lang
stringclasses
8 values
year
stringclasses
27 values
title_en
stringlengths
4
253
title_cs
stringlengths
0
251
abstract_en
stringlengths
49
5.12k
abstract_cs
stringlengths
33
3.92k
authors
sequencelengths
1
577
s2_url
stringlengths
79
79
title_fr
stringclasses
3 values
abstract_fr
stringclasses
3 values
title_ru
stringclasses
13 values
abstract_ru
stringclasses
11 values
title_sk
stringclasses
2 values
abstract_sk
stringclasses
2 values
title_de
stringclasses
4 values
abstract_de
stringclasses
4 values
title_dsb
stringclasses
2 values
abstract_dsb
stringclasses
2 values
title_lt
stringclasses
1 value
abstract_lt
stringclasses
1 value
en
2024
Textual Coverage of Eventive Entries in Lexical Semantic Resources
Textové pokrytí dějových položek v lexikálních sémantických zdrojích
This short paper focuses on the coverage of eventive entries (verbs, predicates, etc.) of some well-known lexical semantic resources when applied to random running texts taken from the internet. While coverage gaps are often reported for manually created lexicons (which is the case of most semantically-oriented lexical ones), it was our aim to quantify these gap.
Tato krátká práce se zaměřuje na pokrytí dějů (sloves, predikátů apod.) některých známých lexikálních sémantických zdrojů při aplikaci na náhodně běžící texty převzaté z internetu. Zatímco mezery v pokrytí jsou často hlášeny u ručně vytvořených lexikonů (což je případ většiny sémanticky orientovaných lexikálních zdrojů), bylo naším cílem kvantifikovat tyto mezery.
[ "Eva Fučíková", "Cristina Fernández-Alcaina", "Jan Hajič", "Zdeňka Urešová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
A Comparison of Human and Machine Performance in Simultaneous Interpreting as a Basis for Further Research in Automatic Simultaneous Speech Translation
Srovnání výkonu člověka a stroje při simultánním tlumočení jako základ pro další výzkum v oblasti automatického simultánního překladu řeči
Simultaneous interpreting (SI) is a cognitively difficult task which requires specialized training and continued learning to deliver the best performance. Even with full attention, occasional errors in interpreting are inevitable, and they are likely to raise in longer sessions. With the advances in deep learning and quality increase in automatic simultaneous speech translation (ASST), we can consider using them as a substitute solution if human interpreters are not available, e.g. for capacity reasons or due to financial constraints. The presented study assesses the state of the art by analysing translation departures in human SI (professional and trainee interpreters) and ASST. Two terminologically rich speeches, delivered by non-native English speakers, were remotely interpreted by two M.A. interpreting students, and one professional interpreter, to Czech. We had the speeches translated also by an in-house speech translation system. Our quality assessment focuses on content transfer from the original speech, extending Barik’s taxonomy (1971), which identifies omissions, additions, substitutions, and semantic errors, with grammatical errors and formal defects (Kopczyński 1980). Our findings reveal that trainee interpreters outperformed the professional one in multiple areas, such as avoiding over-reliance on original structure or excessive use of demonstrative pronouns. In comparison, the ASST model translated more literally, with outputs having fewer translation departures than human interpretations. However, focusing on content transfer only could ignore other qualities of human SI that are more important to the client. Future research aims to expand our sample size and evaluation criteria, incorporating more evaluators, students, and professionals, to further validate our findings.
Simultánní tlumočení (SI) je kognitivně náročný úkol, který vyžaduje specializovaný výcvik a neustálé vzdělávání, aby bylo možné podat co nejlepší výkon. I při plné pozornosti jsou občasné chyby v tlumočení nevyhnutelné a při delších sezeních se pravděpodobně zvýší. Díky pokroku v oblasti hlubokého učení a zvyšování kvality automatického simultánního překladu řeči (ASST) můžeme uvažovat o jejich využití jako náhradního řešení, pokud nejsou k dispozici lidští tlumočníci, např. z kapacitních důvodů nebo kvůli finančním omezením. Předkládaná studie hodnotí současný stav na základě analýzy posunů v překladu u lidských profesionálních tlumočníků a tlumočníků-studentů a ASST. Dva terminologicky bohaté projevy, přednesené nerodilými mluvčími angličtiny, byly na dálku tlumočeny dvěma studenty magisterského studia tlumočnictví a jedním profesionálním tlumočníkem do češtiny. Projevy jsme nechali přeložit také interním systémem pro překlad řeči. Naše hodnocení kvality se zaměřuje na přenos obsahu z originálního projevu a rozšiřuje Barikovu taxonomii (1971), která identifikuje vynechávky, přidávání, záměny a sémantické chyby, o gramatické chyby a formální nedostatky (Kopczyński 1980). Naše zjištění ukazují, že tlumočníci-studenti překonali profesionální tlumočníky v několika oblastech, například v tom, že se vyhýbali přílišnému spoléhání na původní strukturu nebo nadměrnému používání ukazovacích zájmen. Ve srovnání s tím model ASST překládal doslovněji, přičemž výstupy vykazují méně posunů v překladu. Zaměření pouze na přenos obsahu by však mohlo mít za následek, že budeme ignorovat jiné vlastnosti lidského SI, které jsou pro klienta důležitější. Cílem budoucího výzkumu je rozšířit velikost našeho vzorku a kritéria hodnocení a zahrnout více hodnotitelů, studentů a odborníků, aby bylo možné naše zjištění dále ověřit.
[ "Věra Kloudová", "Petra Vavroušová Mračková", "Dávid Javorský", "Ondřej Bojar" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Annotation and automated classification of dramatic situations
Anotace a automatizovaná klasifikace dramatických situací
In this paper, we describe the process of annotation of dramatic situations on 52 selected play scripts. We updated the list of the original Polti’s 36 situations and use 58 situations in total. We show that the task of selection and annotation of dramatic situation is hard, as annotators often disagrees. In the second part of the paper, we propose and evaluate a method of automatic detection of 5 selected situations in the play scripts.
V tomto příspěvku popisujeme proces anotace dramatických situací na 52 vybraných scénářích. Aktualizovali jsme seznam 36 situací původního Poltiho a celkem používáme 58 situací. Ukazujeme, že úkol výběru a anotace dramatických situací je těžký, protože anotátoři často nesouhlasí. V druhé části příspěvku navrhujeme a hodnotíme metodu automatické detekce 5 vybraných situací ve scénářích.
[ "David Mareček", "Marie Nováková", "Klára Vosecká", "Josef Doležal", "Tomáš Musil", "Rudolf Rosa" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
OOVs in the Spotlight: How to Inflect them?
OOV v centru pozornosti: Jak je skloňovat?
We focus on morphological inflection in out-of-vocabulary (OOV) conditions, an under-researched subtask in which state-of-the-art systems usually are less effective. We developed three systems: a retrograde model and two sequence-to-sequence (seq2seq) models based on LSTM and Transformer. For testing in OOV conditions, we automatically extracted a large dataset of nouns in the morphologically rich Czech language, with lemma-disjoint data splits, and we further manually annotated a real-world OOV dataset of neologisms. In the standard OOV conditions, Transformer achieves the best results, with increasing performance in ensemble with LSTM, the retrograde model and SIGMORPHON baselines. On the real-world OOV dataset of neologisms, the retrograde model outperforms all neural models. Finally, our seq2seq models achieve state-of-the-art results in 9 out of 16 languages from SIGMORPHON 2022 shared task data in the OOV evaluation (feature overlap) in the large data condition. We release the inflection system with the seq2seq models as a ready-to-use Python library.
Zaměřujeme se na morfologické skloňování v podmínkách mimo slovní zásobu (OOV), což je nedostatečně prozkoumaná dílčí úloha, v níž jsou nejmodernější systémy obvykle méně efektivní. Vyvinuli jsme tři systémy: retrográdní model a dva sekvenční modely (seq2seq) založené na LSTM a Transformeru. Pro testování v podmínkách OOV jsme automaticky extrahovali rozsáhlý dataset podstatných jmen v morfologicky bohaté češtině s rozdělením dat na lemmata a dále jsme ručně anotovali reálný dataset OOV neologismů. Ve standardních podmínkách OOV dosahuje Transformer nejlepších výsledků, přičemž jeho výkonnost se zvyšuje v ensamblu s LSTM, retrográdním modelem a základními modely SIGMORPHON. V reálném datovém souboru OOV neologismů dosahuje retrográdní model lepších výsledků než všechny neuronové modely. A konečně naše modely seq2seq dosahují nejlepších výsledků v 9 ze 16 jazyků z dat sdílené úlohy SIGMORPHON 2022 při hodnocení OOV (překrývání funkcí) v podmínce velkého množství dat. Inflexní systém s modely seq2seq vydáváme jako knihovnu v jazyce Python připravenou k použití.
[ "Tomáš Sourada", "Jana Straková", "Rudolf Rosa" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Multiword Expressions between the Corpus and the Lexicon: Universality, Idiosyncrasy, and the Lexicon-Corpus Interface
Víceslovné výrazy mezi korpusem a lexikonem: Univerzalita, idiosynkrazie a rozhraní mezi slovníkem a korpusem
We present ongoing work towards defining a lexicon-corpus interface to serve as a benchmark in the representation of multiword expressions (of various parts of speech) in dedicated lexica and the linking of these entries to their corpus occurrences. The final aim is the harnessing of such resources for the automatic identification of multiword expressions in a text. The involvement of several natural languages aims at the universality of a solution not centered on a particular language, and also accommodating idiosyncrasies. Challenges in the lexicographic description of multiword expressions are discussed, the current status of lexica dedicated to this linguistic phenomenon is outlined, as well as the solution we envisage for creating an ecosystem of interlinked lexica and corpora containing and, respectively, annotated with multiword expressions.
Představujeme probíhající práci na rozhraní mezi lexikonem a korpusem, které bude sloužit jako standard pro reprezentaci víceslovných výrazů (různých slovních druhů) ve specializovaných slovnících, a na propojení slovníkových hesel a jejich výskytů v korpusu. Konečným cílem je použití dotyčných zdrojů k automatickému rozpoznávání víceslovných výrazů v textu.
[ "Verginica Barbu Mititelu", "Voula Giouli", "Stella Markantonatou", "Ivelina Stoyanova", "Petya Osenova", "Kilian Evang", "Daniel Zeman", "Simon Krek", "Carole Tiberius", "Christian Chiarcos", "Ranka Stanković" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
A COST Action on Universality, Diversity and Idiosyncrasy in Language Technology
Akce COST o univerzálnosti, diverzitě a idiosynkrazii v jazykových technologiích
This paper presents the objectives, organization and activities of a scientific network dedicated to universality, diversity and idiosyncrasy in language technology. We describe the objectives and organization of this initiative, the people involved, the working groups and the ongoing tasks and activities. This paper is also an open call for participation towards new members and countries.
Tento příspěvek představuje cíle, organizaci a aktivity vědecké sítě zaměřené na univerzalitu, diverzitu a idiosynkrazii v jazykových technologiích. Popisujeme cíle a organizaci této iniciativy, komunitu do iniciativy zapojených výzkumníků, pracovní skupiny a probíhající činnosti. Tento článek je také otevřenou výzvou k účasti nových členů a zemí.
[ "Agata Savary", "Daniel Zeman", "Verginica Barbu Mititelu", "Anabela Barreiro", "Olesea Caftanatov", "Marie-Catherine de Marneffe", "Kaja Dobrovoljc", "Gülşen Cebiroğlu Eryiğit", "Voula Giouli", "Bruno Guillaume", "Stella Markantonatou", "Nurit Melnik", "Joakim Nivre", "Atul Kr. Ojha", "Carlos Ramisch", "Abigail Walsh", "Beata Wójtowicz", "Alina Wróblewska" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
UDMorph: Morphosyntactically Tagged UD Corpora
UDMorph: Morfosyntakticky označené UD Corpora
UDMorph provides an infrastructure parallel to that provided by UD for annotated corpus data that follow the UD guidelines, but do not provide dependency relations: a place where new annotated data-sets can be deposited, and where existing data-sets can be found and downloaded. It also provides a corpus creation environment to easily create annotated data for additional languages and variants. And it provides a REST and GUI interface to a growing collection taggers with a CoNLL-U output, currently for around 150 different languages, where taggers for new data-sets in UDMorph are automatically added.
UDMorph poskytuje infrastrukturu paralelní s infrastrukturou poskytovanou UD pro anotovaná korpusová data, která se řídí pokyny UD, ale neposkytují vztahy závislostí: místo, kam lze ukládat nové anotované datové sady a kde lze najít a stáhnout existující datové sady. . Poskytuje také prostředí pro vytváření korpusů pro snadné vytváření anotovaných dat pro další jazyky a varianty. A poskytuje rozhraní REST a GUI rostoucím taggerům sbírek s výstupem CoNLL-U, aktuálně pro přibližně 150 různých jazyků, kam se automaticky přidávají taggery pro nové datové sady v UDMorph.
[ "Maarten Janssen" ]
https://www.semanticscholar.org/paper/551bdf0f88166089a813afd87371dcd78eb4105f/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Polish-ASTE: Aspect-Sentiment Triplet Extraction Datasets for Polish
Polish-ASTE: Datové sady pro extrakci aspektu a sentimentu pro polštinu
Aspect-Sentiment Triplet Extraction (ASTE) is one of the most challenging and complex tasks in sentiment analysis. It concerns the construction of triplets that contain an aspect, its associated sentiment polarity, and an opinion phrase that serves as a rationale for the assigned polarity. Despite the growing popularity of the task and the many machine learning methods being proposed to address it, the number of datasets for ASTE is very limited. In particular, no dataset is available for any of the Slavic languages. In this paper, we present two new datasets for ASTE containing customer opinions about hotels and purchased products expressed in Polish. We also perform experiments with two ASTE techniques combined with two large language models for Polish to investigate their performance and the difficulty of the assembled datasets. The new datasets are available under a permissive licence and have the same file format as the English datasets, facilitating their use in future research.
Aspect-Sentiment Triplet Extraction (ASTE) je jednou z nejnáročnějších a nejsložitějších úloh v analýze sentimentu. Jedná se o konstrukci trojic, které obsahují aspekt, s ním spojenou polaritu sentimentu a názorovou frázi, která slouží jako zdůvodnění přiřazené polarity. Navzdory rostoucí popularitě této úlohy a mnoha navrženým metodám strojového učení pro její řešení je počet datových sad pro ASTE velmi omezený. Zejména není k dispozici žádný datový soubor pro žádný ze slovanských jazyků. V tomto článku představujeme dva nové datové soubory pro ASTE obsahující názory zákazníků na hotely a zakoupené produkty vyjádřené v polštině. Provádíme také experimenty se dvěma technikami ASTE v kombinaci se dvěma velkými jazykovými modely pro polštinu, abychom prozkoumali jejich výkonnost a obtížnost sestavených datových sad. Nové datové sady jsou k dispozici pod povolenou licencí a mají stejný formát souborů jako anglické datové sady, což usnadňuje jejich použití v budoucím výzkumu.
[ "Marta Lango", "Borys Naglik", "Mateusz Lango", "Iwo Naglik" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Towards a Unified Taxonomy of Deep Syntactic Relations
Na cestě ke sjednocené taxonomii hloubkově syntaktických vztahů
This paper analyzes multiple deep-syntactic frameworks with the goal of creating a proposal for a set of universal semantic role labels. The proposal examines various theoretic linguistic perspectives and focuses on Meaning-Text Theory and Functional Generative Description frameworks and PropBank. For the purpose of this research, data from five languages is used – Spanish and Catalan, Czech, English, and Finnish. This proposal is oriented towards Universal Dependencies, therefore the universal semantic role labels have been applied to the 5 languages described in the paper with the further intention of applying the universal semantic role labels to the full UD data.
Tento článek rozebírá několik systémů hloubkové syntaxe s cílem navrhnout univerzální sadu sémantických rolí. Návrh studuje různé teoreticky-lingvistické pohledy a zaměřuje se na Teorii význam-text (MTT), Funkční generativní popis (FGD) a PropBank. Pro tento výzkum byla použita data z pěti jazyků – španělštiny a katalánštiny, češtiny, angličtiny a finštiny. Návrh je orientován na Universal Dependencies. Univerzální značky sémantických rolí byly aplikovány na 5 jazyků popsaných v článku, nicméně cílem do budoucnosti je aplikovat toto značkování na kompletní data UD.
[ "Kira Droganova", "Daniel Zeman" ]
https://www.semanticscholar.org/paper/c3161fced4033ec94884e4577d71a136bf525e12/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Announcing the Prague Discourse Treebank 3.0
Uvedení Pražského diskurzního treebanku 3.0
We present the Prague Discourse Treebank 3.0 - a new version of annotation of discourse relations marked by primary and secondary discourse connectives in the data of the Prague Dependency Treebank. It brings three major updates: (i) largely revised annotation of discourse relations, (ii) consistency with the Lexicon of Czech Discourse Connectives (CzeDLex), and (iii) data not only in its native format (Prague Markup Language, discourse relations annotated at the top of the dependency trees), but also in the Penn Discourse Treebank 3.0 format and sense taxonomy (plain text plus a stand-off discourse annotation).
Představujeme Pražský diskurzní treebank 3.0 - novou verzi anotace diskurzních vztahů vyjádřených primárními a sekundárními diskurzními konektory v datech Pražského závislostního korpusu. Přináší tři hlavní aktualizace: (i) opravenou anotaci diskurzních vztahů, (ii) konzistenci se Slovníkem českých diskurzních konektorů (CzeDLex) a (iii) data nejen v původním formátu (Prague Markup Language, diskurzní vztahy anotované v závislostních stromech), ale také ve formátu a taxonomii diskurzních typů Penn Discourse Treebanku 3.0 (prostý text plus samostatná diskurzní anotace).
[ "Pavlína Synková", "Jiří Mírovský", "Lucie Poláková", "Magdaléna Rysová" ]
https://www.semanticscholar.org/paper/79b2125873a59f85d01da3cb8540eb9ecab26758/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
How Gender Interacts with Political Values: A Case Study on Czech BERT Models
Jak gender interaguje s politickými hodnotami: Studie na českých modelech typu BERT
Neural language models, which reach state-of-the-art results on most natural language processing tasks, are trained on large text corpora that inevitably contain value-burdened content and often capture undesirable biases, which the models reflect. This case study focuses on the political biases of pre-trained encoders in Czech and compares them with a representative value survey. Because Czech is a gendered language, we also measure how the grammatical gender coincides with responses to men and women in the survey. We introduce a novel method for measuring the model's perceived political values. We find that the models do not assign statement probability following value-driven reasoning, and there is no systematic difference between feminine and masculine sentences. We conclude that BERT-sized models do not manifest systematic alignment with political values and that the biases observed in the models are rather due to superficial imitation of training data patterns than systematic value beliefs encoded in the models.
