The Ukrainian е-anecdotes corpus was created in order to describe structural and linguistic features of Ukrainian e-anecdotes. It contains 500 anecdotes, its volume is 18,582 tokens. Corpus-driven approach with Sketch Engine corpus management allowed to describe some linguistic and quantitative characteristics of Ukrainian e-anecdotes, to identify rare/unusual words in the corpus of e-anecdotes and interpret them, to analyze the keyword collocations and to determine linguistic and genre features of Ukrainian anecdotes, as well as to identify the ethnonymic collocations as the ethnic stereotypes markers. Corpus, the Ukrainian language е-anecdote, frequency, score, Skecth Engine COLINS-2021: 5th International Conference on Computational Linguistics and Intelligent Systems, April 22-23, 2021, Kharkiv, Ukraine
Despite the fact that since the appearance of the term “anecdote” in 550 in Byzantium (the historians
P. Caesarea’s book “Secret History”), many works have been written, the anecdotes genre and linguistic
features study in different languages continues. Ukrainian language studies are not rich in research:
Ukrainian anecdotes are considered from the linguistic, folklore, literary studies, ethnopsychology point
of view [
We have analyzed all the open corpora developed for the Ukrainian language (section 2.1), but none of them contains the anecdotes’ subcorpus, and accordingly, there are no anecdotes’ corpus-based studies, what emphasizes the urgency of this work.
The purpose of our article is to create a corpus of e-anecdotes and to show the corpus capabilities in the study as well as the structural and linguistic features of Ukrainian e-anecdotes through their quantitative parameters prism. This corpus-driven approach is used in the study, which is aimed to interpret corpus data as a whole [12].
First we will present the existing resources and NLP tools developed for the Ukrainian language or those that can be used (section 2) for this purpose, and then introduce the Ukrainian е-anecdotes corpus (section 3) as
well its exploitation for the Ukrainian e-anecdotes linguistic and quantitative characteristics descrition (section 4).
2021 Copyright for this paper by its authors.
First we offer a short review of the open Ukrainian corpora and some NLP tools. 2.1.
Ukrainian web corpus of the University of Leipzig (2019, 1 billion tokens) [13]. In the search results there are the examples, compatibility, three-dimensional connectivity graph. However, the search is possible only by the word form.
Ukrainian language corpus of the “Chtivo” library (600 million words, 6.6 GB of Ukrainian texts from the electronic “Chtivo” library) [14] – these are automatically recognized fiction, scientific, journalistic texts, that do not have the errors correction and corpus annotation.
“Ukrainian laboratory” [15] consists of: web body of Internet texts with syntactic markup of almost 3 billion tokens, “Golden corpus” – it is a manually annotated corpus of the various styles texts with removed homonymy (140 thousand tokens); parallel corpora of fiction (6 million tokens).
Corpora of Ukrainian texts (authors: Dmytro Chaplynsky and Vsevolod Dyomkin) [16] contain The corpus of NER-annotations (has 229 texts from the Ukrainian Brownian corpus of 217,381 tokens from 6,751 annotated named entities), Corpus UberText that is a corpus of Ukrainian periodicals (over 6 GB) where Andriy Rysin and BRUK initiative nlp-uk package was used for texts tokenization and lemmatization. Corpus of laws and legal acts – corpus (over 9 GB) with tokenization and lemmatization as well as the Word Embedding models.
Corpus of the Ukrainian language on the linguistic portal Mova.info (2003–2020) [17] developed in the computer linguistics laboratory of the Institute of Philology of Taras Shevchenko National University of Kyiv under the prof. N. Darchuk direction. The corpus contains the following subcorpora: legislative texts, scientific texts, poetic language, journalism, folklore texts, fiction. It has great functionality.
General Regionally Annotated Corpus of Ukrainian (GRAC) [18] – corpus of texts annotated by regions, containing more than 650 million tokens (v. 10). Authers: M. Shvedova, R. Von Waldenfels, S. Yarygin, A. Rysin, V. Starko, M. Wozniak, M. Kruk and others. The corpus covers the period from 1816 to 2020 and contains more than 900,000 texts by about 26,000 authors. Texts have the following meta-marking: by styles, themes, genres, original language, dating, spelling, information about the author of the text, information about the media, by regions, by the text source. The humor is presented among the genres (search by tag: <doc genre="HUM"/>), but there no anecdotes among the texts. The corpus has a linguistic annotation: morphological (works on the morphological analysis system basis developed by the r2u group specialists Andriy Rysin, Vasyl Starko and others), grammatical homonymy (removal of homonymy is done manually, but inconsistently), semantic for 3000 most frequent words [19, 20].
