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
2013
Prague Arabic Dependency Treebank 1.5
Pražský závislostní korpus arabštiny, verze 1.5
The Prague Arabic Dependency Treebank, version 1.5 (PADT 1.5) is a collection of texts in Modern Standard Arabic, entiched with morphological and analytical (surface-syntax dependency) annotation.
Pražský závislostní korpus arabštiny, verze 1.5, je kolekce textů v moderní arabštině, obohacená o ruční morfologickou a závislostní anotaci.
[ "Otakar Smrž", "Viktor Bielický", "Iveta Smržová", "Petr Pajas", "Zdeněk Žabokrtský", "Daniel Zeman", "Jakub Kráčmar", "Shadi Saleh", "Jan Hajič", "Petr Zemánek" ]
https://www.semanticscholar.org/paper/3aade50704602b6bc2659768ff74bcfb5a0244c3/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
CzechMate: Machine Translation into Czech
CzechMate: strojový překlad do češtiny
Restricted web-based demo version of application of statistical machine translation from English, German, Spanish and French to Czech.
Omezená webová demoverze aplikace statistického strojového překladu z angličtiny, němčiny, španělštiny a francouzštiny do češtiny.
[ "Daniel Zeman" ]
https://www.semanticscholar.org/paper/c8e07e0a61639a2c756477426d7a551f99eb464a/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Linguistic Tools and Data at ÚFAL
Lingvistické nástroje a data na ÚFALu
Accompanying website to the workshop on linguistic tools for students of Czech at Palacký University in Olomouc. Most of the time it describes data and tools provided by ÚFAL, but not solely. Sometimes it draws attention to related software created at a different institute or as a result of a common project of ÚFAL and other institutes.
Tato stránka vznikla jako doprovodný web k semináři o lingvistických nástrojích pro studenty bohemistiky na Univerzitě Palackého v Olomouci. Zabývá se především daty a nástroji poskytovanými ÚFALem, ale nejenom jimi. Občas upozorňuje i na související software vzniklý buď zcela na jiném pracovišti, nebo jako výsledek společného projektu ÚFALu s dalšími ústavy.
[ "Daniel Zeman" ]
https://www.semanticscholar.org/paper/f0f69ade2a6f95ca4650c3477d1a316c635cdbde/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Twenty Flavors of One Text
Dvacet příchutí jednoho textu
The relationship between a sentence in a natural language as written down and its meaning is a very complex phenomenon. Many variations of the sentence preserve the meaning while other superficially very small changes can distort or completely reverse it. In order to process and produce sentences algorithmically, we need to somehow capture the semantic identity and similarity of sentences. The issue has been extensively studied from a number of directions, starting with thesauri and other lexical databases that capture synonymy of indivitual words (most notably the WordNet \cite{wordnet}), automatic paraphrasing of longer phrases or even sentences (e.g. \cite{bannard-callison-burch-2005}, \cite{kauchak-barzilay}, \cite{denkowski-lavie:2010:WMT}) or textual entailment \cite{textual-entailment}. We are still far away from a satisfactory solution. The field of machine translation makes the issue tangible in a couple of ways, most importantly within the task of automatic MT evaluation. Current automatic MT evaluation techniques rely on the availability of some reference translation, i.e. the translation as produced by a human translator. Obtaining such reference translations is relatively costly, so most datasets provide only one reference translation, see e.g. \cite{callisonburch-EtAl:2012:WMT}. While there are many possible translations of a given input sentence, only a handful of them are available as reference translations. The sets of hypotheses considered or finally selected by the MT system can be completely disjoint from the set of reference translations. \cite{bojar-kos:2010:WMT} observed that only about 5--10\% of reference translations were \emph{reachable} for Czech-to-English translation , regardless whether the system would actually search towards them. Relying mostly on unreachable reference translations is detrimental for MT system development. Specifically, automatic MT evaluation methods perform worse and consequently automatic model optimization suffers. Occasionally there are a few more references available. For example, in the NIST OpenMT evaluation\footnote{\url{http://www.itl.nist.gov/iad/mig/tests/mt/2006/doc/mt06eval_official_results.html}} or in the Prague Czech-English Dependency Treebank \cite{pcedt} there are four. Difficult as they may be to obtain, even four references fail to provide an adequate representation of the enormous space of correct translations. We want to trigger discussion about how many and what sort of references we actually need for MT, how can we obtain and use them. Obviously the set of possible translations is exponentional in the length of the sentence. Dreyer and Marcu \cite{dreyer-marcu} generated translation networks for the HyTER evaluation metric; the networks provided them with a compact representation of the large space of possible translations. Along the same lines, we propose unification-based approach to describe the space \cite{tsd-hyter}. Using a Prolog-based annotation tool and applying a 2-hour upper working limit per sentence, our translators were able to generate tens of thousands (or more) Czech references for each English sentence; the record figure was 19 billion translations of one long sentence. The way of actually using such data is an open question. If we just take the \texttt{mteval} script and run it with 19 billion references instead of four, we will probably not live long enough to see the results. Randomly selected and manually checked samples from these generated translations indicated that the translations are indeed plausible. We used a Levenshtein-based string similarity measure to confirm diversity of the generated set. We also measured the string similarity between system outputs and various reference translations: those from the generated set and also two manually post-edited system outputs. Unsurprisingly, each of the two post-edited translations were much closer to the original system outputs than the single reference translation that was originally available for that test set. However, our annotators managed to produce references which are almost exactly as close as the post-edited translations, even though they did not have access to them or the system outputs. \cite{wmt2013} showed that with several references (and with the BLUE and NIST metrics) smaller test set with multiple post-edited references has the same level of correlation with human judgments as a much larger test set with just one reference, created independently of MT. In future we might further investigate that in terms of \textit{which} system outputs to post-edit (average? good? all?) During the past 20 years we have accummulated impressive amount of bitext and it keeps growing. Our current conclusion is that not every bitext is equally valuable. Diversity of translation alternatives in either language is at least as important. So where will we be after the next 20 years? Quite likely, the amount of bitext available will multiply accordingly. However, we also hope that for every sentence ``pair'', the number of available translation variants will also multiply.
Vztah mezi větou v přirozeném jazyce, tak jak je napsaná, a jejím významem je velmi složitý jev. Existuje mnoho variant téže věty, které zachovávají význam, zatímco jiné, povrchově velmi malé změny mohou význam zkreslit nebo zcela otočit. Abychom mohli věty algoritmicky zpracovávat a generovat, musíme najít způsob, jak zachytit jejich sémantickou identitu a podobnost.
[ "Daniel Zeman", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/d667fcb022cdb065809fe3a9f259f9192d537f8a/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
A Gentle Introduction to Machine Learning in NLP using R
Úvod do strojového učení v počítačové lingvistice s použitím R
The course provides a concise introduction to principles and algorithms of machine learning in natural language processing both theoretically and practically. We will focus on fundamental ideas in machine learning and the basic theory behind them. The principles of machine learning will be presented in a gentle way so that students do not have to be afraid of scary mathematical formulas. We will give a brief introduction to the R system for statistical computing, which we use as a tool for practical demonstration. Students will gain practical know-how needed to apply the machine learning techniques to new problems. The presented methods of machine learning will be practically demonstrated on selected tasks from the field of natural language processing. Understanding these tasks will not require any extra linguistic knowledge. We will show how to master NLP tasks using the R system and how to experiment with real data sets.
Úvodní kurz je věnován teoretickému a praktickému pohledu na základní principy a algoritmy strojového učení v počítačovém zpracování přirozeného jazyka. Posluchačům jsou představeny základy statistického systému R.
[ "Barbora Hladká", "Martin Holub" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Feature Engineering in the NLI Shared Task 2013: Charles University Submission Report
Ladění atributů v soutěži NLI Shared Task 2013: zpráva týmu z Univerzity Karlovy
Our goal is to predict the first language (L1) of English essays’s authors with the help of the TOEFL11 corpus where L1, prompts (topics) and proficiency levels are provided. Thus we approach this task as a classification task employing machine learning methods. Out of key concepts of machine learning, we focus on feature engineering. We design features across all the L1 languages not making use of knowledge of prompt and proficiency level. During system development, we experimented with various techniques for feature filtering and combination optimized with respect to the notion of mutual information and information gain. We trained four different SVM models and combined them through majority voting achieving accuracy 72.5%.
Naším cílem je predikovat rodný jazyk (L1) autorů anglických esejí za pomoci korpusu TOEFL11, ve kterém jsou známy jazykové úrovně autorů a témata esejí. Úlohu řešíme jako klasifikační úlohu pomocí řízených metod strojového učení. Zaměřujeme se na ladění atributů, mezi které jazykovou úroveň a témata nezahrnujeme. Atributy navrhujeme napříč jazyky L1. Experimentujeme s několika technikami pro filtrování a kombinaci atributů s ohledem na kritéria z informační teorie. Celkem jsme natrénovali čtyři modely SVM a pomocí většinového hlasování je zkombinovali do modelu dosahujícího úspěšnosti 72.5%.
[ "Barbora Hladká", "Martin Holub", "Vincent Kríž" ]
https://www.semanticscholar.org/paper/59bf7a1a3f3d30a3fce0ae983dff420baf07bce5/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Capek and iPad in a pocket
Čapek a iPad v jedné kapse
Manual tagging and parsing are obligatory parts of Czech lessons, probably the most unpopular. We implement Capek editor for iPhone and iPad (to be published at AppStore in April 2013) to make tagging and parsing cool. Annotations by school children are automatically transformed into the Prague Dependency Treebank annotation framework.
Tvaroslovné a větné rozbory jsou povinnou součástí hodin češtiny, pravděpodobně tou nejméně oblíbenou. Pro zařízení iPhone a iPad připravujeme editor Čapek (publikace na AppStore duben 2013), díky kterému by mohlo být provádění rozborů "hustý". Z akademického pohledu jsou školské rozbory základnou pro rozšíření banky dat, kterou využíváme v aplikacích počítačového zpracování přirozeného jazyka.
[ "Barbora Hladká", "Jiří Hana", "Martin Wenisch" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Annotation on the tectogrammatical level. Additions to annotation manual (with respect to annotation of PDTSC and PCEDT)
Anotace na tektogramatické rovině. Dodatky k anotátorské příručce (s ohledem na anotování PDTSC a PCEDT)
Technical report provides new annotation rules for annotation of czech text on tectogrammatical level of the Prague Dependency Treebanks. The rules supplement main annotation manual (ÚFAL/CKL TR-2005-28) and were formed with respect to annotation of PCEDT 2.0 and PDTSC 2.0.
Technická zpráva shrnuje nová anotační pravidla pro anotovaní českých textů na tektogramatické rovině Pražských závislostních korpusů, která doplňují velký anotační manuál (ÚFAL/CKL TR-2005-28). Pravidla vznikla v souvislosti s anotováním korpusů PCEDT 2.0 a PDTSC 2.0.
[ "Marie Mikulová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
From PDT 2.0 to PDT 3.0 (Modifications and Complements)
Od PDT 2.0 k PDT 3.0 (Úpravy a doplňky)
Technical report describes improvement of the older scenario of PDT 2.0 that was applied in PDT 3.0. The description of changes and additions for the annotation procedure, the motivation for them, the procedure of their annotation and exemplification are given in the first part of this report. The extensions of the textual coreference, inter-sentential relations and genre classification of the texts included in PDT 3.0 are described in the other part of the report.
Technická zprava popisuje inovace, které byly provedeny ve verzi PDT 3.0 oproti verzi PDT 2.0. Vysvětluje jejich lingvistickou podstatu, zdůvodnění, pokyny pro anotaci a exemplifikaci. Její jedna část je věnována nově zavedeným nebo upraveným jevům tektogramatické roviny. V její další části jsou představeny jevy textové koreference, mezivětných vztahů, žánrové klasifikace dat a způsob jejich anotování v PDT 3.0.
