J. Antidze, N. Gulua

On One Approach to Automatized Semantic Analysis of Phrases of  a Natural Language

Presented by Cor. Member of the Academy A.Frangishvili

ABSTRACT. The present work describes the approach for developing an algorithm of automatic semantic analysis of phrases, preparing a relevant knowledge base and  using of our instrumental means for realizing of the algorithm by computer.

Key words: semantic analysis, micro-cosmos, micro-theory, software tool, knowledge base, deduction, semantic graphs.

Scientific researches, translating from one language into another, teaching, identifying faces and solving other problems connected with these processes are nowadays in the scope of human competence. However, people use computers more and more to cope with them, and at present only a miserable part of computers abilities is utilized relevantly. The role of natural language in solving these problems is well known, in particular, solving the question of deducing meanings from a text and formalizing them. This problem is considered to be solvable and is known as semantic analysis of a natural language. Little is done for computerized semantic analysis of Georgian texts [1-6]. Complications of semantic analysis are caused by peculiarities of sentences. Let discuss some of them:

1)      A sentence can be true or false. The total of inner corners of a rectangle is equal to 360, whether it is true or false can be proved logically, but the truth of the sentence which describes some historical event can not be proved logically. It must be accepted by historians as a fact or denied by them as such.

2)      A sentence can be universal or typical, e.g. All humans are mortal is universal, but All men like alcohol is not universal, but is typical

3)      A sentence can be trustworthy or unbelievable.

4)      A sentence can be rare or approximately trust-worthy.

5)      A sentence can be homonymous in sense of its meaning.

The list of such examples can be continued [7], but even the above maintained examples are enough to make sure how complicated the nature of a sentence meaning can be and that it can not be characterized by its truth or falsity only. For some sentences discussion of their truth or falsity is useless. Hence, to define the meaning of a sentence it is necessary to discuss its constituent word. Possible meanings of each word should be studied, word interactions within a sentence should be considered and all these will make possible to deduce the meaning of a sentence. It is essential to know the real world, in which the natural language is used. The real world is described by means of knowledge bases, where for imagining real or possible world the semantic networks are used. Shank’s and Shapiro’s graphical imaginations have been especially widely used and they are known in scientific literature as conceptual graphs[8-9]. Equally widely is used Schmidt’s graphical imagination for presenting meaning of conversation and is known as DRS [10].  Our aim is to define semantic meaning of a whole sentence. To achieve this we should know semantic characterization of certain words with the list of their semantic peculiarities, and with meanings of certain peculiarities. At the same knowledge of the universe, which is hidden in the semantic meaning, at this sentence should be taken into consideration. Words syntactically connected in a sentence should be semantically limited, which is expressed by semantic features of words and by their meanings. By means of them correct semantic relation between syntactically related words can be defined. The same can be said on more complicated syntactic connections, such as syntactic connections between noun groups and verb groups. Limitations themselves represent logical images built on semantic features and their meanings. After this the formalism, which was used by us for analyzing Georgian sentences morphologically and syntactically, can be utilized for semantic analysis, as well [11,12]. To achieve this, it is important to define correctly semantic features of certain word classes. Concepts of micro-cosmos and micro-theory are used by us with this purpose. They are utilized for transforming a text in a natural language into TMR language [13]. According to this theory of micro-cosmos consists of micro theories. Micro theory is compiled for certain word classes. To illustrate better how such a micro theory is compiled, let discuss it for adjectives. It is based first of all on syntax and semantics of adjectives, which is widely accepted in literature. It is based on the following principles:

1.      Attributive adjectives define nouns they are related to,

2.       Not all adjective define nouns, especially if they are used as predicative attributive and predicative usage of adjectives is one of those features, by which adjectives are connected with: attributively / productively; modification by means of adverbs; statically / dynamically; gradable  / non-gradable ; inheritability / non- inheritability.

3.       Taxonomy of adjectives

4.      Predicative / non-predicative adjectives.

5.      Relative adjectives.

6.      Relative qualitative adjectives.

7.      Series of adjectives.

8.      Comparative degrees of qualitative adjectives, based on ontology and lexicology is important too. Ontological approach presupposes characterizing each word according to the mentioned classification with presenting corresponding features and their meanings [14]. In our case, features and relevant meanings are written with a word in a dictionary. Such features and their meanings are used to compile limitations by means of which semantic meaning of a word or a whole syntactic construction will be defined. Micro-theory for other word classes (nouns, verbs and other word classes) must be built in a similar way. Such a presentation enables to use instrumental means for semantic analysis created by as without any modification of these instruments. It enable compilers of algorithms of semantic analysis to use instrumental means created by us to check their algorithm and correct it without programming of the algorithm.

I.Javakhishvili Tbilisi State University

References:

1. N. Gulua. Formalized description of Georgian texts, the software for texts processing and its application for construction of a teaching system, PHD thesis, Tbilisi, 1999.

2. J.Antidze, N. Gulua, On selection of Georgian texts computer analysis formalism, Communications of the Georgian Academy of Sciences, 162, N2, 2000

3. J.Antidze, N. Gulua. Analysis of Georgian texts for construction of a teaching system, Reports of enlarged session of the seminar of IAM TSU, vol. 13, N4, 1998.

4. N. Gulua. On the construction of Georgian interactive teaching system, Reports of enlarged session of the seminar of IAM TSU, vol.13, N4, 1998

5. J.Antidze, N.Gulua. On the method for construction of a teaching systems. Bulletin of Sukhumi branch of Tbilisi State University, N1, 1998.

6. N.Gulua. On Computer Syntactical Analysis of Georgian Texts, Bulletin of the Georgian Academy of Sciences, N3, 1999.

7. D. Kaiser. La semantique des langues naturelles et les logiques, Annales des Telecommunications, Tome 44, N1-2,1989.

8. Shank, Roger C.. Conceptual Information Processing, Amsterdam, 1975.

9. Shapiro, Stuart C, Rapoport W. The SNePS family, In Semantic Networks in Artificial Intelligence, Oxford, 1992.

10. Kamp Hans, Uwe Reyle. From Discource tu Logic, Dordrecht, 1993.

11. J.Antidze, d. Mishelashvili. Software Tools for Natural language Texts processing, VI international Tbilisi Symposium: Language, Logic and Computation, Batumi, 2005

12. J.Antidze, d.Mishelashvili. Software Tools for morphological and Syntactic Analyses of Natural Language Texts, Proceedings of the conference III _ Georgian Language and Computer Technologies, Tbilisi, 2005.

13. Reale, Stephen, Sergei Nirenburg and Gavi Maheseh. Semantic Analysis in the Micro-cosmos Machine Translation Project, in SNLP_ 95, Bangkok.

14. V. Raskin, S. Nirenburg. Lexical Semantic of Adjectives – A Micro-theory of Adjectival Meaning, Computing Research Laboratory, New Mexico State University, Las Cruces.

Bbunebrivi enis Afrazebis avtomaturi semantikuri analizisadmi erTi midgomis Sesaxeb

reziume

 naSromSi aRwerilia midgoma, Tu rogor Sedges frazebis avtomaturi semantikuri analizis algoriTmi, momzaddes saTanado codnis baza da gamoyenebul iqnes Cvens mier SemoTavazebuli instrumentuli saSualebebi misi kompiuterze realizaciisaTvis.