The general structure of the сomputer linguistic system (CLS) processing of textual content in the Ukrainian language and the conceptual scheme/model of the functioning of a typical CLS based on modelling the interaction of the main processes and IS components have been developed. Modelling of the main NLP processes of CLS was carried out due to the interaction of the main processes/components of IS and methods of linguistic processing of text content adapted to the Ukrainian language based on grapheme, morphological, lexical, syntactic, semantic, structural, ontological and pragmatic analysis, which allowed to improve the IT of intellectual analysis of the text flow for solving a specific NLP problem. This ensured the adaptation of NLP processes for the analysis of Ukrainian-language textual content. A formal model of a computer linguistic system for processing Ukrainian-language textual content was developed and described, which made it possible to determine the main structural elements and operators of natural language processing at each level of text analysis such as grapheme/phonological, morphological, syntactic, semantic, referential, structural, ontological and pragmatic. Due to the complexity of the morphology of the Ukrainian language, detailed attention is paid to the description of the model of morphological analysis of textual content. Examples of modelling processes for solving typical NLP problems such as CLS identification of viral news headlines and correction of grammatical and stylistic errors are given.
Computer linguistic system (CLS) based on NLP methods for text/audio data analysis is already an integral part of human everyday life [1-5]. On behalf of the user, some CLSs browse the large volume of Internet information and offer new personalized mechanisms/techniques/tools for interacting with the computer [6-9], for example through spam filters of e-mail traffic [10-15], IISS, virtual personal assistants, automatic translation IS etc. CLS with the support of natural language analysis is at the intersection of experimental research [16-21] and practical development of usually commercial software [22-27]. CLS of speech analysis and text analytics interact directly with the user through the 0000-0001-6417-3689 (V. Vysotska)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). support of feedback, which significantly and continuously affects the functioning of the software and the results of the analysis [28-33].
The potential for implementing natural language analysis in CLS/IS/modules is constantly growing exponentially [34-38]. A disproportionately large volume of NLP applications is usually implemented by large campaigns due to the complexity of the projects and the need for their commercialization [39-45]. As the opportunities for implementation in everyday CLS become more widespread, they become less visible, masking the complexity of their implementation. In parallel, the development of big data science and computer linguistics, especially based on non-English natural text corpora, has not yet reached the level necessary for simplification, optimization, and standardization of the processes of developing appropriate linguistic software [46-51].
CLSs for solving a large volume of specific NLP problems are just starting to spread and will eventually automate more processes that are currently solved through additional forms and selecting/clicking options/buttons. To develop the IT implementation of the appropriate linguistic software and ensure high reliability of CLS, it is necessary to take into account modern promising scientific methods of ML, data analysis, big data, and based on hypothesis analysis [52-55].
To support the functioning of a typical CLS and the operation of the main processes when solving a specific NLP problem, it is necessary and sufficient to implement the main subsystems as client, server and technological (Fig. 1) [56-63].
Input content
Computer linguistic system Technological subsystem
Server subsystem
Client subsystem
Relevant content
User requests
The main processes of the functioning of a typical CLS based on the intellectual analysis of a text stream for the solution of a specific NLP problem [64-72]: technological processing of incoming content streams: a. search and recognition of content from relevant sources; b. accumulation of analyzed content from the source in the cloud; c. saving information about the location of the found content in the corresponding source in the database/datastore; d. preliminary processing of recognized content in the cloud; e. analysis and marking/classification of recognized content according to the degree of relevance to the content and purpose of the CLS; f. integration of content provided that its degree of relevance/relevance is greater than the threshold value; g. saving integrated relevant content in the DB; h. forming an image (descriptive service data) of integrated relevant content and saving it in the DB; content management based on text analysis and processing through the server subsystem (Fig. 2) based on data from the client subsystem and the content support module: a. processing of user request streams from the client subsystem to form the correct IIS expression and subsequent caching of popular content through the server subsystem;
Data from cloud storage as a result of integration Data from the moderator Moderator's rules
Conservation / accumulation Content data
store Intellectual and informational content search Knowledge base Rule base
User requests
Relevant content Searchable content DB
Analysis of requests from regular users
CLS server subsystem
Organization/ support of
access Formation of
answers Formation and filling of content cache
Relevant content User requests
Relevant cached data in cloud storage Cached data in cloud storage storage files for storing relevant content; knowledge base for IIS/modification/maintenance of this content; accumulation and appropriate processing of service content.
The CLS server subsystem is formed from part of the functional components of the management/support/integration/content modules, in particular [73-81]: IIS content based on linguistic analysis of requests; formation and filling of the cache of information blocks of relevant popular content frequently requested by users and visitors for quick access; interactive access to relevant CLS profiles/options; analysis of user requests to accumulate content cache; storage and accumulation of information blocks in the cloud; extraction of cached content from the cloud at the request of the user or its destruction due to the onset of unpopularity; replenishment/modernization by the moderator of rules and IIS knowledge base/content analysis as requests of regular users; analysis of user requests to generate relevant reports.
The annotated CLS database is the basis of the Website IIS module. Operational and highquality IIS in the context of current content ensures its high relevance for the CLS user [8184]. The use of annotated <L in the IIS module helps to implement effective IIS-relevant content without information noise (Fig. 3). CLS should provide [85-91]: generation of Webpage according to the template and content of the Website; preservation and maintenance of cache/filling of Web page/Website according to the needs of the target audience; provision of prompt access to the Website for all types of users.
Mashup-IS consists in forming a set of integrated content from Internet resources according to the needs of the target audience and specific user requests for convenient navigation on the Website/Web page [81-91]:
=< , , , , , >, (
Input data from various sources
Data and rules from the moderator
Technological subsystem
Content integration
Primary processing Content creation Content marking
Server subsystem Content repository
Access management
Content search Annotated database
The integration of a set of data from various sources , including the Website, consists of combining them according to the appropriate collection of conditions in one Website or web page to use different types of content while preserving its main features, presentation characteristics and the possibility of further processing :
= ( , ). (
The integration should provide the user of the Website/Webpage Mashup-IS to perceive the integrated content as a single information space using large DS, including the cloud, and high-quality/operational IIS of relevant content upon request according to the collection of IIS conditions :
= ( , , ). (
Convenient navigation in the Website/Webpage Mashup-IS contributes to the
realization of the possibility of supporting the user to search for relevant and relevant
content for him throughout the available IS information space with the greatest
completeness and accuracy with the least expenditure of effort on his part. CLS is a
specialized IS, DSS or multi-agent system for solving a specific NLP problem based on a set
of integrated content from different sources according to the needs of the target audience
and specific user requests for convenient navigation on the Website/Webpage, taking into
account the statistics of the CLS operation, history of actions and personal profiles of users
and history of requests/transitions from IISS. A typical formal CLS model will be
presented as a tuple:
=< , , , , , ,
, , , 1, 2, 1 , 2 , 3 , 4 , ,
(
Data repository of processed content Integrated raw content
Cloud storage with cached content
User
CLS Website
Visitor IATCS subsystem
Moderator
A collection of integrated raw content is contained in a database based on Non-SQL. A collection of integrated processed content is contained in a SQL-based repository/database. From , the filling for the cloud is formed based on the statistics пof popular requests from users for a certain period. Collection is a specific DB/DS of cached current popular content to optimize the functioning of CLS based on builtin/modified/additional services in the cloud. These services are the result of the work of the moderator, who updates the caching rules in CLS and/or Website, updates data in the SQL database, IIS/IATCS/ management/ integration/ support of integrated and service content, integration of unprocessed content in Non-SQL database and collection/accumulation of CLS/Website operation statistics.
The basis of the IATCS subsystem is the main NLP processes of CLS.