[ "Marie Mikulová", "Eduard Bejček", "Jiří Mírovský", "Anna Nedoluzhko", "Jarmila Panevová", "Lucie Poláková", "Pavel Straňák", "Magda Ševčíková", "Zdeněk Žabokrtský" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Modifications and Complements of Prague Dependency Treebank (From PDT 2.0 to PDT 3.0)
Úpravy a doplňky Pražského závislostního korpusu (Od PDT 2.0 k PDT 3.0)
Technical report describes improvement of the older scenario of PDT 2.0 that was applied in PDT 3.0. The description of changes and additions for the annotation procedure, the motivation for them, the procedure of their annotation and exemplification are given in the first part of this report. The extensions of the textual coreference, inter-sentential relations and genre classification of the texts included in PDT 3.0 are described in the other part of the report.
Technická zprava popisuje inovace, které byly provedeny ve verzi PDT 3.0 oproti verzi PDT 2.0. Vysvětluje jejich lingvistickou podstatu, zdůvodnění, pokyny pro anotaci a exemplifikaci. Její jedna část je věnována nově zavedeným nebo upraveným jevům tektogramatické roviny. V její další části jsou představeny jevy textové koreference, mezivětných vztahů, žánrové klasifikace dat a způsob jejich anotování v PDT 3.0.
[ "Marie Mikulová", "Eduard Bejček", "Jiří Mírovský", "Anna Nedoluzhko", "Jarmila Panevová", "Lucie Poláková", "Pavel Straňák", "Magda Ševčíková", "Zdeněk Žabokrtský" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Do spoken texts differ from the written ones in valency?
Liší se mluvené a psané texty ve valenci?
In this article, we introduce the Prague Dependency Treebank of Spoken Czech. The syntactic and semantic annotation of this corpus has led to the expansion of PDT-Vallex, a valency lexicon of Czech verbs, which has previously been linked only to the annotation of written texts.
V tomto článku představujeme PDTSC a jeho anotaci, která vedla k rozšíření slovníku PDT-Vallex, valenčnímu slovníku českých sloves, původně vytvořenému pro psané texty.
[ "Marie Mikulová", "Jan Štěpánek", "Zdeňka Urešová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Unstructured Data: How to Describe a Natural Language?
Nestrukturovaná data:Jak popsat přirozený jazyk?
The talk focuses on basic syntactic relations, namely on dependency and word order, and on their mutual interplay. We present a formal framework that gives us a possibility of an economic and linguistically adequate description of these relations as well as their automatic processing. Finally, we shortly mention the application of the framework in language resources and NLP tools.
Přednáška se soustřeďuje na základní syntaktické vztahy, jmenovitě na závislostní relace a na slovosled, a dále na jejich vzájemnou souhru. Prezentujeme formální rámec, který nám umožňuje ekonomicky a přitom jazykově adekvátně popsat tyto vztahy a jejich automatizované zpracování. Nakonec krátce zmiňujeme využití tohoto formálního ránce v aplikacích NLP.
[ "Markéta Lopatková" ]
https://www.semanticscholar.org/paper/902e4ede99a84d882473a0ed82fae80737464835/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Describing an elusive: Syntactic relations in a natural languafe and their description.
Zachycení nezachytitelného: Syntaktické vztahy v přirozeném jazyce a jejich formální reprezentace.
The talk focuses on basic syntactic relations, dependency relationship and word order in particular, on their mutual relations, and examples of mutual interplay (including so called non-projective constructions). Then we present a formal frame allowing us to describe these relations ant their automated processing. Finally, we shortly mention various applications of the proposed framework in NLP tasks.
Přednáška se zaměří na základní syntaktické vztahy, především závislost a slovosled, na jejich vzájemné vztahy a příklady vzájemného ovlivňování (včetně tzv. neprojektivních konstrukcí). Dále představíme formální rámec, který umožňuje popis těchto vztahů a jejich automatické zpracování. V závěru bude krátce zmíněno využití tohoto rámce v jazykových zdrojích a (na nich budovaných) nástrojích NLP.
[ "Markéta Lopatková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Formalization of Word-Order Shifts by Restarting Automata
K formalizaci slovosledných posunů ponocí restartovacích automatů
The phenomenon of word order freedom plays an important role in syntactic analysis of many natural languages. This paper introduces a notion of a word order shift, an operation reflecting the degree of word order freedom of natural languages. The word order shift is an operation based upon word order changes preserving syntactic correctness, individual word forms, their morphological characteristics, and their surface dependency relations in a course of a stepwise simplification of a sentence, a so called analysis by reduction. The paper provides linguistic motivation for this operation and concentrates on a formal description of the whole mechanism through a special class of automata, so called restarting automata with the shift operation enhanced with a structured output, and their computations. The goal of this study is to clarify the properties of computations needed to perform (enhanced) analysis by reduction for free word order languages.
Fenomén volného slovosledu hraje důležitou roli v syntaktické analýze mnoha přirozených jazyků. Tento příspěvek zavádí pojem slovosledného posunu ('shiftu'), který odrážející stupeň volnosti slovosledu. Slovosledný posun je operace založená na slovosledných změnách zachovávajících syntaktickou správnost, jednotlivé slovní tvary, jejich morfologické charakteristiky a jejich syntaktické vztahy v průběhu tzv. redukční analýzy, tedy postupného zjednodušování věty. Příspěvek přináší lingvistickou motivaci pro tuto operaci a zaměřuje se na formální popis celkového mechanismu pomocí speciální třídy automatů, tzv. restartovacich automatů s operací shift obohacených strukturovaným výstupem strukturovaný výstup. Cílem této studie je objasnit vlastnosti výpočtů potřebných k provedení (rozšířené) redukční analýzy pro přirozené jazyky s volným slovosledem.
[ "Martin Plátek", "Markéta Lopatková" ]
https://www.semanticscholar.org/paper/30d862381f12827e8dde871353662ecf83a31580/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
May we understand each other?
Budeme si rozumět?
I will talk about a topic called "natural language processing" which deals with ways of getting the computer to understand human languages. I will tell you something about language corpora, different levels and methods of linguistic tagging, electronic dictionaries, methods of analysis and synthesis of natural languages, automatic translation. Everything will be shown on real examples.
Budu mluvit o tématu, kterému se říká "automatické zpracování přirozeného jazyka" a myslí se tím způsoby, jak přimět počítač, aby rozuměl nějaké lidské řeči. Povíme si něco o jazykových korpusech, různých úrovních a způsobech lingvistického značkování, elektronických slovnících, metodách analýzy a syntézy přirozeného jazyka, automatickém překladu. Vše si ukážeme na skutečných příkladech.
[ "Jaroslava Hlaváčová" ]
https://www.semanticscholar.org/paper/392ef8d538b819d9bd7cb24775e02ef4b71b1453/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Special domain data mining through DBpedia on the example of Biology
Dobývání oborových dat pomocí DBpedie na příkladu biologie
Wikipedia is not only a large encyclopedia, but lately also a source of linguistic data for various applications. Individual language versions allow to get the parallel data in multiple languages. Inclusion of Wikipedia articles into categories can be used to filter the language data according to a domain. In our project, we needed a large number of parallel data for training systems of machine translation in the field of biomedicine. One of the sources was Wikipedia. To select the data from the given domain we used the results of the DBpedia project, which extracts structured information from the Wikipedia articles and makes them available to users in RDF format. In this paper we describe the process of data extraction and the problems that we had to deal with, because the open source project like Wikipedia, to which anyone can contribute, is not very reliable concerning consistency.
Wikipedie slouží nejen jako rozsáhlá encyklopedie zasahující do mnoha odvětví, ale v poslední době stále častěji i jako zdroj jazykových dat pro nejrůznější aplikace. Jednotlivé jazykové mutace umožňují získat i paralelní data ve více jazycích. Zařazení článků wikipedie do kategorií potom může sloužit k filtrování jazykových dat. V našem projektu se zabýváme automatickým překladem textů v oboru biologie a lékařství, proto jsme potřebovali větší množství paralelních dat. Jedním ze zdrojů byla právě wikipedie. Pro výběr dat splňujících daná kritéria – tedy dané obory v daných jazycích – jsme využili projektu Dbpedia, který ze stránek wikipedie extrahuje strukturované informace a ve formátu RDF je zpřístupňuje uživatelům. V příspěvku popíšeme postup extrakce dat a problémy, které jsme museli řešit, neboť u otevřeného projektu jako wikipedie, do něhož může přispívat kdokoli, nelze spoléhat na konzistenci.
[ "Jaroslava Hlaváčová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Intensifying Verb Prefix Patterns in Czech and Russian
Slovesné předpony se stupňujícím efektem v češtině a ruštině
In Czech and in Russian there is a set of prefixes changing the meaning of imperfective verbs always in the same manner. The change often (in Czech always) demands adding a reflexive morpheme. This feature can be used for automatic recognition of words, without the need to store them in morphological dictionaries.
V češtině a v ruštině existuje množina prefixů, které mění význam nedokonavých sloves vždy stejným způsobem. Změna často (v českém jazyce vždy) vyžaduje přidání reflexivního morfému. Tato vlastnost nedokonavých sloves může být použita pro automatické rozpoznávání slov, aniž by bylo nutné ukládat takto vzniklé tvary do morfologických slovníků.
[ "Jaroslava Hlaváčová", "Anna Nedoluzhko" ]
https://www.semanticscholar.org/paper/6adbf70dfaec7a63891b6068be23c4cd272b0b3b/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Rule-Based Extraction of English Verb Collocates from a Dependency-Parsed Corpus
Extrakce kolokátů anglických sloves ze závislostního korpusu
We report on a rule-based procedure of extracting and labeling English verb collocates from a dependency-parsed corpus. Instead of relying on the syntactic labels provided by the parser, we use a simple topological sequence that we fill with the extracted collocates in a prescribed order. A more accurate syntactic labeling will be obtained from the topological fields by comparison of corresponding collocate positions across the most common syntactic alternations. So far, we have extracted and labeled verb forms and predicate complements according to their morphosyntactic structure. In the next future, we will provide the syntactic labeling of the complements.
Vytváříme systém pravidel, s jejichž pomocí značkujeme slovesa a jejich kolokáty v závislostním korpusu. Místo klasických syntaktických kategorií používáme šablonu, do níž zařazujeme nalezené kolokáty v přesně daném pořadí. Porovnáním jednotlivých pozic pro nejčastější syntaktické alternace daného slovesa získáme jeho přesnější syntaktický popis.
[ "Silvie Cinková", "Martin Holub", "Ema Krejčová", "Lenka Smejkalová" ]
https://www.semanticscholar.org/paper/59e5f86b28ea4d807203fa708d1f7efc6c72b38e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
A Case Study of a Free Word Order
Studie volného slovosledu
The paper aims at the investigation of the phenomenon of free word order. It concentrates on the relationship between (formal) dependencies and word order. The investigation is performed by means of a semiautomatic application of a method of analysis by reduction to Czech syntactically annotated data. The freedom of word order is expressed by means of a number of necessary shifts in the process of analysis by reduction. The paper shows that this measure provides a new view of the problem, it is orthogonal to measures reflecting the word order freedom based on a number of non-projective constructions or clitics in a sentence.
Článek zkoumá jev volného slovosledu. Soustředí se na vztak mezi (formální) závislostí a slovosledem. Zkoumání je provedeno aplikací poloautomatické metody redukční analýzy na česká syntaktický anotovaná data. Volnost slovosledu je vyjádřena pomocí počtu nutných záměn v průběhu redukční analýzy. Článek ukazuje, že tato míra je ortogonální k mírám vyjadřujícím volnost slovosledu pomocí počtu neprojektivních konstrukcí či klitik ve větě.