Grammatical Error Correction and Fluency Corpus for the Ukrainian Language [21] – this is a new and ambitious project from Grammarly. Now UA-GEC containes 328 779 tokens.
The Ukrainian Brown Corpus [22] is under development (authors: Vasyl Starko, Andriy Rysin and others). Three more corpora, that are under development, can be mentioned: The Medical Corpus in Ukrainian (UKRMED) [23], The multilingual aligned corpus with Ukrainian as the target language [24] and Ukrainian corpus WikiWars-UA [25]. The last one is annotated with temporal information and also proposes a comparison with the English version of annotations. 2.2.
To perform the objectives of this study, we are going to analyze the free lexicons, PoS-taggers and parsing tools developed for the Ukrainian language or another Slavic inflectional language, useful for our work.
MULTEXT-East free lexicons 4.0 is a morphosyntactic lexicons resource which also contains Ukrainian lexicon (300292 entries) [26], which can be used in AntConc for Keyword List.
Mystem+ is an open source of automatic morphological analysis for Russian such as: TreeTagger, Mystem, TnT, Hunpos and MarMoT. There are also exist the comparative study of automatic morphological analysis for Russian [27]. Asiryan A. K. described a comparison of morphological analyzers TreeTagger, MyStem, TnT, pymorphy2 and FreeLing for three Russian corpora [28].
FreeLing is an open source language analysis tool suite providing language analysis functionalities (morphological analysis, named entity detection, PoS-tagging, parsing, Word Sense Disambiguation, Semantic Role Labelling, etc.) for a variety of languages (English, Spanish, Portuguese, Italian, French, German, Russian, Catalan, Galician, Croatian, Slovene , among others) [29]. Ukrainian is not on the list, however FreeLing version 4.2 copes well with parsing as for the Russian language texts.
The greatest interest for the anecdotes’ dialogic speech analysis is Dialogue Annotation and Research Tool (DART) Version 3.0, author: Martin Weisser (May 2019). It is a research environment designed to allow you to annotate and analyse single or multiple dialogues in batch mode, with the ultimate aim to identify speech acts automatically, and thereby create pragmatically annotated corpora. The version 3 identifies 162 speech acts automatically, and also has a number of additional functions and improvements to both the interface and the output options from within the individual analysis modules, as well as a completely new pattern counting facility [30]. The DART Taxonomy version 3.0 is designed for English and consist of 2 parts: speech-act label and function.
Project to generate POS tag dictionary for Ukrainian language. For all files in data/dict the project generates all possible word forms with POS tags by using affix rules from files in data/affix [31].
AntConc [32] is a freeware corpus analysis toolkit for concordancing and text analysis. It is good for Ukrainian but we had to append the Ukrainian alphabet to Token Definition Settings.
In reality, there exist only few resources for studying the Ukrainian language. The most suitable for our tasks is Sketch Engine [33]. It is a commercial corpus management and corpus query software with user friendly interface. It has no PoS-tagging for Ukrainian but can perform parsing.
Collection of Ukrainian e-anecdotes is composed of 500 texts by random sampling from various Ukrainian-language sites. All texts were normalized (for example, the different codes use in a set of words is detected mама≠мама, hіяк≠ніяк, hа≠на (mother, any, take), typing errors (розмляють instead of розмовляють/talk). Further corpus exploitation is done with the Sketch Engine tools. Our corpus statistics is given on the Figure 1.
The WordList option allows you to find the most frequent tokens in anecdotes (and further lemmatization will allow you to identify lemmas). According to A. Shmelev, an anecdote is a formal dialogue, because at the beginning the sender asks a question, but without waiting for the answer from the addressee, he answers it himself [34]. The frequency list (Figure 2) shows the high frequency of personal pronouns different forms (я, ти, мене, він, ви, мені, його, тебе/ I, you, me, he, you, me him, you). This is in line with the dialogue format. Parsing allows to determine the sentences quantitative parameters by intonation (Table 1). Absolute 1007 560 427 138 8 frequency
Relative 46,5 25,9 19,7 1,1 6,4 0,4 frequency,
%
Pseudo-sentences consist only of punctuation marks (an ellipsis or several exclamation marks) and do not contain verbal components. As you can see, half of the sentences in the corpus are intonationally and emotionally colored. 4.2.
