Anna Chutkerashvili

Controlled natural languages (CNLs) are engineered languages that are based on natural language, but have their vocabulary, syntax, and/or semantics restricted [1]. The motivation is to have a language that, on one hand, looks as natural as possible and, on the other hand, is simple and unambiguous. Application areas of CNL are quite broad. They serve improving communication and mutual understanding between humans (especially for people with specific reading and understanding restrictions), facilitate manual or automatic translation, are suitable for reasoning, writing technical documents and legal texts, interchanging business rules among organizations, for efficient communication in emergency situations and crisis management, etc. Examples include technical documentations of Boeing and IBM, special English CNLs for communications between ships and harbors, etc. While English dominates the landscape of CNLs ([1] analyzes 100 English CNLs), controlled (sub)languages have been developed for many other languages (e.g., Bulgarian, French, Spanish, Chinese, Russian, German, Greek, Spanish, and Japanese). However, Georgian is not among them. In general, from the computational point of view, despite some recent and ongoing attempts, Georgian is still pretty under-resourced. In this talk we discuss challenges on constructing Georgian controlled language, which aims at establishing clear and unambiguous communication between different parties in emergency situations

Controlled natural languages are used as representative languages of a high level of knowledge. These languages occupy a very important place in the field of computer linguistics because of their two very interesting properties: first, like natural languages, they have a non-formal structure, therefore, they are simpler to use than formal languages. Second, they are precisely defined as subgroups of natural languages and can be translated automatically into a formal target language and processed after that. They can balance the disadvantages of natural languages and formal languages for the most accurate representation of knowledge and can help domain specialists write specifications in a controlled natural language. Controlled Georgian Language for crisis management will be discussed in this talk

In order to mark the information source, Georgian uses reported speech markers such as -metki, -tko and -o [1]. The -metki (c’avida-metki), which derives from the sequence me vtkvi “I said”, has been grammaticalized as a postverbal clitic. It marks the firsthand information reporting quotations in the first person singular: the speaker reports an utterance s/he had already made or reflected on in the past. In some dialects of Georgian (e.g. Kakhetian) metki is attested as an autonomous lexical unit occupying the initial position (head-position) in a phrase “metki gaetrie akedan!”. According to Harrice, the grammaticalization of metki in phrases like “metki gaetrie akedan!” is not completed, Boeder considers the particle metki in “metki gaetrie akedan!” in an intermediate stage between quotative particle and an autonomous verb. In colloquial language metki sometimes occurs twice in a sentence as an autonomous lexical unit (at the beginning) and as an enclitic (at the end of the sentence) in order to intensify the marking of reported speech “metki c’amoiqvane ege+c-metki”). In some contexts, metki as an autonomous marker loses its original meaning of saying, acquiring epistemic overtones of speaker’s reliability like ‘I thought’, ‘I was sure’ “metki saxlšidamxvdeboda”. Such an interpretation is partially caused by the presence of conditional screeve “metki saxlšidamxvdeboda”, however, the expression of an unsuccessful expectation towards metki can be expressed also with the present screeve (dzalian ggavs, metki šeni da+a).

Controlled natural languages (CNLs) are engineered languages that are based on natural language, but have their vocabulary, syntax, and/or semantics restricted [9]. The motivation is to have a language that, on one hand, looks as natural as possible and, on the other hand, is simple and unambiguous. The application area of CNLs is quite broad. There are general-purpose CNLs (e.g., Attempto Common English [6]), whose vocabulary can be adapted to specific areas, and domain-specific ones, which reflect also the domain syntax. Due to space limitations, we just list some interesting practical applications of the domain-specific CNLs: writing technical documentation [3], mathematical authoring [5], crisis management [12, 13, 2], legal texts [4], interchanging business rules among organizations [11], and editing ontologies in OWL. In this talk we present some of the work done. We present some ambiguity cases and propose alternative clearer solutions. We analyze which of the existing ambiguity cases for other languages appear also in Georgian, and which are typical only for it. Being a complex language not closely related to any well-resourced language, developing a CNL for Georgian is an additional challenge. In fact, Georgian has one of the most complex morphologies: e.g., the verb form is very complex – it has agglutinative and inflection morphemes, and many categories (person, number, tense, mood, voice, etc.) which complicates parsing.