[ "Vladislav Kuboň", "Markéta Lopatková", "Jiří Mírovský" ]
https://www.semanticscholar.org/paper/694d6e26d9ecfce90352de615246ac7668febd93/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Automatic Processing of Linguistic Data as a Feedback for Linguistic Theory
Automatické zpracování jazykových dat jako ověření jazykové teorie
The paper describes a method of identifying a set of interesting constructions in a syntactically annotated corpus of Czech (the Prague Dependency Treebank) by application of an automatic procedure of analysis by reduction to the trees in the treebank. The procedure reveals certain linguistic phenomena that go beyond `dependency nature'. The paper contains discussion and analysis of individual phenomena, as well as the quantification of results of the automatic procedure on a subset of the treebank.
Článek se zabývá metodou identifikace zajímavých konstrukcí v syntakticky anotovaném korpusu češtiny (Pražském závislostním korpusu) pomocí aplikace automatické procedury redukční analýzy na stromy tohoto korpusu. Procedura odhaluje některé lingvistické jevy, které jdou za hranici "závislosti". Článek obsahuje diskuzi a analýzu jednotlivých jevů a rovněž kvantifikuje výsledky automatické procedury na podmnožině korpusu.
[ "Vladislav Kuboň", "Markéta Lopatková", "Jiří Mírovský" ]
https://www.semanticscholar.org/paper/2850e251b44654c77142ac48a315d396639c2235/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Prague Dependency Treebank 3.0
Prague Dependency Treebank 3.0
Dependency corpus annotated at morphological, syntactic, and deep syntactic levels. Previous version 2.5 has been enriched by annotation of discourse relations, bridging anaphora, extended textual coreference, and genres. Morphological level has been improved and some grammatemes have been revised.
Závislostní korpus anotovaný na rovině morfologické (2 miliony slov), syntaktické (1,5 milionu slov) a sémantické (0,8 milionu slov). Oproti verzi 2.5 přibyly opravy morfologie, revize gramatémů, anotace diskurzních vztahů, asociační anafory, rozšířené textové koreference a žánrů.
[ "Eduard Bejček", "Eva Hajičová", "Jan Hajič", "Pavlína Jínová", "Václava Kettnerová", "Veronika Kolářová", "Marie Mikulová", "Jiří Mírovský", "Anna Nedoluzhko", "Jarmila Panevová", "Lucie Poláková", "Magda Ševčíková", "Jan Štěpánek", "Šárka Zikánová" ]
https://www.semanticscholar.org/paper/c2e9552a6d3f683469e547cfba0841ff465c0c50/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Syntactic Identification of Occurrences of Multiword Expressions in Text using a Lexicon with Dependency Structures
Syntaktická identifikace výskytů víceslovných výrazů v textu s pomocí lexikonu se závislostními strukturami
We deal with syntactic identification of occurrences of multiword expression (MWE) from an existing dictionary in a text corpus. The MWEs we identify can be of arbitrary length and can be interrupted in the surface sentence. We analyse and compare three approaches based on linguistic analysis at a varying level, ranging from surface word order to deep syntax. The evaluation is conducted using two corpora: the Prague Dependency Treebank and Czech National Corpus. We use the dictionary of multiword expressions SemLex, that was compiled by annotating the Prague Dependency Treebank and includes deep syntactic dependency trees of all MWEs.
V tomto článku se zabýváme identifikací výskytů víceslovných výrazů z existujícího slovníku v textovém korpusu. Víceslovné výrazy mohou být libovolné délky a přerušeny v povrchovém pořádku slov. Analyzujeme a porovnáváme tři různé přístupy využívající lingvistické analýza na různých rovinách.
[ "Eduard Bejček", "Pavel Straňák", "Pavel Pecina" ]
https://www.semanticscholar.org/paper/07e562362c0d681d9f7c4250e7f53cd1ec1494c4/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Czech adverbs with the suffix -o in theoretical description and in a syntactically annotated corpus
Česká adverbia s příponou -o v teoretickém popisu a v syntakticky anotovaném korpusu
Both an adverb with the suffix -e and an adverb with the suffix -o are derived from many adjectives in Czech. Adverbs with the latter suffix are often a component of the predicate. The predicative syntactic function has become one of the main arguments to isolate the adverbs with the suffix -o as a separate part of speech (Komárek, 1954); nevertheless, this proposal has not been widely accepted. In the present paper, a detailed analysis of the adverbs with the suffix -o is carried out on the basis of material from the Czech National Corpus. The analysis documents that solely a part of the adverbs with the suffix -o specializes in the predicative function, other adverbs with this suffix fulfill an adverbial function as well, or do not fulfill the predicative function at all. Besides the syntactic function and the lexical meaning of these adverbs, their relation to the base adjective and the impact of context is taken into consideration when discussing the representation of these adverbs within the deep-syntactic annotation of the Prague Dependency Treebank.
V příspěvku analyzujeme syntaktické funkce deadjektivních adverbií s příponou -o na základě materiálu z Českého národního korpusu. Pouze část těchto adverbií se specializuje na predikativní funkci, jiná plní kromě této funkce také funkci příslovečného určení a některá se v predikativní funkci neuplatňují vůbec. Kromě syntaktické funkce těchto adverbií zkoumáme také jejich význam a vztah k základovému adjektivu a navrhujeme, jak je reprezentovat v hloubkově syntaktické anotaci Pražského závislostního korpusu.
[ "Magda Ševčíková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Deadjectival derivates in Czech as syntactic vs. lexical derivates
Deadjektivní deriváty v češtině jako deriváty syntaktické vs. lexikální
Deadjectival derivates are a representative sample of Czech derivation. On the basis of the semantic relation to their base adjectives, they can be classified into two groups: into syntactic derivates, which have the same meaning as their base adjectives but fulfill different syntactic functions (e.g. deadjectival adverbs), and lexical derivates, which share the syntactic function with the base adjectives but differ from them in meaning (e.g. diminutive adjectives). The present paper focuses on deadjectival nouns with the suffix -ost, which express a quality. Based on the analysis of data from the corpus SYN2010, we try to document that the meaning of quality is closely interconnected with the meaning of a bearer of the quality, and propose to revise Dokulil’s classification of these nouns as syntactic derivates. The classification of a derivate as a syntactic or lexical one is directly reflected in the deep-syntactic annotation of the Prague Dependency Treebank.
Slova odvozená od adjektiv představují zajímavý, reprezentativní vzorek slovotvorných derivačních procesů v češtině: patří ke slovním druhům substantiv, adjektiv, sloves a adverbií, vzájemně se liší v řadě aspektů, mimo jiné z hlediska sepjetí jejich lexikálního významu s významem slovotvorným. V příspěvku probereme možnost třídění deadjektivních derivátů právě z hlediska jejich vztahu k základovému adjektivu, pracujeme přitom s Kuryłowiczovou (1936) koncepcí derivace syntaktické a lexikální, která je v Dokulilových slovotvorných pracích reflektována. Z klasifikace deadjektivních derivátů jako syntaktických nebo lexikálních vyvodíme přímé důsledky pro zachycení těchto slov v hloubkověsyntaktické anotaci Pražského závislostního korpusu. Syntaktické deriváty je možné reprezentovat základovým adjektivem, protože jeho odlišná syntaktická funkce je v tomto syntakticky anotovaném korpusu zachycena speciálním atributem. Oproti tomu lexikální deriváty se budou od svého základového slova lišit lexikální hodnotou.
[ "Magda Ševčíková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Productivity of selected deadjectival suffixes in Czech
Produktivita vybraných deadjektivních sufixů v češtině
In the present paper, central ideas of the work on productivity by the Czech linguist Miloš Dokulil are presented and placed in the context of recent approaches to productivity. Based on material from large corpora of Czech, productivity of four Czexh suffixes (-ost, -ství/ctví, -ismus, -ita) is analyzed; it is demonstrated that productivity measures used for quantification of productivity give different results.
V příspěvku shrnujeme základní rysy Dokulilovy teorie slovotvorné produktivity a zasazujeme ji do kontextu zkoumání produktivity v posledních desetiletích. Na materiálu z velkých korpusů češtiny zkoumáme produktivitu čtyř českých sufixů (-ost, -ství/ctví, -ismus, -ita) a ukazujeme, že jednotlivé míry používané ke kvantifikaci produktivity vedou k rozdílným výsledkům.
[ "Magda Ševčíková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Determining Productivity in Word-Formation: Deadjectival Suffixes in Czech
Slovotvorná produktivita: deadjektivní přípony odvozující názvy vlastností
Though the term of productivity in word-formation seems to be intuitively clear and the idea (under various terms) has been present in linguistics for centuries, there are many approaches to productivity in contemporary linguistics, which differ in several aspects. In the talk, I briefly sketch selected issues of the study of productivity focusing on the last two or three decades, during which productivity has become one of the central topics of the research into word-formation. In this context, I refer to the main ideas of the work on productivity by Miloš Dokulil, whose word-formation theory has become a widely respected and, in fact, the only common ground of word-formation descriptions in Czech linguistics. I focus on his opposition of the so-called systemic and empirical productivity and present some suggestions on how Dokulil’s ideas could be applied to the current, corpus-based research in productivity, by analyzing the productivity of four suffixes which are used to derive names of qualities in Czech (-ost, -ství/-ctví, -ismus and -ita).
V přednášce se zaměříme na Dokulilovu teorii produktivity, která je integrální součástí jeho široce respektované slovotvorné teorie, připomeneme především jeho distinkci produktivity systémové a empirické. Při analýze přípon -ost, -ství/-ctví, -ismus a -ita, které jsou v češtině používány (mimo jiné) k odvozování názvů vlastností, se kromě jejich systémových rysů pokusíme stanovit míru produktivity také na základě frekvenčních údajů z velkých korpusových dat.
[ "Magda Ševčíková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Working with the Prague Dependency Treebank
Práce s Pražským závislostním korpusem
The workshop presented the Prague Dependency Treebank and basic tools that can be used to search it. Unlike more widely known Czech National Corupus, the PDT contains two layers of syntactic annotation. The theory of the Functional Generative Description upon which the treebank is built was presented, and the aspects were described where the theory and the treebank diverge.
Workshop představil Pražský závislostní korpus a základní nástroje, které lze používat k jeho prohledávání. Na rozdíl od dnes již poměrně známého Českého národního korpusu obsahuje PDT i anotace syntaktických struktur, a to hned na dvou rovinách. Seznámili jsme se s teorií funkčního generativního popisu, na jejímž základě byl korpus vybudován, a popsali jsme si aspekty, kterými se od teoretických východisek odchyluje.
[ "Jan Štěpánek" ]
https://www.semanticscholar.org/paper/19b67fdf9a10055018283dae6e460beae75fd94e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Querying Dependency Treebanks in PML-TQ
Prohledávání Pražského závislostního korpusu v PML-TQ
A tutorial describing PML-TQ, a tool for searching syntactically annotated corpora. Two forms of the client interface were presented, with examples over Czech, English, and German data.
Tutoriál přibližující práci s nástrojem PML-TQ určeným pro prohledávání syntakticky anotovaných korpusů. Představeny byly dvě verze klientského rozhraní, příklady používaly data v češtině, angličtině a němčině.