Neuronové jazykové modely, které dosahují nejlepších výsledků ve většině úloh zpracování přirozeného jazyka, jsou trénovány na rozsáhlých textových korpusech, které nevyhnutelně obsahují hodnotově zatížený obsah a často zachycují nežádoucí zkreslení, které modely odrážejí. Tato studie se zaměřuje na politická zkreslení předtrénovaných kodérů v češtině a porovnává je s reprezentativním hodnotovým průzkumem. Protože čeština používá gramatický rod, měříme také, jak se gramatický rod shoduje s odpověďmi mužů a žen v průzkumu. Zavádíme novou metodu měření modelově vnímaných politických hodnot. Zjišťujeme, že modely nepřiřazují pravděpodobnost výroků podle hodnotově orientovaného uvažování a že neexistuje systematický rozdíl mezi větami v ženském a mužském rodě. Docházíme k závěru, že modely velikosti BERT neprojevují systematický soulad s politickými hodnotami a že zkreslení pozorovaná v modelech jsou spíše důsledkem povrchního napodobování vzorů v trénovacích dat než systematických hodnotových přesvědčení zakódovaných v modelech.
[ "Adnan Al Ali", "Jindřich Libovický" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Uhglish : Corpus of Spoken Contact Center Conversations Annotated in Dialog States
Uhglish : korpus mluvených konverzací kontaktních center s anotovanými k dialogovými stavmi
We present a new corpus, named Uhglish, composed of real-life contact center conversations in English that are annotated in dialog states and Named Entities. This corpus has been set up to develop new systems able to model dialog flows in a 2023 event at the turn of speech level. In the contact center industry, the conversation usually aims at solving the caller’s request. As far as we know, most dialog states databases contain acted conversation and are not representative of contact centers. We hypothesize that the current corpora are either too easy or too far from the real conversations happening in contact centers. To solve this issue, we propose a corpus with a double-rich annotation scheme enabling an investigation of the named entities, dialog states, and dialog flow. Uhglish regroups 245 conversations with a total of approximately 85 000 turns of speech, all recorded in a semi-acted environments collected by an intelligent call tracking company. First analyses show a good consistency of the annotations. We also show that ground truth dialog state information can be retrieve by algorithms.
Představujeme nový korpus nazvaný Uhglish, který se skládá z reálných konverzací v kontaktním centru v angličtině, které jsou anotovány v dialogových stavech a pojmenovaných entitách. Tento korpus byl vytvořen za účelem vývoje nových systémů schopných modelovat dialogové toky v události 2023 na úrovni obratu řeči. V oboru kontaktních center je konverzace obvykle zaměřena na na vyřešení požadavku volajícího. Pokud je nám známo, většina databází stavů dialogů obsahuje hranou konverzaci a jsou nejsou reprezentativní pro kontaktní centra. Předpokládáme, že současné korpusy jsou buď příliš jednoduché, nebo příliš vzdálené od skutečných konverzací probíhajících v kontaktních centrech. Abychom tento problém vyřešili, navrhujeme korpus s dvojnásobně bohatým obsahem anotačním schématem, které umožňuje zkoumat pojmenované entity, stavy dialogu a průběh dialogu. Uhglish přeskupuje 245 konverzací s celkem přibližně 85 000 obraty řeči, které byly zaznamenány v poloaktuálním prostředí. shromážděných inteligentní společností pro sledování hovorů. První analýzy ukazují dobrou konzistenci anotací. Rovněž jsme ukazujeme, že základní pravdivé informace o stavu dialogu lze získat pomocí algoritmů.
[ "Manon Macary", "Patrícia Schmidtová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Compounds in Universal Dependencies: A Survey in Five European Languages
Kompozita v Universal Dependencies: průzkum napříč pěti evropskými jazyky
V Universal Dependencies jsou kompozita, která chápeme jako slova obsahující dva nebo více kořeny, reprezentovány podle tokenizace odrážející ortografické konvence daného jazyka. Uzavřená složenina odpovídá jedinému slovu v Universal Dependencies (např. "vodopád"), zatímco spojovníková složenina ("tchán") a otevřená složenina ("jablkový koláč") se skládají z více slov. Cílem tohoto článku je otevřít diskuzi o tom, jak postoupit směrem k jednotné anotaci složenin. Řešení, které navrhujeme, spočívá v reprezentaci interní struktury všech typů složenin analogicky k syntaktickým frázím. To by nejen zvýšilo srovnatelnost složení uvnitř i mezi jazyky, ale také umožnilo srovnávání složenin a syntaktických frází.
In Universal Dependencies, compounds, which we understand as words containing two or more roots, are represented according to tokenization, which reflects the orthographic conventions of the language. A closed compound corresponds to a single word in Universal Dependencies (e.g., "waterfall") while a hyphenated compound ("father-in-law") and an open compound ("apple pie") to multiple words. The aim of this paper is to open a discussion on how to move towards a more consistent annotation of compounds. The solution we argue for is to represent the internal structure of all compound types analogously to syntactic phrases, which would not only increase the comparability of compounding within and across languages but also allow comparisons of compounds and syntactic phrases.
[ "Emil Svoboda", "Magda Ševčíková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
PaReNT (Parent Retrieval Neural Tool): A Deep Dive into Word Formation Across Languages
PaReNT (Parent Retrieval Neural Tool): Hluboký vhled do slovotvorby napříč jazyky
We present PaReNT (Parent Retrieval Neural Tool), a deep-learning-based multilingual tool performing parent retrieval and word formation classification in English, German, Dutch, Spanish, French, Russian, and Czech. Parent retrieval refers to determining the lexeme or lexemes the input lexeme was based on (e.g. ‘darkness’ is traced back to ‘dark’; ‘waterfall’ decomposes into ‘water’ and ‘fall’). Additionally, PaReNT performs word formation classification, which determines the input lexeme as a compound (e.g. ‘proofread’), a derivative (e.g. ‘deescalate’) or as an unmotivated word (e.g. ‘dog’). These seven languages are selected from three major branches of the Indo-European language family (Germanic, Romance, Slavic). Data is aggregated from a range of word-formation resources, as well as Wiktionary, to train and test the tool. The tool is based on a custom-architecture hybrid transformer block-enriched sequence-to-sequence neural network utilizing both a character-based and semantic representation of the input lexemes, with two output modules – one decoder-based dedicated to parent retrieval, and one classifier-based for word formation classification. PaReNT achieves a mean accuracy of 0.62 in parentretrieval and a mean balanced accuracy of 0.74 in word formation classification.
Představujeme PaReNT (Parent Retrieval Neural Tool), vícejazyčný nástroj založený na hlubokém učení pro vyhledávání slovotvorných předků a slovotvornou klasifikaci v angličtině, němčině, holandštině, španělštině, francouzštině, ruštině a češtině. Vyhledávání slovotvorných předků znamená určení lexému nebo lexémů, na kterých byl založen vstupní lexém (např. „darkness“ se dá vysledovat k ‚dark‘; „waterfall“ se rozloží na „water“ a „fall“). PaReNT navíc provádí slovotvornou klasifikaci, která určuje vstupní lexém jako složený (např. ‚proofread‘), derivát (např. ‚deescalate‘) nebo jako nemotivované slovo (např. ‚dog‘). Těchto sedm jazyků je vybráno ze tří hlavních větví indoevropské jazykové rodiny (germánská, románská, slovanská). Data byla agregována z celé řady slovotvorných zdrojů a Wikislovníku za účelem tréninku a testování nástroje. Nástroj je založen na vlastní hybridní architektuře -- sequence-to-sequence neuronové sítě obohacené o transformerový blok využívající jak znakovou, tak sémantickou reprezentaci vstupních lexémů se dvěma výstupními moduly – jednoho na bázi dekodéru určený pro vyhledávání slovotvorných předů a jednoho klasifikačního modulu určeného pro slovotvornou klasifikaci. PaReNT dosahuje 0,62 accuracy ve vyhledávání slovotvorných předků a balanced accuracy 0,74 ve slovotvorné klasifikaci.
[ "Emil Svoboda", "Magda Ševčíková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Universal Anaphora: The First Three Years
Universal Anaphora: První tři roky
The aim of the Universal Anaphora initiative is to push forward the state of the art in anaphora and anaphora resolution by expanding the aspects of anaphoric interpretation which are or can be reliably annotated in anaphoric corpora, producing unified standards to annotate and encode these annotations, delivering datasets encoded according to these standards, and developing methods for evaluating models that carry out this type of interpretation. Although several papers on aspects of the initiative have appeared, no overall description of the initiative's goals, proposals and achievements has been published yet except as an online draft. This paper aims to fill this gap, as well as to discuss its progress so far.
Cílem iniciativy Universal Anaphora je posunout stav poznání v oblasti anafory a analýzy anafory tím, že rozšíříme aspekty interpretace anafory, které jsou nebo mohou být spolehlivě anotované v anaforických korpusech, vytvoříme jednotné standardy pro takové anotace, poskytneme sady dat s anotacemi podle těchto standardů a vyvineme metody hodnocení modelů, které vykonávají tento typ interpretace. I když už vyšlo několik článků popisujících jednotlivé aspekty této iniciativy, dosud nebyl publikován žádný souhrnný popis jejích cílů, návrhů a dosažených výsledků. Předkládaný článek má za cíl tuto mezeru zaplnit a probrat dosavadní pokrok iniciativy.
[ "Massimo Poesio", "Maciej Ogrodniczuk", "Vincent Ng", "Sameer Pradhan", "Juntao Yu", "Nafise Sadat Moosavi", "Silviu Paun", "Amir Zeldes", "Anna Nedoluzhko", "Michal Novák", "Martin Popel", "Zdeněk Žabokrtský", "Daniel Zeman" ]
https://www.semanticscholar.org/paper/86f8041fac69c30a185a02202a924cf9a0c4b6e1/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Exploring Interpretability of Independent Components of Word Embeddings with Automated Word Intruder Test
Zkoumání interpretovatelnosti nezávislých složek slovních embeddigů pomocí automatizovaného testu nevhodných slov
Analýza nezávislých komponent (ICA) je algoritmus původně vyvinutý pro vyhledávání oddělených zdrojů ve smíšeném signálu, například v nahrávce více osob mluvících ve stejnou dobu ve stejné místnosti. Na rozdíl od analýzy hlavních komponent (PCA) umožňuje ICA reprezentovat slovo jako nestrukturovanou sadu rysů, aniž by byl některý konkrétní rys považován za významnější než ostatní. V tomto článku jsme použili ICA k analýze slovních vložek. Zjistili jsme, že ICA lze použít k nalezení sémantických rysů slov a tyto rysy lze snadno kombinovat a hledat slova, která kombinaci vyhovují. Ukázali jsme, že většina nezávislých komponent představuje takové rysy. Ke kvantifikaci interpretovatelnosti komponent používáme test narušitelů slov, který provádějí jak lidé, tak velké jazykové modely. Navrhujeme použít automatizovanou verzi testu narušitele slov jako rychlý a levný způsob kvantifikace interpretovatelnosti vektorů bez nutnosti lidské práce.
Independent Component Analysis (ICA) is an algorithm originally developed for finding separate sources in a mixed signal, such as a recording of multiple people in the same room speaking at the same time. Unlike Principal Component Analysis (PCA), ICA permits the representation of a word as an unstructured set of features, without any particular feature being deemed more significant than the others. In this paper, we used ICA to analyze word embeddings. We have found that ICA can be used to find semantic features of the words and these features can easily be combined to search for words that satisfy the combination. We show that most of the independent components represent such features. To quantify the interpretability of the components, we use the word intruder test, performed both by humans and by large language models. We propose to use the automated version of the word intruder test as a fast and inexpensive way of quantifying vector interpretability without the need for human effort.
[ "Tomáš Musil", "David Mareček" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Performing AI-Generated Theater Plays
Uvádění divadelních her generovaných umělou inteligencí
In this chapter, we look at the theater play AI: When a Robot Writes a Play, which was created in a human-machine collaboration, and discuss the implications of the creation and production process on our understanding of authorship and performance.
V této kapitole se podíváme na divadelní hru AI: When a Robot Writes a Play, která vznikla ve spolupráci člověka a stroje, a rozebereme důsledky procesu tvorby a produkce na naše chápání autorství a performance.
[ "Tomáš Musil", "Klára Vosecká", "Rudolf Rosa" ]
https://www.semanticscholar.org/paper/8386f580dfaa947384f7835d5b6ab6a9b5fbf75f/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Function Words at Praguian Functional Generative Description
Funkční slova v pražském Funkčním generativním popisu
The present contribution is aimed at a complex description of the treatment of the so-called function words within the framework of a dependency-based Functional Generative Description as proposed in Prague by Petr Sgall and his team and its reflection in the Prague Dependency Treebank, an original annotated corpus of Czech. Both the framework and the treebank are based on a stratificational model of language, a part of which are two levels of dependency-based syntactic structure, one oriented towards the syntactic structure of the sentence on the surface layer called analytical and the other oriented towards the underlying, deep sentence structure called tectogrammatical. The dependency tree structure of the sentence on the analytical level contains all the words present in the sentence as separate nodes, while the dependency representation of a sentence on the tectogrammatical level conceived of as a linguistically structured meaning of the sentence contains only content words as its nodes. On the analytical level, a distinction is made between different classes of function words, the main boundary being between the function words functioning within verbal complexes and containing information about the morpho-syntactic properties of verbs (i.e. auxiliaries) and those being parts of nominal groups (prepositions) or connecting clauses (or, as the case may be, parts of clauses) into one whole (conjunctions). It is argued in the paper that auxiliaries should be considered to be dependents on the verb that is their governor and to which they „belong“, and the prepositions and conjunctions, on the contrary, should be considered to be the heads of the nouns or clauses the form of which they “govern” or “control”. On the tectogrammatical level, the semantic contribution of the function words to the meaning of the sentence is reflected by information attached to the nodes of the tectogrammatical tree in the form of complex labels. Auxiliaries, prepositions and conjunctions are the most evident classes of function words, though there are some groups of words such as particles that stand on the borderline between function words and content words, to which we also pay attention in our study.
Tento příspěvek je zaměřen na komplexní popis zacházení s tzv. funkčními slovy v rámci závislostního funkčního generativního popisu a jeho odraz v Pražském závislostním korpusu. Teoretický rámec i treebank jsou založeny na stratifikačním modelu jazyka, jehož součástí jsou dvě úrovně syntaktické struktury založené na závislosti, jedna orientovaná na povrchovou syntaktickou strukturu věty (analytická rovina) a druhá orientovaná na hloubkovou větnou strukturu (tektogramatická). Závislostní stromová struktura věty na analytické úrovni obsahuje všechna slova přítomná ve větě jako samostatné uzly, zatímco závislostní reprezentace věty na tektogramatické úrovni koncipované jako lingvisticky strukturovaný význam věty obsahuje pouze obsahová slova jako její uzly. Na analytické úrovni se rozlišují různé třídy funkčních slov, přičemž hlavní hranice je mezi funkčními slovy fungujícími v rámci verbálních komplexů a obsahujícími informace o morfosyntaktických vlastnostech sloves a těch, které jsou součástí nominálních skupin (předložky) nebo spojují věty (případně části vět) v jeden celek (spojky). V příspěvku se tvrdí, že pomocná slovesa se považují za závislé na slovesu, ke kterému „patří“, a předložky a spojky se naopak považují za hlavy podstatných jmen, jejichž formu „řídí“ nebo „ovládají“. Na tektogramatické rovině se sémantický přínos funkčních slov k významu věty odráží v informacích připojených k uzlům tektogramatického stromu ve formě komplexních atributů. Pomocná slovesa, předložky a spojky jsou nejzřetelnějšími třídami funkčních slov, existují skupiny také slov, jako jsou částice, které stojí na pomezí mezi funkčními slovy a obsahovými slovy, kterým je ve studii také věnována pozornost.
[ "Eva Hajičová", "Jarmila Panevová", "Marie Mikulová", "Jan Hajič" ]
https://www.semanticscholar.org/paper/3e4b3a1d7e4b5ca488027419378ef55701890593/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Cost-Effective Discourse Annotation in the Prague Czech–English Dependency Treebank
Nákladově efektivní anotace diskurzu v Pražském česko-anglickém závislostním treebanku.
We present a cost-effective method for obtaining a high quality annotation of explicit discourse relations in the Czech part of the Prague Czech-English Dependency Treebank, utilizing three resources: (i) annotation projection from the Penn Discourse Treebank 3.0, (ii) manual tectogrammatical (deep syntax) representation of sentences of the corpus, and (iii) the Lexicon of Czech Discourse Connectives CzeDLex.
Představujeme nákladově efektivní metodu pro získání vysoce kvalitní anotace explicitních diskurzních vztahů v české části Pražského česko-anglického závislostního treebanku s využitím tří zdrojů: (i) projekce anotace z Penn Discourse Treebanku 3.0, (ii) ruční tektogramatické (hloubkové syntaktické) reprezentace vět korpusu a (iii) Slovníku českých diskurzních konektorů CzeDLex.
[ "Jiří Mírovský", "Pavlína Synková", "Lucie Poláková", "Marie Paclíková" ]
https://www.semanticscholar.org/paper/3be12fa75f6a69e0dceb46e1e19fe5fee5e6db4c/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Fine-grained Classification of Circumstantial Meanings within the Prague Dependency Treebank Annotation Scheme
Jemná klasifikace okolnostních významů v rámci anotačního schématu Pražského závislostního korpusu
In the contribution, we propose a formally and semantically based fine-grained classification of circumstantial meanings based on the analysis of a large number of valuable examples from the Prague Dependency Treebanks. The methodology and principles of the presented approach are elaborated in detail and demonstrated on two case studies. The classification of circumstantial meanings is carried out for the Czech language, but the methodology and principles used are language independent. The contribution also addresses the question of language universality and specificity through a comparison with English. The aim of this work is to enrich the annotation in the Prague Dependency Treebanks with detailed information on circumstantial meanings but it may also be useful for other semantically oriented projects. To the best of our knowledge, a similar corpus-based and corpus-verified elaborate classification of circumstantial meanings has not yet been proposed in any annotation project. The contribution presents the results of an ongoing work.
V příspěvku navrhujeme formálně a sémanticky podloženou jemnozrnnou klasifikaci okolnostních významů na základě analýzy velkého množství příkladů z Pražských závislostních korpusů. Metodika a principy prezentovaného přístupu jsou podrobně rozpracovány a demonstrovány na dvou případových studiích. Klasifikace okolnostních významů se provádí pro český jazyk, ale použitá metodika a principy jsou jazykově nezávislé. Příspěvek se také zabývá otázkou jazykové univerzality a specifičnosti prostřednictvím srovnání s angličtinou. Cílem této práce je obohatit anotaci Pražských závislostních korpusů, ale navržená klasifikace může být užitečná i pro jiné sémanticky zaměřené projekty. Pokud je nám známo, podobná korpusově založená a korpusově ověřená propracovaná klasifikace okolnostních významů nebyla dosud navržena v žádném anotačním projektu. Příspěvek prezentuje výsledky probíhající práce.
[ "Marie Mikulová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
The function-to-form principle in the determination of secondary forms (On the case of delimitation of circumstantial meanings)
Uplatňování funkčního přístupu při určování sekundárních předložek (na úloze klasifikace okolnostních významů)vání
Within the task of classifying circumstantial meanings, we revise existing approaches to the determination of secondary prepositions and conjunctions. Introducing a function-to-form approach, we propose three principles for determining secondary forms and their meanings. The functional approach sheds light on the nuanced nature of these formal means, highlighting their capacity to convey semantico-syntactic dependency and emphasising the dynamic and diverse nature of this class of formal means.
