of this one.

MACHINE TRANSLATION MODULE OF THE STAGE-3 SYSTEM

(ARCHITECTURE)

2.4. Implementation

The ETAP-3 system was implemented on a MicroVax computer (VMS operating system). Recently, new software has been created to work with ETAP-3 on personal computers under Windows NT 4.0, which allows the lexicographer to use a number of additional tools and more efficiently maintain and edit dictionaries.

3. Interface for UNL language

3.1 Background and goals

UNL module is being developed as part of a vast international project with a very ambitious goal: to overcome, at least in part, the language barrier that separates Internet users. Despite the fact that with the advent of the Internet, temporal and spatial barriers between people have practically disappeared, Internet users continue to be separated by a language barrier. This, apparently, is the main obstacle to successful international and interpersonal communication in the information society. The diversity of languages ​​spoken by Internet users has been recognized as one of the pressing problems of mankind. In any case, this is evidenced by the fact that the project, which aims to solve this problem, is carried out under the auspices of the UN and is coordinated by the Institute for Advanced Study at the UN University.

The project was founded in 1996. Currently, 15 universities and research institutes from Brazil, Germany, India, Indonesia, Jordan, Spain, Italy, China, Latvia, Mongolia, Russia, Thailand, France and Japan are participating in the project.

It is expected that teams from other countries will join the project in the coming years, so that ultimately it is planned to cover the official languages ​​of all UN member countries

The idea of ​​the project is as follows. A universal intermediary language is proposed that is powerful enough to express all the essential information that natural language texts convey. This language is the Universal Networking Language (UNL) proposed by H. Uchida (United Nations University). For each natural language, it is proposed to develop two systems: a “deconverter” that would translate texts from the UNL language into a given language, and an “enconverter” that would convert texts in a given language into UNL language expressions. It should be emphasized that UNL text generation will not be fully automatic. This procedure is planned as a dialogue between a computer and a human (editor).

Thus, this project is fundamentally different from traditional machine translation. First of all, the input for generating texts in different natural languages ​​is the UNL structure, the quality of which does not depend on the imperfection of text analysis procedures. During interactive UNL building structure editor will review the results of the automatic enconverter, correct errors and resolve the remaining ambiguity. The editor can then run the deconverter and translate what it has edited UNL expression into your native language to check your work and make further changes to the expression if necessary.

Another important difference of the UNL system from machine translation lies in the fact that expressions in the language UNLs can be generated and stored regardless of the natural languages ​​into which these texts will be translated. UNL can be thought of as a generic way to represent a value. To process UNL text automatically, for example, to index it, search it, or extract information from it, it is not necessary to translate this text into natural language. The latter is necessary only if a person will work with the text.

An enconverter and a deconventor for each natural language form a language server that is planned to be hosted on the Internet. All language servers will be connected to a single UNL network, which will allow the Internet user to translate any document from UNL into his own language, as well as translate into UNL those texts that he wants to make publicly available.

3.2 UNL language

In this article, we will not be able to describe the UNL language in detail, since this topic deserves a separate article, which will probably be written by the creator of the language, Dr. Hiroshi Uchida. We will only dwell on those features of the UNL language that will be important for further presentation. The complete UNL language specification is located at http://www.unl.ias.unu.edu/.

UNL is a computer language designed to represent information in such a way that it would be possible to generate texts containing this information in a wide variety of languages. A UNL expression is a directed hypergraph corresponding to a natural language sentence. Graph arcs denote semantic relationships, for example, agent(actor)object(an object),time(time),place(place),instrument(tool),mode(modus operandi) and others. In the nodes of the graph are the so-called Universal Words (CS) denoting concepts, or groups of CS. Nodes can be provided with attributes. Attributes contain additional information about the use of a node in a given sentence, for example, @imperative, @generic, @future, @obligation.

Each US corresponds to some English word. Some words have semantic delimiters that clarify the meanings of these words. In most cases, delimiters indicate the place of the concept in the knowledge base. This is done in the following way. Universal Word of the Kind A (icl>B) interpreted as ‘A belongs to category B’. For example, US coach without any delimiters has the same meanings as the English word coach generally. Delimiters are used to clarify the meaning of a word. Yes, the expression coach (icl>transport) should be understood as ‘coach as a means of transport, that is, bus; expression coach (icl>human) has the interpretation ‘ coach like a man, that is, trainer, and the expression coach (icl>do)– interpretation ‘ coach as a kind of action’, that is, a verb train. In other words, the apparatus of limiters makes it possible to represent the US as an English word taken in exactly one meaning. In addition, delimiters allow you to introduce concepts for which there are no single-word designations in the English language. For example, in Russian there is an extensive group of verbs of movement, the meaning of which includes an indication of the method or means of movement: fly in, swim, crawl, run and others. There are no one-word English correspondences for the verbs of this group. However, on the basis of English words, it is possible to build an ES that is close to them in meaning, for example, come (met>ship) means ‘to arrive, the vehicle being a ship’.

Let us give an example of an expression in the UNL language corresponding to the English sentence

(2) However, language differences are a barrier to the smooth flow of information in our society.

