Search

This paper presents the work in progress on ELEXIS-SR corpus, the Serbian addition to the ELEXIS multilingual annotated corpus (Martelli et al., 2023), comprising semantic annotations and word sense repositories. The ELEXIS corpus has parallel annotations in ten European languages, serving as a cross-lingual benchmark for evaluating low and medium-resourced European languages. The focus in this paper is on multiword expressions (MWEs) and named entities (NEs), their recognition in the ELEXIS-SR sentence set, and comparison with annotations in other languages. The first steps in building the Serbian sense inventory are discussed, and some results concerning MWEs and NEs are analysed. Once completed, the ELEXIS-SR corpus will be the first sense annotated corpus using the Serbian WordNet (SrpWN). Finally, ideas to represent MWE lexicon entries as Linguistic Linked-Open Data (LLOD) and connect them with occurrences in the corpus are presented.

Within the project Mainstreaming Climate Resilience in the Road Transportation Management in Serbia (CliRtheRoads), a complex software solution was developed to support Public Enterprise Roads of Serbia in climate change adaptation planning and management. The software solution comprises 1) web portal for data entry and management for authorised users 2) publicly available web GIS part 3) mobile GIS application and 4) back-end database (Figure 1). In order to store and process the data collected during field visits and surveys, the data base developed in project first phase (Valjevo test area), was modified to include new datasets and to store new datatypes (as floods), and new modules (as activity cost) for Kraljevo test area. Both test areas have been chosen due theirs diversified characteristics, many climate related hazards, and the fact they suffered greatly from disasters recently (Abolmasov et al. 2017, Abolmasov et al. 2021). The database was upgraded to new model aimed for maintaining legacy data from both test areas, but also for future infrastructure resilience projects based on same approach, to enable that all data will be stored and mapped in a unified manner. The data model include 1) main concepts (entity types) for the instabilities, with detailed data comprising common and specific attributes, 2) storage of multimedia (mainly photo from field work, but possible video as well), 3) knowledge base with cost of activities, including catalogue per each entity type with job type and description, unit and total price, maintenance type (regular maintenance, rehabilitation, urgent maintenance), including recommending activities (system allow user to add specific activities, description and cost that is further calculated and aggregated with other data); 4) secondary data from other sources (other projects, legacy data and external resources), or from interpreted data. The data was stored in the PostgreSQL Database, a web application was developed (using PHP

The aim of our paper is to research inter- and intralinguistic similarities/differences in the use of simile rhetorical figures (Niculae and Yaneva, 2013). We aimed at differences in the source concept in different languages, as well as in different means for expressing the comparison (simile, superlatives, compounds etc.). Our research relies on four bilingual aligned corpora containing mostly literary texts and involving English, French (Stankovi´c et al., 2017), German (Andonovski et al., 2019) and Italian (Periši´c et al., 2023) as one of the languages and Serbian as the other.

In this paper we present a model for selection of good dictionary examples for Serbian and the development of initial model components. The method used is based on a thorough analysis of various lexical and syntactic features in a corpus compiled of examples from the five digitized volumes of the Serbian Academy of Sciences and Arts (SASA) dictionary. The initial set of features was inspired by a similar approach for other languages. The feature distribution of examples from this corpus is compared with the feature distribution of sentence samples extracted from corpora comprising various texts. The analysis showed that there is a group of features which are strong indicators that a sentence should not be used as an example. The remaining features, including detection of non-standard and other marked lexis from the SASA dictionary, are used for ranking. The selected candidate examples, represented as featurevectors, are used with the GDEX ranking tool for Serbian candidate examples and a supervised machine learning model for classification on standard and non-standard Serbian sentences, for further integration into a solution for present and future dictionary production projects.