Natural languages are determined not by rules, but by the context of the application, which
is reconstructed for computer processing. We often identify the meaning of the words used
in combination with other interlocutors. The phrase золота рибка [zolota rybka] (goldfish)
means both a sea creature and a person with a short memory, or a wish-fulfilling
creature/person, so the interlocutors must agree to a common understanding of the
context. Accordingly, speech/language is limited by mentality, region, society and level of
education. Conveying content/meaning is easiest for interlocutors with similar life
experiences, education, place of residence, etc. Therefore, automating the understanding of
speech to solve a specific NLP problem through the appropriate CLS is a rather complex and
painstaking process, especially for synthetic languages, particularly for the Ukrainian
language. The general formal model of CLS is given by the collection:
=< , , , , , , , , , , , , , >,
(
Compared to formal languages (subject/thing/object), natural languages are more universal, but less formalized. We often use one word to describe several meanings (for example, краб [krab] (a crab) is a sea creature, a dish, a nebula in the constellation of Taurus and a cockade on a sailor's cap) depending on the content of the dialogue (for example, a description of the emotions of diving, dinner, a book read, visiting a museum, watching a historical film, etc. only for the word краб). To store multiple meanings for each word, the language must be redundant (exceeding the amount of information to transmit/store a message over its entropy). That is, it is not possible to determine in advance the exact meaning of the content for each association (every linguistic variable is ambiguous by default). Lexical and structural ambiguity is a great achievement of natural language, for example, for generating new ideas, and manifestations of creativity.
Regardless of the NLP task, the process of processing Ukrainian-language texts in arbitrary CLS is presented as a sequence of mandatory operators for meaningful structural analysis of input text content:
information source input text grapheme analysis (GA) or phonological analysis (PHA) morphological analysis (MA) lexical analysis (LA) syntactic analysis (SYA) semantic analysis (SEM) structured text content
Additional operators are analyzed such as pragmatic (extraction of knowledge), ontological and referential (formation of interphrase units). Their application depends on the complexity and purpose of solving the NLP problem.
The main process of linguistic analysis of textual content is presented:
(
The main linguistic process of processing textual Ukrainian-language information to solve a specific NLP problem consists of nine stages:
Stage 1. Grapheme analysis of textual Ukrainian-language information :
= ( , , ), (
Stage 2. Morphological analysis of text content consists in the identification, analysis and determination of the form and structure of words, in particular:
= ( , , ), = 3 ∘ 2 ∘ 1, or = 3 ∘ 4 ∘ 1, where 1 is the morphological segmentation operator of a graphemically recognized chain of symbols (words/tokens); 2 is lemmatization operator of lexemes; 3 is POST operator (marking of parts of speech) for segmented words; 4 is words stemming operator.
Classical general algorithm of morphological analysis.
Step 1. Morphological segmentation of the input chain of symbols (replacing GA for short English-language messages, and supplementing GA for large corpora of English-language texts, and for Ukrainian-language texts of all kinds - a separate step for marking words in two sets as immediately identifiable (for example, prepositions) or impossible to identify (the noun is not in the nominative case).
Step 2. Lemmatization (reduction to normal form based on dictionary analysis) or stemming – determination of bases (word forms with endings cut off).
Step 3. Identification of the grammatical category of each word and the collection of their corresponding properties in relation to the use in a specific place of the text. (for example, a collection for a noun: gender, case, person, etc.).
Step 4. Formation of a linear sequence of morphological structures.
Stage 3. Lexical analysis of text content С in the intermediate stage of token sequence analysis for generating a parsing tree at the SYA level:
= ( , , ), (12) ′ = 2 ∘ 1, , ′ = 5 ∘ 4 ∘ 3, or ′ = 5 ∘ 4. (13) where 1 is the Speech Segmentation operator for identification/clarification of words/phrases/tokens after MA or in case of incorrect interpretation during PHA (usually performed in parallel with PHA and MA in a cyclic process); 2 is speech recognition operator (SR) or speech-to-text (STT) depending on the content of the NLP task; 3 is Optical Character Recognition operator (OCR) as the second part after GA and MA for clarifying incorrect moments of recognition taking into account the recognized neighboring tokens; 4 is word tokenization/segmentation operator as data preparation for building a parsing tree in SYA; 5 is Text-To-Speech operator (TTS).
Stage 4. The syntactic analysis of the text content С consists in building a tree for parsing the dependencies of words in a sequence of tokens based on their categories: = ( , , ), or = 3 ∘ 2 ∘ 1, (14) where 1 is the implementation operator of Grammar induction; 2 is operator of identification/elimination of boundary ambiguity or sentence violation (Sentence Breaking, Sentence Boundary Disambiguation); 3 is operator of syntactic Parsing of phrases/sentences for building a SYA tree.
Stage 5. Semantic analysis of textual content С is
= ( , , ), or = 2 ∘ 1, (15) where 1 is the identification operator of lexical semantics with the generation of a collection of values of each lexeme of the text; 2 is the relational semantics identification operator of the interdependencies of the lexeme content of the text.
A classic general algorithm for semantic analysis.
Step 1. Lexemes are compared with meaningful dictionary values.
Step 2. Formation of probabilistic sets for each fragment of text/sentence/phrase with alternative sems, respectively, for lexemes.
Step 3. Preliminary interconnection of the content of tokens into a single structure.
Step 4. Generation of an ordered collection of logical records of superpositions from semantic classes of lexemes and basic lexical functions.
Step 5. Finding/marking inaccuracies, contradictions, incorrectness and ambiguity of the content of the obtained result based on the lexical dictionary.
Semantic analysis is currently not used in most CLS.
Stage 6. Referential analysis formation of interphrase units .
= ( , , ). (16)
Referential analysis is often a part of semantic analysis. For Slavic languages, when analyzing large text corpora, it is best to take it as a separate stage (for example, to analyze the correspondence of a social group/community in social networks or other dialogues to identify logical meaningful connections between the posts of different participants due to the subjectivity of the speech of each. The classic general algorithm of reference analysis:
Step 1. Contextual analysis of marked fragments of textual content , for example, analysis of the pronoun/conjunction що [shcho] (that) analysis depending on the context to separate the center of unity or local references such as його [yoho] (his), який [yakyy] (which), цей [tsey] (this).
Step 2. Actual sentence segmentation of marked fragments of textual content , for identification of thematic structures based on themes/rem.
Step 3. Identification of regular recurrence of context/theme/rheme.
Step 4. Highlighting the duplicated nomination of lexical units of the text.
Step 5. Identification of synonymization of lexical units of the text.
Step 6. Isolation of implications based on situational connections.
Step 7. Identification of the identity of the reference (for example, the comparison of lexical units of the text with the object/subject/phenomenon of dialogue/image).
Stage 7. Structural analysis of text content С based on the degree of coincidence of lexical terminological units of the unity of text fragments.
= ( , , ), or = ( , , ). (17)
Similarly, referential analysis is often part of SEM for short texts/messages, or not used at all. For large corpora of texts as an additional stage of elimination of marked inaccuracy in SEM. Classical general algorithm of structural analysis.
Step 1. Formation/replenishment of the basic set of rhetorical relations of interphrase units based on the results of reference analysis and/or SEM.
Step 2. Generation of a non-linear network/graph of interphrase units.
Stage 8. Ontological analysis of textual content С based on SEM results and reference/structural analyzes if necessary: С = ( , , ), С = ( , , ) or С = ( , , ). (18)
Stage 9. Pragmatic analysis of text content С is used to determine the structure of the text taking into account the context of sentences when forming paragraphs, sections and dialogues. PA is an essential addition to SEM, reference and structural analyzes if they did not contribute to the elimination of marked inaccuracy. In some cases, it is sufficient to apply PA immediately after SEM. It is also an indispensable stage of data preparation for extracting knowledge from text corpora.
= ( , , , , [С, , ], ), or = 2 ∘ 1, (19) where 1 is the semantics identification operator outside individual sentences/phrases; 2 is text processing operator through higher-level NLP applications, for example, to simulate intelligent behavior and apparent understanding of natural language.