Na úkolu klasifikace okolnostních významů revidujeme dosavadní přístupy k určování sekundárních předložek a spojek. Zavedením přístupu od funkce k formě navrhujeme tři principy pro určování sekundárních forem a jejich významů. Funkční přístup poukazuje na nuanční povahu těchto formálních prostředků a zdůrazňuje jejich schopnost zprostředkovat sémantiko-syntaktickou závislost.
[ "Marie Mikulová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
ReadabilityCzech
ReadabilityCzech - softwarová knihovna pro jazyk R
ReadabilityCzech computes readability of texts written in Czech with language-specific adaptations of the classic readability formulas Flesch Reading Ease and Flesch-Kincaid Grade Level. Both readability formulas use counts of syllables, word counts per sentence, and sentence counts. The syllable counts per word are computed by a dedicated script in this package. The word and sentence counts are obtained from an automatic text analysis implemented by the udpipe package by J. Wijffels.
ReadabilityCzech počítá pochopitelnost česky psaného textu pomocí klasických vzorců Flesch Reading Ease a Flesch-Kincaid Grade Level adaptovaných na češtinu. Oba vzorce počítají s počtem slabik ve slově a počtem slov ve větě. Slabiky se počítají speciálním skriptem vyvinutým pro tento balíček. Počty vět a počty slov ve větách se získávají pomocí výstupu z největšího českého jazykového modelu parseru UDPipe.
[ "Silvie Cinková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Training data for the shared task Ideology and Power Identification in Parliamentary Debates
Trénovací data pro sdílený úkol Ideologie a identifikace moci v parlamentních debatách
This dataset contains a selection of speeches from ParlaMint corpora (version 4.0) as the training set for the shared task on "Ideology and Power Identification in Parliamentary Debates" in CLEF 2024.
Tento datový soubor obsahuje výběr projevů z ParlaMint corpora (verze 4.0) jako tréninkový soubor pro sdílený úkol „Ideologie a identifikace moci v parlamentních debatách“ v CLEF 2024.
[ "Çağrı Çöltekin", "Matyáš Kopp", "Vaidas Morkevičius", "Nikola Ljubešić", "Katja Meden", "Tomaž Erjavec" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2024
More than ChatGPT
Více než jen ChatGPT
More than just ChatGPT: Principles and capabilities of current artificial intelligence tools
Více než jen ChatGPT: Principy a možnosti současných nástrojů umělé inteligence
[ "Rudolf Rosa" ]
https://www.semanticscholar.org/paper/1c77f87b4e34b2ffb3c7e6b9cd659fa1601241b6/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2024
AudioPSP 24.01: Audio recordings of proceedings of the Chamber of Deputies of the Parliament of the Czech Republic
AudioPSP 24.01: Audio recordings of proceedings of the Chamber of Deputies of the Parliament of the Czech Republic
Audio recordings of proceedings of the Chamber of Deputies of the Parliament of the Czech Republic. Recordings cover all available audio files from 2013-11-25 to 2023-07-26.
Zvukové záznamy z jednání Poslanecké sněmovny Parlamentu České republiky. Záznamy pokrývají všechny dostupné zvukové soubory od 2013-11-25 do 2023-07-26.
[ "Matyáš Kopp" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2024
ParCzech 4.0
ParCzech 4.0
The ParCzech 4.0 corpus consists of stenographic protocols that record the Chamber of Deputies' meetings in the 7th term (2013-2017), the 8th term (2017-2021) and the current 9th term (2021-Jul 2023). The protocols are provided in their original HTML format, Parla-CLARIN TEI format. The corpus is automatically enriched with the morphological, syntactic, and named-entity annotations using the procedures UDPipe 2 and NameTag 2. The audio files are aligned with the texts in the annotated TEI files.
Korpus ParCzech 4.0 se skládá ze stenografických protokolů, které zaznamenávají schůze Poslanecké sněmovny v 7. volebním období (2013-2017), 8. volebním období (2017-2021) a aktuálním 9. volebním období (2021-7. 2023). Protokoly jsou poskytovány v původním HTML formátu, Parla-CLARIN TEI formátu. Korpus je automaticky obohacen o morfologické, syntaktické a pojmenované anotace pomocí postupů UDPipe 2 a NameTag 2. Zvukové soubory jsou sladěny s texty v komentovaných TEI souborech.
[ "Matyáš Kopp" ]
https://www.semanticscholar.org/paper/514a94683ac55dbb27266e608a40e245b2ad321c/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Function Words in Universal Dependencies
Funkční slova v Universal Dependencies
Languages differ in the way they encode grammatical structure, drawing on a wide range of strategies including the linear arrangement of words and phrases, morphological processes such as inflection, derivation, compounding and incorporation, as well as the use of specialized particles which may be realized as clitics or independent words. The latter are commonly known as function words.
Jazyky se liší v tom, jakým způsobem kódují gramatickou strukturu, přičemž využívají širokou škálu strategií včetně lineárního uspořádání slov a frází, morfologických procesů jako je ohýbání, odvozování, skládání slov a inkorporace, ale také použitím zvláštních částic, které mohou být realizovány jako příklonky nebo samostatná slova. Pokud jde o samostatná slova, používá se pro ně pojem funkční (též synsémantická) slova.
[ "Marie-Catherine de Marneffe", "Joakim Nivre", "Daniel Zeman" ]
https://www.semanticscholar.org/paper/a74e2ff8f4f0c5075fe00433b7bc5637a1431107/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Building an Infrastructure for Uniform Meaning Representations
Budování infrastruktury pro UMR
This paper reports the first release of the UMR (Uniform Meaning Representation) data set. UMR is a graph-based meaning representation formalism consisting of a sentence-level graph and a document-level graph. The sentencelevel graph represents predicate-argument structures, named entities, word senses, aspectuality of events, as well as person and number information for entities. The document-level graph represents coreferential, temporal, and modal relations that go beyond sentence boundaries. UMR is designed to capture the commonalities and variations across languages and this is done through the use of a common set of abstract concepts, relations, and attributes as well as concrete concepts derived from words from individual languages. This UMR release includes annotations for six languages (Arapaho, Chinese, English, Kukama, Navajo, Sanapana) that vary greatly in terms of their linguistic properties and resource availability. We also describe on-going efforts to enlarge this data set and extend it to other genres and modalities. We also briefly describe the available infrastructure (UMR annotation guidelines and tools) that others can use to create similar data sets.
Tento článek informuje o prvním vydání datové sady UMR (Uniform Meaning Representation). UMR je na grafu založený reprezentační formalismus, který se skládá z grafu na úrovni věty a grafu na úrovni dokumentu. Graf na úrovni věty reprezentuje struktury predikátových argumentů, pojmenované entity, slovní smysly, aspektualitu událostí, jako i informace o osobách a číslech pro entity. Graf na úrovni dokumentu reprezentuje koreferenční, časové, a modální vztahy, které přesahují hranice věty. UMR je navrženo tak, aby zachycovalo společné rysy a variace napříč jazyky, a to pomocí společné sady abstraktních pojmů, relací a atributů i konkrétních konceptů odvozených ze slov z jednotlivých jazyků. Tato verze UMR obsahuje anotace pro šest jazyků (Arapaho, čínština, angličtina, Kukama, Navajo, Sanapana), které se značně liší z hlediska jejich jazykových vlastností a dostupnosti zdrojů. Popisujeme také probíhající snahy o rozšíření této datové sady a rozšířit ji na další žánry a modality. Dále stručně popisujeme dostupnou infrastrukturu (UMR anotace pokyny a nástroje), které mohou ostatní použít k vytvoření podobných datových souborů.
[ "Julia Bonn", "Matthew Buchholz", "Jayeol Chun", "Andrew Cowell", "William Croft", "Lukas Denk", "Jens E L Van Gysel", "Jan Hajič", "Kenneth Lai", "James Martin", "Skatje Myers ", "Alexis Palmer", "Martha Palmer", "James Pustejovsky", "Zdeňka Urešová", "Nianwen Xue", "Jin Zhao", "Bennet Post", "Kristine Stenzel", "Haibo Sun", "Rosa Vallejos", "Sijia Ge" ]
https://www.semanticscholar.org/paper/3fc04fd33886e9e0977007111c4c9045ac7ce2ee/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Towards Unified Uni- and Multi-modal News Headline Generation
Směrem ke sjednocenému uni- a multimodálnímu vytváření titulků zpráv
Thanks to the recent progress in vision-language modeling and the evolving nature of news consumption, the tasks of automatic summarization and headline generation based on multimodal news articles have been gaining popularity. One of the limitations of the current approaches is caused by the commonly used sophisticated modular architectures built upon hierarchical cross-modal encoders and modality-specific decoders, which restrict the model's applicability to specific data modalities – once trained on, e.g., text+video pairs there is no straightforward way to apply the model to text+image or text-only data. In this work, we propose a unified task formulation that utilizes a simple encoder-decoder model to generate headlines from uni- and multi-modal news articles. This model is trained jointly on data of several modalities and extends the textual decoder to handle the multimodal output.
Díky nedávnému pokroku v modelování vizuálního jazyka a vyvíjející se povaze spotřeby zpráv získávají na oblibě úkoly automatické sumarizace a generování titulků na základě multimodálních zpravodajských článků. Jedno z omezení současných přístupů je způsobeno běžně používanými sofistikovanými modulárními architekturami postavenými na hierarchických crossmodálních kodérech a dekodérech specifických pro modalitu, které omezují použitelnost modelu na konkrétní datové modality – jakmile jsou natrénovány např. na párech text+video. neexistuje žádný přímý způsob, jak aplikovat model na text+obrázek nebo pouze textová data. V této práci navrhujeme jednotnou formulaci úkolu, která využívá jednoduchý model kodér-dekodér pro generování titulků z uni- a multimodálních zpravodajských článků. Tento model je trénován společně na datech několika modalit a rozšiřuje textový dekodér o zpracování multimodálního výstupu.
[ "Mateusz Krubiński", "Pavel Pecina" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Evaluating Optimal Reference Translations
Hodnocení optimálních referenčních překladů
The overall translation quality reached by current machine translation (MT) systems for high-resourced language pairs is remarkably good. Standard methods of evaluation are not suitable nor intended to uncover the many translation errors and quality deficiencies that still persist. Furthermore, the quality of standard reference translations is commonly questioned and comparable quality levels have been reached by MT alone in several language pairs. Navigating further research in these high-resource settings is thus difficult. In this paper, we propose a methodology for creating more reliable document-level human reference translations, called “optimal reference translations,” with the simple aim to raise the bar of what should be deemed “human translation quality.” We evaluate the obtained document-level optimal reference translations in comparison with “standard” ones, confirming a significant quality increase and also documenting the relationship between evaluation and translation editing.
Celková kvalita překladu, které dosahují současné systémy strojového překladu (MT) pro jazykové páry s velkým množstvím zdrojů, je pozoruhodně vysoká. Standardní metody hodnocení nejsou vhodné ani nejsou určeny k odhalování chyb a nedostatků v kvalitě překladu, které stále přetrvávají. Kromě toho je kvalita standardních referenčních překladů běžně zpochybňována a srovnatelné úrovně kvality dosáhlysystémy MT pouze v několika jazykových párech. Orientace v dalším výzkumu v těchto podmínkách s velkým množstvím zdrojů je tedy obtížná. V tomto článku navrhujeme metodiku pro vytváření spolehlivějších referenčních dokumentů na úrovni lidských překladů, nazývaných "optimální referenční překlady", s jednoduchým cílem zvýšit laťku toho, co by mělo být považováno za "kvalitu lidského překladu". Získané optimální referenční překlady na úrovni dokumentů vyhodnocujeme v porovnání se "standardními" překlady, přičemž potvrzujeme výrazné zvýšení kvality a také dokumentujeme vztah mezi hodnocením a úpravou překladu.
[ "Vilém Zouhar", "Věra Kloudová", "Martin Popel", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/fbac541ee6dc5af8d036ef619928474b84528927/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Leak, Cheat, Repeat: Data Contamination and Evaluation Malpractices in Closed-Source LLMs
Kontaminace dat a nekalé praktiky při vyhodnocování komerčních jazykových modelů
Natural Language Processing (NLP) research is becoming increasingly focused on the use of Large Language Models (LLMs), with some of the most popular ones being either fully or partially closed-source. The lack of access to model details, especially regarding training data, has repeatedly raised concerns about data contamination among researchers. Several attempts have been made to address this issue, but they are limited to anecdotal evidence and trial and error. Additionally, they overlook the problem of indirect data leaking, where models are iteratively improved by using data coming from users. In this work, we conduct the first systematic review of work using OpenAI's ChatGPT and GPT-4, the most prominently used LLMs today, in the context of data contamination. By analysing 255 papers and considering OpenAI's data usage policy, we extensively document how much data has been leaked to ChatGPT in the first year after the model's release. At the same time, we document a number of evaluation malpractices emerging in the reviewed papers, including unfair or missing baseline comparisons, reproducibility issues, and authors' lack of awareness of the data usage policy. Our work provides the first quantification of the ChatGPT data leakage problem.
Výzkum zpracování přirozeného jazyka (NLP) se stále více zaměřuje na používání velkých jazykových modelů (LLM), přičemž některé z nejpopulárnějších modelů jsou buď plně, nebo částečně closed-source. Nedostatečný přístup k údajům o modelu, zejména pokud jde o trénovací data, opakovaně vyvolává mezi výzkumníky obavy z kontaminace dat. Bylo učiněno několik pokusů o řešení tohoto problému, které se však omezují na anekdotické důkazy a pokusy a omyly. Navíc přehlížejí problém nepřímého úniku dat, kdy jsou modely iterativně vylepšovány pomocí dat pocházejících od uživatelů. V této práci provádíme první systematický přehled prací využívajících ChatGPT a GPT-4 společnosti OpenAI, dnes nejpoužívanější LLM, v kontextu kontaminace dat. Na základě analýzy 255 prací a s přihlédnutím k politice používání dat společnosti OpenAI rozsáhle dokumentujeme, kolik dat uniklo do modelu ChatGPT v prvním roce po jeho vydání. Současně dokumentujeme řadu chybných praktik při evaluaci, které se objevují v recenzovaných článcích, včetně nekorektních srovnání nebo chybějících baselines, problémů s reprodukovatelností a nedostatečného povědomí autorů o politice používání dat. Naše práce poskytuje první kvantifikaci problému úniku dat v ChatGPT.
[ "Simone Balloccu", "Patrícia Schmidtová", "Mateusz Lango", "Ondřej Dušek" ]
https://www.semanticscholar.org/paper/798feda076ad710df65d509a7884bd15937c8056/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
LatinPipe at the EvaLatin 2024 Dependency Parsing Shared Task
Příspěvek LatinPipe ve sdílené úloze EvaLatin 2024 v závislostním parsingu
We present LatinPipe, the winning submission to the EvaLatin 2024 Dependency Parsing shared task. Our system consists of a fine-tuned concatenation of base and large pre-trained LMs, with a dot-product attention head for parsing and softmax classification heads for morphology to jointly learn both dependency parsing and morphological analysis. It is trained by sampling from seven publicly available Latin corpora, utilizing additional harmonization of annotations to achieve a more unified annotation style. Before fine-tuning, we train the system for a few initial epochs with frozen weights. We also add additional local relative contextualization by stacking the BiLSTM layers on top of the Transformer(s). Finally, we ensemble output probability distributions from seven randomly instantiated networks for the final submission. The code is available at https://github.com/ufal/evalatin2024-latinpipe.
Představujeme LatinPipe, vítězný soutěžní systém do EvaLatin 2024 sdíleného úkolu v závislostním parsingu. Náš systém se skládá z fine-tunovaného předtrénovaného jazykového modelu a modeluje jak závislostní parsing pomocí dot-product attention hlavy, tak morfologickou analýzu pomocí softmaxových klasifikačních hlav. Systém je trénován vzorkováním ze sedmi veřejně dostupných latinských korpusů, využívajících dodatečnou harmonizaci anotací k dosažení jednotnějšího stylu anotací. Před doladěním trénujeme systém po několik počátečních epoch se zafixovanými váhami. Přidáváme také dodatečnou lokální relativní kontextualizaci přidáním BiLSTM vrstev nad Transformer(y). Nakonec ve finálním systému skládáme výstupní pravděpodobnostní rozdělení jako ensemble ze sedmi náhodně instancovaných sítí. Zdrojový kód je dostupný na https://github.com/ufal/evalatin2024-latinpipe.
[ "Milan Straka", "Jana Straková", "Federica Gamba" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Developing a Rhetorical Structure Theory Treebank for Czech
Vývoj treebanku dle teorie rétorické struktury pro češtinu
We introduce the first version of the Czech RST Discourse Treebank, a collection of Czech journalistic texts manually annotated using the Rhetorical Structure Theory (RST), a global coherence model proposed by Mann and Thompson (1988).
Představujeme první verzi korpusu Czech RST Discourse Treebank, soubor českých žurnalistických textů ručně anotovaných na základě lingvistické teorie globální koherence Rhetorical Structure Theory (RST, Mann a Thompsonová, 1988).
[ "Lucie Poláková", "Jiří Mírovský", "Šárka Zikánová", "Eva Hajičová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Conversion in languages with different morphological structures: A semantic comparison of English and Czech
Konverze v jazycích s jinými morfologickými strukturami: Kognitivně-sémantické srovnání angličtiny a češtiny
This article presents a comparative study of the semantics of conversion between verbs and nouns in two languages with different morphological structures – English and Czech. To make the cross-linguistic comparison of semantic relations possible, a cognitive approach is used to provide conceptual semantic categories applicable across both languages. The semantic categories, based on event schemata introduced by Radden & Dirven (2007) primarily for syntactic description, are applied to data samples of verb–noun conversion pairs in both languages, using a dictionary-based approach. We analyse a corpus sample of 300 conversion pairs of verbs and nouns in each language (e.g., run.v – run.n, pepper.n – pepper.v; běhat ‘run.v’– běh ‘run.n’, pepř ‘pepper.n’ – pepřit ‘pepper.v’) annotated for the semantic relation between the verb and the noun. We analyse which relations appear in the two languages and how often, looking for significant differences to answer the question of whether the morphological characteristics of a language influence the semantics of conversion. The analysis of the annotated samples documents that the languages most often employ conversion for the same concepts (namely, instance of action and result) and that the range of semantic categories in English and Czech is generally the same, suggesting that the typological differences between the two languages do not affect the range of possible meanings that conversion can express. The data also show a marked difference in the number of conversion pairs with multiple semantic relations between the verb and the noun, which was found to be larger in English than in Czech, and also in the frequency with which certain semantic relations occur, and these differences seem to be at least partially related to the morphological characteristics of Czech and English.