[ "Jan Štěpánek" ]
https://www.semanticscholar.org/paper/b3982035612762b0ec1eda4c9f73771afeb0217f/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Alex Dialogue Systems Framework - v0.1
Alex Dialogue Systems Framework - v0.1
This is the Alex Dialogue Systems Framework - 2013 release. This framework is being developed by the dialogue systems group at UFAL - http://ufal.mff.cuni.cz/ - the Institute of Formal and Applied Linguistics, Faculty of Mathematics and Physics, Charles University in Prague, Czech Republic. The purpose of this work is to facilitate research into and development of spoken dialogue systems. The main goals are: to provide baseline components need for a building spoken dialogue systems (SDSs), to provide example implementations of SDSs for several domains, to provide tools for processing dialogue system interactions logs, e.g. for audio transcription, semantic annotation, or SDSs evaluation. Implemented features: VOIP using PJSIP 2.1 with some modifications, ASR using OpenJulius, GoogleASR or KALDI, VAD using Gaussian Mixture Models or Feed-Forward Neural Networks, SLU using a set of logistic regression classifiers for detecting dialogue acts, DM using probabilistic discriminative dialogue state tracking and handcrafted policies, NLG using template based generation possibly with efficient inflection into the correct surface form for morphologically rich languages, TTS using flite, VoiceRSS and SpeechTech, evaluation of dialogue systems using Amazon Mechanical Turk crowdsourcing platform, building acoustic models using the HTK and KALDI toolkits.
Toto je vydání Alex Dialogue Systems Framework pro rok 2013. Tento framework je vyvýjen skupinou zabývající se výzkumem v oblasti hlasových dialogových systémů na ÚFAL - http://ufal.mff.cuni.cz/ - Ústav formální a aplikované lingvistiky, MFF, UK. Hlavní cíle vývoje jsou: poskytnout základní komponenty pro vývoj SDS, poskytnout příklady hlasových dialogových systémů, poskytnout nástroje pro zpracopvání záznamů dialogů, například audio přepisy, sémantické anotace, nebo vyhodnocení dialogových systémů. Implemntované vlastnosti: VOIP používající PJSIP 2.1, ASR používající OpenJulius, GoogleASR nebo KALDI, VAD používající Gaussian Mixure Models nebo Feed-Forward Neural Networks, SLU používající množinu klasifikátorů založených na logistické regresi, DM používající diskriminativní pravěpodobnostní aktualizaci stavu a ručně psané strategie řízení, NLG používající šablon s možností využití šablon založených na tektogramatických stromech pro správnou syntézu povrchové realizace, TTS používající flite, VoiceRSS nebo SpeechTech, vyhodnocení dialogových systémů používajáící Amazon Mechanical Turk, tvorba akustických modelů pomocí HTK a KALDI.
[ "Filip Jurčíček", "Ondřej Dušek", "Matěj Korvas", "Lukáš Žilka", "Ondřej Plátek", "David Marek" ]
https://www.semanticscholar.org/paper/14a6c0a97c82c8d502ce53a60698ce630bafb294/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Automatic post-editing of phrase-based machine translation outputs
Automatická post-editace výstupů frázového strojového překladu
We present Depfix, a system for automatic post-editing of phrase-based English-to-Czech machine translation outputs, based on linguistic knowledge. First, we analyzed the types of errors that a typical machine translation system makes. Then, we created a set of rules and a statistical component that correct errors that are common or serious and can have a potential to be corrected by our approach. We use a range of natural language processing tools to provide us with analyses of the input sentences. Moreover, we reimplemented the dependency parser and adapted it in several ways to parsing of statistical machine translation outputs. We performed both automatic and manual evaluations which confirmed that our system improves the quality of the translations.
Představujeme Depfix, systém pro samočinnou post-edititaci výstupů frázových strojových překladů z angličtiny do češtiny, založený na jazykovědných znalostech. Nejprve jsme rozebrali druhy chyb, kterých se dopouští typický strojový překladač. Poté jsme vytvořili sadu pravidel a statistickou komponentu, které opravují takové chyby, které jsou běžné nebo závažné a může přicházet v úvahu jejich oprava pomocí našeho přístupu. Používáme řadu nástrojů pro zpracování přirozeného jazyka, které nám poskytují rozbor vstupních vět. Navíc jsme reimplementovali závislostní analyzátor a několika způsoby jej upravili pro provádění rozboru výstupů statistických strojových překladačů. Provedli jsme automatická i ruční vyhodnocení, která potvrdila, že kvalita překladů se zpracováním v našem systému zlepšuje.
[ "Rudolf Rosa" ]
https://www.semanticscholar.org/paper/a6dfb61b255bfe585acacaab16360a633b44d226/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Depfix: Automatic post-editing of phrase-based machine translation outputs
Depfix: Automatická post-editace výstupů frázového strojového překladu
Depfix is a system for automatic post-editing of phrase-based English-to-Czech machine translation outputs, based on linguistic knowledge. We analyzed the types of errors that a typical machine translation system makes, and created a set of rules and a statistical component that correct some of the errors. We use a range of natural language processing tools to provide us with analyses of the input sentences. Moreover, we reimplemented the dependency parser and adapted it in several ways to parsing of statistical machine translation outputs. We performed both automatic and manual evaluations which confirmed that our system improves the quality of the translations.
Depfix je systém pro samočinnou post-edititaci výstupů frázových strojových překladů z angličtiny do češtiny, založený na jazykovědných znalostech. Rozebrali jsme druhy chyb, kterých se dopouští typický strojový překladač, a vytvořili jsme sadu pravidel a statistickou komponentu, které opravují některé z těchto chyby. Používáme řadu nástrojů pro zpracování přirozeného jazyka, které nám poskytují rozbor vstupních vět. Navíc jsme reimplementovali závislostní parser a několika způsoby jej upravili pro provádění rozboru výstupů statistických strojových překladačů. Provedli jsme automatická i ruční vyhodnocení, která potvrdila, že kvalita překladů se zpracováním v našem systému zlepšuje.
[ "Rudolf Rosa" ]
https://www.semanticscholar.org/paper/a6dfb61b255bfe585acacaab16360a633b44d226/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Error Correction of PB SMT Outputs with automatic post-editing shown on English to Czech translation
Oprava chyb ve výstupech frázového statistického strojového překladu ukázaná na překladu z angličtiny do češtiny
Some things are hard for PB SMT -- negation, agreement, passives... Automatic post-editing can often help -- a handful of simple rules might do the job very well. NLP tools are typically useful -- tagger, parser, morphological generator; adapt them to SMT outputs if possible.
Některé veci jsou pro PB SMT obtížné -- negace, shoda, trpný rod... Automatická post-editace může často pomoci -- i hrstka jednoduchých pravidel může fungovat velmi dobře. NLP nástroje jsou typicky užitečné -- tagger, parser, morphologický generátor; adaptujte je pro zpracování výstupů SMT, je-li to možné.
[ "Rudolf Rosa" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Deepfix: Statistical Post-editing of Statistical Machine Translation Using Deep Syntactic Analysis
Deepfix: Statistická post-editace statistického strojového překladu s využitím hloubkové syntaktické analýzy
Deepfix is a statistical post-editing system for improving the quality of statistical machine translation outputs. It attempts to correct errors in verb-noun valency using deep syntactic analysis and a simple probabilistic model of valency. On the English-to-Czech translation pair, we show that statistical post-editing of statistical machine translation leads to an improvement of the translation quality when helped by deep linguistic knowledge.
Deepfix je statistický post-editovací systém pro zlepšování kvality výstupů statistického strojového překladu. Pokouší se opravovat chyby ve slovesné a substantivní valenci za použití hloubkové syntactické analýzy a jednoduchého pravděpodobnostního valenčního modelu. Na jazykovém páru angličtina-čeština ukazujeme že, pokud je podporována hlubokou lingvistickou znalostí, statistická post-editace statistického strojového překladu vede ke zlepšení kvality překladu.
[ "Rudolf Rosa", "David Mareček", "Aleš Tamchyna" ]
https://www.semanticscholar.org/paper/ce1e458dfafb80b5f2773c617780c7ea3125b9c9/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Translation of "It" in a Deep Syntax Framework
Překlad anglického "it" pomocí tektogramatiky
We present a novel approach to the translation of the English personal pronoun it to Czech. We conduct a linguistic analysis on how the distinct categories of it are usually mapped to their Czech counterparts. Armed with these observations, we design a discriminative translation model of it, which is then integrated into the TectoMT deep syntax MT framework. Features in the model take advantage of rich syntactic annotation TectoMT is based on, external tools for anaphoricity resolution, lexical co-occurrence frequencies measured on a large parallel corpus and gold coreference annotation. Even though the new model for it exhibits no improvement in terms of BLEU, manual evaluation shows that it outperforms the original solution in 8.5% sentences containing it.
Článek se věnuje překladu anglického zájmena "it" do češtiny. Navrhli jsme diskriminativní model a zapojili ho do překladu přes tektogramatickou rovinu systémem TectoMT. Výsledkem je zlepšení kvality překladu v 8.5% vět obsahujícich "it".
[ "Michal Novák", "Anna Nedoluzhko", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/c2281b813e6b5c8f0effd72898e3572f498ff53d/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Two Case Studies on Translating Pronouns in a Deep Syntax Framework
Dvě připadové studie překladu zájmen přes tektogramatickou rovinu
We focus on improving the translation of the English pronoun it and English reflexive pronouns in an English-Czech syntax-based machine translation framework. Our evaluation both from intrinsic and extrinsic perspective shows that adding specialized syntactic and coreference-related features leads to an improvement in trans- lation quality.
Příspěvek se zaměřuje na zlepšování překladu anglického zájmena "it" a anglických reflexivních zájmen v systému strojového překladu z angličtiny do češtiny přes tektogramatickkou rovinu.
[ "Michal Novák", "Zdeněk Žabokrtský", "Anna Nedoluzhko" ]
https://www.semanticscholar.org/paper/c1c1fd3aab35bb4fbef9b7becd62014a9c9f4e09/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
A Trilingual Learner Corpus illustrating European Reference Levels
Trojjazyčný žákovský korpus ilustrující CEFR
Since its publication in 2001, the Common European Framework of Reference for Languages (CEFR) has gained a leading role as an instrument of reference for language teaching and certification and for the development of curricula (cf. CEFR 2001). At the same time, there is a growing concern about CEFR reference levels being insufficiently illustrated, leaving practitioners without comprehensive empirical characterizations of the relevant distinctions. This is particularly the case for languages other than English (cf. e.g. Hulstijn 2007, North 2000). The MERLIN project addresses this demand for the first time for Czech, German and Italian by developing a didactically motivated online platform that enables CEFR users to explore authentic written learner productions that have been related to the CEFR levels in a methodologically sophisticated and rigorous way. The core of the multilingual online platform is a trilingual learner corpus relying on a cross-linguistic design and composed of roughly 2300 learner texts produced in standardized language certifications (as used by telc, DE, UJOP-Charles University in Prague, CZ as well as UNIcert, DE) validly related to the CEFR, covering the levels A1-C1. The aim of this paper is both to present the MERLIN project and its corpus, and to discuss its current state. In addition to providing preliminary statistics, we detail three key aspects: (1) the data collection and transcription, (2) the creation of the annotation schemata and (3) the technical design of the workflow devised to compile and query the corpus. For each of these aspects, we highlight its requirements and the means employed for the underlying tasks. We also discuss the challenges induced by the cross-linguistic nature of the project: particularities of the languages from three different families (Slavic, Germanic and Romance) have to be covered, both on a linguistic and a technical perspective.
Od publikace v roce 2011, ziskala CEFR vedoucí úlohu ve výuce a certifikaci cizích jazyků. Bohužel jednotlivé úrovně CEFR nejsou dostatečně ilustrovány. Cílem projektu Merlin je vyřešit tento problém pro češtinu, němčinu a italštinu.