Sketch Engine comprises only one Ukrainian corpus. It is Ukrainian Web 2014 (ukTenTen14) amounting to 1,388,494,043 tokens. So, it can reflect the general language trends. It was chosen as the reference corpus. The anecdotes corpus is a focus corpus. If a word with frequency 1 in focus corpus has frequency 0 in the reference corpus, it is considered as rare or unusual word (Figure 3). In order to perform this task advanced settings were used.
As a result, 997 such tokens (5%) were detected. It: surzhykisms (валідольчік, балуєшся, лавірував/validol, roistering, showed dexterity) dialectisms (Йсусу, вдівся/ Jesus, dressed) slang vocabulary (крякнути/to crack) graphically unstable neologisms (смска, смс-ку/sms) the accusative case forms (Романівно, доцю, куме, Русланчику/Romanivno, daughter, crony, Ruslanchiku) Ukrainian graphic forms of Russian words (московскоє/moskow) specific exclamations (уіііі, шшшшш/ уіііі,shshshshsh) the reproduction of emphasis (тааат, га-а-ади-и-и, мааам, о-о-ооо, уф-ф-ф, таак/dadyyy, gaa-adi-i-i, maaam, o-o-ooo, uf-f-f, yeees) the reproduction of the character's speech in a state of altered consciousness (for example: А коли ти зрозумів, що вона п'яна? — Коли прийшла СМС “Повзони мені, а то я не можу найти свій тефелоон” / And when did you realize that she was drunk? — When the SMS came “Coull me, and for I can not find my tefeloon”) the reproduction of the character’s distorted articulation (фклітинку/flaid – children's speech; мяфко, фамо, прифкакало/meat, itself, jumped – consonant changes because of a stuffed mouth, дддома, сість, масиніст/at home, six, train draiver – due to physical condition) the reproduction of Ukrainian words pronunciation by a Russian-speaking character (хлопци, останівка/ guys, stop) barbarisms (lakalut, dont) words used in a certain grammatical form only one time (комарів, устриць/mosquitoes, oysters).
Thus, the anecdote as a conversational genre reflects the full oral non-formal speech linguistic means range. 4.3.
With the same tool but on the different criteria words which are very frequent in the focus corpus were determined, what allows to compare with the frequency in the referent corpus (Figure 4). In such case it is relevant to compare the frequency per million.
Comparison of the sorting by frequency results in the focus corpus and separately in the reference corpus revealed that in the reference corpus the first 18 positions are occupied by prepositions and conjunctions, and in the focus corpus by prepositions, conjunctions, personal pronouns and exclamations. Thus, the anecdote as a conversational genre is characterized by a high frequency of expressive means and personal pronouns, the use of which creates the narrator-eyewitness effect for the events depicted in the anecdote.
In the obtained result, all nouns (N) and verbs (V) with a frequency of 10 were marked with the help of MS Excel tools, semantically close (запитує=питає/asks) were lemmatized and combined. Table 2 represents the most frequent nouns and verbs list.
The anecdote is of an exclusively anthropocentric direction. Ukrainian anecdotes are thematically diverse: anecdotes about the family, friends and godparents, neighbors, school and university, professions, about Vovochka (they became especially relevant due to an allusion to the leader of the Russian Federation), personalized animals, gender roles, different ethnic groups. The most frequent keywords indicate the anecdotes about family and gender predominance in the corpus. 4.4.
The use of concordance for the most frequency keywords made it possible to identify some linguistic features. For example, the predominance of the accusative case for the token кум/crony (Figure 5). Accordingly, it appears in the speech of the characters. For each concordance we verified collocations with such criteria as: range from –7 to 7, custom range, minimum frequency in corpus is 1, minimum frequency in a given range is 3. It allows to reveal paired characters: дружина — чоловік, мама — тато/wife – husband, mom – dad, etc., since these tokens have the highest LogDice as a statistic score for identifying collocations. Figure 6 shows the collocations search results for the lexeme чоловік/husband.