Tento článek představuje srovnávací studii sémantiky konverze mezi slovesy a podstatnými jmény ve dvou jazycích s odlišnou morfologickou strukturou - angličtině a češtině. Aby bylo možné mezijazykové srovnání sémantických vztahů, je použit kognitivní přístup, který poskytuje konceptuální sémantické kategorie použitelné v obou jazycích. Sémantické kategorie, založené na konceptuálních schématech událostí zavedených Raddenem & Dirvenem (2007) primárně pro syntaktický popis, jsou aplikovány na datové vzorky konverzních dvojic sloves a podstatných jmen v obou jazycích, a to na základě slovníků. Analyzujeme korpusový vzorek 300 konverzních dvojic sloves a podstatných jmen v každém jazyce (např. run.v - run.n, pepper.n - pepper.v; běhat - běh, pepř - pepřit), u nichž je anotován sémantický vztah mezi slovesem a podstatným jménem. Analyzujeme, které vztahy se v obou jazycích objevují a jak často, a hledáme významné rozdíly, abychom odpověděli na otázku, zda morfologické charakteristiky jazyka ovlivňují sémantiku konverze. Analýza anotovaných vzorků dokládá, že oba jazyky nejčastěji používají konverzi pro stejné koncepty (konkrétně instanci děje a výsledek) a že rozsah sémantických kategorií v angličtině a češtině je obecně stejný, což naznačuje, že typologické rozdíly mezi oběma jazyky nemají vliv na rozsah možných významů, které konverze může vyjádřit. Data také ukazují výrazný rozdíl v počtu konverzních dvojic s více sémantickými vztahy mezi slovesem a podstatným jménem, který je větší v angličtině než v češtině, a také ve frekvenci, s níž se určité sémantické vztahy vyskytují - tyto rozdíly zřejmě alespoň částečně souvisejí s morfologickými charakteristikami češtiny a angličtiny.
[ "Hana Hledíková", "Magda Ševčíková" ]
https://www.semanticscholar.org/paper/2f50a2d836bd1db96915461ce8cdffae62a7a66b/
null
null
null
null
null
null
null
null
null
null
null
null
en
2024
Charles Translator: A Machine Translation System between Ukrainian and Czech
Charles Translator: překladový systém mezi ukrajinštinou a češtinou
We present the Charles Translator, a machine translation system between Ukrainian and Czech. The translator is implemented as an online web interface and as an Android app with speech input, both featuring Cyrillic-Latin script transliteration. The system translates directly, in comparison to other available systems that use English as a pivot.
Představujeme Charles Translator, překladový systém mezi ukrajinštinou a češtinou. Překladač je implementován jako online webová služba a jako aplikace pro Android s řečovým vstupem, včetně transliterace mezi azbukou a latinkou. Systém překládá přímo, ve srovnání s jinými dostupnými překladači, které pivotují přes angličtinu.
[ "Martin Popel", "Lucie Poláková", "Michal Novák", "Jindřich Helcl", "Jindřich Libovický", "Pavel Straňák", "Tomáš Krabač", "Jaroslava Hlaváčová", "Mariia Anisimova", "Tereza Chlaňová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Simultaneouly translates into more than 40 languages. Czech scientists developed a unique translation system
Simultánně tlumočí do více než 40 jazyků současně. Čeští vědci vyvinuli unikátní překladatelský systém
The abstract is only in Czech: Jeden řečník, pět tlumočníků a simultánní tlumočení do více než 40 jazyků. To dokáže systém Elitr, evropský on-line překladatel, u jehož zrodu stáli odborníci z Matematicko-fyzikální fakulty Univerzity Karlovy. Je určený pro konference a každé slovo pronesené řečníkem musí během několika milisekund proběhnout skoro celou Evropou, aby si je lidé mohli vzápětí přečíst na mobilu.
Jeden řečník, pět tlumočníků a simultánní tlumočení do více než 40 jazyků. To dokáže systém Elitr, evropský on-line překladatel, u jehož zrodu stáli odborníci z Matematicko-fyzikální fakulty Univerzity Karlovy. Je určený pro konference a každé slovo pronesené řečníkem musí během několika milisekund proběhnout skoro celou Evropou, aby si je lidé mohli vzápětí přečíst na mobilu.
[ "Eva Kézrová", "Ondřej Bojar", "Dominik Macháček" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Overview of the Second Shared Task on Automatic Minuting (AutoMin) at INLG 2023
Přehled druhého úkolu o automatickém zápisu ze schůzek (AutoMin) na konferenci INLG 2023
In this article, we report the findings of the second shared task on Automatic Minuting (AutoMin) held as a Generation Challenge at the 16th International Natural Language Generation (INLG) Conference 2023. The second Automatic Minuting shared task is a successor to the first AutoMin which took place in 2021. The primary objective of the Au- toMin shared task is to garner participation of the speech and natural language process- ing and generation community to create au- tomatic methods for generating minutes from multi-party meetings. Five teams from diverse backgrounds participated in the shared task this year. A lot has changed in the Genera- tive AI landscape since the last AutoMin espe- cially with the emergence and wide adoption of Large Language Models (LLMs) to differ- ent downstream tasks. Most of the contribu- tions are based on some form of an LLM and we are also adding current outputs of GPT- 4 as a benchmark. Furthermore, we examine the applicability of GPT-4 for automatic scor- ing of minutes. Compared to the previous in- stance of AutoMin, we also add another do- main, the minutes for EU Parliament sessions, and we experiment with a more fine-grained manual evaluation. More details on the event can be found at https://ufal.github.io/ automin-2023/.
V tomto článku informujeme o výsledcích druhé sdílené úlohy o automatickém zzápisu ze schůzek (AutoMin), která se konala jako Generation Challenge na 16. mezinárodní konferenci o generování přirozeného jazyka (INLG) 2023. Druhá sdílená úloha Automatic Minuting navazuje na první úlohu AutoMin, která se konala v roce 2021. Hlavním cílem sdílené úlohy AutoMin je získat účast komunity zpracovávající a generující řeč a přirozený jazyk pro vytvoření automatických metod pro generování zápisů z jednání více stran. Letos se sdílené úlohy zúčastnilo pět týmů z různých oblastí. Od posledního AutoMinu se v oblasti generativní umělé inteligence hodně změnilo, zejména s nástupem a širokým rozšířením velkých jazykových modelů (Large Language Models, LLM) pro různé navazující úlohy. Většina příspěvků je založena na nějaké formě LLM a přidáváme také aktuální výstupy GPT- 4 jako benchmark. Dále zkoumáme použitelnost GPT-4 pro automatické skórování protokolů. Oproti předchozímu institutu AutoMin přidáváme také další dokumentaci, zápisy ze zasedání Parlamentu EU, a experimentujeme s jemnějším manuálním hodnocením. Více podrobností o této akci naleznete na adrese https://ufal.github.io/ automin-2023/.
[ "Tirthankar Ghosal", "Ondřej Bojar", "Marie Hledíková", "Tom Kocmi", "Anna Nedoluzhko" ]
https://www.semanticscholar.org/paper/a26f0b9f2495220533b15526e5292976338e2936/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Multi-Feature Clustering of Search Results
Shlukování výsledků vyhledávání s ohledem na více rysů
Search engines are a helpful tool used by many people on a daily basis. However, their usability is limited: Search results are returned as flat ranked lists and additional user input is required in order to explore the subtopics of the given search query. One way to bring more structure into these search results is clustering. This thesis proposes methods to automatically cluster search results into different subtopics by making use of query-specific features. These features include the topics of documents, which can be determined from their paragraphs with the help of language models, as well as named entities, which are abundant in news documents, and metadata such as publication dates. Hereby, the focus lies on computationally viable feature extraction and embedding techniques using recent technologies in natural language processing that generalize to different domains and languages. In addition to traditional document clustering methods, novel approaches for temporal clustering are proposed, which are based on bin-wise gradients and the hierarchical merging of histograms of timestamps. Experiments on English search results and on a corpus of German news documents determine baseline models for topical clustering and named entity clustering, and show the potential for trend identification by temporal clustering. They further conclude that the order of operations significantly impacts the results when search results are clustered hierarchically with respect to multiple features on different levels of the hierarchy. The main takeaway for constructing a search clustering pipeline architecture is that the majority of its computation time can be hidden by precomputing document embeddings and retrieving them with the input search results.
Vyhledávače jsou užitečným nástrojem, který denně používá mnoho lidí. Jejich použitelnost je však omezená: Výsledky vyhledávání jsou vraceny jako ploché řazené seznamy a pro prozkoumání dílčích rysů daného vyhledávacího dotazu je nutný dodatečný uživatelský vstup. Jedním ze způsobů, jak do výsledků vyhledávání vnést větší strukturu, je shlukování. Tato práce navrhuje metody pro automatické seskupování výsledků vyhledávání do různých dílčích rysů s využitím vlastností specifických pro dotaz. Mezi tyto vlastnosti patří témata dokumentů, která lze určit z jejich odstavců pomocí jazykových modelů, stejně jako pojmenované entity, které jsou hojně zastoupeny v diskusních dokumentech, a metadata, jako jsou data publikace. Tím je kladen důraz na výpočetně životaschopné techniky extrakce a vkládání funkcí s využitím nejnovějších technologií ve zpracování přirozeného jazyka, které se zobecňují do různých domén a jazyků. Kromě tradičních metod shlukování dokumentů jsou navrhovány nové přístupy k časovému shlukování, které jsou založeny na přechodech podle bin a hierarchickém slučování histogramů časových razítek. Experimenty s výsledky vyhledávání v angličtině a na korpusu německých zpravodajských dokumentů určují základní modely pro aktuální shlukování a shlukování pojmenovaných entit a ukazují potenciál pro identifikaci trendů pomocí časového shlukování. Dále dochází k závěru, že pořadí operací významně ovlivňuje výsledky, když jsou výsledky vyhledávání shlukovány hierarchicky s ohledem na více funkcí na různých úrovních hierarchie. Hlavní myšlenkou pro konstrukci architektury potrubí pro shlukování vyhledávání je, že většina jeho výpočetního času může být skryta předkomprimováním vložení dokumentů a jejich načtením se vstupními výsledky vyhledávání.
[ "Christopher Brückner" ]
https://www.semanticscholar.org/paper/93eda335ee7d344cca868915aee4b3b9cbe73205/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
DocMarker
DocMarker
DocMarker is an annotation tool for creating training data for the text-to-form information retrieval NLP task. Say you have a free-form text (rich-text maybe) that contains some information that should be filled out into some structured form. This tool lets you record and annotate this form-filling process.
DocMarker je anotační nástroj pro tvorbu trénovacích dat pro úlohu získávání informací z volného textu do strukturovaného formuláře.
[ "Jiří Mayer", "Pavel Pecina" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Czech Language Olympics in the changes of time
Olympiáda v českém jazyce v proměnách času
The paper presents an analysis of the Czech Language Olympics tasks.
Článek předkládá analýzu soutěžních úkolů Olympiády v českém jazyce.
[ "Kateřina Rysová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Analysis of Corpus-based Word-Order Typological Methods
Analýza typologických metod korpusového zkoumání slovosledu
This article presents a comparative analysis of four different syntactic typological approaches applied to 20 different languages. We compared three specific quantitative methods, using parallel CoNLL-U corpora, to the classification obtained via syntactic features provided by a typological data-base (lang2vec).
Tento článek představuje srovnávací analýzu čtyř různých syntaktických typologických přístupů aplikovaných na 20 různých jazyků. Porovnali jsme tři specifické kvantitativní metody s využitím paralelních korpusů ve formátu CoNLL-U, na klasifikaci prostřednictvím syntaktických rysů získaných z typologické databáze (lang2vec).
[ "Diego Alves", "Božo Bekavac", "Daniel Zeman", "Marko Tadić" ]
https://www.semanticscholar.org/paper/6f06a19535e1af10448c4962fdb8a94e77145c74/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Corpus-based Syntactic Typological Methods for Dependency Parsing Improvement
Na korpusu založené syntaktické typologické metody zlepšení závislostní syntaktické analýzy
This article presents a comparative analysis of four different syntactic typological approaches applied to 20 different languages to determine the most effective one to be used for the improvement of dependency parsing results via corpus combination. We evaluated these strategies by calculating the correlation between the language distances and the empirical LAS results obtained when languages were combined in pairs. From the results, it was possible to observe that the best method is based on the extraction of word order patterns which happen inside subtrees of the syntactic structure of the sentences.
Tento článek představuje srovnávací analýzu čtyř různých syntaktických typologických metod použitých na 20 různých jazyků za účelem stanovení, která metoda může nejvíce přispět ke zlepšení výsledků automatické syntaktické analýzy (parsingu) při kombinování korpusů. Tyto strategie jsme vyhodnotili tak, že jsme spočítali korelaci mezi vzdáleností jazyků na jedné straně a úspěšností parsingu (LAS), získanou při kombinaci jazyků po dvojicích. Ve výsledcích bylo možné pozorovat, že nejlepší metoda je založena na extrakci slovosledných vzorů, které se objevují uvnitř podstromů v syntaktické struktuře vět.
[ "Diego Alves", "Božo Bekavac", "Daniel Zeman", "Marko Tadić" ]
https://www.semanticscholar.org/paper/33e1f2c0a8c350faf9a2f4a35ffc7305b16bffae/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Corpus-Based Multilingual Event-type Ontology: annotation tools and principles
Vícejazyčná ontologie synonym založená na korpusu: anotační nástroje a principy
In the course of building a multilingual Event-type Ontology resource called SynSemClass, it was necessary to provide the maintainers and the annotators with a set of tools to facilitate their job, achieve data format consistency, and in general obtain high-quality data. We have adapted a previously existing tool, developed to assist the work in capturing bilingual synonymy. This tool needed to be both substantially expanded with some new features and fundamentally changed in the context of developing the resource for more languages, which necessarily is to be done in parallel. We are thus presenting here the tool, the new data structure design which had to change at the same time, and the associated workflow.
V průběhu budování vícejazyčného zdroje Event-type Ontology s názvem SynSemClass bylo nutné poskytnout správcům a anotátorům sadu nástrojů, které jim usnadní práci, dosáhnou konzistence datového formátu a obecně získají vysoce kvalitní data. Upravili jsme dříve existující nástroj, vyvinutý tak, aby pomáhal při práci se zachycením dvojjazyčných synonym. Tento nástroj bylo potřeba jak podstatně rozšířit o některé nové funkce, tak zásadně změnit v souvislosti s vývojem zdroje pro více jazyků, což se nutně musí dít paralelně. Prezentujeme zde tedy nástroj, návrh nové datové struktury, která se musela zároveň změnit, a související workflow.
[ "Eva Fučíková", "Jan Hajič", "Zdeňka Urešová" ]
https://www.semanticscholar.org/paper/08cba6e26311defe8b4372fdc4f8fbb9df0d62ea/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
HaVQA: A Dataset for Visual Question Answering and Multimodal Research in Hausa Language
HaVQA: Datová sada pro vizuální zodpovídání otázek a multimodální výzkum v jazyce hausa
This paper presents HaVQA, the first multimodal dataset for visual question-answering (VQA) tasks in the Hausa language. The dataset was created by manually translating 6,022 English question-answer pairs, which are associated with 1,555 unique images from the Visual Genome dataset. As a result, the dataset provides 12,044 gold standard English- Hausa parallel sentences that were translated in a fashion that guarantees their semantic match with the corresponding visual information. We conducted several baseline experiments on the dataset, including visual question answering, visual question elicitation, text-only and multimodal machine translation.
Tento článek představuje HaVQA, první multimodální soubor dat pro úlohy vizuálního odpovídání na otázky (VQA) v hauském jazyce. Dataset byl vytvořen ručním překladem 6 022 anglických dvojic otázka-odpověď, které jsou přiřazeny k 1 555 unikátním obrázkům z datové sady Visual Genome. Výsledkem je 12 044 paralelních anglicko-hauských vět, které byly přeloženy způsobem zaručujícím jejich sémantickou shodu s odpovídající vizuální informací. Na tomto souboru dat jsme provedli několik základních experimentů, včetně vizuálního zodpovídání otázek, vizuální elicitace otázek, překladu pouze textu a multimodálního strojového překladu.
[ "Shantipriya Parida", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/a563f823a887b75dd61adc96c556a8bd83c6e4c3/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Symbol Generation via Autoencoders for Handwritten Music Synthesis
Generování symbolů pomocí autoenkodérů pro syntézu ručně psaných notopisů
Optical Music Recognition is one of the fields where synthetic data is effectively utilized for training deep learning recognition models. Due to the lack of manually annotated data, the training data is generated by an automatic procedure which produces real-looking images of music scores in large quantities. Mashcima, a system for synthesizing training data for handwritten music recognition, generates complete music scores but the individual symbols are not synthetic, they are sampled from real symbol datasets. In this paper, we explore the impact of utilizing an adversarial autoencoder within the symbol synthesis pipeline. We show that in some cases the use of an autoencoder may not only be motivated by the creation of latent-space symbol embeddings but also by improved recognition accuracy.
Optické rozpoznávání notopisů je jednou z oblastí, kde jsou syntetická data efektivně využívána pro trénování rozpoznávacích modelů hlubokého učení. Vzhledem k nedostatku ručně anotovaných dat jsou trénovací data generována automatickým postupem, který vytváří reálně vypadající obrázky notopisů ve velkém množství. Mashcima, systém pro syntézu trénovacích dat pro rozpoznávání hudebních rukopisů, generuje kompletní notopisy, ale jednotlivé symboly nejsou syntetické, jsou vzorkovány z reálných datových souborů symbolů. V tomto článku zkoumáme dopad využití adversariálního autoenkodéru v rámci syntézy symbolů. Ukazujeme, že v některých případech může být použití autoenkodéru motivováno nejen vytvořením latentního prostoru symbolů, ale také zlepšenou přesností rozpoznávání.
[ "Jonáš Havelka", "Jiří Mayer", "Pavel Pecina" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Possibilities and limitations of optimal reference translations: Exploring ‘translationese’ in professional translations of newspaper articles
Možnosti a meze tvorby tzv. optimálních referenčních překladů: po stopách „překladatelštiny“ v profesionálních překladech zpravodajských textů
The study explores translation quality by analysing two different Czech translations of English newspaper articles, done by professionals hired by a translation company. The original idea was for a tandem of translators-cum-theoreticians to synthesise the best of the two existing translations, while introducing slight to moderate modifications where necessary, to arrive at an optimal reference translation, i.e. a translation thought to be the best possible that can be achieved by a team of human translators; optimal reference translations can then be used in assessments of excellent machine translations. It soon became apparent, however, that a considerable amount of editing and creativity was needed from the team striving for an optimal reference translation, prompting the present authors to subject the original translations to a detailed quality assessment. The primary focus is on the formal aspect of the existing translations and the phenomenon known as translationese, which is understood here – in line with Translation Studies authorities (Newmark, Baker) – to refer to a lack of sensitivity to target language usage. The problems identified fall into a wide range of linguistic categories such as spelling, morphosyntax, grammar, lexicon and word formation. Special attention is paid to the presence of source-language interference; having reviewed existing theoretical discussions of interference, the authors drafted a typology which was then expanded to include several other types of errors recurrent in the translations analysed.
Studie zkoumá kvalitu překladů, zejména styl včetně přítomnosti rušivých vlivů zdrojového jazyka, a to na základě analýzy kvality českých překladů dvou anglických novinových článků. Překlady vyhotovili tři překladatelé najatí překladatelskou agenturou. Další oblastí, kterou studie zkoumá, je tvorba tzv. otimálních překladů, tj. překladů považovaných za nejlepší možné za daných podmínek. Hodnocení kvality se zaměřuje především na styl stávajících překladů a zjištěné problémy spadají do široké škály jazykových kategorií, jako je pravopis, morfologie, gramatika, slovní zásoba a tvorba slov. Zvláštní důraz je kladen na interference a návrh typologie sestavený autory lze rozšířit o několik dalších typů interference, které se v analyzovaných překladech opakují. Optimální překlady, vyplývající z týmové práce dvou překladatelů a teoretiků, kteří syntetizovali to nejlepší ze tří existujících překladů a přitom využívali značnou část vlastní kreativity, mohou být v budoucnu použity při hodnocení vynikajících strojových překladů.