During translation, derivative words (derivatives) complicate the situation. The problem is in derivational forms, since the addition of word-forming affixes causes linguistic changes in words. Some affixes are synonymous, others are homonymous, and it is very important to solve this problem in the process of building computer models. It is necessary to pay special attention to phonetic phenomenon and root changes when creating translation algorithms and further programming. To solve this problem, first of all, we created a database of word-building affixes of the Georgian language. It combines morphemes that are native to the modern literary Georgian language or introduced from other languages. For the automatic formation of the corresponding Georgian-English derivational forms, models of word formations of different semantic groups were created for both languages. A database of English derivational affixes was also created. The Georgian language is known for an abundance of morphological forms. For perfect automatic translation, a morphological processor of the Georgian language was connected to the system. Thus, automatic lemmatization of Georgian words is performed, after which algorithms for derivation model recognition and automatic translation are used. The same process, only in reverse, is used for the English derivational form of the word. In both cases, we get one or several (in the case of homonymy in the original form and synonymy in the final form) lemma of the derivational form of the word.

Fundamental scientific researches in Computational Linguistics have been carried out at the Archil Eliashvili Institute of Control Systems, Georgian Technical University, for many years. Various combination methods (lexical functions, synonymous series, semantic roles and superparadigms) were developed for the Georgian language; A computer dictionary of the Georgian language has been created, which, at the same time, carries out functions of a morphological generator, in other words, it produces the full paradigm for each lexical unit. Within the framework of the fundamental issues of language modeling, a means of presenting linguistic algorithms has been developed, that allows the formulation of a bi-directional analysis-combined processor. For some languages, the filling-widening process of dictionaries ​​has been simplified with the help of a grammar compiler, which is the most modern tool for the automatic realization of a formal language model. It is possible to compile morphological processor libraries of individual languages ​​for different variations of any language (according to time, space, origin, genre, etc.) and so on. Automatic machine translation can be considered as the main achievement. To solve this task completely it is necessary to create lexical translator. This type of system is rather valued among the ordinary users, as it makes it easier for them to learn foreign-language texts intensively thus is much more useful while composing text.The strategy of our team is to provide reliable support for future language technologies by the theoretical and practical key issues that have been worked up in separate projects for the past years. The computer products created by our team are used in various linguistic areas. It is a challenge for linguists to create computational model of a language, taking into account its multi detections and changes. We have created main components to compile a national corpus manager.

In order to solve theoretical and applied tasks of Georgian language it is very important to draw out deeply annotated text corpora. While syntactically annotated corpora are now available for English, Czech, Russian and other languages, for Georgian they are rare. The environment, developed by our research group, offers several NLP applications, including a module of morphologic, syntactic and semantic level, a Universal Networking Language interface and a natural language interface to access SQL type databases. In this article, we research the automatic syntactic parser of Georgian Language. It includes syntactic level as well as morphologic level of Georgian language model. The basis of the linguistic model of Georgian text syntax annotation is the dependency grammar.

According to Melchuk, the left side of the “Meaning↔Text” language model should be represented by a set of synonymic utterances and some most simple, “neutral” member of it can be considered as a “dominant” of this set, which represents a semantic core of the set and correspondingly of the “Meaning”. Proceeding from this, the basis of pragmatic-semantic level functioning should be implemented by quasi-synonymic transfigurations, which lead from the dominant to the arbitrary utterance of synonymic set (during synthesis) or in an opposite direction (in the case of analysis), taking into account in the course of these operations the pragmatic-semantic nuances, which make difference between initial and final arguments of each transformation. The possibility and result of these transformations are defined by same special information, which should be added to the morphologic and syntactic data of dictionary entities. We shall mention three kinds of such information. The first of them is represented by the spectrum of “lexical functions”, which can be considered as some version of the lexical paradigm: they define for each lexeme set of content in which it acquires one of semantic nuances, defined by same concrete lexical function. The “Synonymic rows” can be considered as a generalization of one of the lexical functions, particularly, of Syn, which implies the unity of words synonymic to the lexeme of the given dictionary unit. The concept of “synonymic rows” implies more “base” conditions of quasi-synonymy, which at the same time must be followed by description of all features, which define the semantic differences between these quasi-synonyms. The Georgian verbal super-paradigms, in their turn, may be considered as a particular case of quasisynonymic rows: these entities unite all paradigms derived from the one and the same lexeme. The main advantage of such unification is the stability of possibilities and results of transformations between utterances based on the different members of superparadigm. 