Depending on the tricky NLP task, grapheme (text analytics) or phonological (speech recognition) analysis of text content is used. PHA consists in the study of the structure, organization, and interpretation of speech sounds of a specific natural language (Table 1) based on phonemic, phonetic, and prosodic rules of and dictionaries of analogs. GA is a recursive parsing of the text , taking into account the linguistic features of graphemes of various languages (including non-natural ones, for example, mathematical, programming, pseudo-languages, etc.) based on the rules for recognizing strings of a certain language and dictionaries of reference grapheme models, in particular: ′ = ( , , , ), ′ ⊇ . (20)
GA can be an OCR part - encoding/recognizing text on an image into a string of characters for e-submission.
Depending on the tasks, there are the following methods for forming PHA rules:
General phonology ′1 (rules of phoneme organization in different languages). Descriptive phonology ′2 (identification of the phoneme of a language or dialect). Historical phonology ′3 (changes in phonemes, language structure during the period).
Segmental phonology ′4 (analysis of phonemes, syllables, phonetic words, syntagms and phrases).
Supersegmental phonology ′5 (analysis of intonation, tone, stress, rhythm, tempo and pauses).
Phonetic analysis ′6 (analysis of the sound structure of the language).
Phonemic analysis ′7 of the smallest unit of the phonological level.
GA is an elementary text parsing (Fig. 5) taking into account the features of graphs of different languages and the use of special symbols, objects and marks. A grapheme is an atomic meaningful linguistic (grapheme) unit of a written text (sign, symbol, special symbol, object as a picture, etc.). The purpose of GA is to form a model of the grapheme structure of the input text and generate grapheme rules (regular expressions) for the identification/classification of grapheme units in the sequence of character strings/graphemes and the connections between them. The purpose of the first level of GA – grapheme identification – is to label meaningfully independent sequences of symbols, tokens in these sequences and to determine the main language of the input text content/fragments based on a priori grapheme standards (Fig. 5). The tuple of reference grapheme models is best described by a formal grammar (abbreviations of the criteria are given in Table 2). In parallel with the parsing/identification of graphemes, they are classified/marked according to established rules (Table 3).
Decoding Official symbol
Brackets Mathematical symbol
Capital letter
Small letter Latin capital letter
Latin small letter Cyrillic capital letter Cyrillic small letter English capital letter English small letter German capital letter German small letter
Polish letter
Polish small letter Ukrainian capital letter
String classifier
Sign classifier
Names dictionary Geography names dictionary Abbreviations dictionary
Software modules Grapheme analysis as grapheme parsing Lexical analysis Syntactic analysis Semantic analysis Reference analysis Ontological analysis
Pragmatic analysis Morphological analysis
Output data Highlighted text
Fragments Lexemes: - language, - speechless, - conditional language Grapheme structure
Fragments Sentence Syntagms Lexemes Relation Let's consider the classical Chomsky with the ℎ =< ℎ ,
, ℎ =< , >, ,
Space Space >,
– Symbol Space Symbol Space and : Space – – – – (21) (22)
Table 4 provides a list of grapheme classification/marking rules for reference models (Table 2) according to production rules: The set of russian capital letters; A set of russian lowercase letters; Latin capital consonant letters;
Latin capital vowels; Terminal Latin small consonant letters;
Terminal Latin small vowel letters; The set of terminal Cyrillic capital consonants;
Cyrillic capital vowels;
Cyrillic small consonant letters; Terminal Cyrillic lowercase vowels. production rules are used to identify, classify and mark meaningful grapheme units of analysis of the input text of content (words, abbreviations, stable phrases as idioms and metaphors, sentence boundaries and quotations/sarcasms by punctuation, emoticons, geographical and proper names, abbreviations, words with apostrophes, etc.) at the parsing stage, taking into account the language of text fragments. Requirements for identifying a grapheme unit in a sequence of symbols for further morphological analysis of words: 1) the character sequence is easily identified and classified; 2) sequence too large to identify value by dictionary; 3) the sequence is too small to identify many values; 4) the number of grapheme units is too large to split the sample.
MA consists in identifying, analyzing and determining the form and structure of words in a natural language text using methods such as
Morphological Segmentation , 1 Lemmatization , POST (marking of parts of speech) and Stemming (Table 5), in 2 3 4 particular: ′ = ( , , ), (23) where ′ ⊆ , ′ = ∘ ∘ (enough for English-language short texts of a certain 3 2 1 topic) or ′ = ∘ ∘ (for most cases of messages of various topics).
3 4 1 Classification of natural language lexeme stemming algorithms
Name Word={приватизаційний} Stemming={приватизац}; Word={цивілізаційний} Stemming={цивілізац}; Word={інформаційний} Stemming={інформац}; ………………………… Word={проголошую, наголошувати, виголошував} Stemming={голошу}.
Stemming={інформац} Word={інформаційний, інформаційна, інформаційне, інформаційним, інформаційними, інформаційних, інформаційні, інформаційній, інформаційнім, інформаційного, безпритульної, інформаційному, інформаційною, інформаційну} KnowledgeBase={чорн, чорняв} Word={чорнява} Count={4, 6} Stemming={чорнява}.
The algorithm will choose the longer option.
If English stemming is a simple task, then Ukrainian stemming is several levels more difficult.
A detailed description of the non-commercial stemming algorithm for Ukrainian is a matter of time.
Word={особистість} Stemming={особист} End={ість}; Word={спогади} Stemming={спогад} End={и}; Word={дивними} Stemming={дивн} A hybrid approach part of speech is determined without taking into account the context in which this word was used in the sentence.
A combination of the above algorithms is used. likely part of speech for that word is preferred.
The probability of stemming errors increases with an incorrect description of the rules and formation of the table of endings the learning base, the better the result of their work. The knowledge base for these algorithms is a set of logical rules and IIS tables.
The table does not contain all word forms, but exceptions to the rules that are incorrectly processed by the clipping algorithm. End={ими}, where End – the result of learning the algorithm, i.e Word(кияни) {End(ість) = FALSE, End(и) = TRUE, End(ими) = FALSE} Cut (и) or Word(чуйними) {End (ість) = FALSE, End (и) = TRUE, End (ими) = TRUE} Cut (и) OR Cut (ими).
For example, the algorithm can use the method of cutting off endings and suffixes, but at the first stage, perform IIS on the table.
The best way for Slavic languages: (24) 1 Lematization is the transformation of a word form into a lemma (normal dictionary 2 form). Usually, during the transformation, a dictionary is used to present words in their actual form. Otherwise, remove only inflections and return to the lemma.
Morphological segmentation is the division of words into separate morphemes to identify their class (Table 6-8). The complexity is directly proportional to the complexity of the morphology (word structure) of a specific natural language (Table 9-10). Linguistic characteristics of some classes of verb stem morphemes
Verb фарбувати(ся) усміхнутися стогнати спитати(ся) сміятися розфарбувати(ся) привести(ся) поділити(ся) побудувати(ся) нести(ся) молоти(ся) малювати(ся) любити(ся) кохати(ся) змарніти запізнюватися → запізнитися досліджувати(ся) → дослідити(ся) втручатися втратити → втрачати вести(ся)
Analysis фарб-ува-ти(-ся) усміх-ну-ти-ся стогн-а-ти спит-а-ти(-ся) сміj-а-ти-ся розфарб-ува-ти(-ся) привес-ти(-ся) поділ-и-ти(-ся) побуд-ува-ти(-ся) нес-ти(-ся) мол-о-ти(-ся) мал-юва-ти(-ся) люб-и-ти(-ся) кох-а-ти(-ся) змарн-і-ти запізн-юва-ти-ся → запізн-и-ти-ся дослідж-ува-ти(-ся) → дослід-и-ти(-ся) втруч-а-ти-ся втрат-и-ти → втрач-а-ти вес-ти(-ся)
Basics , у, ся − ся) , н, ся) фарб-( , ̄, , усміх-( ̄, , , 1. If there is a base with the sign – TE -и(і,ї)-. 2. If there is a base with the sign – TE -а-/-я-. 3. In the presence of a base with the sign of and a suffix of a participle beginning with a consonant - either TE, or a suffix for the formation of perfect and imperfect forms of verbs mainly of foreign origin. 4. If there is an infinitive base with the sign -а (-я), -ува- (-юва-), -овува- – the suffix -н- (-ий, а, -е, -і), is added to it, for example, посія-(ти) – посіяний, чита-(ти) – читаний, розпиля(ти) – розпиляний, писа-(ти) – писаний, зігна-(ти) – зігнаний; загоювати – загоюваний, оспівувати – оспівуваний, застосовувати – застосовуваний; the suffix -ува- (-юва-), if the stress moves to the first vowel, it changes to -ова-, for example, роздрукува(ти) – роздрукований, сформулюва(ти) – сформульований, реконструюва(ти) – реконструйований, запрограмува(ти) – запрограмований.