[ "Věra Kloudová", "David Mraček", "Ondřej Bojar", "Martin Popel" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Optimal reference translation of English-Czech WMT2020
Optimální referenční překlady z angličtiny do češtiny WMT2020
We define "optimal reference translation" as a translation thought to be the best possible that can be achieved by a team of human translators. Optimal reference translations can be used in assessments of excellent machine translations. We selected 50 documents (online news articles, with 579 paragraphs in total) from the 130 English documents included in the WMT2020 news test (http://www.statmt.org/wmt20/) with the aim to preserve diversity (style, genre etc.) of the selection. In addition to the official Czech reference translation provided by the WMT organizers (P1), we hired two additional translators (P2 and P3, native Czech speakers) via a professional translation agency, resulting in three independent translations. The main contribution of this dataset are two additional translations (i.e. optimal reference translations N1 and N2), done jointly by two translators-cum-theoreticians with an extreme care for various aspects of translation quality, while taking into account the translations P1-P3. We publish also internal comments (in Czech) for some of the segments. Translation N1 should be closer to the English original (with regards to the meaning and linguistic structure) and female surnames use the Czech feminine suffix (e.g. "Mai" is translated as "Maiová"). Translation N2 is more free, trying to be more creative, idiomatic and entertaining for the readers and following the typical style used in Czech media, while still preserving the rules of functional equivalence. Translation N2 is missing for the segments where it was not deemed necessary to provide two alternative translations. For applications/analyses needing translation of all segments, this should be interpreted as if N2 is the same as N1 for a given segment. We provide the dataset in two formats: OpenDocument spreadsheet (odt) and plain text (one file for each translation and the English original). Some words were highlighted using different colors during the creation of optimal reference translations; this highlighting and comments are present only in the odt format (some comments refer to row numbers in the odt file). Documents are separated by empty lines and each document starts with a special line containing the document name (e.g. "# upi.205735"), which allows alignment with the original WMT2020 news test. For the segments where N2 translations are missing in the odt format, the respective N1 segments are used instead in the plain-text format.
„Optimální referenční překlady“ definujeme jako nejlepší možné překlady, jakých může dosáhnout tým lidských překladatelů. Optimální referenční překlady lze použít při hodnocení vynikajících strojových překladů. Ze 130 anglických dokumentů zařazených do datasetu WMT2020 (http://www.statmt.org/wmt20/) jsme vybrali 50 dokumentů (online zpravodajské články, celkem 579 odstavců) s cílem zachovat rozmanitost (styl, žánr atd.) výběru. Kromě oficiálního referenčního českého překladu poskytnutého organizátory WMT (P1) jsme prostřednictvím profesionální překladatelské agentury najali další dva překladatele (P2 a P3, rodilí mluvčí češtiny), čímž vznikly tři nezávislé překlady. Hlavním přínosem tohoto souboru dat jsou dva další překlady (tj. optimální referenční překlady N1 a N2), které společně provedli dva překladatelé a zároveň teoretici překladu s mimořádnou péčí o různé aspekty kvality překladu, přičemž zohlednili překlady P1-P3. K některým segmentům zveřejňujeme i interní komentáře (v češtině). Překlad N1 by se měl co nejvíce blížit anglickému originálu (co do významu i jazykové struktury). Ženská příjmení přechýlena (např. „Mai“ je přeloženo jako „Maiová“). Překlad N2 je volnější, snaží se být kreativnější, idiomatičtější, pro čtenáře čtivější a řídí se typickým stylem používaným v českých médiích, přičemž zachovává pravidla funkční ekvivalence. Překlad N2 chybí u segmentů, kde nebylo považováno za nutné uvést dva alternativní překlady. U aplikací/analýz, které vyžadují překlad všech segmentů, by se to mělo interpretovat tak, že N2 je pro daný segment stejný jako N1. Soubor dat poskytujeme ve dvou formátech: OpenDocument spreadsheet (odt) a prostý text (jeden soubor pro každý překlad a anglický originál). Některá slova byla při tvorbě optimálních referenčních překladů zvýrazněna různými barvami; toto zvýraznění a komentáře jsou přítomny pouze ve formátu odt (některé komentáře odkazují na čísla řádků v souboru odt). Dokumenty jsou odděleny prázdnými řádky a každý dokument začíná zvláštním řádkem obsahujícím název dokumentu (např. "# upi.205735"), což umožňuje zarovnání s originálem datasetu WMT2020. Pro segmenty, kde chybí překlady N2 ve formátu odt, se místo nich použijí příslušné segmenty N1 ve formátu prostého textu.
[ "Věra Kloudová", "David Mraček", "Ondřej Bojar", "Martin Popel" ]
https://www.semanticscholar.org/paper/abaf9b037961bf8e5e020a1c2845c34bf7ed982b/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Throw it in the machine. Will machine translation work in fiction?
Hoď to do stroje. Uplatní se strojový překlad v umělecké literatuře?
"Machine translation is improving all the time and translators will eventually not be needed." Is this true, or does this improvement have its limits? What all is lost in machine translation and are we content with that? Věra Kloudová and Ondřej Bojar from ÚFAL MFF UK discuss with literary translators Dagmar Heeg and Viktor Janiš. Moderated by Martin Severýn. The debate will take place at the World of Books Prague in the Professional Forum section.
„Strojový překlad se stále zlepšuje a překladatelé časem nebudou potřeba.“ Je to pravda, nebo má toto zlepšování své limity? Co všechno se ve strojovém překladu ztrácí a spokojíme se s tím? Věra Kloudová a Ondřej Bojar z ÚFAL MFF UK diskutují s literárními překladateli Dagmar Heeg a Viktorem Janišem. Moderuje Martin Severýn. Debata proběhne v rámci veletrhu Svět Knihy Praha v sekci Profesní fórum.
[ "Věra Kloudová", "Ondřej Bojar" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Sustaining the European Language Grid: Towards the ELG Legal Entity
Udržitelnost European Language Grid: Směřování k právnímu subjektu ELG
When preparing the European Language Grid EU project proposal and designing the overall concept of the platform, the need for drawing up a long-term sustainability plan was abundantly evident. Already in the phase of developing the proposal, the centrepiece of the sustainability plan was what we called the “ELG legal entity”, i. e., an independent organisation that would be able to take over operations, maintenace, extension and governance of the European Language Grid platform as well as managing and helping to coordinate its community. This chapter describes our current state of planning with regard to this legal entity. It explains the different options discussed and it presents the different products specified, which can be offered by the legal entity in the medium to long run. We also describe which legal form the organisation will take and how it will ensure the sustainability of ELG.
Popisujeme současný stav plánování udržitelnosti platformy ELG včetně možných právních forem ELG a diskuse, jakým způsobem by byly schopny převzít správu, údržbu a rozšiřování platformy European Language Grid. Představujeme a diskutujeme různé možnosti a specifické produkty, které může právnická osoba nabízet ve střednědobém až dlouhodobém horizontu.
[ "Georg Rehm", "Katrin Marheinecke", "Stefanie Hegele", "Stelios Piperidis", "Kalina Bontcheva", "Jan Hajič", "Khalid Choukri", "Andrejs Vasiljevs", "Gerhard Backfried", "Katja Prinz", "José Manuel Gómez-Pérez", "Ulrich Germann" ]
https://www.semanticscholar.org/paper/938ed52dd825ee28753ea2863a8f386278916b5f/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Spanish Verbal Synonyms in the SynSemClass Ontology
Španělská Synonyma v Ontologii SynSemClass
This paper presents ongoing work in the expansion of the multilingual semantic event-type ontology SynSemClass (Czech-English-German)to include Spanish. As in previous versions of the lexicon, Spanish verbal synonyms have been collected from a sentence-aligned parallel corpus and classified into classes based on their syntactic-semantic properties. Each class member is linked to a number of syntactic and/or semantic resources specific to each language, thus enriching the annotation and enabling interoperability. This paper describes the procedure for the data extraction and annotation of Spanish verbal synonyms in the lexicon.
Tato práce prezentuje probíhající práce v rozšiřování vícejazyčného sémantického event-type ontology SynSemClass (česko-anglicko-německy) zahrnout španělštinu. Stejně jako v předchozích verzích lexikonu, španělská slovní synonyma byla sebrána z paralelníhokorpusu a zařazena do tříd založených na jejich syntakticko-sémantických vlastnostech. Každý člen třídy je spojen s určitým počtem syntactických a/nebo sémantických zdrojů specifických pro každý jazyk, čímž se obohatí anotace a umožní interoperabilitu. Tato práce popisuje postup pro extrakci dat a annotaci španělských slovních synonym v lexikonu.
[ "Cristina Fernández-Alcaina", "Eva Fučíková", "Jan Hajič", "Zdeňka Urešová" ]
https://www.semanticscholar.org/paper/6cb63773f1924add840f59357e988cf2481672a6/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
The SynSemClass lexicon: A resource for multilingual synonymy
Slovník SynSemClass: Zdroj pro vicejazyčnou synonymii
This paper presents ongoing work on the multilingual lexicon SynSemClass (henceforth, SSC). Other related projects have addressed the issue of synonymy in multilingual contexts from various perspectives, e.g., WordNet (Vossen 2004; Fellbaum and Vossen 2012) or Predicate Matrix (López de Lacalle et al. 2016). However, to the best of our knowledge, only SSC formalizes multilingual verbal synonymy in terms of syntactic and semantic properties.
Tato práce představuje probíhající práci na vícejazyčném lexikonu SynSemClass (napříště SSC). Další související projekty se zabývaly problematikou synonymie ve vícejazyčných kontextech z různých pohledů, např. WordNet (Vossen 2004; Fellbaum a Vossen 2012) nebo Predicate Matrix (López de Lacalle et al. 2016). Podle našeho nejlepšího vědomí však pouze SSC formalizuje vícejazyčné slovní synonymum z hlediska syntaktických a sémantických vlastností.
[ "Cristina Fernández-Alcaina", "Eva Fučíková", "Jan Hajič", "Zdeňka Urešová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Live speech translation demonstration
Demonstrace živého překladu řeči
Abstract is only in Czech: Demonstrovali jsme živý překlad řeči vycházející z projektu ELITR (elitr.eu) na Jednom dni s informatikou a matematikou, popularizační akci pro středoškoláky na MFF UK. Měli jsme stánek, u kterého účastníci akce mohli mluvit do mikrofonu v některém ze čtyř podporovaných jazyků a sledovat živý překlad do 42 jazyků.
Demonstrovali jsme živý překlad řeči vycházející z projektu ELITR (elitr.eu) na Jednom dni s informatikou a matematikou, popularizační akci pro středoškoláky na MFF UK. Měli jsme stánek, u kterého účastníci akce mohli mluvit do mikrofonu v některém ze čtyř podporovaných jazyků a sledovat živý překlad do 42 jazyků.
[ "Dominik Macháček" ]
https://www.semanticscholar.org/paper/21e653b2f44edc931a3e067fd4cb610f9ba600dc/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Demonstration of live speech translation and of Whisper-Streaming
Demonstrace živého překladu řeči a Whisper-Streaming
Demonstration of live speech translation from ELITR project. The system uses Whisper-Streaming tool.
Demonstrace živého překladu řeči vycházejího z projektu ELITR a nástroje Whisper-Streaming.
[ "Dominik Macháček" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Whisper-Streaming demonstration
Demonstrace Whisper-Streaming
We demonstrated Whisper-Streaming -- a tool for live speech transcription and translation of Whisper, big offline speech-to-text model. The event participants could speak into microphone in any of 96 supported languages and observe live speech transcript and translation into English.
Demonstrovali jsme Whisper-Streaming -- nástroj na živý přepis a překlad řeči velkým offline speech-to-text modelem Whisper. Účastníci akce mohli mluvit do mikrofonu v některém z 96 podporovaných jazyků a sledovat živé přepisy a překlady do angličtiny.
[ "Dominik Macháček" ]
https://www.semanticscholar.org/paper/bfb71dc924c565702aac1d1c5a8de069d5b0467a/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Live speech translation
Živý překlad řeči
Abstract is only in Czech: Demonstrace a prezentace živého překladu řeči vycházejícího z projektu ELITR a nástroje Whisper-Streaming.
Demonstrace a prezentace živého překladu řeči vycházejícího z projektu ELITR a nástroje Whisper-Streaming.
[ "Dominik Macháček" ]
https://www.semanticscholar.org/paper/21e653b2f44edc931a3e067fd4cb610f9ba600dc/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Live speech translation demonstration
Demonstrace živého překladu řeči
Abstract is only in Czech. Demonstrovali jsme živý překlad řeči vycházející z projektu ELITR (elitr.eu) na konferenci CzeDER. Překládali jsme všechny projevy na jednodenní konferenci z angličtiny do češtiny a dalších více než 40 jazyků automatickým systémem, a také jsme živě opravovali chyby v přepisech. Dále jsme měli stánek, na kterém účastníci konference mohli vyzkoušet překlad své řeči a dozvědět se více informací.
Demonstrovali jsme živý překlad řeči vycházející z projektu ELITR (elitr.eu) na konferenci CzeDER. Překládali jsme všechny projevy na jednodenní konferenci z angličtiny do češtiny a dalších více než 40 jazyků automatickým systémem, a také jsme živě opravovali chyby v přepisech. Dále jsme měli stánek, na kterém účastníci konference mohli vyzkoušet překlad své řeči a dozvědět se více informací.
[ "Dominik Macháček", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/21e653b2f44edc931a3e067fd4cb610f9ba600dc/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
MT Metrics Correlate with Human Ratings of Simultaneous Speech Translation
Automatické metriky strojového překladu korelují s lidským hodnocením simultánního překladu řeči
There have been several meta-evaluation studies on the correlation between human ratings and offline machine translation (MT) evaluation metrics such as BLEU, chrF2, BertScore and COMET. These metrics have been used to evaluate simultaneous speech translation (SST) but their correlations with human ratings of SST, which has been recently collected as Continuous Ratings (CR), are unclear. In this paper, we leverage the evaluations of candidate systems submitted to the English-German SST task at IWSLT 2022 and conduct an extensive correlation analysis of CR and the aforementioned metrics. Our study reveals that the offline metrics are well correlated with CR and can be reliably used for evaluating machine translation in simultaneous mode, with some limitations on the test set size. We conclude that given the current quality levels of SST, these metrics can be used as proxies for CR, alleviating the need for large scale human evaluation. Additionally, we observe that correlations of the metrics with translation as a reference is significantly higher than with simultaneous interpreting, and thus we recommend the former for reliable evaluation.
Abstrakt je pouze v angličtině. There have been several meta-evaluation studies on the correlation between human ratings and offline machine translation (MT) evaluation metrics such as BLEU, chrF2, BertScore and COMET. These metrics have been used to evaluate simultaneous speech translation (SST) but their correlations with human ratings of SST, which has been recently collected as Continuous Ratings (CR), are unclear. In this paper, we leverage the evaluations of candidate systems submitted to the English-German SST task at IWSLT 2022 and conduct an extensive correlation analysis of CR and the aforementioned metrics. Our study reveals that the offline metrics are well correlated with CR and can be reliably used for evaluating machine translation in simultaneous mode, with some limitations on the test set size. We conclude that given the current quality levels of SST, these metrics can be used as proxies for CR, alleviating the need for large scale human evaluation. Additionally, we observe that correlations of the metrics with translation as a reference is significantly higher than with simultaneous interpreting, and thus we recommend the former for reliable evaluation.
[ "Dominik Macháček", "Ondřej Bojar", "Raj Dabre" ]
https://www.semanticscholar.org/paper/26f10cea3d27f90315d8ace9834aff049d7cb463/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Turning Whisper into Real-Time Transcription System
Whisper jako systém pro živý přepis řeči
Whisper is one of the recent state-of-the-art multilingual speech recognition and translation models, however, it is not designed for real time transcription. In this paper, we build on top of Whisper and create Whisper-Streaming, an implementation of real-time speech transcription and translation of Whisper-like models. Whisper-Streaming uses local agreement policy with self-adaptive latency to enable streaming transcription. We show that Whisper-Streaming achieves high quality and 3.3 seconds latency on unsegmented long-form speech transcription test set, and we demonstrate its robustness and practical usability as a component in live transcription service at a multilingual conference.
Abstrakt je pouze v angličtině: Whisper is one of the recent state-of-the-art multilingual speech recognition and translation models, however, it is not designed for real time transcription. In this paper, we build on top of Whisper and create Whisper-Streaming, an implementation of real-time speech transcription and translation of Whisper-like models. Whisper-Streaming uses local agreement policy with self-adaptive latency to enable streaming transcription. We show that Whisper-Streaming achieves high quality and 3.3 seconds latency on unsegmented long-form speech transcription test set, and we demonstrate its robustness and practical usability as a component in live transcription service at a multilingual conference.
[ "Dominik Macháček", "Raj Dabre", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/d574b48117abfddb40e901ee14762763862900e4/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Robustness of Multi-Source MT to Transcription Errors
Robustnost vícezdrojového překladu vůči chybám transkripce
Automatic speech translation is sensitive to speech recognition errors, but in a multilingual scenario, the same content may be available in various languages via simultaneous interpreting, dubbing or subtitling. In this paper, we hypothesize that leveraging multiple sources will improve translation quality if the sources complement one another in terms of correct information they contain. To this end, we first show that on a 10-hour ESIC corpus, the ASR errors in the original English speech and its simultaneous interpreting into German and Czech are mutually independent. We then use two sources, English and German, in a multi-source setting for translation into Czech to establish its robustness to ASR errors. Furthermore, we observe this robustness when translating both noisy sources together in a simultaneous translation setting. Our results show that multi-source neural machine translation has the potential to be useful in a real-time simultaneous translation setting, thereby motivating further investigation in this area.
Abstrakt je pouze v angličtině. Automatic speech translation is sensitive to speech recognition errors, but in a multilingual scenario, the same content may be available in various languages via simultaneous interpreting, dubbing or subtitling. In this paper, we hypothesize that leveraging multiple sources will improve translation quality if the sources complement one another in terms of correct information they contain. To this end, we first show that on a 10-hour ESIC corpus, the ASR errors in the original English speech and its simultaneous interpreting into German and Czech are mutually independent. We then use two sources, English and German, in a multi-source setting for translation into Czech to establish its robustness to ASR errors. Furthermore, we observe this robustness when translating both noisy sources together in a simultaneous translation setting. Our results show that multi-source neural machine translation has the potential to be useful in a real-time simultaneous translation setting, thereby motivating further investigation in this area.
[ "Dominik Macháček", "Peter Polák", "Ondřej Bojar", "Raj Dabre" ]
https://www.semanticscholar.org/paper/93e0092d3e4bb0c1a1f0df2faab1395fde97dde3/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Deep Dive Machine Translation
Hloubkový rozbor strojového překladu
Machine Translation (MT) is one of the oldest language technologies having been researched for more than 70 years. However, it is only during the last decade that it has been widely accepted by the general public, to the point where in many cases it has become an indispensable tool for the global community, supporting communication between nations and lowering language barriers. Still, there remain major gaps in the technology that need addressing before it can be successfully applied in under-resourced settings, can understand context and use world knowledge. This chapter provides an overview of the current state-of-the-art in the field of MT, offers technical and scientific forecasting for 2030, and provides recommendations for the advancement of MT as a critical technology if the goal of digital language equality in Europe is to be achieved.
