• Publicationsclose
• Research dataclose
• Research softwareclose
• 2014-2023close
• Open Access close
• Article close
• Slovenščina 2.0: Empirične, aplikativne in interdisciplinarne raziskave close

Z jezikom nismo vedno zmožni neposredno ubesediti vsega, kar mislimo, zato za razlago pojavnosti uporabljamo različne jezikovno-kognitivne postopke, med drugim metafore in metonimije. Prepoznavanje vrednosti in razširjenosti metaforičnih in metonimičnih izrazov v jeziku je v zadnjih dvajsetih letih vodilo k povečanemu zanimanju za sistematično identifikacijo in luščenje tovrstnih figurativnih izrazov v korpusih posameznih jezikov. Izraze, pri katerih potekajo konceptualne preslikave, ki sodelujejo pri metaforičnih in metonimičnih procesih, je namreč težko izluščiti iz korpusa, ki niso posebej označeni za namene raziskovanja figurativnega jezika. V članku opredelim razumevanje konceptualne metafore in konceptualne metonimije, predstavim najpogostejše metode luščenja metaforičnih in metonimičnih izrazov iz jezikovnih korpusov ter na primeru korpusa g-KOMET, ki je ročno označen za metaforične izraze in metonimične prenose, ponazarjam poskus sistematizacije nekaterih najbolj prisotnih metonimičnih prenosov v slovenskem govorjenem jeziku.

Pričujoča tematska številka revije Slovenščina 2.0 se posveča digitalnemu jezikoslovju, hitro rastočemu interdisciplinarnemu področju raziskav na stičišču tradicionalnega jezikoslovja, informacijskih tehnologij in družboslovnih ved. V ospredju digitalnojezikoslovnih raziskav je ohranjanje, analiza in uporaba jezikovnih podatkov, digitalnih artefaktov z jezikom kot nosilcem medčloveškega sporazumevanja. Digitalno jezikoslovje tako pri nas kot po svetu postaja vse pomembnejše ne samo v akademskih in izobraževalnih krogih, temveč tudi v javnem in zasebnem sektorju, ki za uspešno delovanje v sodobni družbi in gospodarstvu vse bolj potrebujeta strokovnjake, vešče upravljanja z digitalnimi jezikovnimi podatki.

Namen prispevka je predstaviti raziskavo preverjanja razumljivosti nerevidiranih strojno prevedenih spletnih besedil. Primarni udeleženci v raziskavi so bili splošni bralci in ne izurjeni prevajalci ali popravljalci strojnih prevodov. Gre za prvo tovrstno raziskavo, ki je bila izvedena za slovenski jezik. Cilj raziskave je bil preveriti, v kolikšni meri so nerevidirani strojni prevodi razumljivi splošnemu bralstvu, pri čemer sem se posvetil tudi vplivu besedilnega in slikovnega konteksta. Preverjal sem prevode prevajalnikov Google Translate in eTranslation. Raziskava je bila izvedena z anketo, v kateri so udeleženci odgovarjali na vprašanja, ki so preverjala razumevanje spremljajočega besedilnega segmenta, v katerem je bila napaka. Rezultati nudijo vpogled v trenutno stopnjo razvoja strojnih prevajalnikov, ne z vidika storilnosti pri njihovem popravljanju, ampak z vidika, koliko jih razume ciljno bralstvo. Na koncu članka nudim novo evalvacijo izvornih segmentov, ki sem jih v začetku leta 2023 ponovno prevedel, tokrat še s prevajalnikom DeepL.

This paper is a follow-up and elaboration of the paper published in the JTDH 2022 Conference Proceedings on manual semantic annotation of named entities based on a proposed set of annotations for a corpus of modernist literary texts. We first briefly describe the corpus and introduce the annotation scheme, then focus on the results of additional analyses, and conclude with further challenges and issues we identified with respect to established NER systems and practices of related projects. Overall, we identify several categories of proper names, foreign language elements, and bibliographic citations, but focus here on the challenges of annotating names of literary characters and place names, and provide examples of the results of preliminary analyses of these entities in the corpus.

Developing effective question answering (QA) models for less-resourced languages like Slovene is challenging due to the lack of proper training data. Modern machine translation tools can address this issue, but this presents another challenge: the given answers must be found in their exact form within the given context since the model is trained to locate answers and not generate them. To address this challenge, we propose a method that embeds the answers within the context before translation and evaluate its effectiveness on the SQuAD 2.0 dataset translated using both eTranslation and Google Cloud translator. The results show that by employing our method we can reduce the rate at which answers were not found in the context from 56% to 7%. We then assess the translated datasets using various transformer-based QA models, examining the differences between the datasets and model configurations. To ensure that our models produce realistic results, we test them on a small subset of the original data that was human-translated. The results indicate that the primary advantages of using machine-translated data lie in refining smaller multilingual and monolingual models. For instance, the multilingual CroSloEngual BERT model fine-tuned and tested on Slovene data achieved nearly equivalent performance to one fine-tuned and tested on English data, with 70.2% and 73.3% questions answered, respectively. While larger models, such as RemBERT, achieved comparable results, correctly answering questions in 77.9% of cases when fine-tuned and tested on Slovene compared to 81.1% on English, fine-tuning with English and testing with Slovene data also yielded similar performance.