Сollocation with ethnonyms as ethnic stereotypes markers
Stereotypes are the different assumptions and standardized norms put by the society on people belonging to certain groups, based on their gender, race, ethnicity or age etc. These opinions or assumptions do not correspond to the truth as they discriminate the individuals without any regard for their feelings or emotions [35]. There are exist stereotypes of different nature: sexual, gender, racist, ethnic, group.
V. Samokhina distinguishes between two types of ethnic jokes: that of the first type describe one ethnic group, of the second – several. The latter often depict other ethnic groups on the negative side [36]. In addition, the factor whether a person characterizes himself/herself (autostereotype) or another ethnic group or ethnic groups (heterostereotype) should be considered.
The ethnonym українець/Ukrainian was found among the frequency KeyWords. This indicates the predominance in the corpus of anecdotes with the autostereotype, however, perhaps in contrast to other ethnic groups. In order to find out with which ethnocharacters Ukrainians most often correlate in anecdotes, we have used the Typically Score (Figure 7). It shows how strong the collocation is. Obtained results for minimum score 5 showed only one case:
Simple search in Concordance allows to find in anecdotes the following ethno-characters, which are compared with Ukrainians: грузин, китаєць, японець, француз/Georgian, Chinese, Japanese, French. Anecdotes about one ethnic group present the following ethnic characters: американець, єврей, москаль/American, Jew, Muscovite.
Created Ukrainian e-anecdotes corpus contains 500 anecdotes, body volume 18 582 tokens, parsing is done by sentences. Corpus-driven approach with Sketch Engine tools allowed to identify the following linguistic and genre characteristics of the anecdote within the corpus: • various forms personal pronouns’ high frequency, due to the anecdotes’ dialogic form • in terms of intonation, about 50% of sentences are intonationally and emotionally colored • anecdote as a conversational genre reflects the full range of oral non-formal speech linguistic means • KeyWords in the corpus testify the anthropocentric direction of anecdotes and the most frequent keywords indicate the family and gender anecdotes predominance in the corpus
• analysis of collocations by LogDice of keywords revealed some features: the predominance of the accusative case for the lexeme кум/crony, paired characters: wife – husband, mother – father, collocations of the ethnonym Ukrainian in the e-anecdotes corpus correlates with the word German.
So as GRAC is the largest in volume and functionality corpus and gives the possibility to create your own subcorpus of selected texts, it was chosen to download the created Ukrainian e-anecdotes corpus for future research (see GRAC-11). This is the practical significance of this work on creating the eanecdotes corpus.
As the GRAC corpus has a semantic meta-markup that gives the possibility for proper names search, the onym space of Ukrainian anecdotes can be explored in the future. 6. References [12] E. Tognini-Bonelli, Corpus Linguistics at Work, John Benjamins Publishing Amsterdam–
Philadelphia, 2001. [13] Ukrainian web corpus of the University of Leipzig. URL: http://corpora.informatik.unileipzig.de/de?corpusId=ukr_mixed_2014 [14] Ukrainian language corpus of the “Chtivo” library. URL: http://korpus.org.ua/ [15] Ukrainian laboratory. URL: https://mova.institute [16] Corpora of Ukrainian texts. URL: https://lang.org.ua/uk/corpora/ [17] Corpus of the Ukrainian language. URL: http://www.mova.info/corpus.aspx?l1=209 [18] General Regionally Annotated Corpus of Ukrainian (GRAC). URL: uacorpus.org [19] M. Shvedova, The General Regionally Annotated Corpus of Ukrainian (GRAC, uacorpus.org): Architecture and Functionality, in: Proceedings of the 4th International Conference on Computational