Rudarska proizvodnja obuhvata kompleksne tehnološke sisteme, što nameće potrebu za uspostavljanjem i unapređivanjem sistema upravljanja rizikom. Heterogenost i obim podataka neophodnih za upravljanje rizikom zahtevaju sistem koji ih na fleksibilan način integriše i omogućava njihovo optimalno korišćenje. Osnovni cilj ove disertacije je razvoj ontologije za domen rudarstva i na njoj zasnovanog modela za upravljanje rizikom. Njegova realizacija podrazumeva i implementaciju algoritama ekstrakcije informacija za popunjavanje ontologije, kao i odgovarajuće softversko rešenje. Razvoj modela obuhvata i značajno proširenje rudarskog korpusa, kao i kreiranje terminološke baze podataka, realizovano korišćenjem metoda računarske lingvistike i korpusa dokumenata iz oblasti rudarstva (planova, izveštaja, zakona, udžbenika i monografija). Korišćena je i deskriptivna metoda za sistematizaciju podataka, zatim konačni automati i statističke analize za ekstrakciju informacija, kao i komparativna i analitička istraživačka metoda za vrednovanje i interpretaciju dobijenih rezultata. Za razvoj modela korišćeni su alati informacionih tehnologija: UML za modeliranje koncepata , OWL za razvoj ontologije, SWRL pravila za mehanizam zaključivanja, upitni jezici CQL nad korpusom i SPARQL nad ontologijom . Rezultati istraživanja pokazuju da je moguće formalizovati informacije i znanje o rizicima u rudarstvu, te razviti model koji će unaprediti efikasnost upravljanja rizikom i pomoći menadžmentu rudnika u donošenju odluka o primeni mera za smanjenje uticaja rizika identifikovanih u rudniku. Ostvarenjem ciljeva ove disertacije dat je doprinos povećanju efikasnosti u identifikaciji, analizi i reagovanju na rizik kroz izgradnju specifične domenske ontologije za rizike u rudarstvu., Mining production involves complex technological systems, which calls for the need to create and improve risk management systems. The heterogeneity and scope of data necessary for risk management require a system that integrates them in a flexible way and enables their optimal use. The main goal of this dissertation is to develop an ontology for the mining domain and a risk management model based on it. Its realization includes the implementation of information extraction algorithms for improving the ontology, as well as an appropriate software solution. The development of the model includes a significant expansion of the mining corpus, as well as the creation of a terminological database, realized using methods of computational linguistics and a corpus of documents from the mining domain (plans, reports, laws, textbooks and monographs). For systematization of data a descriptive method was used, finite automata and statistical analyzes for information extraction, and comparative and analytical research methods for evaluation and interpretation of the obtained results. Information technology tools were used for model development: UML for concept modeling, OWL for ontology development, SWRL rules for inference mechanism, query languages CQL for corpus and SPARQL for ontology. The research results show that it is possible to formalize information and knowledge about risks in mining and develop a model that will improve the efficiency of risk management and assist mine management in making decisions on implementing measures to reduce the impact of risks identified in a mine. Achieving the goals of this dissertation has contributed to increasing efficiency in identification, analysis and response to risk by developing a specific domain ontology for risks in mining.

Abusive speech in social media, including profanities, derogatory and hate speech, has reached the level of a pandemic. A system that would be able to detect such texts could help in making the Internet and social media a better and more respectful virtual space. Research and commercial application in this area were so far focused mainly on the English language. This paper presents the work on building AbCoSER, the first corpus of abusive speech in Serbian. The corpus consists of 6,436 manually annotated tweets, out of which 1,416 were labelled as tweets using some kind of abusive speech. Those 1,416 tweets were further sub-classified, for instance to those using vulgar, hate speech, derogatory language, etc. In this paper, we explain the process of data acquisition, annotation, and corpus construction. We also discuss the results of an initial analysis of the annotation quality. Finally, we present an abusive speech lexicon structure and its enrichment with abusive triggers extracted from the AbCoSER dataset.

In this paper we discuss the project of digitization of the Dictionary of the Serbo-Croatian Standard and Vernacular Language. Scanning and character recognition were a particular challenge, since various non-standard character set encoding was used in the course of the almost 60-year long production of the dictionary. The first aim of the project was to formalize the micro-structure of the dictionary articles in order to parse the digitized text of and transform it into structured data stored in relational lexical database. This approach is compatible with several standard structured forms and ontologies (TEI, LMF, Ontolex, LexInfo). A lexical database model was designed in compliance with these structured forms, following mostly the lemon model. Mapping of the lexical entry markers to LexInfo and TEI enabled export of the lexical data to the mentioned formats. A software solution for the dictionary text analysis, parsing and lexical database population was developed and tested on the first and the last published volumes of the dictionary (which contain 27,141 articles in total). An evaluation of the results shows that the developed model and software solution can be successfully used for the other volumes as well.