Strojový překlad (MT), jedna z nejstarších jazykových technologií, je zkoumán již více než 70 let. Až v posledním desetiletí však začal být přijímán širokou veřejností, a to do té míry, že v mnoha případech se stal nepostradatelným nástrojem pro globální komunitu podporující komunikaci mezi národy a snižování jazykových bariér. Tato kapitola podává přehled současného stavu technologie v oblasti MT, nabízí technické a vědecké prognózy do roku 2030 a doporučuje MT jako klíčovou technologii, pokud má být dosaženo digitální jazykové rovnosti v Evropě.
[ "Inguna Skadina", "Andrejs Vasiljevs", "Marcis Pinnis", "Aivars Bērziņš", "Nora Aranberri", "Joachim van den Bogoaert", "Sally O’Connor", "Mercedes García-Martínez", "Iakes Goenaga", "Jan Hajič", "Manuel Herranz", "Christian Lieske", "Martin Popel", "Maja Popović", "Sheila Castilho", "Federico Gaspari", "Rudolf Rosa", "Riccardo Superbo", "Andy Way" ]
https://www.semanticscholar.org/paper/fbef461f31252393091178ec35075af5e575c76b/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
What’s the Meaning of Superhuman Performance in Today’s NLU?
Jaký je význam lepší-než-lidé výkonnosti v dnešních úlohách NLP?
In the last five years, there has been a significant focus in Natural Language Processing (NLP) on developing larger Pretrained Language Mod- els (PLMs) and introducing benchmarks such as SuperGLUE and SQuAD to measure their abilities in language understanding, reasoning, and reading comprehension. These PLMs have achieved impressive results on these bench- marks, even surpassing human performance in some cases. This has led to claims of superhu- man capabilities and the provocative idea that certain tasks have been solved. In this position paper, we take a critical look at these claims and ask whether PLMs truly have superhuman abili- ties and what the current benchmarks are really evaluating. We show that these benchmarks have serious limitations affecting the compar- ison between humans and PLMs and provide recommendations for fairer and more transpar- ent benchmarks.
V posledních pěti letech se v Natural Language Processing (NLP) výrazně zaměřili na vývoj větších předtrénovaných jazykových modelů (PLM) a zavedení benchmarků jako SuperGLUE a SQuAD pro měření jejich schopností v porozumění jazyku, uvažování a porozumění čtení. Tyto PLM dosáhly v těchto testech působivých výsledků, v některých případech dokonce překonaly výkon člověka. To vedlo k tvrzení o nadlidských schopnostech a provokativní myšlence, že určité úkoly byly vyřešeny. V tomto pozičním dokumentu se na tato tvrzení kriticky podíváme a ptáme se, zda PLM skutečně mají nadlidské schopnosti a co současné benchmarky skutečně hodnotí. Ukazujeme, že tyto benchmarky mají vážná omezení ovlivňující srovnání mezi lidmi a PLM a poskytují doporučení pro spravedlivější a transparentnější benchmarky.
[ "Simone Tedeschi", "Johan Bos", "Thierry Declerck", "Jan Hajič", "Daniel Hershcovich", "Eduard Hovy", "Alexander Koller", "Simon Krek", "Steven Schockaert", "Rico Sennrich", "Ekaterina Shutova", "Roberto Navigli" ]
https://www.semanticscholar.org/paper/90df9c6924425d7366d16731a34bfa4e52ad5b2e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Shortening of the results of machine translation using paraphrasing dataset
Zkracování strojového překladu pomocí datasetu parafrází
As machine translation applications continue to expand into the realm of real-time events, the need for faster and more concise translation becomes increasingly important. One such application is simultaneous speech translation, an emission of subtitles in the target language given speech in the source language. In this work, we focus on easing reader's comprehension of subtitles by making the translation shorter while preserving its informativeness. For this, we use the S, M and L version of the Paraphrase Database (PPDB), and exploit their property that some of the paraphrasing rules differ in length of the left and right side. Selecting rules that make the output shorter, we fine-tune an MT model to naturally generate shorter translations. The results show that the model's conciseness improves by up to 0.61%, which leaves the space for improvements using bigger versions of PPDB in future work.
S tím, jak se aplikace strojového překladu stále rozšiřují do oblasti událostí v reálném čase, je potřeba rychlejšího a výstižnějšího překladu stále důležitější. Jednou z takových aplikací je simultánní překlad řeči, vysílání titulků v cílovém jazyce dané řeči ve zdrojovém jazyce. V této práci se zaměřujeme na usnadnění čtenářského porozumění titulkům tím, že překlad zkrátíme a zároveň zachováme jeho informativnost. K tomu využíváme S, M a L verzi databáze parafrází (PPDB) a využíváme jejich vlastnosti, že některá pravidla parafrázování se liší délkou levé a pravé strany. Výběrem pravidel, která zkrátí výstup, doladíme model MT tak, aby přirozeně generoval kratší překlady. Výsledky ukazují, že stručnost modelu se zlepšuje až o 0,61 %, což ponechává prostor pro vylepšení pomocí větších verzí PPDB v budoucí práci.
[ "Andrej Perković", "Jernej Vičič", "Dávid Javorský", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/bb923bfdf08b3f41ac60429e734a5fe06568f726/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Data-to-text Generation with Neural Language Models
Generování textu z dat pomocí neuronových jazykových modelů
We introduce the problems of data-to-text generation and the current state of the art, i.e. pretrained language models. We further detail our experiments with using pretrained language models in a pipeline setup, thus increasing explainability and accuracy of the outputs. We also discuss our experiment in collecting a special dataset for verbalizing individual facts and evaluating pretrained language models, including ChatGPT, on this data.
Představíme problematiku generování textu z dat a současný stav techniky, tj. předtrénované jazykové modely. Dále podrobně popisujeme naše experimenty s využitím předtrénovaných jazykových modelů v pipeline nastavení, které zvyšuje vysvětlitelnost a přesnost výstupů. Pojednáváme také o našem experimentu se sběrem speciální datové sady pro verbalizaci jednotlivých faktů a o hodnocení předtrénovaných jazykových modelů, včetně ChatGPT, na těchto datech.
[ "Ondřej Dušek" ]
https://www.semanticscholar.org/paper/ecbb9c929069a39b9e2754707ac084875253d6f2/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Dialogue Systems (introduction)
Dialogové systémy (úvod)
This lecture presents an introduction into the state of the art in current dialogue systems, mostly focusing on neural architectures, end-to-end models, and pretrained language models.
V této přednášce se seznámíte s nejnovějším stavem současných dialogových systémů, především se zaměřením na neuronové architektury, end-to-end modely a předtrénované jazykové modely.
[ "Ondřej Dušek" ]
https://www.semanticscholar.org/paper/eb7e846c9a03af77c334f1b766a39f4cd9bc2b93/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Robust Data-to-text Generation with Pretrained Language Models
Robustí generování textu z dat s předtrénovanými jazykovými modely
The task of data-to-text generation amounts to describing structured data, such as RDF triples, in fluent natural language sentences. The state-of-the-art approach in research systems today is finetuning pretrained language models (PLMs). This often leads to overfitting the data and may produce hallucinations, i.e. situations where the PLM generates outputs that are not grounded in the input, typically replicating (or amplifying) training data noise. Rather than applying a PLM as black box for the whole data-to-text task, we aim at using PLMs for simple individual subtasks, aiming to achieve broad generalization and minimize hallucination. First, we use a pipeline approach where the PLMs only work as text "editors", rather than generators, taking advantage of their high output fluency. The data is converted into text in an initial preprocessing step, where we use simple handcrafted templates (one per RDF relation). This results in very short sentences recounting the individual input facts, corresponding to RDF triples. The PLMs then order the individual facts and fuse them into fluent sentences. This helps us generate without in-domain training data (except the simple templates) and achieve good fluency and accuracy. We further examine the capability of PLMs to produce accurate descriptions of individual RDF triples, in order to remove the last handcrafted step. Using a specially collected dataset of varied RDF triple descriptions, we show that PLMs finetuned with a variety of relations are very robust in verbalizing novel, unseen relations. The key to PLMs' usability here is providing clear and meaningful relation labels.
Úkolem generování textu z dat je popsat strukturovaná data, například trojice RDF, plynulými větami v přirozeném jazyce. Nejnovějším přístupem ve výzkumných systémech je dnes dolaďování předtrénovaných jazykových modelů (PLM). To často vede k nadměrnému přizpůsobení dat a může vést k halucinacím, tj. situacím, kdy PLM generuje výstupy, které nemají oporu ve vstupních datech, typicky replikují (nebo zesilují) šum tréninkových dat. Namísto použití PLM jako černé skříňky pro celou úlohu generování textu z dat se snažíme využít PLM pro jednoduché jednotlivé dílčí úlohy, přičemž cílem je dosáhnout širokého zobecnění a minimalizovat halucinace.
[ "Ondřej Dušek" ]
https://www.semanticscholar.org/paper/1459db606c56cf762c0034c069d70493fa52a650/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Robust Data-to-text Generation with Pretrained Language Models
Robustí generování textu z dat s předtrénovanými jazykovými modely
The task of data-to-text generation amounts to describing structured data in fluent natural language sentences. The state-of-the-art approach in research systems today is finetuning pretrained neural language models (PLMs). This often leads to overfitting and hallucinations, i.e. situations where the PLM generates outputs that are not grounded in the input, replicating or amplifying training data noise. Rather than applying a PLM as black box for the whole data-to-text task, we aim at using PLMs for simple subtasks, aiming to achieve broad generalization and minimize hallucination. First, we use a pipeline approach where the PLMs only work as text “editors”, rather than generators, taking advantage of their high output fluency. The data is converted into text in an initial preprocessing step, where we use simple handcrafted templates recounting the individual input facts (i.e. relations between entities). The PLMs then order the facts and fuse them into fluent sentences. This helps us generate without in-domain training data and achieve good fluency and accuracy. We further examine the capability of PLMs to produce accurate descriptions of individual facts from the data, in order to remove the last handcrafted step. Using a specially collected dataset, we show that PLMs finetuned to describe a variety of relations are very robust in verbalizing novel, unseen relations. The key to PLMs’ usability here is providing clear relation names on the input.
Úkolem generování textu z dat je popsat strukturovaná data plynulými větami v přirozeném jazyce. Nejnovějším přístupem ve výzkumných systémech je dnes jemné ladění předtrénovaných neuronových jazykových modelů (PLM). To často vede k overfittingu a halucinacím, tj. k situacím, kdy PLM generuje výstupy, které nemají oporu ve vstupních datech, a replikuje nebo zesiluje šum tréninkových dat. Spíše než na použití PLM jako černé skříňky pro celou úlohu převodu dat na text se zaměřujeme na použití PLM pro jednoduché dílčí úlohy s cílem dosáhnout široké generalizace a minimalizovat halucinace. Nejprve použijeme pipeline přístup, kdy PLM pracují pouze jako textové "editory", nikoliv generátory, přičemž využijeme jejich vysokou plynulost výstupu. Data jsou převedena na text v úvodním kroku předzpracování, kde používáme jednoduché ručně vytvořené šablony přepočítávající jednotlivá vstupní fakta (tj. vztahy mezi entitami). PLM pak fakta seřadí a spojí je do plynulých vět. To nám pomáhá generovat bez tréninkových dat v doméně a dosáhnout dobré plynulosti a přesnosti. Dále zkoumáme schopnost PLM vytvářet z dat přesné popisy jednotlivých faktů, abychom odstranili poslední ručně vytvořený krok. Pomocí speciálně shromážděné sady dat ukazujeme, že PLM vyladěné pro popis různých vztahů jsou velmi robustní při verbalizaci nových, dosud neviděných vztahů. Klíčem k použitelnosti PLM je zde poskytování jasných názvů relací na vstupu.
[ "Ondřej Dušek" ]
https://www.semanticscholar.org/paper/1459db606c56cf762c0034c069d70493fa52a650/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Artificial intelligence and language learning
Umělá inteligence a výuka jazyků
Functioning principles of artificial intelligence and especially large language models like ChatGPT. What to look out for with AI tools. Legal issues of AI tools. Different ways to use ChatGPT or tips and tricks for prompting. Recommendations on the use of AI in language learning.
Principy fungování umělé inteligence a zejména velkých jazykových modelů jako je ChatGPT. Na co si dát pozor s AI nástroji. Právní otázky AI nástrojů. Různé způsoby použití ChatGPT aneb tipy a triky na promptování. Doporučení ohledně využití AI ve výuce jazyků.
[ "Ondřej Dušek", "Rudolf Rosa", "Lucie Poslušná" ]
https://www.semanticscholar.org/paper/8c2db6774609dc696f7bf41b54a7d5a905523897/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Speaking Multiple Languages Affects the Moral Bias of Language Models
Vícejazyčnost ovlivňuje morální tendence jazykových modelů
Pre-trained multilingual language models (PMLMs) are commonly used when dealing with data from multiple languages and cross-lingual transfer. However, PMLMs are trained on varying amounts of data for each language. In practice this means their performance is often much better on English than many other languages. We explore to what extent this also applies to moral norms. Do the models capture moral norms from English and impose them on other languages? Do the models exhibit random and thus potentially harmful beliefs in certain languages? Both these issues could negatively impact cross-lingual transfer and potentially lead to harmful outcomes. In this paper, we (1) apply the MORALDIRECTION framework to multilingual models, comparing results in German, Czech, Arabic, Chinese, and English, (2) analyse model behaviour on filtered parallel subtitles corpora, and (3) apply the models to a Moral Foundations Questionnaire, comparing with human responses from different countries. Our experiments demonstrate that, indeed, PMLMs encode differing moral biases, but these do not necessarily correspond to cultural differences or commonalities in human opinions. We release our code and models.
Předtrénované vícejazyčné jazykové modely (PMLM) se běžně používají při práci s daty z více jazyků a při mezijazykovém přenosu. PMLM se však pro každý jazyk trénují na různém množství dat. V praxi to znamená, že často fungují mnohem lépe v angličtině než v mnoha jiných jazycích. V tomto článku zkoumáme, do jaké míry to platí i pro schopnost zachytit morální normy a intuice. Zachycují modely morální normy z angličtiny a vnucují je ostatním jazykům? Vykazují modely v některých jazycích náhodná, a tedy potenciálně škodlivá morální přesvědčení? Oba tyto problémy by mohly negativně ovlivnit mezijazykový přenos a potenciálně vést ke škodlivým výsledkům. V tomto článku (1) aplikujeme metodu MORALDIRECTION na vícejazyčné modely a porovnáváme výsledky v němčině, češtině, arabštině, čínštině a angličtině, (2) analyzujeme chování modelů na paralelních korpusech z filmových titulků a (3) aplikujeme modely na dotazník Moral Foundation Questionaire a porovnáváme je s odpověďmi lidí z různých zemí. Naše experimenty ukazují, že PMLM skutečně zachycují různé morální intuice, které však nutně neodpovídají kulturním rozdílům nebo obvyklým lidským názorům. Spolu s článkem zveřejňujeme zdrojový kód a modely.
[ "Katharina Hämmerl", "Björn Dieseroth", "Patrick Schramowski", "Jindřich Libovický", "Constantin A. Rothkopf", "Alexander Fraser", "Kristian Kersting" ]
https://www.semanticscholar.org/paper/d47ba83dc89193bbd816b0b929dd1319fba52779/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Exploring Anisotropy and Outliers in Multilingual Language Models for Cross-Lingual Semantic Sentence Similarity
Zkoumání anizotropie a odlehlých hodnot ve vícejazyčných jazykových modelech pro mezijazykovou sémantickou podobnost vět
Previous work has shown that the representations output by contextual language models are more anisotropic than static type embeddings, and typically display outlier dimensions. This seems to be true for both monolingual and multilingual models, although much less work has been done on the multilingual context. Why these outliers occur and how they affect the representations is still an active area of research.We investigate outlier dimensions and their relationship to anisotropy in multiple pre-trained multilingual language models. We focus on cross-lingual semantic similarity tasks, as these are natural tasks for evaluating multilingual representations. Specifically, we examine sentence representations. Sentence transformers which are fine-tuned on parallel resources (that are not always available) perform better on this task, and we show that their representations are more isotropic. However, we aim to improve multilingual representations in general. We investigate how much of the performance difference can be made up by only transforming the embedding space without fine-tuning, and visualise the resulting spaces. We test different operations: Removing individual outlier dimensions, cluster-based isotropy enhancement, and ZCA whitening. We publish our code for reproducibility.
Předchozí práce ukázaly, že reprezentace, které jsou výstupem kontextových jazykových modelů, jsou více anizotropní než statické slovní embedingy a obvykle mají dimenze s odlehlými hodnotami. Zdá se, že to platí jak pro jednojazyčné, tak pro vícejazyčné modely, ačkoli jim bylo věnováno mnohem méně práce. Proč se tyto odlehlé hodnoty objevují a jak ovlivňují reprezentaci, je stále aktivní oblastí výzkumu. V tomto článku zkoumáme odlehlé dimenze a jejich vztah k anizotropii v několika předtrénovaných vícejazyčných modelech. Zaměřujeme se na úlohy mezijazykové sémantické podobnosti, protože se jedná o přirozené úlohy pro hodnocení vícejazyčných reprezentací. Konkrétně zkoumáme reprezentace vět, tzv. Sentence Transformers, které jsou dotrénované na paralelních datech (které nejsou vždy k dispozici). Dosahují v této úloze lepších výsledků jejich reprezentace jsou izotropnější. Naším cílem je však zlepšit vícejazyčné reprezentace obecně. Zkoumáme, jak velký rozdíl ve výkonnosti lze dohnat pouhou transformací embedingů bez dotrénování, a vizualizujeme výsledné prostory. Testujeme různé operace: Odstranění jednotlivých odlehlých dimenzí, vylepšení izotropie na základě shluků a whitening pomocí ZCA.
[ "Katharina Hämmerl", "Alina Fastowski", "Jindřich Libovický", "Alexander Fraser" ]
https://www.semanticscholar.org/paper/086ed6f81cabfbd5f8794d4f275847347ed56f42/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Valency structure of complex predicates with Light Verbs, The case of Czech
Valenční struktura komplexních predikátů s kategoriálními slovesy, Případ češtiny
This chapter deals with light verb constructions. It is shown that what differentiates individual complex predicates from each other is the interaction between valency complementations of the Light Verb and those of the predicative noun through sharing semantic roles, which is manifested as their coreference. It is shown that the coreference, affecting the deep syntactic structure of the complex predicate, has consequences for both the semantic and the surface syntactic layer.
Tato kapitola se zabývá konstrukcemi s kategoriálním slovesem. Ukazuje, že to, co odlišuje jednotlivé komplexní predikáty, je interakce mezi valenčními doplněními kategoriálního slovesa a predikátového jména. Tato interakce, která se v hloubkové rovině projevuje jako koreference valenčních doplnění, má pak sémantické i syntaktické důsledky.