Lexical functions LF () may be considered as something similar to the lexical “paradigm” of the corresponding vocabulary unit C0; which includes stable lexical units associated with C0 and connected to it by some meaningful and standard semantic relations. Particularly, the inclusion of lexical functions both in the dictionary units and in the system of a language is that this information often supplies the possibility of correct choice of a word combination in the text: gasca brÂaneba - ordered; gaak’eta gancxadeba - announced, miiγo gadac’qvet’ileba - decided; misca rčeva - advised;… (1). The Oper1 (C0), represented by verbs of (1), cannot be replaced by each other in spite of the semantic likeness of their arguments, and thus, the correct choice of these verbs fully depends on the corresponding C0. On the other hand, the components of lexical functions (LF) may serve as one of the tools for quasi-synonymous transformations.

The paper discusses the meaning and functions of the Georgian lexical items mxolod and mart’o. It shows that both are focus particles which can substitute each other in particular contexts and that mxolod unlike mart’o, can be negated and imply an additive reading. the paper further shows that mart’o can be categorized as an adverb additionally to its categorization as focus particle. As adverb it can function as adverbial or secondary predicate. whereas mxolod and mart’o as focus particles relate to physical, countable items mart’o can denote physical and spiritual properties. It follows from the bi-categorial status of mart’o that mart’o is more productive in word formation than mxolod.  

The paper considers the Georgian particles k’i and -c. these particles are frequently used, separately, as well as together. They have different meanings but both of them can have a focusing function: emphasizing a word or a phrase they are attached to. In spite of having similar or even the same semantic features, the particles k’i and -c cannot substitute each other. One reason for this is that -c is a bound form and k’i is not. they never substitute for each other but they very often occur together and they are much more emphatic when they are used together. The dominating element in building up the meaning of -c k’i is -c, which is stronger in emphasis. -c k’i is used to emphasize something unexpected or surprising. These particles are rendered in English by “even”. 

The semantic and grammatic peculiarities of conveying the primitive concept of “happen” in Georgian verbs are studied in the paper. Three main groups are distinguished: 1. Ambient verbs; 2. one actant verbs like “I am dying” or “I am sleeping” and 3 two actant verbs, their actants being defined as an unconscious agent and a patient. There are three morphological structures for expressing the forms of the last type. Some ambiguous cases of two valence “happen” verbs are also considered with the patient and the third actant (in Tesniere’s sense). There are a lot of deviations from the regular cases, and further study could reveal more interesting cases.

Doctoral Thesis Referee

Master Theses Supervisor

Doctoral Thesis Supervisor/Co-supervisor

Scientific editor of monographs in foreign languages

Scientific editor of a monograph in Georgian

Editor-in-Chief of a peer-reviewed or professional journal / proceedings

Review of a scientific professional journal / proceedings

Member of the editorial board of a peer-reviewed scientific or professional journal / proceedings

Participation in a project / grant funded by an international organization

Participation in a project / grant funded from the state budget

Patent authorship

Membership of the Georgian National Academy of Science or Georgian Academy of Agricultural Sciences

Membership of an international professional organization

Membership of the Conference Organizing / Program Committee

National Award / Sectoral Award, Order, Medal, etc.

Honorary title

Monograph

Handbook

Research articles in high impact factor and local Scientific Journals

Publication in Scientific Conference Proceedings Indexed in Web of Science and Scopus