[ "Andrea Abel", "Lionel Nicolas", "Jirka Hana", "Barbora Štindlová", "Katrin Wisniewski", "Detmar Meurers", "Serhiy Bykh" ]
https://www.semanticscholar.org/paper/1c15138c7699e98ab52b3cffd46d3e1ae39094bb/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Determining Compositionality of Word Expressions Using Various Word Space Models and Measures
Určování kompozicionality slovních výrazů pomocí různých vektorových modelů a metod
This paper presents a comparative study of 5 different types of Word Space Models (WSMs) combined with 4 different compositionality measures applied to the task of automatically determining semantic compositionality of word expressions. Many combinations of WSMs and measures have never been applied to the task before. The study follows Biemann and Giesbrecht (2011) who attempted to find a list of expressions for which the compositionality assumption – the meaning of an expression is determined by the meaning of its constituents and their combination does not hold. Our results are very promising and can be appreciated by those interested in WSMs, compositionali
Tento článek prezentuje komparativní studii 5 různých typů slovních vektorových modelů a 4 různých měr pro měření kompozicionality slovních výrazů.
[ "Lubomír Krčmář", "Karel Ježek", "Pavel Pecina" ]
https://www.semanticscholar.org/paper/53befa780540191af005fb2906f523baa15a62b7/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Determining Compositionality of Word Expressions Using Word Space Models
Určování kompozicionality slovních výrazů pomocí vektorových modelů
This research focuses on determining semantic compositionality of word expressions using word space models (WSMs). We discuss previous works employing WSMs and present differences in the proposed approaches which include types of WSMs, corpora, preprocessing techniques, methods for determining compositionality, and evaluation testbeds. We also present results of our own approach for determining the semantic compositionality based on comparing distributional vectors of expressions and their components. The vectors were obtained by Latent Semantic Analysis (LSA) applied to the ukWaC corpus. Our results outperform those of all the participants in the Distributional Semantics and Compositionality (DISCO) 2011 shared task.
Tento článek shrnuje dosavadní modely pro určování kompozicionality slovních výrazů pomocí vektorových modelů a prezentuje náš vlastní přístup založený na latentní sémantické analýze. Experimenty jsou provedeny na testovacích datech DISCO 2011.
[ "Lubomír Krčmář", "Karel Ježek", "Pavel Pecina" ]
https://www.semanticscholar.org/paper/6f78e1d45491039e5a199994b01295b90cae66bd/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
EnTam: An English-Tamil Parallel Corpus
EnTam: Angličtina-Tamilština Paralelní Korpus
We have collected English-Tamil bilingual data from some of the publicly available websites for NLP research involving Tamil. The standard set of processing has been applied on the the raw web data before the data became available in sentence aligned English-Tamil parallel corpus suitable for various NLP tasks. The parallel corpora cover texts from bible, cinema and news domains.
Shromáždili jsme anglicky Tamil dvojjazyčných data z některé z veřejně dostupných internetových stránek pro NLP výzkum zahrnující Tamil.Standardní sada zpracování byla použita na nezpracovaná data, než webových údaje byly k dispozici ve větě vyrovnaném anglicky Tamil paralelní korpus vhodný pro různé úlohy NLP. Paralelní korpusy krycí texty z bible, kino a zpravodajských domén.
[ "Loganathan Ramasamy", "Ondřej Bojar", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/b9afc8918fa6a8dc76d0ebfefdf51f93beea12e5/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
TamilTB - Tamil Dependency Treebank 1.0
TamilTB - závislostní korpus tamilštiny, verze 1.0
TamilTB 1.0 improves over the very first release of the Tamil dependency treebank (TamilTB v0.1). The current version introduces many improvements over the orthography and morphological tagset of the data. Now, the primary data representation of the treebank uses UTF-8 encoding.
TamilTB 1.0 zlepšuje v průběhu prvního vydání Tamil závislostní korpus (TamilTB v0.1). Aktuální verze přináší řadu vylepšení oproti pravopis a morfologické tagset údajů. Nyní, primární datový reprezentace korpusu používá kódování UTF-8.
[ "Loganathan Ramasamy", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/9f0db0e4322735482bff63935a7533514ffa7e9a/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Developing Sentiment Annotator in UIMA – the Unstructured Management Architecture for Data Mining Applications
Vývoj anotátoru pro postojovou analýzu v prostředí UIMA
In this paper we present UIMA – the Unstructured Information Management Architecture, an architecture and software framework for creating, discovering, composing and deploying a broad range of multi-modal analysis capabilities and integrating them with search technologies. We describe the elementary components of the framework and how they are deployed into more complex data mining applications. The contribution is based on our experience in work on the sentiment analysis task for IBM Content Analytics project.
V článku představujeme naši práci na rozšíření projektu IBM Content Analytics o analýzu sentimentu za využití softwaru UIMA. Popisujeme jednotlivé komponenty a jejich propojení v rámci komplexnějších dataminigových aplikací.
[ "Jan Hajič, jr.", "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/6e6bfa397f7495f01d837d1510832789c7fc8ce7/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Comparison of Bayesian Discriminative and Generative Models for Dialogue State Tracking
Srovnání Baysovských diskriminativiních a generativních modelů pro sledování dialogového stavu
In this paper, we describe two dialogue state tracking models competing in the 2012 Dialogue State Tracking Challenge (DSTC). First, we detail a novel discriminative dialogue state tracker which directly estimates slot-level beliefs using deterministic state transition probability distribution. Second, we present a generative model employing a simple dependency structure to achieve fast inference. The models are evaluated on the DSTC data, and both significantly outperform the baseline DSTC tracker.
V tomto článku Srovnání Baysovských diskriminativiních a generativních modelů pro sledování dialogového stavu. Bylo zjistěno že oba prezentované modely mají přibližně stejnou přesnost, ale diskriminativní model je zhruba 20 krát výpočetně efektivnější.
[ "Lukáš Žilka", "David Marek", "Matěj Korvas", "Filip Jurčíček" ]
https://www.semanticscholar.org/paper/64a4cab5c79e4fe33e1aecf3d64d559f1c02aeda/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Detecting Entity Relations in Texts
Detekce vztahů mezi entitama v textech
We present our initial steps towards a linguistic processing of texts to detect semantic relations there. Our work is an essential part of the INTLIB project whose aim is to provide a more efficient and user-friendly tool for querying textual documents other than full-text search. This tool is proposed as a general framework which can be modified and extended for particular data domains. Currently, we focus on Czech legal texts. We present the application JTagger which detects references in court decisions and the application RExtractor which extracts relations between entities from Czech laws.
Prezentace prvních kroků směrem ke lingvistickému zpracování textů pro detekci sémantických vztahů v nich. Naše práce je klíčovou součastí projektu INTLIB, kterého cílem je vytvořit efektivní a uživatelsky přívětivý nástroj pro dotazování nad textovými dokumenty, odlišný od full-textu. Tento nástroj je navrhován jako obecní framework, který lze snadno modifikovat a rozšiřovat pro další domény. Pilotní doménou projektu jsou legislativní texty. Prezentujeme aplikaci JTagger, která detekuje reference v soudních rozhodnutích a aplikaci RExtractor, která extrahuje vztahy mezi entitami v českých zákonech.
[ "Vincent Kríž" ]
https://www.semanticscholar.org/paper/e14fa05cf95d0a54ac8ff296eef50e5d608392c0/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Detecting Semantic Relations in Texts and Their Integration with External Data Resources
Detekce sémantických vazeb v textech a jejich integrace s externími datovými zdroji
We present our initial steps towards a linguistic processing of texts to detect semantic relations in them. Our work is an essential part of the INTLIB project whose aim is to provide a more efficient and user-friendly tool for querying textual documents other than full-text search. This tool is proposed as a general framework which can be modified and extended for particular data domains. Currently, we focus on Czech legal texts.
Prezentujeme první kroky směrem k lingvistickému zpracování textů pro detekci sémantických vazeb. Naše práce je důležitou částí projektu INTLIB, který má za cíl vytvořit efektivnější a uživatelsky více přívětivý nástroj na dotazování v textových dokumentech než full-textové vyhledávání. Tento nástroj je navrhnut jako obecný framework, který bude možno modifikovat a rozšiřovat pro konkrétní datové domény. Aktuálně se zaměřujeme na české legislativní dokumenty.
[ "Vincent Kríž" ]
https://www.semanticscholar.org/paper/819a432a1944f096a5baf453cffc1470c56daab6/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
JTagger - Automatic Reference Recognition in Court Decisions
JTagger - Automatická detekce referencí v soudních rozhodnutích
We address the task of reference detection and classification in Czech court decisions. This task is a typical named-entity recognition task where the entities are references (links) to other documents. We apply a supervised machine learning approach, namely Hidden Markov Models. A supervised methodology requires manual annotation of training data so we annotated 300 court decisions.
Aplikace detekuje reference v českých soudních rozhodnutích. Tento úkol chápeme jako typický úkol pro identifikaci jmenných entit, kde entity jsou reference (odkazy) na jiné dokumenty. Aplikujeme přístupy strojového učení s učitelem, konkrétně skryté Markovovy modely. Protože metody strojového učení s učitelem vyžadují manuálně anotovaná trénovací data, anotovali jsme 300 soudních rozhodnutí.
[ "Vincent Kríž", "Barbora Hladká" ]
https://www.semanticscholar.org/paper/3c38604169f3a84933b3a0785a55073ed90d98b5/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Halfway through the INTLIB project
Projekt INTLIB v polovině řešení
The project INTLIB, an INTelligent LIBrary, aims to provide a more efficient and user-friendly tool for querying documents other than full-text [1]. The work is funded by the Technology Agency of the Czech Republic (grant no. TA02010182). On the input we assume a collection of documents related to a particular problem domain (e.g., legislation, medicine, environment, architecture, etc.). In the first phase we extract a knowledge base from the documents using natural language processing tools. In the second phase we deal with efficient and user friendly visualization and browsing (querying) of the extracted knowledge. The project team consists of researchers coming from MFF UK ([2], [3]) and businessmen from Sysnet Ltd. [4]. We will present the results we achieved so far on legislative domain. In addition, we, researchers, will describe our experience gained while cooperating with the commercial partner. [1] http://ufal.mff.cuni.cz/intlib [2] http://www.ksi.mff.cuni.cz/xrg/ [3] http://ufal.mff.cuni.cz [4] http://www.sysnet.cz/
Cílem projektu INTLIB, an INTelligent LIBrary, je poskytnout efektivní a uživatelsky přívětivý nástroj pro inteligentnější dotazování než nabízí full text [1]. Projekt je financován Technolofgickou agenturou ČR, grant č. TA02010182. Na vstupu uvažujeme kolekci dokumentů určité domény (např. legislativa, medicína, životní prostředí, architektura). V první fázi jsou z této kolekce extrahovány znalosti pomocí nástrojů počítačového zpracování přirozeného jazyka. Ve druhé fázi je navrženo inteligentní vyhledávání spolu s vizualizací. Projektový tým tvoří badatelé z MFF UK ([2], [3]) a podnikatelé ze společnosti Sysnet Ltd. [4]. Představíme výsledky, které jsme doposud dosáhli na legislativní doméně. [1] http://ufal.mff.cuni.cz/intlib [2] http://www.ksi.mff.cuni.cz/xrg/ [3] http://ufal.mff.cuni.cz [4] http://www.sysnet.cz/
[ "Martin Nečaský", "Barbora Hladká" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Linked Open Data for Legislative Domain - Ontology and Experimental Data
Propojená otevřená data pro legislativní doménu -- ontologie a experimentální data
In this paper, we present an approach of publishing legislative content as Linked Open Data (LOD). LOD is a set of principles of publishing data on the Web in a machine-readable way so that links between different data sets, possibly published by different publishers, can be created. Therefore, LOD enable not only to publish data but also enrich with other existing data published according to the principles. We present what is the motivation for publishing legislation as LOD and what benefits can be gained. We then introduce a legislative ontology which builds on existing commonly used ontologies. We also show how we converted existing sources of legislation in Czech Republic to LOD.