Morphological and phonological rules A 1 2 B 1 2 3 4 5 6 7 8 9 10 11 12 C 1 2 D 1 2 3 E 1 2
Rule
Example
The basic rules of alternation of consonants in personal forms Conjugation (declension) I - consonants засвистати – засвищу, хотіти – хочу, чесати – чешу, колихати – change at the end of the stem, if there is an колишу, мазати – мажу, могти – можу, полоскати – полощу, пекти – alternation in the 1st person singular – г- печу – печений; ж, з-ж, к-ч, х-ш, с-ш, т-ч, ст-щ, ск-щ Conjugation (declension) II - we have sound changes only in the 1st person singular – д-дж, т-ч, з-ж, с-ш, зд-ждж, ст-щ where 1 and 2 are arbitrary vowels.
ІХ. : + и → і, where is sound designation [j] (yot).
Х. . : + → я, → я. Х. . : + у → ю, у → ю Х. . : + е → є, е → є Х. . : Х′ + → Х + я, Х. . : Х′ + у → Х + ю, ……………………………………………………………………………..
ІХ. : о + ( , ) + → + ( , ) + , (96) where Z is arbitrary sequence no longer than 3 characters (alternation o/a in the base of the type скочити/скакати [skochyty/skakaty] (jump), ломити/ламати [lomyty/lamaty] (break), кроїти/краяти [kroyity/krayaty] (cut), клонити/кланятися [klonyty/klanyatysya] (bow), котити/катати [kotyty/kataty] (roll), схопити/хапати [skhopyty/khapaty] (grab), гонити/ганяти [honyty/hanyaty] (chase), допомогти/допомагати [dopomohty/dopomahaty] (help)); group of consonants after that alternating symbol -о- (that is, what separates it from TE -а-/-я- before -у(ю)ва- ), cannot contain more than 3 letters.
ІХ. . : с′ + ІХ. . : в′ + ІХ. . : б′ + ІХ. . : д′ + → ш + , → вл′ + , → бл′ + , → дж′ + , (111) Х. . : Х′ + е → Х + є, (112)
XI. Rules for erasing the indicator of the boundary between verb morphemes in text content in the Ukrainian language.
+ → , (113) where and are any morphemes that none of the rules of groups IX-X apply to + . Such a restriction on and prevents untimely destruction of the boundary between morphemes before the application of the corresponding morphological rules. If any morphonological rule can be applied, then it must be applied to prevent the formation of nonsense words, for example, *котаючий [*kotayuchyy] (*rolling) from котити [kotyty] (rolling) or *качений [*kachenyy] (*rolling) from катати [kataty] (rolling). Main properties [404, 882]:
1. Transitivity of derivability. If there is a sequence 0, 1, . . . , , in which each i-th chain is directly derived from i-1 according to the hypothetical syllogism ( → , → ├ → ), is derived from 0, and the sequence 0, 1, . . . , is the derivation of from 0 [882].
2. Direct derivability. If there are 2 sequences and : = 1 2, = 1 2, (114) where 1 and/or 2 can be empty and the grammar has the rule → , the is directly derived from , for example, from the sequence according to rule VI.3О( , ̄, , , ∅, ся − ся)а + С( , , , ̄) + Ф the sequence is directly derived О( , ̄, , , ∅, ся − ся)а + юч + Ф.
XII. Rules for marking a lexeme as an adjective with multiple features in the corresponding sentence/phrase (gender, tense, case, etc.) in the Ukrainian text. (XІI) А′х,у, → лийх, ,
This chain cannot be further generated for the Ukrainian language.
LA is preliminary processing of text or speech, transformation of a chain of symbols into a sequence of tokens (tuples of symbols according to appropriate patterns): where ′ ⊆ , ′ = 2 ∘ 1 (enough to transform the sound series of speech into printed text) or ′ = 5 ∘ 4 ∘ 3 (to transform scanned text into speech), but is sufficient for most cases: ′ = ( , , , ), ′ = 5 ∘ 4. (116) (117) 1. Speech Segmentation 1 is the division of the sound stream of human speech into separate words. In human conversation or speech, pauses between consecutive words are almost unidentifiable, so segmentation is a necessary task for speech recognition. In most spoken languages, sounds as consecutive letters are combined with each other in the process of coarticulation, so converting an analog signal into discrete symbols is a rather complex NLP process in CLS systems for technical implementation. 2. Speech Recognition 2 (SR) or Speech-To-Text (STT) is transformation of the speech signal into e-text. Different people pronounce words in each language with different stresses, speeds, and intonations. CLS should recognize a wide range of input unstructured data as identical to a single equivalent. 3. Optical Character Recognition 3 (OCR) is translation of scanned handwritten or printed text after GA and MA into a sequence of codes for e-submission with correction of simple errors based on statistics and theory probabilities of using a sequence of N-grams/words/endings instead of obscure symbols. 4. Tokenization or Word Segment 4 is the demarcation and categorization of sections of the chain of input symbols for MA. For English or Ukrainian languages, this is a fairly trivial situation, since words are usually separated by spaces (problems only if there are stylistic and grammatical errors in the text). However, some written languages such as Chinese, Korean, Japanese, and Thai do not mark word boundaries in this way. Then tokenization is an important task based on vocabulary and word morphology. Sometimes the method is used to form a Bag-of-words model (BOW) in Data Mining. 5. Text-To-Speech 5 (TTS) is transformation of handwritten, typewritten or printed text into a speech signal for oral presentation, for example, for people with visual impairments.
SYA is the basis of semantic analysis: ′ = ( , , ), (118) where ′ ⊆ , ′ = 3 ∘ 2 ∘ 1: 1. Grammar induction 1 is generation of formal grammar to describe language syntax. 2. Sentence Breaking or Sentence Boundary Disambiguation 2 is analysis of the presence/absence of appropriate punctuation marks and the text between them (a dot not only marks the end of a sentence, but also a contraction). 3. Parsing 3 is the generation of sentences from the input sequence of symbols of the parsing tree (grammatical analysis) for the analysis of the grammatical structure according to the given formal grammar (Table 16). There are hundreds or thousands of analyzes for a typical sentence, most of which are absolutely meaningless to a native speaker. There are two main types of parsing: Dependency Parsing 13 and Constituency Parsing 32, or in the form of some combination of these methods 33. Dependency parsing focuses on the relationship between words in a sentence (primary objects and predicates), and constituent parsing focuses on building a SYA tree using probabilistic context-free (stochastic) grammar (PCFG).
