Purpose and tasks

The main goal of master's work is to improve the efficiency of testing systems. The development component semantic models of natural language answers analyze is the basis for this.

To achieve this purpose it is necessary to solve the following tasks:

- An analyze the semantic analysis techniques of sentences and natural language text models;

- Analyze the principles of natural-language communication in testing systems;

- To develop a component model questions and answers;

- To develop algorithms and software for knowledge testing system in a given subject area;

- To conduct an experimental test of results.

The relevance of the research topic

At present, the testing system based on modifications of the choice the correct answer from the proposed set. This is a significant limitation of testing systems. On the one hand, we can guess the answer, on the other hand, the amount of tested knowledge is limited. Entry the answer would avoid guessing, increase flexibility and diversity of questions and the objectively of the answers estimate. Therefore, the use of natural language text analysis in testing systems is important.

The scientific novelty

-   The improving method of the component analysis model is proposed.

-   An algorithm that carries out a non direct search for meaning in the text is proposed.

-   The estimation criterion of the reliability natural language answers is proposed.

Expected practical results

The main expected results is development of a new semantic model for the analyze of natural language utterance, which can be used in knowledge testing systems.

This model will be implemented as a software package and used by the cathedra, "Artificial Intelligence Systems" in Institute of Informatics and Artificial Intelligence Donetsk National Technical University to test student’s knowledge on disciplines that are reading by the cathedra. The program will analyze natural-language answers to questions that need to provide a definition of the discipline term, and to assess the extent of their compliance with the correct answer.

Interaction with the model will be implemented through a user interface that displays the knowledge level of the testing user.

A review of research and development on the subject. Global level

Currently, the development of many models of linguistic analyzer. These models are able to perform the analysis of natural language text, to determine the meaning and generate statements, within certain limits. But this approach of modeling the communication process is very diverse. The main difference between these approaches lie in the methods of the component implementation understanding  the meaning, analysis tools, as well as the extent and methods of knowledge representation.

Currently, can highlight such models of selection and presentation of meaning: a component analysis, a network of conceptualizations, meaning identification of the pattern, an integrated approach (look draw 1).

.

Draw 1 – models of linguistic analyzer.

Component approach is based on the premise that the semantics of natural languages ​​can be expressed in terms of a semantic factor set (atoms of the meaning). In reviewing the words stand out features (animate, inanimate and so on), which divide the word into separate groups. The best known representative of this trend is Charles Fillmore [3].

The basis of the semantic representation of the model of "conceptual dependency" (R. Schenk [4]) is a network of conceptualizations. The network conceptualizations are kvazigraf similar marked up oriented graph, in which, except for binary relations, present ternary and kvarnarnye, and the arcs are connected not only to the top, but the other arc.

In the model identification of the pattern meaning (Wilks), the text is characterized by the following entities: the meaning of words, messages, text fragments, and semantic interoperability. For each word, which is part of a piece of text can be selected by one of the meanings of words, whereby ambiguity is removed.

Model of an integrated approach, (I. Mel'chuk [12]) is a multi-layered transformer meanings in the text and vice versa. There are four basic levels: phonetic, morphological, syntactic and problematic. Each of them, except the problematic is divided into two other levels - superficial and deep.

A review of research and development on the subject. National level

Among the works of Ukrainian scientists in this field, an important contribution was made Svyatogor L. [11] and Gladun V. [8]. In their paper offers an expanded interpretation of "natural language text." In addition, this work contains a complete scheme of development of the semantic resources of language due to "computer meaning" and the dialogue. In this article means of achieving this purpose in the process of semantic word processing is presented. Word Processing is performed by using a three-level ontology to extract from the text the ontological sense. It is also proposed putting additional feedback to refine the content through dialogue.

A summary of the results of their own

Further development was chosen model-case Charles Fillmore [3]. A deep review of the model, allowed to come to the conclusion that it can be expanded and applied to complex in natural language sentences. These proposals are answers to questions-definitions in the knowledge testing systems. This model remain valid only within a certain domain. This limitation ensures the absence of ambiguity in obtaining the semantic content of the text answers in testing systems.