In this paper, we present recent developments on a new version (v3.0) of DirKorp (Korpus direktivnih govornih činova hrvatskoga jezika), the first Croatian corpus of directive speech acts developed for the purposes of pragmatic research. The corpus contains 800 elicited speech acts collected via an online questionnaire with role-playing tasks, a method of simulated communication that is implemented under pre-set conditions. This method is suitable for researching speech acts due to the ability to collect a great number of examples of such acts of equal propositional content and illocutionary purpose used in the same controlled situations. The presented situations are classified into two categories with regard to the relationship between the participants of the communication act: (1) situations involving interlocutors who are not in a familiar relationship; (2) situations involving interlocutors in a familiar relationship. Assignments of the two categories are organized into four pairs, asking respondents to share a speech act of similar propositional content. The respondents were 100 Croatian speakers, all undergraduate (63%) or graduate students (37%) of the Faculty of Humanities and Social Sciences (University of Zagreb). The corpus has been manually annotated on the speech act level, each speech act containing up to 14 features: (1) respondent ID, (2) familiarity/unfamiliarity, (3) utterance type, (4) directive performative verb in 1st person, (5) illocutionary force, (6) propositional content, (7) T/V form, (8) exhortative, (9) lexical marker of request, (10) lexical marker of apology, (11) lexical marker of gratitude, (12) honorific title, (13) grammatical mood, and (14) modal verb in 2nd person. It contains 12,676 tokens and 1,692 types. The corpus is encoded according to the TEI P5: Guidelines for Electronic Text Encoding and Interchange, developed and maintained by the Text Encoding Initiative Consortium (TEI). DirKorp is available for download under the CC BY-SA 4.0 license from GitHub in TEI format. We describe applied pragmatic annotation as well as the structure of the corpus.

Universal Dependencies (UD) je mednarodno usklajena označevalna shema za medjezikovno primerljivo oblikoslovno in skladenjsko označevanje besedil po načelih odvisnostne slovnice, ki je bila ob več kot 130 drugih svetovnih jezikih uspešno uporabljena tudi za označevanje besedil v slovenščini. V prispevku predstavimo rezultate nedavnih aktivnosti v povezavi s shemo UD znotraj projekta Razvoj slovenščine v digitalnem okolju, v okviru katerega smo obstoječo infrastrukturo nadgradili s prenovo in podrobno dokumentacijo označevalnih smernic UD za slovenščino, razširitvijo drevesnice SSJ-UD za pisno slovenščino z novimi povedmi iz korpusov ssj500k in ELEXIS-WSD, izdelavo testne množice iz besedil korpusa SentiCoref za spletni portal SloBENCH ter polavtomatsko pretvorbo oblikoslovnih oznak referenčnih učnih korpusov SUK in Janes-Tag. Na razširjeni drevesnici SSJ-UD je bil naučen tudi novi napovedni model za skladenjsko razčlenjevanje v orodju CLASSLA-Stanza, ki ga v prispevku v podporo nadaljnjim jezikoslovnim aplikacijam podrobneje ovrednotimo z vidika splošne natančnosti razčlenjevanja in najpogostejših tipov napak.

Večina sodobnih sistemov za strojno prevajanje temelji na arhitekturi nevronskih mrež. To velja za spletne ponudnike strojnega prevajanja, za raziskovalne sisteme in za orodja, ki so lahko v pomoč poklicnim prevajalcem v njihovi praksi. Čeprav lahko sisteme nevronskih mrež uporabljamo na običajnih centralnih procesnih enotah osebnih računalnikov in strežnikov, je za delovanje s smiselno hitrostjo potrebna uporaba grafičnih procesnih enot. Pri tem smo omejeni z velikostjo slovarja, kar zmanjšuje kakovost prevodov. Velikost slovarja besednih enot je še posebej pereč problem visoko pregibnih jezikov. Rešujemo ga z uporabo podbesednih enot, s katerimi dosežemo večjo pokritost jezika. V članku predstavljamo različne metode razcepljanja besed na podbesedne enote z različno velikimi slovarji in primerjamo njihovo uporabo v strojnem prevajalniku za jezikovni par slovenščina-angleščina. V primerjavo vključujemo še prevajalnik brez razcepljanja besed. Predstavljamo rezultate uspešnosti prevajanja z metriko BLEU, hitrosti učenja modelov in hitrosti prevajanja ter velikosti modelov. Dodajamo pregled praktičnih vidikov uporabe podbesednih enot v strojnem prevajalniku, ki ga uporabljamo skupaj z orodji za računalniško podprto prevajanje.

 The present moment raises many questions about the workings and resilience of parliamentary democracy in Western-type democracies, including the former socialist states of the East Central European region, where various forms of populism and illiberal democracy are taking shape. Among these, Slovenia is taken as a case study, since it is not only a former socialist state, but was also for a long time acknowledged as a post-socialist success story. Focusing on the central state institution in systems of parliamentary democracy, i.e. the parliament, and its members (MPs) this paper considers speech as performed during parliamentary sessions by MPs from populist and non-populist political parties between the years 1992 and 2018, the period of a fully democratic Slovene national parliament. It combines the methodological approaches of cultural history with corpus linguistics in order to map any possible differences in populist and non-populist discourse of MPs. Special attention is given to situations where MPs mentioned the public, thus testing the hypothesis that populist MPs engage more with the public as a part of their populist political style.