For example, when parsing/generating sentences/phrases, the choice of case inflection of a specific word in the Ukrainian language directly depends on the type of base and part of speech, in particular, for noun groups taking into account the context (Table 9-10, Table 16):
1 1 1 2, 2 2 3 4,
A set of linguistic units of one type, together with a set of linguistic units 1 of another type, are transformed in a different way 1 2, than together with a set of linguistic units 2 of the third type – 3 4. Without taking into account the context, a more fractional classification should be introduced:
1 1 2, 2 3 4,
Verb group
Word replacement 3) 1 ̃ , , , 2 → 1 з,ай,м, 2, 1 , , , 2 ̃ ′; 4) ̃ , , ,3 → , , . 1) ̃ ,тепер, → ,тепер, ̃ ′, ′,зн, ′ ̃ ″, ″,ор, ″; 2) ̃ ,тепер, → ,тепер, ̃ ′, ′,ор, ′ ̃ ″, ″,зн, ″;
Each record is a rules set, for example, II.1 forms 648 rules: ̃ч,од,н,3 → ̃ч,од,н,3 ̃ч,од,р,1; ̃ч,од,р,3 → ̃ч,од,р,3 ̃ч,од,р,1; ...; ̃сер,мн,м,3 → ̃сер,мн,м,3 ̃сер,мн,род,3 (Table 17). To generate a sentence tree in Ukrainian, the inflections agreement rules corresponding are used (Fig. 6Fig. 7).
Each step is an expansion of a symbol of the sequence (for example, ̃ од,тп,3 – Rод,тп,3 Šч,од,з,1 Šс,од,о,3) or replacement (so, Šч,од,з,1 – чз,аойдм,з,1). For such deployment, it is necessary to form more detailed types of linguistic units to take into account the location in the context of the sentence, for example: мн,род → мн,род, (121) , , (4.1) where is the word form of a sentence, is the base of a word of type ( = 1,2,3, . . . ), мн,род is the inflection of the genitive plural in Ukrainian languages, for example: 2.(I) # Sж,од,н,3 Vод,тепер,3 # 3.(III.1) # Sж,oд,н,3 Vод,тепер,3 Sч,од,зн,1 Sсер,од,ор,3 # 4.(II.1) # Sж,од,н,3 Sч,од,р,3 Vод,тепер,3 Sч,од,зн,1 Sсер,од,ор,3 # 5.(II.2) # Аж,од,н Sж,од,н,3 Sч,од,р,3 Vод,тепер,3 Sч,од,зн,1 Sсер,од,ор,3 # 6.(II.2) # Аж,од,н Sж,од,н,3 Ач,од,р Sч,од,р,3 Vод,тепер,3 Sч,од,зн,1 Sсер,од,ор,3 # 7.(II.2) # Аж,од,н Sж,од,н,3 Ач,од,р Sч,од,р,3 Vод,тепер,3 Sч,од,зн,1 Асер,од,ор Sсер,од,ор,3 # 8-9 .............................................................................................................................. 10.(II.4) # Аж,од,н Sж,од,н Ач,од,р Sч,од,р Vод,тепер,3 Sч,од,зн,1 Асер,од,ор Sсер,од,ор # 11.(II.3) # Аж,од,н Sж,од,н Ач,од,р Sч,од,р Vод,тепер,3 Sчз,аойдм,зн,1 Асер,од,ор Sсер,од,ор # 12-18 ............................................................................................................................
IV.1 IV.7 IV.4 IV.6 IV.3 сина наповнює мене безмежним щастям
IV.6 IV.2 IV.6 # весела посмiшка твого # 1 мн,род → ів(друз − ів), м1н,род → ів, 2 мн,род → ок(іграш − ок), м2н,род → ок, 3 мн,род → ей(діт − ей), м3н,род → ей, the Ukrainian language on the presence/absence of negation:
̃ → ̃ , ̃ → ̃1, ̃ → ¬ ̃ , ̃ → ¬ ̃2.
≠¬ where ̃ is a verb group in a sentence, is a transitive verb in a verb group, ̃ is a direct object, ̃ is a noun group, ¬ is a negation, in particular, for sentences, школяр пише есе [shkolyar pyshe ese] (the student writes an essay) and школяр не пише есе [shkolyar ne pyshe ese] (the student does not write an essay) with relevant conclusions: The use of the instrumental subject ̃
with a verbal noun depends on the presence of the object ̃ (аналіз змісту системою [analiz zmistu systemoyu] - content analysis by ≠¬ the system): 4 мн,род → их(знайом − их), м4н,род → их, 5 мн,род → (машин −), м5н,род → .
………………………………………………………………… The choice of the case of the direct complement in a word form or sentence depends in (126) (127) (128) (129) (130) (131) (132) ̃ → ̃′ ̃
̃ , ̃ → ̃′ ̃ ̃ 1,
It is impossible to completely abandon the semantics of the context during the correct identification and correction of grammatical and stylistic errors, it is necessary to take into account not only one symbol in the left part of the rules (30)-(40), which ensures the permutation of symbols.
Semantic features are investigated using a set of lexical and linguistic resources in the form of dictionaries and libraries, dictionary management tools, semantic role marking 23 and the process of word embeddings 3. Word embedding consists in mapping words, phrases or phrases from the dictionary into vectors of real numbers for ease of processing, for example, based on Word2Vec.
= 23(3( , ), ).
Taking semantics into account significantly simplifies the grammatical tree (Fig. 8). If we connect the symbols (ancestors) directly to the final results (descendants) during expansion, replacement or rewriting, we get a tree of components, or a syntactic structure. The specified rules (Table 16) are capable of generating other phrases that are not necessarily meaningful, as rules II.1 and II.2 are cyclical. Along with the sequence as весела посмішка [vesela posmishka] (funny smile), you can get весела весела посмішка [vesela vesela posmishka] (funny funny smile), etc. The number of phrases in natural language must be finite. With the correct construction of the sentence parsing tree (Fig. 6-Fig. 8) and further shortening (Fig. 9), it is possible to match cases. Аж,од,н Sж,од,н,3
Ач,од,р Sж,од,н
Sч,од,р,3 Sч,од,р Vод,тепер,3 Sч,од,зн,1
Sсер,од,ор,3 Асер,од,ор
Sсер,од,ор,3 Sчз,аойдм,зн,1
Sсер,од,ор
сина наповнює мене безмежним щастям безмежним Sсер,од,ор (133) весела
посмiшка наповнює безмежним щастям
The agreement of cases between the linguistic units of the sentence affects the subsequent semantic analysis of the text. For example, in the Ukrainian language it is possible to generate sequences of linguistic units of the type 1 2 2 2 3 2′ 1′ 2′ 3′, 1 3 2 1 1′ 3′ 2′ 1′ etc. (or as ′): Саша, Софія, Катя, Данило, … – а спортсмен, співачка, художниця,поет, … відповідно. In particular, sequence of proper names; ′ ( ′ ′ ′ ′. . . ) is a sequence of professions agreed with proper names in the family; is a punctuation mark. с . . . ) is a 1. → ′ ,
2. ′ → ′ , , = 1,2,3,
3. 4. → , → .
} where , ′, are basic linguistic units; , are auxiliary linguistic units; is the initial symbol as an indicator of the chain generation type. For a more effective study of errors, meta-data analysis of linguistic features and characteristics of the original text is used, in particular, content genre, presence/absence of dialect, slang, terminology, and the probability of writing by a native speaker or the result of a translation.
Semantic analysis is currently not used in most CLS, but with the gradual introduction of AI into the everyday life of the average person, this task should be solved and simplified (Fig. 10). The more complex the grammar of the language, the more difficult it is to conduct CEM: ′ = ( , , ), where ′ ⊆ , ′ = 2 ∘ 1.
Linguistic semantics 1 (individual words in context) consists of: ′′ = 16 ∘ 15 ∘ 14 ∘ 13 ∘ 12 ∘ 1, ′′ ⊆ ′.