Паралелни корпуси представљају врсту вишејезичних корпуса који су последњих деценија постали изузетно значајни у области обраде природних језика (Natural Language Processing - NLP) и један од важнијих ресурса за истраживаче у различитим областима лингвистике и сродним језичким дисциплинама. Са израдом ове докторске дисертације започет је рад на паралелном српско-немачком корпусу књижевних текстова, СрпНемКор. У току рада на дисертацији обрађено је четрнаест романа написаних у другој половини 20. и првој половини 21. века на српском и немачком језику. За садржај корпуса одабрано је седам романа оригинално написаних на српском и њихови еквиваленти на немачком језику и седам романа оригинално написаних на немачком (четири припадају аустријској књижевности, три припадају немачкој књижевности) и њихови еквиваленти на српском језику. У дисертацији је детаљно анализиран поступак прикупљања и одабира материјала за корпус, затим обрада текстова применом расположивих језичких алата и ресурса за оба језика, као и паралализација коришћењем одговарајућег софтвера. Произведени паралелни корпус књижевних текстова, СрпНемКор, смештен је у дигиталну библиотеку Библиша која омогућава двојезичну претрагу комплетног текста паралелних колекција уз могућност морфолошког и семантичког проширење упита позивањем различитих лексичких и терминолошких ресурса. У дисертацији је анализирана могућност семантичког проширења упита заснована на синонимима позивањем терминолошке базе Терми. Терми је терминолошка вишејезична база која подржава развој терминолошких речника из различитих домена (математика, рачунарство, рударство, библиотекарство, рачунарска лингвистика и многи други) а до сада је омогућавала проширење упита само на српском и енглеском језику. На основу екстракције лексичких јединица из паралелне колекције СрпНемКор, база Терми је допуњена новим лексичким јединицама на српском, њиховим еквивалентима на немачком језику као и синонимима. Добијена листа преводних парова је искоришћена и VI, Aligned multilingual corpora have become essential resources in multilingual Natural Language Processing (NLP) in the last decades, as well as one of the major resources for researchers in various areas of linguistics and related language disciplines. This doctoral dissertation presents a new aligned Serbian-German literary corpus, SrpNemKor, for which fourteen novels written in the second half of the 20th and the first part of the 21st century were selected. From selected novels, seven are originally written in Serbian and have their equivalents in the German language, and other seven are originally written in German (four belong to Austrian literature, three belong to German literature) and have their equivalents in the Serbian language. In the dissertation in detail is analyzed the process of the collection and selection of novels for the corpus, then texts processing using available language tools and resources for both languages, as well as parallelization using appropriate software. The new aligned corpus, SrpNemKor, is stored in digital library Bibliša. Bibliša enables bilingual full-text search of aligned collections with the possibility of morphological and semantic query expansion by invoking appropriate lexical and terminological resources. In this research, it was specifically analyzed the possibility of semantic extension of the search queries based on synonyms by invoking the terminological database Termi. Termi is a multilingual database launched as a support for the development of terminological dictionaries in various domains (mathematics, computer science, mining, library science, computational linguistics, etc.). Until now Termi supported only the processing and representation of terms in Serbian and English. Based on the extraction of lexical units from the SrpNemKor, Termi is enriched with new lexical units in Serbian and their German equivalents with their synonyms which enables Bibliša to expand queries in German as well. The obtained list of

Aligning senses across resources and languages is a challenging task with beneficial applications in the field of natural language processing and electronic lexicography. In this paper, we describe our efforts in manually aligning monolingual dictionaries. The alignment is carried out at sense-level for various resources in 15 languages. Moreover, senses are annotated with possible semantic relationships such as broadness, narrowness, relatedness, and equivalence. In comparison to previous datasets for this task, this dataset covers a wide range of languages and resources and focuses on the more challenging task of linking general-purpose language. We believe that our data will pave the way for further advances in alignment and evaluation of word senses by creating new solutions, particularly those notoriously requiring data such as neural networks. Our resources are publicly available at https://github.com/elexis-eu/MWSA., Aligning senses across resources and languages is a challenging task with beneficial applications in the field of natural language processing and electronic lexicography. In this paper, we describe our efforts in manually aligning monolingual dictionaries. The alignment is carried out at sense-level for various resources in 15 languages. Moreover, senses are annotated with possible semantic relationships such as broadness, narrowness, relatedness, and equivalence. In comparison to previous datasets for this task, this dataset covers a wide range of languages and resources and focuses on the more challenging task of linking general-purpose language. We believe that our data will pave the way for further advances in alignment and evaluation of word senses by creating new solutions, particularly those notoriously requiring data such as neural networks. Our resources are publicly available at https://github.com/elexis-eu/MWSA.