Představujeme přístup k publikování legislativních dokumentů ve formě Propojených otevřených dat(LOD). LOD je sada principů, jak publikovat data na webu ve strojově čitelné podobě, aby bylo možné propojit různé zdroje dat.
[ "Martin Nečaský", "Tomáš Knap", "Jakub Klímek", "Irena Holubová", "Barbora Hladká" ]
https://www.semanticscholar.org/paper/556c8038332823882bf139557f383ce28bbf4f9e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Tools for Machine Translation Quality Inspection
Nástroje pro kontrolu kvality strojového překladu
This report describes Addicter, tool for automatic error detection and evaluation, providing its user also with graphical interface useful for browsing through the dataset.
Tato zpráva popisuje Addicter, nástroj pro automatickou detekci a vyhodnocení chyb, který dává uživateli k dispozici také grafické rozhraní, užitečné pro prohlížení dat.
[ "Jan Berka", "Ondřej Bojar", "Mark Fishel", "Maja Popović", "Daniel Zeman" ]
https://www.semanticscholar.org/paper/1f53a3c05a42183d6fa33f36851ed87b4bcf845a/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Results of the WMT13 Metrics Shared Task
Výsledky úlohy metrik strojového překladu na WMT13
This paper presents the results of the WMT13 Metrics Shared Task. We asked participants of this task to score the outputs of the MT systems involved in WMT13 Shared Translation Task. We collected scores of 16 metrics from 8 research groups. In addition to that we computed scores of 5 standard metrics such as BLEU, WER, PER as baselines. Collected scores were evaluated in terms of system level correlation (how well each metric’s scores correlate with WMT13 official human scores) and in terms of segment level correlation (how often a metric agrees with humans in comparing two translations of a particular sentence).
Článek přináší výsledky společné úlohy metrik strojového překladu na workshopu WMT13. Účastníci úlohy nechali svými systémy vyhodnotit výstupy strojového překladu. Sebrali jsme hodnocení pomocí celkem 16 metrik od 8 týmů a doplnili hodnocení pomocí 5 standardních metrik (BLEU, WER, PER ad.). Hodnocení pak byla porovnána z hlediska korelace s lidským hodnocením a to jak na úrovni jednotlivých vět, tak na úrovni celého testovacího textu.
[ "Matouš Macháček", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/ac67ee39cbb30e12aa4257b8d6ee4f1c104f7aba/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
How the MT system from Matfyz beated Google
Jak překladač z Matfyzu porazil Google
A popularizing description of Chimera, our MT system, for HNFuture, the scientific supplement of the online journal iHNed.
Popis soutěžního systému Chiméra populární formou pro HNFuture, vědeckou přílohu elektronického deníku iHNed.
[ "Michal Kalina", "Ondřej Bojar" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Building a Corpus of Evaluative Sentences in Multiple Domains
Budování korpusu evaluativních vět z mnoha významových domén
The paper describes the process of building a Czech language corpus of annotated sentences for sentiment analysis tasks, the SubLex 1.0 subjectivity lexicon, the data, the process of annotation and evaluating experiments.
V příspěvku je popsán proces vytváření korpusu anotovaných vět pro účely postojové analýzy v češtině. Dále je popsán vznik slovníku subjektivních výrazů Sublex 1.0, struktura použitých dat, proces anotace evaluativních stavů a komputační experimenty hodnotící kvalitu anotace.
[ "Jana Šindlerová", "Kateřina Veselovská" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Verb Valency and Argument Non-correspondence in a Bilingual Treebank
Valence sloves a provázání nestejných valenčních doplnění v dvoujazyčném závislostním korpusu
In this paper we present a contrastive study of one interesting non-correspondence between deep syntactic valency structures of two different languages. On the material of the Prague Czech-English Dependency Treebank we observe sentences in which an Addressee argument in one language is linked translationally to a Patient argument in the other one, particularly we aim our attention at the class of judgement verbs (in a broad sense).
V tomto příspěvku představujeme kontrastivní studii jednoho případu provázání nestejných valenčních doplnění v rámci paralelních syntaktických struktur dvou jazyků. Na materiálu Pražského česko-anglického závislostího korpusu (PCEDT) sledujeme věty, v nichž je adresátové valenční doplnění z jednoho jazyka překladově provázáno s patientovým doplněním v jazyce druhém. Konkrétně se zabýváme zejména slovesy ze sémantické třídy "Judgement" (slovesa soudu) v širokém pojetí.
[ "Jana Šindlerová", "Zdeňka Urešová", "Eva Fučíková" ]
https://www.semanticscholar.org/paper/af10a7c98fc3de7b5f14c011db839b2ae774a34c/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Text annotations in the Prague Dependency Treebank
Textové anotace v Pražském závislostním korpusu
Description of text annotations and some selected problems of the annotations (annotation on dependency trees and on a plain text; the discrepancy between syntactic structure and discourse structure; an unexpressed thought as a discourse argument)
Popis textových anotací a jejich vybraných problémů (anotace na stromech a na lineárním textu; nesoulad mezi syntaktickou a diskurzní strukturou; nevyjádřená myšlenka jako diskurzní argument)
[ "Šárka Zikánová" ]
https://www.semanticscholar.org/paper/804848d67185d1125bc30c93a71c3e2b6ca90ce1/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Automatic Preannotation of TFA in the Czech Part of PCEDT
Automatická předanotace TFA v české části PCEDT
The presentation introduces the individual points of the automatic pre-annotation of topic-focus articulation in the Czech part of the corpus PCEDT.
Příspěvek představuje jednotlivé body automatické předanotace aktuálního členění v české části korpusu PCEDT.
[ "Magdaléna Rysová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
On Explicative Relationships in Czech
K explikativním vztahům v češtině
The paper deals with distinguishing of explicative and causal relationships in Czech.
Příspěvek se zabývá rozlišením explikativních a kauzálních vztahů v češtině.
[ "Magdaléna Rysová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Utilizing Source Context in Statistical Machine Translation
Využití zdrojového kontextu ve statistickém strojovém překladu
Current methods for statistical machine translation typically utilize only a limited context in the input sentence. Many language phenomena thus remain out of their reach, for example long-distance agreement in morphologically rich languages or lexical selection often require information from the whole source sentence. In this work, we present an overview of approaches for including wider context in SMT and describe our first experiments.
Současné metody pro statistický strojový překlad obvykle využívají pouze omezený kontext ve zdrojové větě. Mnoho jazykových jevů tak zůstavá mimo jejich dosah, např. gramatická shoda v morfologicky bohatých jazycích nebo lexikální výběr často vyžadují informace z celé zdrojové věty. V této práci představujeme přehled metod modelování širšího kontextu ve strojovém překladu a popisujeme naše první experimenty.
[ "Aleš Tamchyna" ]
https://www.semanticscholar.org/paper/168676a3dcb5895d1c27a147cd3a6aab340f71a2/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
MTMonkey 1.0
MTMonkey 1.0
MTMonkey is a web service which handles and distributes JSON-encoded HTTP requests for machine translation (MT) among multiple machines running an MT system, including text pre- and post processing. It consists of an application server and remote workers which handle text processing and communicate translation requests to MT systems. The communication between the application server and the workers is based on the XML-RPC protocol.
MTMonkey je webová služba, která obsluhuje a distribuuje žádosti na strojový překlad předávané přes HTTP ve formátu JSON. Rozděluje je na několik strojů, na kterých běží systém pro strojový překlad, a zařizuje preprocessing a postprocessing textu. Sestává z aplikačního serveru a jednotlivých workerů, které provádějí zpracování textu a komunikaci s překladovým systémem. Komunikace mezi aplikačním serverem a workery je založena na protokolu XML-RPC.
[ "Aleš Tamchyna", "Ondřej Dušek", "Rudolf Rosa" ]
https://www.semanticscholar.org/paper/8b6af73986daff6d7c861c1522bf1fd552f24c9e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
MTMonkey: A Scalable Infrastructure for a Machine Translation Web Service
MTMonkey: škálovatelná infrastruktura pro webovou službu poskytující strojový překlad
We present a web service which handles and distributes JSON-encoded HTTP requests for machine translation (MT) among multiple machines running an MT system, including text pre- and post processing. It is currently used to provide MT between several languages for cross-lingual information retrieval in the Khresmoi project. The software consists of an application server and remote workers which handle text processing and communicate translation requests to MT systems. The communication between the application server and the workers is based on the XML-RPC protocol. We present the overall design of the software and test results which document speed and scalability of our solution. Our software is licensed under the Apache 2.0 licence and is available for download from the Lindat-Clarin repository and Github.
Představujeme webovou službu, která zpracovává a distribuuje požadavky na strojový překlad (posílané přes HTTP ve formátu JSON). V současné době se používá pro poskytování strojového překladu mezi několika jazyky ve vícejazyčném vyhledávání informací v rámci projektu Khresmoi. Software sestává z aplikačního serveru a vzdálených workerů, kteří zajišťují zpracování textu a komunikují požadavky na překlad systémům strojového překladu. Komunikace mezi aplikačním serverem a workery je založena na protoklolu XML-RPC. Představujeme celkový návrh softwaru a výsledky testů, který dokumentují rychlost a škálovatelnost našeho řešení. Software podléhá licenci Apache 2.0 a je dostupný ke stažení v repozitáři Lindat-Clarin a na serveru GitHub.
[ "Aleš Tamchyna", "Ondřej Dušek", "Rudolf Rosa", "Pavel Pecina" ]
https://www.semanticscholar.org/paper/8b6af73986daff6d7c861c1522bf1fd552f24c9e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Moses Toolkit and our Eman, experiment manager
Nástroj Moses a Eman, náš správce experimentů
We give a brief introduction to phrase-based machine translation and the Moses toolkit. We present Eman, an experiment manager, and show how to use it to train several simple MT systems.
Stručně představujeme frázový strojový překlad a nástroj Moses. Popisujeme Emana, nástroj pro správu experimentů, a ukazujeme, jak ho lze využít k trénování několika jednoduchých systémů pro strojový překlad.
[ "Aleš Tamchyna", "Ondřej Bojar" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
No Free Lunch in Factored Phrase-Based Machine Translation
Ve faktorovém frázovém strojovém překladu není nic zdarma
Factored models have been successfully used in many language pairs to improve translation quality in various aspects. In this work, we analyze this paradigm in an attempt at automating the search for well-performing machine translation systems. We examine the space of possible factored systems, concluding that a fully automatic search for good configurations is not feasible. We demonstrate that even if results of automatic evaluation are available, guiding the search is difficult due to small differences between systems, which are further blurred by randomness in tuning. We describe a heuristic for estimating the complexity of factored models. Finally, we discuss the possibilities of a "semi-automatic" exploration of the space in several directions and evaluate the obtained systems.
Faktorové modely byly úspěšně nasazeny pro řadu jazykových párů, kde zlepšily různé aspekty kvality překladu. V této práci analyzujeme toto paradigma a pokoušíme se automatizovat hledání kvalitních překladových systémů. Zkoumáme prostor možných faktorových systémů a shledáváme, že plně automatické hledání není proveditelné. Ukazujeme, že i když jsou k dispozici výsledky automatické evaluace, zacílení prohledávání je obtížné kvůli malým rozdílům mezi jednotlivými systémy. Tyto rozdíly jsou dále rozmazány náhodností ladění vah. Popisujeme heuristiku pro odhad složitosti faktorových modelů. Závěrem popisujeme možnosti "polo-automatického" prohledávání prostoru v několika směrech a vyhodnocujeme získané systémy.