1 Semantic analysis (134) (135) Linguistic semantics Лексична семантика
Lexical semantics Recognition of named
entities
Sentiment analysis Extracting terminology Ambiguity of the word's meaning
Categorization and analysis of words
Template Common and
distinctive characteristics
Relational semantics
Removal of relations
Semantic parsing Semantic marking of roles coherence (integrity) of the text (Coherence), Anaphora Resolution and final conclusion (Eng. Inference). 4. Named-Entity Recognition 14 (NER) or identification of an object entity, fragmentation of an object entity, extraction of an object entity is extraction from unstructured text of information about the presence of certain named entities of the corresponding categories such as dates, percentages, monetary values, quantities, time, geographical locations, names of organizations or people, etc. Having a capital letter at the beginning of a word does not solve this problem - the beginning of a sentence or a line of poetry also begins with a capital letter. In German, all nouns begin with a capital letter. In addition, named entities often include several words, only some of which are capitalized. French, Ukrainian, and Spanish do not capitalize adjective names. In German, all nouns are capitalized. Chinese, Korean, Japanese, or Arabic have no capital letters at all. 5. Terminology Mining 15 (Terminology Extraction, Term Recognition, Glossary Extraction, Term Extraction) is the automatic extraction of relevant terms from the relevant corpus. One of the first steps towards knowledge domain modeling is the collection of a vocabulary of domain terms as linguistic features of text content. Approaches to term extraction use linguistic processors (marking parts of speech, fragmentation of parts of texts) to extract terminological candidates, that is, syntactically plausible terminological noun phrases or keywords, for example for rubrication. 6. Tonality analysis of the text or multimodal sentiment analysis 16 (Opinion Mining, Sentiment Analysis) is the detection of a subjective emotionally-colored set of content (positive, negative or neutral) in a text array data of a specific author in relation to the corresponding object, subject, event or phenomenon of thematic subject area. It is useful for identifying trends of public opinion for online marketing, in social networks, forming political propaganda, etc.
Relational semantics 2 (semantics of individual sentences) consists of: ′′′ = 2 ∘ 22 ∘ 2, ′′′ ⊆ ′.
3 1 (136) 1. Relationship Extraction 12 is identification of relationships of nominal entities (for example, family ties, colleagues, enemies, etc.). 2. Semantic Parsing 22 is presentation of the semantics of a part of the text (usually a sentence) in the form of a logical formalism (DRT parsing, Discourse Representation Theory) or a graph (AMR parsing, Abstract Meaning Representation), for example: , , : ( , хотіти01) ( , гуляти01) ( , дитина)
g0(a, b)g1(a, e)g0(e, b).
AMR format: (a / хотіти-01 : arg0 (b / дитина) : arg1 (e / гуляти-01 : arg0 b)). (137) 3. Semantic Role Labelling 23 (Implicit Semantic Role Labeling Below) is assignment of semantic role labels to words or word combinations in a sentence, for example, the role of goal, agent, or result according to the algorithm: ′′′′= 325 ∘ 234 ∘ 233 ∘ 232 ∘ 231, ′′′′ ⊆ ′′′. (138) a. Selection of a sentence or a fragment of text 231. b. Definition of semantic predicates in a sentence232 (verb and noun frames). c. Disjunction of defined semantic predicates 233. d. Identification of the elements of the defined frame 234. e. Classification of the identified elements of the frame 235– assigning a semantic role in the sentence.
PA is used to determine the structure of the text taking into account the context of sentences when forming paragraphs, sections and dialogues. The main task is the identification of the context of such linguistic units as sentences and the formation of the semantic relationship of these linguistic elements. PA is a mandatory component in AIS for the interpretation of human dialogue, speech and its corresponding analysis (Fig. 11).
= ( , , ), = 2 ∘ 1.
Pragmatic analysis
Discourse Coreference resolution
Discursive analysis Non-semantic marking of
roles Identification of text
indentation
Arguments mining Thematic segmentation and recognition
Correction of grammatical errors Dialogue management
Natural language
generation AI document
High-level NLP applications
Summarizing the text
Book generation Understanding natural
language Machine translation Question-and-answer systems (139)
Discourse 1 (semantics outside individual sentences) consists of: ′ = 6 ∘ 15 ∘ 14 ∘ 13 ∘ 12 ∘ 1, ′ ⊆ .
1 1 (140) 1. Coreference Resolution 11 is identification in the next text of the following wordsreferences or expressions about objects, subjects, phenomena and events, which are
Natural language processing through higher-level NLP-applications 2 simulate intelligent behavior and obvious understanding of natural language and are currently generally divided into the following classes: 7. Text Summarization 12 (Automatic Summarization) is generation of a readable digest/annotation as a summary in the form of a text fragment from the general analyzed text (for example, a scientific article, publication in a newspaper or magazine). 8. Grammatical Error Correction 22 is identification and correction of grammatical/stylistic errors at all levels of linguistic analysis (phonology/orthography, morphology, vocabulary, syntax, semantics, pragmatics). 9. Machine Translation 32 is automatic translation of text from one human language to another using all levels of linguistic analysis, especially grammar, semantics and facts about the real world, etc., based on the solution of AI-complete (AI-hard) tasks of abstract content translation. 10. Dialogue Management 42 (Dialogue System, Conversational Agent, CA) is the organization of communication between a program and a person, other than chatbots, using text, speech, graphics, tactility, gestures, etc. for two-way communication. 11. Question Answering systems 52 is determination of the correct answer based on the analysis of a typical human question (for example, "What is the capital of Ukraine?") and an open/complex question (for example, "In what the meaning of existence?"). 12. Natural Language Generation 62 (NLG) is conversion of content from databases/databases or semantic intentions into readable specific human language through an algorithm: a. Determination of the content of 621 (what information to mention in the text). b. Document structuring 622 (content transfer template). c. Aggregation 623 (combining similar sentences to improve the readability and naturalness of the text content). d. Lexical selection 624 (adding words to concepts). e. Generation of reference expressions 65 (generation of expressions 2 identifying objects and regions). This task also involves making decisions about pronouns and other types of anaphora. f. Implementation of 66 (text generation taking into account the rules of 2 syntax, morphology and spelling, for example, the use of verbs in the necessary tenses). g. When necessary, use 627 machine learning methods (most often LSTM) on a large set of input data and corresponding (human-written) output texts, for example to generate text captions for images. 13. Natural Language Understanding 72 (NLU) is transformation of text into logical structures for controlling NLP programs, i.e. identification of semantics from a set of possible notations in the form of natural language concepts. The introduction and creation of speech meta-model and ontology is efficient and empirical. To build a formalization of semantics, explicit formalization is used in contrast to implicit assumptions, for example, about a Closed-World Assumption (CWA - any statement whose truth is not known is false) against an open one (Open-World Assumption OWA – the truth of a statement does not depend on the observer's knowledge of its truth), or subjective YES/NO versus objective TRUE/FALSE. 14. Book Generation 82 is creation of full-fledged books based on NLG technology 62, production/associative rules, neural network, factual knowledge and generalization of text 12. 15. Document AI 92 is a platform for training an agent to extract specific necessary data from various types of documents. Designed for users without AI, ML and NLP experience to quickly access the necessary content hidden in texts, for example, for lawyers, business analysts and accountants. (141) (142) (143)
The CLS model for identifying viral news headlines is presented as: ℎ = 16 ∘
∘ 1 ∘ ∘ 3 ∘ 14 ∘ 4, where 4 is tokenization; 14 is recognition of named entities; 3 is marking of parts of speech; is Ngrams (sequences of elements and their frequencies); 1 is clustering; is ML based on Neural Networks (NN); 16 is application of SentiWordNet (a lexical-semantic thesaurus for the analysis of text tonality).
Another relevant NLP technology is error correction, the main tasks of which are error identification, error correction, and user training. Grammatical error correction 22 is one of the sub-processes of correcting various types of errors. Bug fixes provided: = 2( ),
2 where is the result of identifying and correcting grammatical/stylistic errors at all levels of linguistic analysis (phonology/orthography, morphology, vocabulary, syntax, semantics, pragmatics); 22 is the basic error correction process in the text data array X.