The ideas of the research included in proceeding of the international scientific-technical conference of students and young scientists 16 April 2012, and the international scientific-practical conference of young scientists April 21 2012, Donetsk.

Conclusion

In research of existing semantic analysis models of natural language text, highlighted the following ways to get meaning: component analysis, a network of conceptualizations, identification meaning on the pattern, an integrated approach. To date, these models are only able to partially extract knowledge from a given text and to construct correct sentences of natural language for given meaning; to paraphrase these sentences, evaluate them in terms of connectivity and perform other tasks. However, sphere of their application is very narrow.

The use of such models in the knowledge testing systems was showed that the basis for the analysis of the text is splitting it into separate fixed lexical units. At the same time, these units have a formative elements, which complicates the analysis. Identification of these lexical units in the text and handling them is the basis for text analyze.

The application component model for the implementation of knowledge testing has the greatest prospects. Despite the limitations, this model can be improved. This will allow to process complex sentences that are answer-definitions in natural language.

The disadvantage is the limitation of testing a subject area and the types of questions. This restriction is necessary for eliminate ambiguity, the extraction of meaning from the answers.

References

1.     Manning C.D. Foundation of ststustical nature language processing / C.D. Manning. – 1992. – Vol.12, N 4. – P.89-94.

2.     Воропаев А.С. Эволюция вычислительных машин [Электронный ресурс]. – Режим доступа: http://www.snkey.net/books/samxp/ch1-1.html

3.     Filmor C. Frames and the semantics of understanding / C. Filmor // Quaderni di semantica.Vol 4 / 2 December 1985 – p. 222-254

4.     Shank R. Conceptual information processing / R. Shank – Norh-Holland Publishing Company, 1988 – p. 237.

5.     Святогор Л. Семантический анализ текстов естесственного языка: цели и средства / Л. Святогор, В. Гладун // XV th International Conference “Knowledge-Dialogue-Solution” KDS-2 2009, Киев, Украина, Октябрь, 2009.

6.    Палагин А. К анализу естественно-языковых объектов / А. Палагин, С. Крывый, В. Величко, Н. Петренко // International Book Series, Number 9. Intelligent Processing. Supplement to the International Journal “Information Technologies & Knowledge” Volume 3 / 2009. – ITHEA, Sofia, 2009. – p.36.

7.     Кибрик А.Е. Семантическая проблематика гетерологического кодирования / А.Е. Кибрик – М.: Наука 1965. – С. 67-83

8.  Гладун В. Структурирование онтологии ассоциаций для конспектирования естественно–языковых текстов / В. Гладун, В. Величко, Л. Святогор // International Book Series, Number 2. Advanced Research in Artificial Intelligence. Supplement to the International Journal “Information Technologies & Knowledge” Volume 2 / 2008. – ITHEA, Sofia, 2008. – p. 153.

9.     Поспелов Г.С. Искусственный интеллект – основа новой информационной технологии / Г.С. Поспелов – М.: Наука, 1988. – 279 с.

10.  Гаврилова Т.А. Базы знаний интеллектуальных систем / Т.А. Гаврилова, В.Ф. Хорошевский – СПб.: Питер, 2001. – 384 с.

11.  Святогор Л. Определение понятия «Смысл» через онтологию. Семантический анализ текстов естественного языка. / Л. Святогор, В. Гладун // International Book Series, Number 9. Intelligent Processing. Supplement to the International Journal “Information Technologies & Knowledge” Volume 3 / 2009. – ITHEA, Sofia, 2009. – p.53.

12. Мельчук И.А. Опыт теории лингвистических моделей «Смысл – Текст» / И.А. Мельчук – М.: Школа «Языки русской культуры», 1999. – 346 с.

13. Лукьяненко С.А., Бессонов А.В., Казакова Е.И. Моделирование семантики естественно языковых высказываний в автоматизированных информационных системах [Электронный ресурс]. – Режим доступа: http://masters.donntu.ru/2006/fema/lukyanenko/library/art03.htm