[ "Aleš Tamchyna", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/5d921c936962c8d69550923479be63b6bb13311a/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
A learner corpus of Czech: current state and future directions
Žákovský korpus češtiny: současný stav a budoucí vývoj
The paper describes CzeSL, a learner corpus of Czech as a Second Language, together with its design properties. We start with a brief introduction of the project within the context of AKCES, a programme addressing Acquisition Corpora of Czech; in connection with the programme we are also concerned with the groups of respondents, including differences due to their L1; further we comment on the choice of the sociocultural metadata recorded with each text and related both to the learner and the text production task. Next we describe the intended uses of CzeSL. The core of the paper deals with transcription and annotation. We explain issues involved in the transcription of handwritten texts and present the concept of a multi-level annotation scheme including a taxonomy of captured errors. We conclude by mentioning results from an evaluation of the error annotation and presenting plans for future research.
Článek popisuje CzeSL, žákovský korpus češtiny jako druhého jazyka. Nejdříve je popsán v kontextu Akces, souboru akvizičních korpusů češtiny. Dále je diskutováno zamýšlené využití CzeSLu. Jádro příspěvku se zabývá transkripcí a anotací.
[ "Barbora Štindlová", "Svatava Škodová", "Jirka Hana", "Alexandr Rosen" ]
https://www.semanticscholar.org/paper/6aaa1f578f88bc5428a0f6aa26cb2c825f54717e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Evaluating and automating the annotation of a learner corpus
Evaluace a automatizace anotace žákovského korpusu
The paper describes a corpus of texts produced by non-native speakers of Czech. We discuss its annotation scheme, consisting of three interlinked tiers, designed to handle a wide range of error types present in the input. Each tier corrects different types of errors; links between the tiers allow capturing errors in word order and complex discontinuous expressions. Errors are not only corrected, but also classified. The annotation scheme is tested on a data set including approx. 175,000 words with fair inter-annotator agreement results. We also explore the possibility of applying automated linguistic annotation tools (taggers, spell checkers and grammar checkers) to the learner text to support or even substitute manual annotation.
Popis a analýza anotace žákovského korpusu češtiny. Anotační schéma sestává ze tří propojených rovin, každá opravující jiný typ chyb. Anotační schéma bylo testováno na cca 175.000 slovech s rozumnou výslednou IAA. Na závěr diskutujeme možnost automatické anotace.
[ "Alexandr Rosen", "Jiří Hana", "Barbora Štindlová", "Anna Feldman" ]
https://www.semanticscholar.org/paper/8080fb3e7d3ab457c1c3ac8ffa5caa8f3c300e0e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Lexical analysis: Norms and Exploitations
Lexikální analýza: normy a využití
A radically new lexical, corpus-driven approach to linguistic theory and the practical analysis of language. The book argues that a natural langauge consists, basically, of a number of regular clause patterns ("norms"), in which paradigm sets of lexical items map into clause roles or "arguments", given normal valency structure. The picture is complicated because every pattern allows certain alternations, and every aspect of normal usage can be "exploited" creatively by language users. The book offers a typology of linguistic exploitations.
Radikálně nový lexikální, na korpusech založený přístup k lingvistické teorii a praktické analýze jazyka. Kniha ukazuje, že přirozený jazyk se v zásadě skládá z množství pravidelných vzorů klauzí ("norem"), ve kterých paradigmatické množiny lexikálních položek odpovídají klauzálním rolím neboli "argumentům", s ohledem na běžnou valenční strukturu.
[ "Patrick Hanks" ]
https://www.semanticscholar.org/paper/1f4c4549030506b702ba8583ff3d14bc8bd45cc8/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2013
Computers are unlikely to ever replace the human editor and translator
Počítače asi nikdy nahradí lidského korektora ani překladatele
An interview for Radio Czech describing various approaches to machine translation and the difficulties with handling natural languages in general.
Rozhovor pro stanici Rádio Česko popisující různé přístupy ke strojovému překladu a potíže s přirozeným jazykem obecně.
[ "Tereza Burianová", "Marián Vojtek", "Ondřej Bojar" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Introducing the Prague Discourse Treebank 1.0
Pražský diskurzní korpus 1.0 - představení
We present the Prague Discourse Treebank 1.0, a collection of Czech texts annotated for various discourse-related phenomena "beyond the sentence boundary". The treebank contains manual annotations of (1), discourse connectives, their arguments and senses, (2), textual coreference, and (3), bridging anaphora, all carried out on 50k sentences of the treebank. Contrary to most similar projects, the annotation was performed directly on top of syntactic trees (from the previous project of the Prague Dependency Treebank 2.5), benefiting thus from the linguistic information already existing on the same data. In this article, we present our theoretical background, describe the annotations in detail, and offer evaluation numbers and corpus statistics.
Představujeme Pražský diskurzní korpus 1.0, soubor českých textů anotovaných na rovině textových jevů, tj. "nad úrovní popisu věty". Korpus obsahuje ruční anotace (1)textových konektorů, jejich argumentů a významů, (2) textové koreference a (3) tzv. bridging anafory, to vše na 50k větách korpusu.
[ "Lucie Poláková", "Jiří Mírovský", "Anna Nedoluzhko", "Pavlína Jínová", "Šárka Zikánová", "Eva Hajičová" ]
https://www.semanticscholar.org/paper/c61ecbc52a549ac92c73d9ce83a9a1e2255216a2/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Machine Translation between Czech and Russian
Strojový překlad mezi češtinou a ruštinou
This talk described several Machine Translation systems between Czech and Russian that were developed or implemented by UFAL researchers. We discuss which of the approaches - Statistical or Rule-Based is more suitable for MT between related morphologically rich languages. The focus is made on analyzing valency discrepancies between the two languages and their impact on MT.
V přednášce byl prezentován popis systémů strojového překladu mezi češtinou a ruštinou, a to statistických a pravidlových.
[ "Natalia Klyueva" ]
https://www.semanticscholar.org/paper/378121ec481c48818954f6025acd7f6bc6a26004/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Usage of some non-finite constructions in Czech and Russian
Užití některých nefinitních konstrukcí v češtine a ruštině
This paper describes one of the discrepancies between the languages that occur in non-finite clauses, namely in participial clauses and transgressives. We will concentrate mainly on transgressives because they demonstrate greater discrepancies. The usage of Russian transgressives(gerunds) is very similar to that of English, they are typical mostly of a written language rather than spoken, whereas in Czech they are nowadays archaic in any genres and generally not used at all. This discrepancy can pose some challenge for language learners as well as for translators or machine translation systems. We will present some examples of Russian participial/gerund clauses and the respective translations into Czech from a parallel corpus of news and belletristic texts.
V článku je popsana frekvence užití nefinitních konstrukcí s přechodníky a příčestí v češtině a ruštině.
[ "Natalia Klyueva" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Machine Translation Zoo: Tree-to-tree transfer and Discriminative learning
Zoo strojového překladu: tree-to-tree transfer a diskriminativní učení
Machine Translation Taxonomy Online Learning Guided Learning Easy-First Decoding in MT Guided Learning in MT
Taxonomie strojového překladu Online učení Guided učení Easy-First dekódování ve strojovém překladu Guided učení ve strojovém překladu
[ "Martin Popel" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Treex vs. NIF
Treex versus NIF
Treex is a framework for NLP tools. NIF (NLP Interchange format) is an RDF/OWL-based format that aims to achieve interoperability between NLP tools, language resources and annotations. The presentation describes Treex and other tools/treebanks developed at ÚFAL. It focuses on its similarities and differences compared to NIF (and the NLP2RDF initiative).
Treex je platforma pro nástroje zpracování přirozeného jazyka. NIF je formát založený na RDF, který by měl sloužit k propojování takovýchto nástrojů. Přednáška popisuje Treex a další nástroje a treebanky vyvíjené na ÚFAL. Zaměřuje se též na podobnosti a rozdíly s formátem NIF a iniciativou nlp2rdf.
[ "Martin Popel" ]
https://www.semanticscholar.org/paper/880f55384b0f00b24784603dfdfca20eb7e23c9c/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Coordination Structures in Dependency Treebanks
Souřadné struktury v závislostních korpusech
Paratactic syntactic structures are notoriously difficult to represent in dependency formalisms, which has painful consequences such as high frequency of parsing errors related to coordination. In other words, coordination is a pending problem in dependency analysis of natural languages. This paper tries to shed some light on this area by bringing a systematizing view of various formal means developed for encoding coordination structures. We introduce a novel taxonomy of such approaches and apply it on treebanks across a typologically diverse range of 26 languages. Empirical observations on convertibility between selected styles of representations are shown too.
O parataktických syntaktických strukturách je známo, že jejich zachycení pomocí závislostních mechanismů je obtížné, což má bolestivé důsledky jako značné množství chyb v automatické syntaktické analýze, které souvisí s koordinacemi. Jinými slovy, koordinace je nevyřešeným problémem závislostní analýzy přirozených jazyků. Tento článek se snaží vnést trochu světla do této oblasti a přináší systematický přehled různých formálních prostředků vyvinutých k zachycení souřadných struktur. Přinášíme novou taxonomii takových přístupů a aplikujeme ji na treebanky z typologicky rozličné sady 26 jazyků. Dále prezentujeme empirická pozorování, pokud jde o vzájemnou převoditelnost mezi vybranými styly zachycení.
[ "Martin Popel", "David Mareček", "Jan Štěpánek", "Daniel Zeman", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/9d485c041736454a40a9600f439b965aaafdedd3/
null
null
null
null
null
null
null
null
null
null
null
null
en
2013
Treex Tutorial
Seminář o platformě Treex
Hands-on tutorial on using Treex - a modular framework for Natural Language Processing. The tutorial was preceded by a presentation about TectoMT deep-syntactic MT system developed in Treex.
Praktický seminář představení platformu Treex, která slouží k integraci nástrojů pro zpracování přirozeného jazyka. Semináři předcházela přednáška o překladovém systému TectoMT vyvinutém v platformě Treex.
[ "Martin Popel", "Rudolf Rosa" ]
https://www.semanticscholar.org/paper/905d6ac39e4021743f46ec0197c9a19180523e43/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Building an Error-tagged Learner Corpus of Czech
Budování žákovského korpusu s chybovou anotací
We describe a corpus of texts produced by non-native speakers of Czech. We discuss its annotation scheme, consisting of three interlinked levels to cope with a wide range of error types present in the input. Each level corrects different types of errors; links between the levels allow capturing errors in word order and complex discontinuous expressions. Errors are not only corrected, but also classified. The annotation scheme is tested on a doubly-annotated sample of approx. 10,000 words with fair inter-annotator agreement results. We also explore options of application of automated linguistic annotation tools (taggers, spell checkers and grammar checkers) on the learner text to support or even substitute manual annotation.
Přednáška představuje žákovský korpus češtiny a jeho anotační schéma sestavající ze propojených tří rovin. Každá rovina zachycuje a opravuje jiný typ chyb.
[ "Jirka Hana" ]
https://www.semanticscholar.org/paper/b325505216ffa4e9c458ce7138259ee115b54997/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Building a Corpus of Old Czech
Budování korpusu staré češtiny
In this paper we describe our efforts to build a corpus of Old Czech. We report on tools, resources and methodologies used during the corpus development as well as discuss the corpus sources and structure, the tagset used, the approach to lemmatization, morphological analysis and tagging. Due to practical restrictions we adapt resources and tools developed for Modern Czech. However, some of the described challenges, such as the non-standardized spelling in early Czech and the form and lemma variability due to language change during the covered time-span, are unique and never arise when building synchronic corpora of Modern Czech.
Tento článek popisuje kompilaci a anotaci korpusu staré češtiny.