The detailed error correction process is given as:
= 224 ∘ 223 ∘ 222 ∘ 221, coreference resolution 11;
where ′ = 221( , , , , , , 3, 3, 11) is a pattern matching check based on complex multilayer NLP- rules based on regular expressions , lexical vand grammatical victionaries, POS-tags and part-of-speech marking 3, parsing trees 3 and ′′ = 222( ′, , , , , 2212, 2222) statistical methods for refining corrections ( is– Ngrams analysis on based on the set of tokens , POS-tags , and the history of analogues ; 2212 is error identification through the usage statistics with similar words/errors in the text; 2222 is correction to the most likely variant of possible analogues); classifiers for a specific language, analysis N-grams (for example, bigrams with appropriate analysis of left and right context followed by replacement of the best option from POS Ngrams of the model), machine learning rules as a choice from several correct options or identification correct but rare application, ML-based error detection 2213 and error correction 2223 processes, data annotation for training 2233, feature selection for training 2243 and classifier training 2253 using random method forest, logistic regression or other (sometimes pattern matching and simple statistical data cannot generalize decision options, for example, identification and selection of a preposition or article in English, and an adjective in Ukrainian; derivational word-formation morphology; run-on sentences, i.e. independent or subordinate clauses are not joined by a conjunction or punctuation); = 224( ′′′, , , , , 2214, 2224) is neural machine translation based on Noisy channel translation processes 2241 (spelling check, question answers, speech recognition and machine translation based on finding a predicted word with a given word, where symbols are somehow encrypted) and Round-trip translation 2224 (two-way translation from the source language to the target language to assess the quality/accuracy of the result).
The N-gram model assigns probabilities to sentences and sequences of words based on counting N-grams, for example, according to the Markov assumption: =1 ( ) ∏ ( | −1), (144) where is the -th chain or sentence of words; is the current word in the -th chain or sentence; −1 is the previous word in the -th chain or sentence. To identify an error, it is necessary to determine grammatical or stylistic features using the rules of marking parts of speech 3, parsing: dependencies 13 and constituents 32 or in the form of some combination of these methods 33.
The general structure of CLS processing of textual content in the Ukrainian language and the conceptual scheme/model of the functioning of a typical CLS based on modeling the interaction of the main processes and IS components have been developed.
Modeling of the main NLP processes of CLS was carried out due to the interaction of the
main processes/components of IS and methods of linguistic processing of text content
adapted to the Ukrainian language on the basis of grapheme, morphological, lexical,
syntactic, semantic, structural, ontological and pragmatic analysis, which allowed to
improve the IT of intellectual analysis of the text flow for solving a specific NLP problem.
This ensured the adaptation of NLP processes for the analysis of Ukrainian-language textual
content. A formal model of a computer linguistic system for processing Ukrainian-language
textual content was developed and described, which made it possible to determine the main
structural elements and operators of natural language processing at each level of text
analysis such as grapheme/phonological, morphological, syntactic, semantic, referential,
structural, ontological and pragmatic. Due to the complexity of the morphology of the
Ukrainian language, detailed attention is paid to the description of the model of
morphological analysis of textual content. Examples of modeling processes for solving
typical NLP problems such as CLS identification of viral news headlines and correction of
grammatical and stylistic errors are given.
[11] N. Liubchenko, A. Podorozhniak, V. Oliinyk, Research Application of the Spam Filtering
and Spammer Detection Algorithms on Social Media, CEUR Workshop Proceedings
3171 (2022) 116-126.
[12] V.B. Fernandes, et al., Corrigendum to “A spam filtering multi-objective optimization
study covering parsimony maximization and three-way classification”, Applied Soft
Computing Journal 55 (2017) 565.
[13] V. Basto-Fernandes, et al., A spam filtering multi-objective optimization study covering
parsimony maximization and three-way classification, Applied Soft Computing Journal
48 (2016) 111–123.
[14] I Yevseyeva, V. Basto-Fernandes, D. Ruano-Ordás, J.R. Méndez, Optimising anti-spam
filters with evolutionary algorithms, Expert Systems with Applications 40(
Corpus(UKRMED), CEUR Workshop Proceedings 2870 (2021) 402-412. [17] V. Bobicev, O. Kanishcheva, O. Cherednichenko, Sentiment Analysis in the Ukrainian and Russian News, in: Procedeengs of the First Ukraine Conference on Electrical and Computer Engineering (UKRCON), 2017, pp. 1050-1055. [18] O. Cherednichenko, N. Babkova, O. Kanishcheva, Complex Term Identification for
Ukrainian Medical Texts, CEUR Workshop Proceedings 2255 (2018) 146-154.
[19] I. Balush, V. Vysotska, S. Albota, Recommendation System Development Based on
Intelligent Search, NLP and Machine Learning Methods, CEUR Workshop Proceedings
2917 (2021) 584-617.
[20] V. Vysotska, S. Mazepa, L. Chyrun, O. Brodyak, I. Shakleina, V. Schuchmann, NLP Tool
for Extracting Relevant Information from Criminal Reports or Fakes/Propaganda
Content, in: Procedeengs of the IEEE 17th International Conference on Computer
Sciences and Information Technologies (CSIT), 2022, pp. 93-98, doi:
10.1109/CSIT56902.2022.10000563.
[21] O. Prokipchuk, V. Vysotska, P. Pukach, V. Lytvyn, D., Uhryn, Y. Ushenko, Z. Hu, Intelligent
Analysis of Ukrainian-language Tweets for Public Opinion Research based on NLP
Methods and Machine Learning Technology, International Journal of Modern Education
and Computer Science 15(
Luchkevych, O. Brodovska, The Content Analysis Method for the Information Resources Formation in Electronic Content Commerce Systems, CEUR Workshop Proceedings 2870 (2021) 1632-1651. [26] R. Bekesh, L. Chyrun, P. Kravets, A. Demchuk, Y. Matseliukh, T. Batiuk, I. Peleshchak, R.
Bigun, I. Maiba, Structural modeling of technical text analysis and synthesis processes, CEUR Workshop Proceedings 2604 (2020) 562–589. [27] O. Artemenko, V. Pasichnyk, N. Kunanets, K. Shunevych, Using sentiment text analysis of user reviews in social media for e-tourism mobile recommender systems, CEUR workshop proceedings 2604 (2020) 259-271. [28] N. Grabar, T. Hamon, Automatic Detection of Temporal Information in Ukrainian
General-language Texts, CEUR Workshop Proceedings 2136 (2018) 1-10. [29] V. Shyrokov, "Accuracy" vs "Unambiguity" in Linguistics, CEUR Workshop Proceedings 2870 (2021) 1-5. [30] O. Kozlov, Structural Optimization of Fuzzy Systems based on Determination of
Linguistic Terms Number, CEUR Workshop Proceedings 2917 (2021) 174-185. [31] O. Yurchenko, N. Ugolnikova, Linguistic Methods in Social Media Marketing, CEUR
Workshop Proceedings 2870 (2021) 743-754. [32] E. Fedorov, O. Nechyporenko, Linguistic Constructions Translation Method Based on
Neural Networks, CEUR Workshop Proceedings 3396 (2023) 295-306. [33] E. Fedorov, O. Nechyporenko, Dynamic Stock Buffer Management Method Based on
Linguistic Constructions, CEUR Workshop Proceedings 2870 (2021) 1742-1753. [34] A. Taran, The Role of Keyword Language in the Database of World Slavic linguistics "iSybislaw", CEUR Workshop Proceedings 3171 (2022) 266-276. [35] A. Taran, Terminology of Computational Linguistics in Terms of Indexing and Information Retrieval in the System "iSybislaw", CEUR Workshop Proceedings 2870 (2021) 225-234. [36] А. Taran, Information-retrieval System "Base of the World Slavic Linguistics (iSybislaw)" in Language Education, CEUR workshop proceedings 2604 (2020) 590599. [37] I. Bekhta, N. Hrytsiv, Computational Linguistics Tools in Mapping Emotional
Dislocation of Translated Fiction, CEUR Workshop Proceedings 2870 (2021) 685-699. [38] D. Sitnikov, P. Sytnikova, A. Kovalenko, Methods of Eliminating Features from Linguistic
Equations, CEUR Workshop Proceedings 2870 (2021) 877-889. [39] S. Albota, Linguistically Manipulative, Disputable, Semantic Nature of the Community
Reddit Feed Post, CEUR Workshop Proceedings 2870 (2021) 769-783. [40] S. Albota Resolving conflict situations in reddit community driven discussion platform.