[ "Václava Kettnerová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
On the reciprocity of adjectives in Czech
K reciprocitě adjektiv v češtině
This paper aims to extend the knowledge of reciprocity outside the largely examined verbal domain, focusing on the reciprocity of adjectives in Czech. Our main aim is to show how syntactic reciprocalization is applied to constructions of adjectives where one of their valency complementations is subject to a systemic surface ellipsis. We argue that the reciprocity of adjectives is mostly encoded by the same inventory of language means as the reciprocity of verbs, in particular by pluralization of one of the valency complementations affected by reciprocity (where the only difference from verbs is the form mezi ‘between’+Instr) and by an anaphoric expression occupying the other valency complementation involved in reciprocity. The main difference between the formal marking of verbal and adjectival constructions lies in the fact that adverbials (e.g., navzájem ‘mutually’), serving as the so-called specifiers in reciprocal constructions of verbs, often take over the role of the primary marker of reciprocity in adjectival constructions. Further, we show that – similarly to verbs – adjectives bearing the semantic feature of mutuality require less linguistic marking in their reciprocal constructions, relying on pluralization only, while reciprocal constructions of adjectives without this feature are marked by both pluralization and an anaphoric expression or adverbials.
Práce má za cíl rozšířit studium problematiky reciprocity mimo oblast sloves a ukázat, jak je operace syntaktické reciprokalizace aplikována v adjektivních konstrukcích, v nichž jedno z valenčních doplnění podléhá systémové povrchové elipse. Reciprocita adjektiv vykazuje v mnohém shodné rysy s reciprocitou sloves. Jako podstatná se i v případě adjektiv jeví distinkce mezi adjektivy, která nesou rys vzájemnosti ve svém významu, a adjektivy, která takový rys nenesou, ale vyjádření vzájemnosti mezi určitými participanty umožňují. U adjektiv bez rysu vzájemnosti i u inherentně recipročních adjektiv se na vytváření recipročních konstrukcí podílejí stejné jazykové prostředky, avšak oba typy adjektiv se liší mírou jejich využití. Společným formálním rysem recipročních konstrukcí obou typů adjektiv je pluralizace jednoho ze zasažených valenčních doplnění (jediný rozdíl oproti slovesům zde představuje možnost vyjádření pomocí předložkové skupiny mezi+Instr). Povrchová pozice druhého doplnění zasaženého reciprocitou (obsazená reflexivním osobním zájmenem, příp. výrazem jeden druhý) nemusí být v recipročních konstrukcích obou typů adjektiv přítomna. Zatímco tento rys mají inherentně reciproční adjektiva společný s inherentně recipročními slovesy, adjektiva bez rysu vzájemnosti se v tomto bodě se slovesy bez rysu vzájemnosti rozcházejí: u těchto sloves je přítomnost anaforického výrazu v reciproční konstrukci (až na výjimky, kdy má druhé zasažené doplnění formu s+Instr) nutná. V recipročních konstrukcích adjektiv bez rysu vzájemnosti pak pozorujeme, že roli primárního prostředku vyjádření reciprocity přebírají ve vysoké míře adverbia, příp. předložková skupina mezi sebou, které jsou – na základě jejich funkce v recipročních konstrukcích sloves – řazeny k tzv. specifikátorům, tedy k výrazům, které samy o sobě nemohou reciproční význam vyjádřit, pouze ho v souvýskytu s jinými jazykovými prostředky zdůrazňují, příp. desambiguují. Jak zde však ukazujeme, v recipročních konstrukcích adjektiv sehrávají tyto prostředky mnohem podstatnější úlohu – bez jejich přítomnosti by konstrukce neměly reciproční čtení. Reciproční konstrukce adjektiv tak ve vyšší míře než reciproční konstrukce sloves dokládají přechod adverbií a předložkové skupiny mezi sebou od funkce „pouhého“ specifikátoru k funkci primárního markeru reciprocity.
[ "Václava Kettnerová", "Veronika Kolářová" ]
https://www.semanticscholar.org/paper/27fbbcaf95f8b9caa96df039d9d8564a0f866de3/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
A contribution to mark the occasion of Jarmila Panevová's birthday
K narozeninám Jarmily Panevové
In order to mark the occasion of professor Jarmila Panevová’s birthday we present research activities she was involved in over the last fifteen years.
Při příležitosti významného životního jubilea profesorky Jarmily Panevové představujeme jubilantčinu výzkumnou činnost za posledních patnáct let.
[ "Václava Kettnerová", "Veronika Kolářová", "Marie Mikulová", "Magda Ševčíková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Valence of reflexive verbs and its representation in the VALLEX lexicon
Valence reflexivních sloves a její zachycení ve valenčním slovníku VALLEX
As reflexive verbs, we refer here to verbs with the word-forming morpheme se or si in the lemma. These verbs include reflexiva tantum, i.e. verbs without a non-reflexive base counterpart (e.g. smát se `to laugh', stýskat si to complain), verbs derived by word-forming reflexivization from non-reflexive base verbs (e.g. odlišovat se - odlišit se `to differ', povídat si `to talk') and also verbs with the reflexive in the lemma is only optional (e.g. končit (se) `end', naříkat (si) `lament'). In this talk, we present the representation of individual types of reflexive verbs in the VALLEX lexicon. We focus especially on derived reflexive verbs, which are characterized by changes in their valency and meaning, compared to their non-reflexive base counterparts. Based on these changes, we define the following types of derived reflexive verbs: decausative reflexive verbs (e.g. kazit se ← kazit `to spoil'), autocausative reflexive verbs (e.g. nadchnout se ← nadchnout `to get excited'), reflexive verbs with a so-called partitive object (e.g. ovládat se - ovládnout se ← ovládat - ovládnout `to control'), reciprocal reflexive verbs (e.g. důvěřovat si ← důvěřovat `to trust'), conversion reflexive verbs (e.g. plnit se ← plnit `to fill') and deaccusative reflexive verbs (e.g. tajit se ← tajit `to melt'); the last group consists of the so-called quasi-conversion reflexive verbs, which have a systemic relationship with decausatives. For defined types of reflexive verbs, we describe changes in their syntactic behavior, taking into account changes in their semantics, and further show their representation in VALLEX, including basic statistical data.
Jako reflexivní slovesa zde označujeme slovesa se slovotvorným morfémem se nebo si v lemmatu. K těmto slovesům náleží reflexiva tantum, tedy slovesa bez nereflexivního základového protějšku (např. smát se, stýskat si), slovesa odvozená slovotvornou reflexivizací od nereflexivních základových sloves (např. odlišovat se – odlišit se, povídat si) a dále slovesa, u nichž je reflexivum v lemmatu pouze fakultativní (např. končit (se), naříkat (si)). V příspěvku představíme zachycení jednotlivých typů reflexivních sloves ve valenčním slovníku VALLEX. Soustředíme se zejména na odvozená reflexivní slovesa, která vykazují oproti svým nereflexivním základovým protějškům (systémové i nesystémové) změny ve valenci a významu. Na základě těchto změn vymezujeme následující typy odvozených reflexivních sloves: dekauzativní reflexivní slovesa (např. kazit se ← kazit), autokauzativní reflexivní slovesa (např. nadchnout se ← nadchnout), reflexivní slovesa s tzv. partitivním objektem (např. ovládat se – ovládnout se ← ovládat – ovládnout), reciproční reflexivní slovesa (např. důvěřovat si ← důvěřovat), konverzní reflexivní slovesa (např. plnit se ← plnit) a deakuzativní reflexivní slovesa (např. tajit se ← tajit); poslední skupinu tvoří tzv. kvazikonverzní reflexivní slovesa, která mají systémový vztah k dekauzativům. U vymezených typů reflexivních sloves popisujeme změny v jejich syntaktickém chování s přihlédnutím ke změnám v jejich sémantice a dále ukazujeme jejich reprezentaci ve VALLEXu, včetně základních statistických údajů.
[ "Václava Kettnerová", "Markéta Lopatková", "Jiří Mírovský" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Course “Artificial Intelligence”
Kurz “Umělá inteligence”
The aim of the course is to show artificial intelligence from the perspective of various disciplines of science. To reflect on how artificial intelligence affects our daily lives in various fields. The course will focus mainly on contexts/perspectives: AI in psychology, pedagogy, transport, health, linguistics, journalism and theatre. Within all parts of the course, emphasis will be placed on strengthening the multiperspective view of artificial intelligence, providing a broad pension of information on the use of artificial intelligence, its role, impacts and vision for the future.
Cílem kurzu je ukázat umělou inteligenci z perspektivy různých vědních disciplín. Zamyslet nad tím, jak umělá inteligence ovlivňuje náš každodenní život v různých oblastech. Kurz bude zaměřen především na kontexty/perspektivy: AI v psychologie, pedagogice, dopravě, zdravotnictví, lingvistice, žurnalistice a divadle. V rámci všech částí kurzu bude kladen důraz na posilování multiperspektivního pohledu na umělou inteligenci, poskytnutí širokého penza informací o využití umělé inteligence, její roli, dopadech a vizi do budoucnosti.
[ "Ondřej Hrách", "Rudolf Rosa" ]
https://www.semanticscholar.org/paper/1bfe1e6626067e9075570c575153e2b366f9d6cf/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
UFAL Parallel Corpus of North Levantine 1.0
UFAL Paralelní Korpus Severní Levanštiny 1.0
This is the first release of the UFAL Parallel Corpus of North Levantine, compiled by the Institute of Formal and Applied Linguistics (ÚFAL) at Charles University within the Welcome project (https://welcome-h2020.eu/). The corpus consists of 120,600 multiparallel sentences in English, French, German, Greek, Spanish, and Standard Arabic selected from the OpenSubtitles2018 corpus [1] and manually translated into the North Levantine Arabic language. The corpus was created for the purpose of training machine translation for North Levantine and the other languages.
Toto je první vydání UFAL Parallel Corpus of North Levantine, sestaveného Ústavem formální a aplikované lingvistiky (ÚFAL) Univerzity Karlovy v rámci projektu Welcome (https://welcome-h2020.eu/). Korpus se skládá ze 120 600 multiparalelních vět v angličtině, francouzštině, němčině, řečtině, španělštině a standardní arabštině vybraných z korpusu OpenSubtitles2018 [1] a ručně přeložených do severolevantské arabštiny. Korpus byl vytvořen za účelem trénování strojového překladu pro severní levantinu a další jazyky.
[ "Hashem Sellat", "Shadi Saleh", "Mateusz Krubiński", "Adam Pospíšil", "Petr Zemánek", "Pavel Pecina" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Bad MT Systems are Good for Quality Estimation
Špatné MT systémy jsou dobré pro odhad kvality
Quality estimation (QE) is the task of predicting quality of outputs produced by machine translation (MT) systems. Currently, the highest-performing QE systems are supervised and require training on data with golden quality scores. In this paper, we investigate the impact of the quality of the underlying MT outputs on the performance of QE systems. We find that QE models trained on datasets with lower-quality translations often outperform those trained on higher-quality data. We also demonstrate that good performance can be achieved by using a mix of data from different MT systems.
Odhad kvality (Quality Estimation, QE) je úkol předpovídat kvalitu výstupů vytvořených systémy strojového překladu (MT). V současné době jsou nejvýkonnější systémy QE pod dohledem a vyžadují trénování na datech se zlatým skóre kvality. V tomto článku zkoumáme vliv kvality podkladových výstupů MT na výkonnost systémů QE. Zjistili jsme, že modely QE natrénované na datových sadách s méně kvalitními překlady často překonávají modely natrénované na datech s vyšší kvalitou. Ukazujeme také, že dobrého výkonu lze dosáhnout použitím kombinace dat z různých systémů MT.
[ "Iryna Tryhubyshyn", "Aleš Tamchyna", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/79bbe4a5d0eb2c2fede697158eef4f60a5bbed2e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Extending an Event-type Ontology: Adding Verbs and Classes using Fine-tuned LLMs Suggestions
Rozšiřování ontologie: přidávání sloves a tříd s použitím LLM
In this project, we have investigated the use of advanced machine learning methods, specifically fine-tuned large language models, for pre-annotating data for a lexical extension task, namely adding descriptive words (verbs) to an existing (but incomplete, as of yet) ontology of event types. Several research questions have been focused on, from the investigation of a possible heuristics to provide at least hints to annotators which verbs to include and which are outside the current version of the ontology, to the possible use of the automatic scores to help the annotators to be more efficient in finding a threshold for identifying verbs that cannot be assigned to any existing class and therefore they are to be used as seeds for a new class. We have also carefully examined the correlation of the automatic scores with the human annotation. While the correlation turned out to be strong, its influence on the annotation proper is modest due to its near linearity, even though the mere fact of such pre-annotation leads to relatively short annotation times.
V tomto projektu jsme zkoumali využití pokročilých metod strojového učení, konkrétně vyladěných velkých jazykových modelů, pro předanotační data pro úlohu lexikálního rozšíření, konkrétně přidání popisných slov (sloves) do existující (ale zatím neúplné) ontologie typů událostí. Zaměřili jsme se na několik výzkumných otázek, od zkoumání možných heuristik pro poskytnutí alespoň náznaků anotátorům, která slovesa zahrnout a která jsou mimo současnou verzi ontologie, až po možné využití automatických skóre pro pomoc anotátorům, aby byli efektivnější při hledání prahu pro identifikaci sloves, která nemohou být přiřazena do žádné existující třídy, a proto mají být použita jako semena pro novou třídu. Pečlivě jsme také zkoumali korelaci automatických skóre s lidskou anotací. I když se korelace ukázala jako silná, její vliv na vlastní anotaci je vzhledem k její téměř linearitě skromný, i když pouhý fakt takové pre-anotace vede k relativně krátkým časům anotace.
[ "Jana Straková", "Eva Fučíková", "Jan Hajič", "Zdeňka Urešová" ]
https://www.semanticscholar.org/paper/8efe809ccd5eff63727103c1dbbe0ba8b4c5d77a/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2023
Human, soul, artificial intelligence
Člověk, duše, umělá inteligence
In cooperation with the Research Group on Theology and Contemporary Culture of the Evangelical Theological Faculty of Charles University, the Faculty of Mathematics and Physics of Charles University and the Multidisciplinary Working Group on AI in Context, we have prepared an unconventional full-length programme “Man, Soul and Artificial Intelligence” for visitors at the Church of the Infant Jesus for this year. On prepared posters in the space of the church, people could reflect on the possibility of linking the abilities of artificial intelligence and the spirituality of man. Through the application, people could then try to answer a series of questions on their smartphones and let artificial intelligence create an individual image of the soul of man using the artistic techniques of several world-famous artists.
Ve spolupráci s výzkumnou skupinou pro teologii a současnou kulturu Evangelické teologické fakulty Univerzity Karlovy, Matematicko-fyzikální fakultou Univerzity Karlovy a multidisciplinární pracovní skupinou AI v kontextu jsme na letošní rok v kostele Pražského Jezulátka připravili pro návštěvníky netradiční celovečerní program “ Člověk, duše a umělá inteligence“. Na připravených posterech v prostoru kostela se lidé mohli zamyslet nad možností propojení schopností umělé inteligence a spirituality člověka. Prostřednictvím aplikace si pak lidé mohli na svých chytrých telefonech zkusit odpovědět na sérii otázek a nechat umělou inteligenci, aby vytvořila individuální obraz duše člověka s použitím výtvarných technik několika světoznámých umělců.
[ "František Štěch", "Rudolf Rosa", "Jiří Dosoudil", "Tereza Hannemann", "Pavel Pola" ]
https://www.semanticscholar.org/paper/8cd3adc8752c9084ad545ad01e0474e5b0055d97/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Exploratory Analysis of the Applicability of Formalised Knowledge to Personal Experience Narration
Průzkumná analýza aplikovatelnosti formalizovaných znalostí na výpovědi o osobních prožitcích
Some of the victims of Nazi prosecution have consigned their personal experiences in the form of diaries of their internment in concentration camps. Such human-centric texts may contrast with the organisation of knowledge about such events that, for example, historians and archivists make. In this work, we analyse six such narrations with the use of Entity Extraction and Named Entity Recognition techniques, present the results of the corresponding exploration, and discuss the suitability of such tools on this corpus. We show that knowledge tools, that have been successfully used to organise documents, can be lacking when describing personal accounts, and we suggest ways to alleviate this.
Některé z obětí nacistického stíhání předávají své osobní zkušenosti v podobě deníků z internování v koncentračních táborech. Takové texty zaměřené na člověka mohou kontrastovat s uspořádáním znalostí o takových událostech, které dělají například historici a archiváři. V této práci analyzujeme šest takových vyprávění pomocí technik extrakce entit a rozpoznávání pojmenovaných entit, prezentujeme výsledky odpovídajícího zkoumání a diskutujeme o vhodnosti takových nástrojů na tomto korpusu. Ukazujeme, že znalostní nástroje, které se úspěšně používají k organizaci dokumentů, mohou při popisu osobních účtů chybět, a navrhujeme způsoby, jak to zmírnit.
[ "Victor Mireles", "Stephanie Billib", "Artem Revenko", "Stephan Jänicke", "Frank Uiterwaal", "Pavel Pecina" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Tackling Hallucinations in Neural Chart Summarization
Řešení halucinací při neuronové sumarizaci grafů
Hallucinations in text generation occur when the system produces text that is not grounded in the input. In this work, we tackle the problem of hallucinations in neural chart summarization. Our analysis shows that the target side of chart summarization training datasets often contains additional information, leading to hallucinations. We propose a natural language inference (NLI) based method to preprocess the training data and show through human evaluation that our method significantly reduces hallucinations. We also found that shortening long-distance dependencies in the input sequence and adding chart-related information like title and legends improves the overall performance.
K halucinacím při generování textu dochází tehdy, když systém produkuje text, který nemá opodstatnění na vstupu. V této práci se zabýváme problémem halucinací v neuronové sumarizaci grafů. Naše analýza ukazuje, že cílová strana trénovacích datových sad pro sumarizaci grafů často obsahuje dodatečné informace, což vede k halucinacím. Navrhujeme metodu založenou na inferenci v přirozeném jazyce (NLI) pro předzpracování trénovacích dat a pomocí lidského hodnocení ukazujeme, že naše metoda významně snižuje výskyt halucinací. Zjistili jsme také, že zkrácení závislostí na dlouhé vzdálenosti ve vstupní sekvenci a přidání informací souvisejících s grafem, jako je název a legenda, zlepšuje celkový výkon.
[ "Saad Obaid ul Islam", "Iza Škrjanec", "Ondřej Dušek", "Vera Demberg" ]
https://www.semanticscholar.org/paper/c1da40df79d5ec14c9ffac83d7bcedf543daa91c/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Open Calls and Pilot Projects
Otevřené výzvy a pilotní projekty
We describe the two ELG open calls for pilot projects, the objective of which was to demonstrate the use and the advantages of ELG in providing basic LT for applications and as a basis for more advanced LT-based modules or components useful to industry. Our main goal was to attract SMEs and research organisations to either contribute additional tools or resources to the ELG platform (type A pilot projects) or develop applications using Language Technologies available in the ELG platform (type B pilot projects). We start with the detailed description of the submission and evaluation processes, followed by a presentation of the open call results. Afterwards we describe the supervision and evaluation of the execution phase of the projects, as well as lessons learned. Overall, we were very satisfied with the setup and with the results of the pilot projects, which demonstrate an enormous interest in ELG and the Language Technology topic in general.