Each UNL line structure is an expression of the form relation (RS1, RS2). For simplicity, semantic delimiters for universal words are omitted.

aoj( [email protected]@[email protected]@however, [email protected])

mod( [email protected]@[email protected]@however, [email protected])

mod( [email protected], language)

aoj(smooth, [email protected])

mod( [email protected], information)

scn( [email protected], society)

pos(society, we)

3.3. Translation from UNL into Russian in the ETAP-3 system

As already noted in Section 1, ETAP-3 is a transfer system, and the actual translation is carried out at the stage of the normalized syntactic structure (NormSS). At this level, it is most convenient to establish a correspondence between the Russian language and UNL, since the expressions of the UNL language and normalized syntactic structures show many similarities. Here are the most significant ones:

1. Both UNL language expressions and NormSS occupy an intermediate position between the surface and semantic representations of a sentence and approximately correspond to the so-called deep-syntactic level. At this level, the meaning of lexical units is not decomposed into primitives, and the relationships between lexical units are the same for all languages;
2. In both UNL and NormSS expressions, nodes are terminal elements (lexical units) rather than syntactic categories;
3. Nodes contain additional characteristics (attributes);
4. In both UNL and NormSS expressions, arcs are directed dependencies.

At the same time, there are significant differences between UNL and NormSS expressions:

1. In NormSS, all nodes are lexical units, and in UNL, a node can be a subgraph.
2. In NormSS, a node always corresponds to a single word value, and the SS value can be wider or narrower than the value of the corresponding English word:

2.1. The value of the CSS can correspond to several values ​​of one word at once (see above).

2.2. They can match a free phrase (for example, computer-based or high quality).

2.3. They may correspond to some form of the word (for example, the word best is a word form good or well).

2.4. They can denote a concept for which there is no direct equivalent in English.

1. NormSS is the simplest of all connected graphs, namely a tree, while a UNL expression is a hypergraph.
2. In UNL, arcs can form loops and link individual subgraphs.
3. Nodes in NormSS are connected by purely syntactic relations that do not carry any meaning, while relations in UNL denote semantic roles.
4. Attributes in NormSS correspond to grammatical characteristics, while the meaning of many UNL attributes is conveyed by lexical means, both in English and in Russian (for example, modal verbs).
5. NormSS contains information about the order of words in a sentence, and such information is absent in a UNL expression.

NormSS proposal (2) is as follows:

1. 
   Transition from UNL Expression to Intermediate Representation (IR)
2. Transition from PP to Russian NormSS (NormSSR).
3. Synthesis of the Russian proposal for NormSSR.

The first of these steps is the interface between UNL and the ETAP-3 system, and the rest are carried out by standard means of the English-Russian module of the ETAP-3 system.

The translation algorithm from UNL into Russian is shown in Scheme 3.

As follows from the above, the transition from an expression in the UNL language to NormSS should solve the following five tasks:

1. Replace all CLs with English words wherever possible. Russian lexemes will appear at the stage of English-Russian translation when referring to the English dictionary. If no English equivalent has been found for the RS, the value of this RS should be expressed by other means.
2. Translate UNL syntactic relations into ETAP-3 syntactic relations, either directly or by lexical means.
3. Translate UNL language attributes into STAGE-3 grammatical characteristics, either directly or using lexical means.
4. Convert the UNL graph to a dependency tree.
5. Determine the word order in the sentence.

The first and (partly) the second task are solved with the help of UNL dictionaries - English and English combinatorial. For all other tasks, the rules written in the formal-logical FORET language are responsible.

Thus, all these tasks are solved either with the help of dictionaries or with the help of rules. The rules are divided into three classes depending on the degree of universality: GENERAL, Stencil and VOCABULARY rules are distinguished. General rules can be activated when any offer is processed. The other two types of rules apply only if the sentence being processed contains a word that contains a reference to some rule (in the case of a template rule) or the rule itself (in the case of a dictionary rule). Such an organization of rules ensures automatic system tuning: only those rules that are required to process a particular offer are activated.

3.4. Present state of affairs and plans for the future

An experimental version of the UNL to Russian translation module is available at http://proling.iitp.ru/Deco . By the summer of 2000, we plan to make the module available for general use. Our next task is to create an interactive enconverter.

As it is clear from Scheme 3, the interface between UNL and the structures with which the ETAP-3 machine translation module works is carried out at the level of the English NormSS. The same diagram shows that the English translation of the original UNL expressions is a natural by-product of such an architecture. To do this, it is enough to send the English NormSS for synthesis. A number of successful experiments have already been carried out in this direction.

Literature

Yu.D. Apresyan, I.M. Boguslavsky, L.L. Iomdin and etc. (1992 a ). Linguistic processor for complex information systems. Science, 256 p. M.

Ju. D. Apresjan, I. M. Boguslavsky, L. L. Iomdin etal. (1992b). ETAP-2: The Linguistics of a Machine Translation System. // META, Vol. 37, No. 1, pp. 97-112.

Igor Boguslavsky (1995). A bi-directional Russian-to-English machine translation system (ETAP-3). // Proceedings of the Machine Translation Summit V. Luxembourg.

Leonid Iomdin & Oliver Streiter. (1999). Learning from Parallel Corpora: Experiments in Machine Translation. // Dialogue"99: Computational Linguistics and its Applications International Workshop. Tarusa, Russia, June 1999. Vol.2, pp. 79-88.

The study to which this article is devoted was carried out with partial financial support from the Russian Foundation for Basic Research (grant no. 99-06-80277).