Linguistics and Intelligent Systems (COLINS 2020), Volume I: Main Conference, Lviv Ukraine, 2020, pp. 489–506. [20] V. Starko, Semantic Annotation for Ukrainian: Categorization Scheme, Principles, and Tools, in: Proceedings of the 4th International Conference on Computational Linguistics and Intelligent Systems (COLINS 2020), Volume I: Main Conference, Lviv Ukraine, 2020, pp. 239–248. [21] Grammatical Error Correction and Fluency Corpus for the Ukrainian Language. URL: https://github.com/grammarly/ua-gec [22] The Ukrainian Brown Corpus. URL: https://github.com/brown-uk/corpus [23] O. Cherednichenko, O. Kanishcheva, O. Yakovleva, D. Arkatov, Collection and Processing of a Medical Corpus in Ukrainian, in: Proceedings of the 4th International Conference on Computational Linguistics and Intelligent Systems (COLINS 2020), Volume I: Main Conference, Lviv Ukraine, 2020, pp. 272-282. URL: http://ceur-ws.org/Vol-2604/paper21.pdf [24] N. Grabar, T. Hamon, Creation of a multilingual aligned corpus with Ukrainian as the target language and its exploitation, in: Proceedings of the 1th International Conference on Computational Linguistics and Intelligent Systems (COLINS 2017), Kharkiv Ukraine. URL: http://colins.in.ua/wp-content/uploads/2017/12/COLINS-2017_conference_proceedings.pdf [25] N. Grabar, T. Hamon, WikiWars-UA: Ukrainian corpus annotated with temporal expressions, in: Proceedings of the 3d International Conference on Computational Linguistics and Intelligent Systems (COLINS 2019), volume II: Workshop, Kharkiv Ukraine, 2019. URL: http://colins.in.ua/wp-content/uploads/2017/04/preface_volume2_2019.pdf [26] T. Erjavec, MULTEXT-East Version 4: Multilingual Morphosyntactic Specifications, Lexicons and Corpora, in: the Proceedings of the Seventh International Conference on Language Resources and Evaluation, LREC'10 ELRA, Paris, 2010. [27] O. V. Dereza, D. A. Kayutenko, A. S. Fenogenova, Automatic Morphological Analysis for Russian: a Comparative Study, in: Computational Linguistics and Intellectual Technologies: Proceedings of the International Conference “Dialogue”, 2016. URL: http://www.dialog21.ru/media/3473/dereza.pdf [28] A. K. Asiryan, Sravneniye instrumentov morfologicheskoy razmetki [Comparison of tools for morphological marking], Nauchnyy vzglyad v budushcheye. 1.7: Tekhnicheskiye nauki (2017) 27–34. URL: https://doi.org/10.30888/2415-7538.2017-07-01-027 (in Russian) [29] L. Padró, E. Stanilovsky, FreeLing 3.0: Towards Wider Multilinguality, in: Proceedings of the Language Resources and Evaluation Conference (LREC 2012) ELRA, Istanbul Turkey, 2012.
URL: http://nlp.lsi.upc.edu/freeling/node/ [30] M. Weisser, Dialogue Annotation & Research Tool (DART). Version 3.0 (May 2019). URL: http://martinweisser.org [31] A. Rysin, V. Starko, Large Electronic Dictionary of Ukrainian (VESUM), 2005–2021. URL: https://github.com/brown-uk/dict_uk [32] AntConc. Build 3.4.4. Laurence Anthony. URL: http://www.laurenceanthony.net/software/ antconc/ [33] Sketch Engine. Corpus query and management system. URL: https://www.sketchengine.eu/nosketch-engine/ [34] A. D. Shmelev, Russkiy anekdot: Tekst i rechevoy zhanr: monografiya [Russian anecdote: Text and speech genre: monograph], Moscow, 2002. (in Russian) [35] R. Charchika, Gender stereotyping, 2020.URL:https://www.researchgate.net/publication/344727936_
Gender_Stereotypes. doi: 10.13140/RG.2.2.34843.28969. [36] V. O. Samokhina, Zhart u suchasnomu komunikatyvnomu prostori Velykoyi Brytaniyi i SSHA: tekstualʹnyy ta dyskursyvnyy aspekty [Joke in the modern communicative space of Great Britain and the United States: textual and discursive aspects] (Germanic languages), Ph.D. thesis, Taras Shevchenko National University of Kyiv, Кyiv, 2010 (in Ukrainian) [37] A. Kilgarriff, V. Baisa, J. Bušta, M. Jakubíˇcek, V. Kováˇr, J. Michelfeit, P. Rychlý, V. Suchomel, The Sketch Engine: ten years on, Lexicography, 1(1) (2014) 7–36. doi: 10.1007/s40607-014-00099.