Popisujeme dvě otevřené výzvy ELG pro pilotní projekty, jejichž cílem bylo demonstrovat využití a výhody ELG při poskytování základního LT pro aplikace a jako základ pro pokročilejší LT moduly nebo komponenty užitečné pro průmysl. Naším hlavním cílem bylo přilákat malé a střední podniky výzkumné organizace k tomu, aby buď přispěly dalšími nástroji nebo zdroji do platformy ELG pilotní projekty typu A), nebo vyvíjely aplikace s využitím jazykových technologií dostupných v platformě ELG (pilotní projekty typu B). Začneme podrobným popisem procesu podání a vyhodnocení, po kterém následuje prezentace výsledků otevřené výzvy. Poté popíšeme dozor a vyhodnocení realizační fáze projektů, jakož i získané poznatky. Celkově jsme byli velmi spokojeni s nastavením a s výsledky pilotních projektů, které prokazují enormní zájem o ELG a téma jazykových technologií obecně.
[ "Lukáš Kačena", "Jana Hamrlová", "Jan Hajič" ]
https://www.semanticscholar.org/paper/c1b20df276c7c9c118d39e2446abef2553d427d9/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Meaning Representations for Natural Languages: Design, Models and Applications
Reprezentace významu pro přirozené jazyky: design, modely a aplikace
Meaning representations vary along many dimensions • How meaning is connected to text • Anchoring, alignment, multi‐layer vs. text‐span only • Relationship to logical and/or executable form • Mapping to Lexicons/Ontologies • General, task‐oriented • Relationship to discourse and discourse‐like phenomena We’ll overview these in this tutorial.
Významové reprezentace se liší v mnoha dimenzích • Jak je význam spojen s textem • Ukotvení, zarovnání, pouze vícevrstvé vs. text‐span • Vztah k logické a/nebo spustitelné formě • Mapování na lexikony/Ontologie • Obecné, zaměřené na úlohy • Vztah k diskurzu a diskurzu-podobným jevům V tomto tutoriálu si ukážeme jejich přehled.
[ "Nianwen Xue", "Julia Bonn", "Jan Hajič" ]
https://www.semanticscholar.org/paper/5f02c45168ebe30d05f12b5178a93cb7186cd4df/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Findings of the 2023 Conference on Machine Translation (WMT23): LLMs Are Here but Not Quite There Yet
Výsledky WMT23, Konference o strojovém překladu 2023: LLM jsou tu, ale ještě nedosáhly cíle zcela
This paper presents the results of the General Machine Translation Task organised as part of the 2023 Conference on Machine Translation (WMT). In the general MT task, participants were asked to build machine translation sys- tems for any of 8 language pairs (correspond- ing to 14 translation directions), to be evaluated on test sets consisting of up to four different do- mains. We evaluate system outputs with profes- sional human annotators using a combination of source-based Direct Assessment and scalar quality metric (DA+SQM).
(lindat translation) Tato práce prezentuje výsledky Obecné úlohy strojového překladu organizované v rámci Konference o strojovém překladu (WMT) v roce 2023. V obecné úloze MT byli účastníci požádáni o sestavení systémů strojového překladu pro kteroukoli z 8 jazykových dvojic (odpovídajících 14 směrům překladu), které budou vyhodnoceny na testovacích souborech skládajících se až ze čtyř různých oblastí. Vyhodnocujeme systémové výstupy s profesionálními lidskými anotátory pomocí kombinace přímého hodnocení založeného na zdroji a metriky skalární kvality (DA+SQM).
[ "Tom Kocmi", "Eleftherios Avramidis", "Rachel Bawden", "Ondřej Bojar", "Anton Dvorkovich", "Christian Federmann", "Mark Fishel", "Markus Freitag", "Thamme Gowda", "Roman Grundkiewicz", "Barry Haddow", "Philipp Koehn", "Benjamin Marie", "Christof Monz", "Makoto Morishita", "Kenton Murray", "Makoto Nagata", "Toshiaki Nakazawa", "Martin Popel", "Maja Popović", "Mariya Shmatova" ]
https://www.semanticscholar.org/paper/f04389f20ce73729b05d7eb61c673e435be42d65/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Dynamically Chaining APIs: from Dracor to TEITOK
Dynamicky řetězené API: od Dracoru po TEITOK
The FAIR principles are meant to ensure that (corpus) data can be reused for other purposes. But reusability is typically only considered from the perspective of static resources, while nowadays, many important data sources are regularly updated. In this paper, we will show how creating one resource from another using the API of both sides can make for a dynamic setup in which two resources can be kept in-sync programmatically, even in the case of changing source data. We will illustrate this by pulling data from the Dracor corpus via the API, and uploading and processing data into a TEITOK corpus also via the API. In the setup in this paper, syncing is done by periodically checking for modified data, but in a more involved integration, an on-update trigger could be used on the source side.
Principy FAIR mají zajistit, aby (korpusová) data mohla být znovu použita pro jiné účely. Ale opětovná použitelnost je obvykle zvažována pouze z pohledu statických zdrojů, zatímco v dnešní době je mnoho důležitých datových zdrojů pravidelně aktualizováno. V tomto článku si ukážeme, jak může vytvoření jednoho zdroje z druhého pomocí API obou stran vytvořit dynamické nastavení, ve kterém mohou být dva zdroje programově synchronizovány, a to i v případě změny zdrojových dat. To si ukážeme na vytažení dat z korpusu Dracor přes API a nahrání a zpracování dat do korpusu TEITOK také přes API. V nastavení v tomto dokumentu se synchronizace provádí periodickou kontrolou upravených dat, ale při složitější integraci lze na zdrojové straně použít spouštěč při aktualizaci.
[ "Maarten Janssen" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Deep Dive Speech Technology
Hloubkový rozbor řečových technologií
This chapter provides an in-depth account of current research activities and applications in the field of Speech Technology (ST). It discusses technical, scientific, commercial and societal aspects in various ST sub-fields and relates ST to the wider areas of Natural Language Processing and Artificial Intelligence. Furthermore, it outlines breakthroughs needed, main technology visions and provides an outlook towards 2030 as well as a broad view of how ST may fit into and contribute to a wider vision of Deep Natural Language Understanding and Digital Language Equality in Europe. The chapter integrates the views of several companies and institutions involved in research and commercial application of ST.
Tato kapitola popisuje do hloubky současné výzkumné aktivity a existující aplikace v oblasti řečových technologií (ST). Pojednává o technických, vědeckých, komerčních i společenské aspektech v nejrůznějších dílčích oblastech ST a vztahuje ST k širší oblasti zpracování přirozeného jazyka a umělé inteligence. Dále nastiňuje potřebné technologické průlomy v ST, hlavní technologické vize a poskytuje výhled směrem k roku 2030 včetně nastínění způsobu, jak může ST zapadat do širší vize hlubokého porozumění přirozenému jazyku a digitální jazykové rovnosti v Evropě. Kapitola integruje pohledy několika firem a institucí podílejících se na výzkumu a tvorbě komerčních aplikaci ST.
[ "Marcin Skowron", "Gerhard Backfried", "Eva Navas", "Aivars Bērziņš", "Joachim van den Bogoaert", "Franciska de Jong", "Andrea DeMarco", "Inma Hernáez", "Marek Kováč", "Peter Polák", "Johan Rohdin", "Michael Rosner", "Jon Sanchez", "Ibon Saratxaga", "Petr Schwarz" ]
https://www.semanticscholar.org/paper/31bc364a4070c869640343c8c83c5acc246d562b/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
A Unified Taxonomy of Deep Syntactic Relations
Sjednocená taxonomie hloubkově syntaktických vztahů
This paper analyzes multiple deep-syntactic frameworks with the goal of creating a proposal for a set of universal semantic role labels. The proposal examines various theoretic linguistic perspectives and focuses on Meaning-Text Theory and Functional Generative Description frameworks.
Tento článek rozebírá dvě existující anotační schémata pro hloubkovou syntax s cílem navrhnout sadu univerzálních sémantických rolí. Návrh zkoumá různé lingvisticko-teoretické pohledy se zvláštním zřetelem k Teorii smysl-text a k Funkčnímu generativnímu popisu.
[ "Kira Droganova", "Daniel Zeman" ]
https://www.semanticscholar.org/paper/c3161fced4033ec94884e4577d71a136bf525e12/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Results of the Forward-looking Community-wide Consultation
Výsledky konzultace v rámci komunity týkající se výhledů do budoucna
Within the ELE project three complementary online surveys were designed and implemented to consult the Language Technology (LT) community with regard to the current state of play and the future situation in about 2030 in terms of Digital Language Equality (DLE). While Chapters 4 and 38 provide a general overview of the community consultation methodology and the results with regard to the current situation as of 2022, this chapter summarises the results concerning the future situation in 2030. All of these results have been taken into account for the specification of the project’s Strategic Research, Innovation and Implementation Agenda (SRIA) and Roadmap for Achieving Full DLE in Europe by 2030.
V rámci projektu ELE byly navrženy a uskutečněny tři online průzkumy za účelem konzultace současného stavu a budoucí situace kolem roku 2030 v oblasti digitální jazykvé rovnosti (DLE) s komunitou jazykových technologií (LT). Zatímco kapitoly 4 a 38 poskytují obecný přehled metodiky komunitních konzultací a výsledky s ohledem na současnou situaci k roku 2022, tato kapitola shrnuje výsledky týkající se budoucí situace v roce 2030.
[ "Emma Daly", "Jane Dunne", "Federico Gaspari", "Teresa Lynn", "Natalia Resende", "Andy Way", "Maria Giagkou", "Stelios Piperidis", "Tereza Vojtěchová", "Jan Hajič", "Annika Grützner-Zahn", "Stefanie Hegele", "Katrin Marheinecke", "Georg Rehm" ]
https://www.semanticscholar.org/paper/9f06d6d21bcac547c957de69fbde8f96352662c3/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Understanding Borrowing through Derivational Morphology: A Case Study of Czech Verbs
Zkoumání výpůjček pohledem derivační morfologie: Případová studie českých sloves
The transfer of morphemes across languages in language contact situations may lead to an alteration in the morphology of the recipient language. One of the possible outcomes can be the introduction of newer word forms or a formation of newer morphological variants of the existing word forms in the recipient language. Languages often borrow nominal roots and morphologically derive them into verbs and are thus integrated into respective derivational classes. This corpus-based analysis for Czech tries to show how synchronic derivational resources can be used to probabilistically analyze the effects of borrowing in language evolution by focusing on morphological integration of the borrowed nominal roots in verb formations.
Přenos morfémů mezi jazyky v situacích jazykového kontaktu může mít dopady v morfologii cílového jazyka. Jedním z možných výsledků může být zavádění novějších slovních tvarů nebo tvoření novějších morfologických variant existujících tvarů slov v cílovém jazyku. Jazyky si často vypůjčují nominální kořeny, odvozují od nich slovesa a integrují je do derivačních tříd. Předložená korpusová analýza se na materiálu pro češtinu ukazuje, jak lze využít synchronní derivační zdroje pro pravděpodobnostní analýzu výpůjček z hlediska vývoje jazyka.
[ "Abishek Stephen", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/9d8d3db467bb56a2219e660fa79bb7b8a7791d68/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Universal Dependencies for Malayalam
Universal Dependencies pro malajálamštinu
Treebanks can play a crucial role in developing natural language processing systems and to have a gold-standard treebank data it becomes necessary to adopt a uniform framework for the annotations. Universal Dependencies (UD) aims to develop cross-linguistically consistent annotations for the world's languages. The current paper presents the essential pivots of the UD based syntactically annotated treebank for Malayalam. Sentences extracted from the IndicCorp corpus were manually annotated for morphological features and dependency relations. Language-specific properties are discussed which shed light on many of the grammatical areas in the Dravidian language syntax which needs to be examined in-depth. This paper also discusses some pertaining issues in UD taking into consideration the Dravidian languages and provides insights for further improvements in the existing treebanks.
Treebanky mohou hrát zásadní roli při vývoji systémů automatického zpracování přirozeného jazyka a k získání ručně anotovaných syntaktických dat (treebanku) je nezbytné použít uniformní anotační rámec. Universal Dependencies (UD) se snaží vyvinout kroslingvisticky konzistentní anotace pro jazyky světa. Tento článek představuje základy syntakticky anotovaného korpusu založeného na UD pro malajálamštinu.
[ "Abishek Stephen", "Daniel Zeman" ]
https://www.semanticscholar.org/paper/fbe91a46130a2b99c5d5eab78b999c2b62ba4605/
null
null
null
null
null
null
null
null
null
null
null
null
en
2023
Spanish Synonyms as Part of a Multilingual Event-Type Ontology
Španělská slovesná synonyma jako součást SynSemClass ontologie
This paper presents ongoing work on the multilingual event-type ontology SynSemClass, where multilingual verbal synonymy is formalized in terms of syntactic and semantic properties. In the ontology, verbs are grouped into synonym classes, both monolingually and cross-lingually. Specifically, verbs are considered to belong to the same class if they both express the same meaning in a specific context, and theirvalency frame can be mapped to the set of roles defined for a particular class. SynSemClass is built following a bottom-up approach where translational equivalents are automatically extracted from parallel corpora and annotated by human annotators. The task of the annotators consists in mapping the valency frame of a particular verb with the set of roles defined for the class where the verb is included as a potential class member, establishing links to external resources, and selecting relevant examples. The Spanish part of the ontology contains 99 classes enriched with Spanish synonyms (as of March 2023). The resulting resource provides fine-grained syntactic-semantic information on multilingual verbal synonyms and at the same time it links data to other existing monolingual and multilingual resources.
Tato práce představuje probíhající práci na vícejazyčné ontologii typu událostí SynSemClass, kde je vícejazyčné slovní synonymum formalizováno z hlediska syntaktických a sémantických vlastností. V ontologii jsou slovesa seskupena do synonymních tříd, a to jak jednojazyčně, tak i vícejazyčně. Konkrétně jsou slovesa považována za patřící do stejné třídy, pokud obě vyjadřují stejný význam v určitém kontextu, a jejich valenční rámec může být mapován do množiny rolí definovaných pro konkrétní třídu. SynSemClass je postavena na základě přístupu zdola nahoru, kde jsou translační ekvivalenty automaticky extrahovány z paralelních korpusů a anotovány lidskými anotátory. Úkol anotátorů spočívá v mapování valenčního rámce konkrétního slovesa se množinou rolí definovaných pro třídu, kde je sloveso zahrnuto jako potenciální člen třídy, vytvoření odkazů na externí zdroje a výběru relevantních příkladů. Španělská část ontologie obsahuje 99 tříd obohacených o španělská synonyma (k březnu 2023). Výsledný zdroj poskytuje jemnozrnné syntakticko-sémantické informace o vícejazyčných slovních synonymech a zároveň propojuje data s dalšími existujícími jednojazyčnými a vícejazyčnými zdroji.
[ "Cristina Fernández Alcaina", "Eva Fučíková", "Jan Hajič", "Zdeňka Urešová" ]
https://www.semanticscholar.org/paper/91771c38e34b7da4acde66807884ac7ddb48503d/
null
null
null
null
null
null
null
null
null
null
null
null

ÚFAL Bilingual Abstracts Corpus

This is a parallel (bilingual) corpus of Czech and mostly English abstracts of scientific papers and presentations published by authors from the Institute of Formal and Applied Linguistics, Charles University in Prague. For each publication record, the authors are obliged to provide both the original abstract (in Czech or English), and its translation (English or Czech) in the internal Biblio system. The data was filtered for duplicates and missing entries, ensuring that every record is bilingual. Additionally, records of published papers which are indexed by SemanticScholar contain the respective link. The dataset was created from March 2024 image of the Biblio database and is stored in JSONL format, with each line corresponding to one record.

You can run statistics.py from the original repo to get an overview for the dataset:

Total records 3079
Average sents (words) per abstract 4.1 (98.7)
Total sents (words) (en) 12k (303771)
Langs en 2404 (78.1%), cs 650 (21.1%), ru 13 (0.4%), ... (long tail of less-represented languages)
Papers with S2 links 47.2%
Publication year 2024: 35, 2023: 246, 2022: 197, 2021: 210, 2020: 163, 2019: 147, 2018: 173, 2017: 183, 2016: 206, 2015: 169, 2014: 187, 2013: 155, 2012: 146, 2011: 123, 2010: 146, 2009: 132, 2008: 133, 2007: 103, 2006: 113, 2005: 79, 2004: 13, 2003: 4, 2002: 5, 2001: 3, 2000: 4, 1998: 3, 1997: 1

Example

{
    "lang": "en", "year": "2022",
    "title_en": "CorefUD 1.0: Coreference Meets Universal Dependencies",
    "title_cs": "CorefUD 1.0: Setkání koreference a Universal Dependencies",
    "abstract_en": "Recent advances in standardization for annotated language resources have led to successful large scale efforts, such as the Universal Dependencies (UD) project for multilingual syntactically annotated data. By comparison, the important task of coreference resolution, which clusters multiple mentions of entities in a text, has yet to be standardized in terms of data formats or annotation guidelines. In this paper we present CorefUD, a multilingual collection of corpora and a standardized format for coreference resolution, compatible with morphosyntactic annotations in the UD framework and including facilities for related tasks such as named entity recognition, which forms a first step in the direction of convergence for coreference resolution across languages.",
    "abstract_cs": "Nedávný pokrok ve standardizaci anotovaných jazykových zdrojů vedl k úspěšným velkým projektům jako Universal Dependencies (UD), kde se syntakticky anotují data pro mnoho jazyků. Anotace koreference, která spojuje opakované zmínky téže entity v textu a je pro porozumění jazyku velmi důležitá, je zatím standardizačním úsilím relativně nepoznamenaná. V tomto článku prezentujeme CorefUD, mnohojazyčnou sbírku korpusů a standardizovaný formát pro anotaci koreference, kompatibilní s morfosyntaktickou anotací v UD a rozšiřitelný na příbuzné úlohy, jako je rozpoznávání pojmenovaných entit. Jde o první krok směrem ke konvergenci koreferenčních zdrojů napříč jazyky.", 
    "authors": ["Anna Nedoluzhko", "Michal Novák", "Martin Popel", "Zdeněk Žabokrtský", "Amir Zeldes", "Daniel Zeman"],
    "s2_url": "https://www.semanticscholar.org/paper/33336cdc37455107ca34636d844ab352e410eb1a/"
}

Citation

The corpus was created by Rudolf Rosa and Vilém Zouhar. Contact the authors if you are experiencing any difficulties while using this dataset or have any related questions. For citation, please use the following BibTeX entry:

@misc{11234/1-4922,
    title = {Czech and English abstracts of {{\'U}FAL} papers (2022-11-11)},
    author = {Rosa, Rudolf and Zouhar, Vil{\'e}m},
    url = {http://hdl.handle.net/11234/1-4922},
    note = {{LINDAT}/{CLARIAH}-{CZ} digital library at the Institute of Formal and Applied Linguistics ({{\'U}FAL}), Faculty of Mathematics and Physics, Charles University},
    copyright = {Creative Commons - Attribution 4.0 International ({CC} {BY} 4.0)},
    year = {2022}
}
Downloads last month
0
Edit dataset card