CEUR Workshop Proceedings 2604 (2020) 215–226. [41] S. Albota, Linguistic and Psychological Features of the Reddit News Post, in: Procedenngs of the IEEE 15th International Scientific and Technical Conference on Computer Sciences and Information Technologies, CSIT, 2020, 1, pp. 295–299. [42] S. Albota, A. Peleshchyshyn, Contradictory statement as a basis for conflict resolution strategies, CEUR Workshop Proceedings 2588 (2020) 336–345. [58] N. Hrytsiv, I. Bekhta, M. Tkachivska, V. Byalyk, Sylvia Plath’s I felt-Narrative Label of The Bell Jar in Ukrainian Translation: Tagging Textness Features, CEUR Workshop Proceedings 3171 (2022) 240-255. [59] Z. Kunch, O. Lytvyn, I. Mentynska, Modern Ukrainian Electronic Dictionaries: the Problem of Implementing Spelling Changes, CEUR Workshop Proceedings 3396 (2023) 32-47. [60] O. Levchenko, M. Dilai, Key Colour Terms in the Ukrainian Prose Fiction of the 21st
Century, CEUR Workshop Proceedings 3171 (2022) 49-60. [61] O. Levchenko, N. Romanyshyn, D. Dosyn, Method of Automated Identification of Metaphoric Meaning in Adjective + Noun Word Combinations (Based on the Ukrainian Language), CEUR Workshop Proceedings 2386 (2019) 370-380. [62] V. Starko, A. Rysin, VESUM: A Large Morphological Dictionary of Ukrainian As a
Dynamic Tool, CEUR Workshop Proceedings 3171 (2022) 61-70. [63] V. Starko, Implementing Semantic Annotation in a Ukrainian Corpus, CEUR Workshop
Proceedings 2870 (2021) 435-447. [64] O. Synchak, V. Starko, Ukrainian Feminine Personal Nouns in Online Dictionaries and
Corpora, CEUR Workshop Proceedings 3171 (2022) 775-790. [65] N. Romanyshyn, Application of Corpus Technologies in Conceptual Studies (based on the Concept Ukraine Actualization in English and Ukrainian Political Media Discourse), CEUR workshop proceedings 2604 (2020) 472-488. [66] M. Shvedova, The General Regionally Annotated Corpus of Ukrainian (GRAC, uacorpus.org): Architecture and Functionality, CEUR workshop proceedings 2604 (2020) 489-506. [67] M. Shvedova, N. Prydvorova, I. Skibina, Normalization of Early Modern Ukrainian in GRAC: the Case of Lesia Ukrainka's Works, CEUR Workshop Proceedings 3171 (2022) 71-80. [68] S. Kubinska, R. Holoshchuk, S. Holoshchuk, L. Chyrun, Ukrainian Language Chatbot for Sentiment Analysis and User Interests Recognition based on Data Mining, CEUR Workshop Proceedings 3171 (2022) 315-327. [69] N. Borysova, K. Melnyk, N. Babkova, Z. Kochuieva, V. Melnyk, Gender Classification of
Surnames: Ukrainian aspect, CEUR Workshop Proceedings 3171 (2022) 354-364. [70] A. Dmytriv, S. Holoshchuk, L. Chyrun, R. Holoshchuk, Comparative Analysis of Using Different Parts of Speech in the Ukrainian Texts Based on Stylistic Approach, CEUR Workshop Proceedings 3171 (2022) 546-560. [71] K. S. Mandziy, U. V. Yurlova, M. P. Dilai, English-Ukrainian Parallel Corpus of IT Texts: Application in Translation Studies, CEUR Workshop Proceedings 3171 (2022) 724736. [72] K. Vyrodov, A. Chupryna, R. Kotelnykov, Detecting of Anti-Ukrainian Trolling Tweets,
CEUR Workshop Proceedings 3396 (2023) 48-62. [73] L. Kobylyukh, Z. Rybchak, O. Basystiuk, Analyzing the Accuracy of Speech-to-Text APIs in Transcribing the Ukrainian Language, CEUR Workshop Proceedings 3396 (2023) 217-227. [74] K. Datsyshyn, Z. Haladzhun, N. Kunanets, O. Hotsur, N. Veretennikova, Neologisms with the Prefix Anti- in the Ukrainian Online Media in the Covid-19 Pandemic Period, CEUR Workshop Proceedings 3171 (2022) 192-211. [75] Z. Haladzhun, K. Datsyshyn, Y. Bidzilya, N. Kunanets, N. Veretennikova, “Antivaccinationists&Anti-vax”: Linguistic Means of Actualizing Assessment in the Headlines and Leads of Ukrainian Text Media, CEUR Workshop Proceedings 3396 (2023) 118-129. [76] Y. Chychkarov, O. Zinchenko, Handwritten Ukrainian Character Recognition using a Convolutional Neural Networks and Synthetic Dataset, CEUR Workshop Proceedings 3426 (2023) 109-121. [77] T. Basyuk, A. Vasyliuk, Peculiarities of an Information System Development for Studying Ukrainian Language and Carrying out an Emotional and Content Analysis, CEUR Workshop Proceedings 3396 (2023) 279-294. [78] A. Dmytriv, V. Vysotska, M. Bublyk, The Speech Parts Identification for Ukrainian Words Based on VESUM and Horokh Using, in: Proceedings of the IEEE 16th International Conference on Computer Sciences and Information Technologies (CSIT), 22-25 Sept., Lviv, Ukraine, 2021, vol. 2, pp. 21–33. [79] V. Vysotska, O. Markiv, S. Teslia, Y. Romanova, I. Pihulechko, Correlation Analysis of Text Author Identification Results Based on N-Grams Frequency Distribution in Ukrainian Scientific and Technical Articles, CEUR Workshop Proceedings 3171 (2022) 277-314. [80] V. Lytvyn, P. Pukach, V. Vysotska, M. Vovk, N. Kholodna, Identification and Correction of Grammatical Errors in Ukrainian Texts Based on Machine Learning Technology, Mathematics 11 (2023) 904. doi:10.3390/math11040904. [81] P. Kryndach, V. Vysotska, S. Chyrun, L. Chyrun, S. Goloshchuk, R. Holoshchuk, Analysis of Semantic Relationships in Ukrainian Text Content Based on Word2Vec and Machine Learning, in: Proceedings of the IEEE 18th International Conference on Computer Sciences and Information Technologies (CSIT), Lviv, 19-21 October, 2023. [82] B. Rusyn, V. Vysotska, L. Pohreliuk, Model and architecture for virtual library information system, in: Proceedings of the International Conference on Computer Sciences and Information Technologies, CSIT, 2018, pp. 37-41. doi: 10.1109/STCCSIT.2018.8526679. [83] B. Rusyn, V. Lytvyn, V. Vysotska, M. Emmerich, L. Pohreliuk, The Virtual Library System Design and Development, Advances in Intelligent Systems and Computing 871 (2019) 328-349. doi: 10.1007/978-3-030-01069-0_24. [84] L. Chyrun, Y. Burov, B. Rusyn, L. Pohreliuk, O. Oleshek, A. Gozhyj, І. Bobyk, Web resource changes monitoring system development, CEUR Workshop Proceedings 2386 (2019) 255–273. [85] B. Rusyn, L. Pohreliuk, A. Rzheuskyi, R. Kubik, Y. Ryshkovets, L. Chyrun, S. Chyrun, A.
Vysotskyi, V.B. Fernandes, The mobile application development based on online music library for socializing in the world of bard songs and scouts’ bonfires, Advances in Intelligent Systems and Computing 1080 (2020) 734–756. doi: 10.1007/978-3-03033695-0_49.