[ "Jirka Hana", "Boris Lehečka", "Anna Feldman", "Alena Černá", "Karel Oliva" ]
https://www.semanticscholar.org/paper/dfc174ef867342827811cf5c044db13ec4ae683a/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Getting more data – Schoolkids as annotators
Získání více dat – školáci jako anotátoři
We present a new way to get more morphologically and syntactically annotated data. We have developed an annotation editor tailored to school children to involve them in text annotation. Using this editor, they practice morphology and dependency-based syntax in the same way as they normally do at (Czech) schools, without any special training. Their annotation is then automatically transformed into the target annotation schema. The editor is designed to be language independent, however the subsequent transformation is driven by the annotation framework we are heading for. In our case, the object language is Czech and the target annotation scheme corresponds to the Prague Dependency Treebank annotation framework.
Článek popisuje nový způsob jak získat více morfologicky a syntakticky oanotovaných dat.
[ "Jirka Hana", "Barbora Hladká" ]
https://www.semanticscholar.org/paper/0390f5c788ecc4e29cbd99e3a9a85cf71f0d5873/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Prague Markup Language Framework
Prague Markup Language
In this paper we describe the Prague Markup Language (PML), a generic and open XML-based format intended to define format of linguistic resources, mainly annotated corpora. We also provide an overview of existing tools supporting PML, including annotation editors, a corpus query system, software libraries, etc.
V tomto článku popisujeme Prague Markup Language (PML), generický a otevřený formát založený na XML vytvořený za účelem definice formátů lingvistických zdrojů, zvláště anotovaných korpusů.
[ "Jirka Hana", "Jan Štěpánek" ]
https://www.semanticscholar.org/paper/5c737ef589715567279c5f61bd1d9b42d1f76062/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Building a learner corpus
Budování žákovského korpusu
The paper describes a corpus of texts produced by non-native speakers of Czech. We discuss its annotation scheme, consisting of three interlinked levels to cope with a wide range of error types present in the input. Each level corrects different types of errors; links between the levels allow capturing errors in word order and complex discontinuous expressions. Errors are not only corrected, but also classified. The annotation scheme is tested on a doubly-annotated sample of approx. 10,000 words with fair inter-annotator agreement results. We also explore options of application of automated linguistic annotation tools (taggers, spell checkers and grammar checkers) on the learner text to support or even substitute manual annotation.
Článek popisuje žákovský korpus češtiny a jeho anotační schéma sestavající ze propojených tří rovin. Každá rovina zachycuje a opravuje jiný typ chyb.
[ "Jirka Hana", "Alexandr Rosen", "Barbora Štindlová", "Petr Jäger" ]
https://www.semanticscholar.org/paper/b285627b937d01b74ab82f0a056c352af4db0a35/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Resource-light Approaches to Computational Morphology Part 1: Monolingual Approaches
Počítačová morfologie nenaročná na zdroje část 1: Monolingvální přístupy
This article surveys resource-light monolingual approaches to morphological analysis and tagging. While supervised analyzers and taggers are very accurate, they are extremely expensive to create. Therefore, most of the world languages and dialects have no realistic prospect for morphological tools created in this way. The weakly-supervised approaches aim to minimize time, expertise and/or financial cost needed for their development. We discuss the algorithms and their performance considering issues such as accuracy, portability, development time and granularity of the output.
Tento článek shrnuje jednojazyčné přístupy k morfologické analýze a tagování nenáročné na zdroje.
[ "Jirka Hana", "Anna Feldman" ]
https://www.semanticscholar.org/paper/03a3241c00db8fc4328774abeb4ffcb8e50be3f8/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2012
Possibilities of Free Verbal Modifications to be an Obligatory Sentence Element
Možnosti jednotlivých volných slovesných doplnění být obligatorním členem věty
The paper describes which types of free verbal modifications may be obligatory and how often.
Článek popisuje, které typy volných slovesných doplnění vystupují jako obligatorní a jak často.
[ "Kateřina Rysová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
On sentence information structure in Czech: the usual sentence position of particular sentence components
K informační struktuře věty v češtině: obvyklé slovosledné umístění větných participantů
Poster se týká aktuálního členění větného v češtině - prezentuje tendenci jednotlivých větných participantů objevovat se v české větě jako téma nebo réma.
The poster deals with the sentence information structure (functional sentence perspective) in Czech, focusing on the tendency of the particular sentence constituents to appear in Czech sentence as topic or focus.
[ "Kateřina Rysová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Concurrent effects of lexical status and letter-rotation during early stages of visual word recognition: evidence from ERPs
Souběžné projevy lexikálního statutu a rotace písmen v průběhu časných stádií vizuálního rozpoznání slova: ERP důkazy
Recent studies report that the occipito-temporal N170 component of the ERP is enhanced by letter strings, relative to non-linguistic strings of similar visual complexity, with a left-lateralized distribution. This finding is consistent with underlying mechanisms that serve visual word recognition. Conclusions about the level of analysis reflected within the N170 effects, and therefore the timecourse of word recognition, have been mixed. Here, we investigated the timing and nature of brain responses to putatively low- and high-level processing difficulty. Low-level processing difficulty was modulated by manipulating letter-rotation parametrically at 0 degrees, 22.5 degrees, 45 degrees, 67.5 degrees, and 90 degrees. Higher-level processing difficulty was modulated by manipulating lexical status (words vs. word-like pseudowords). Increasing letter-rotation enhanced the N170 led to monotonic increases in P1 and N170 amplitude up to 67.5 degrees but then decreased amplitude at 90 degrees. Pseudowords enhanced the N170 over left occipital-temporal sites, relative to words. These combined findings are compatible with a cascaded, interactive architecture in which lower-level analysis (e.g., word-form feature extraction) leads higher-level analysis (e.g., lexical access) in time, but that by approximately 170 ms, the brain's response to a visual word includes parallel, interactive processing at both low-level feature extraction and higher-order lexical access levels of analysis.
Podle současných studií je okcipito-temporální ERP komponenta N170 levostranně modulována a posílena řetězci znaků, narozdíl od nelingvistických vstupů podobné vizuální komplexity. Tento výsledek je konzistentní s neurálními mechanismy, které zajišťují vizuální rozpoznávání slov. Avšak závěry o úrovních analýzy, které ovlivňují projevy N170, jsou nejednoznačné. V této studii jsme zkoumali časový průběh a kvalitu reakcí mozku na podněty předpokládané nižší a vyšší úrovně zpracování. Podněty nižší úrovně byly modulovány manipulací rotace písmen parametricky v 0 stupních, 22,5 stupních, 45 stupních, 67,5 stupních a 90 stupních. Vyšší úroveň zpracování byla modulována manipulací s lexikálnim statutem (slova vs. pseudoslova). Při zvyšování úhlu rotace písmen se amplituda N170 monotonicky zvyšovala až do úhlu 67,5, ale pak při úhlu 90 stupňů se opět snížila. Pseudoslova v porovnání se slovy zvětšila amplitudu N170 v levé okcipito-temporální oblasti. Tyto kombinované výsledky jsou konzistentní s kaskádovou, interaktivní architekturou, ve které nižší úroveň analýzy (např. rozpoznání tvaru slov) časově předchází vyšší úrovni analýzy (např. lexikální status), ale již v čase 170 ms reakce mozku na vizuální slovní podnět obsahuje paralelní, interaktivní zpracování na obou úrovních.
[ "Albert Kim", "Jana Straková" ]
https://www.semanticscholar.org/paper/1dae6370c2cd47556bc0364d72986e3ff812c7b5/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2012
Most Common Connectives of Causal Relation in Prague Dependency Treebank
Nejčastější konektivní prostředky kauzálního vztahu v Pražském závislostním korpusu
This contribution deals with properties of most common connective means of the causal relation in the Prague Dependency Treebank (namely with expressions protože (because), neboť (because), proto (therefore), takže (so), tak (so), tedy (thus) and totiž (because, namely, sometimes untranslatable)). Some properties of these connectives are described independently of context (e.g. position in the sentence, compatibility with other connectives), the majority of properties is characterized also with regard to frequency (how often these expressions connect parts of one sentence and how often parts of text across sentence boundaries, how often they are used to connect verbal phrases and how often something else, etc.). Its general aim is to show which properties are shared by all these connectives (or by some of them) and which are, on the contrary, specific for each of them. This paper also wants to contribute to the discussion of the part-of-speech characteristics of these connectives – a relation of the observed properties of connectives to the characteristics reported in the relevant Czech literature is described and some criteria for their part-of-speech characteristics are stated. As the material base, data released in the Prague Dependency Treebank for linguistic research (43,955 sentences (740,532 words)) were chosen, because the current state of tagging in this corpus enables us to measure the properties of all connectives regardless of whether they connect parts of one sentence or parts of text across sentence boundaries.
Příspěvek se zabývá vlastnostmi nejčastějších konektivních prostředků kauzálního vztahu v Pražském závislostním korpusu (konkrétně jde o výrazy protože, neboť, proto, takže, tak, tedy a totiž). Některé vlastnosti těchto konektivních prostředků popisuje nezávisle na kontextu (např. pozici ve větě, kombinovatelnost s jinými spojovacími výrazy), u většiny vlastností si všímá i charakteristik frekvenčních (jak často spojují dané výrazy části textu v rámci jednoho větného celku, jak často přes jeho hranice, jak často se užívají ke spojování verbálních frází a jak často jinde apod.). Jeho cílem je ukázat, jaké vlastnosti tyto prostředky sdílejí a které vlastnosti jsou naopak pro jednotlivé prostředky specifické. Zároveň chce přispět k diskusi o slovnědruhové charakteristice zkoumaných výrazů – pozorované vlastnosti jsou vztaženy k charakteristikám uváděným ve vybrané odborné literatuře a jsou uvedena kritéria, podle kterých je možné zkoumané výrazy slovnědruhově popisovat. Jako materiálová základna výzkumu byla zvolena data uvolněná v Pražském závislostním korpusu pro lingvistický výzkum (43 955 vět (740 532 slov)), protože současná podoba značkování v tomto korpusu umožňuje měřit vlastnosti zkoumaných výrazů v rámci jednoho větného celku i přes jeho hranice.
[ "Pavlína Jínová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Analyzing the Most Common Errors in the Discourse Annotation of the Prague Dependency Treebank
Analýza nejčastějších chyb v anotaci textových vztahů v PDT
We present an analysis of the inter-annotator discrepancies of the Czech discourse annotation in the Prague Dependency Treebank 2.0. Having finished the annotation of the inter-sentential semantic discourse relations with explicit connectives in the treebank, we report now on the results of the evaluation of the parallel (double) annotations, which is an important step in the process of checking the quality of the data. After we shortly describe the annotation and the method of the inter-annotator agreement measurement, we present the results of the measurement and, most importantly, we classify and analyze the most common types of annotators’ disagreement.
Článek prezentuje analýzu nejčastějších anotátorských chyb v projektu anotace mezivýpovědních textových vztahů v PDT
[ "Pavlína Jínová", "Jiří Mírovský", "Lucie Poláková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Semi-Automatic Annotation of Intra-Sentential Discourse Relations in PDT
Poloautomatická anotace vnitrovnětných textových vztahů v PDT
In the present paper, we describe in detail and evaluate the process of semi-automatic annotation of intra-sentential discourse relations in the Prague Dependency Treebank, which is a part of the project of otherwise mostly manual annotation of all (intra- and inter-sentential) discourse relations with explicit connectives in the treebank. The rich annotation of the treebank allowed us to automatically detect the intra-sentential discourse relations, their connectives and arguments in most of the cases.
Článek popisuje a vyhodnocuje proces poloautomatické anotace vnitrovětných textových vztahů v Pražském závislostním korpusu, což je část projektu jinak především manuální anotace všech (vnitro- i mezi-větných) diskurzních vztahů s explicitními konektory v tomto korpusu. Bohatá anotace korpusu nám umožnila ve většině případů automaticky najít vnitrovětné diskurzní vztahy, jejich konektory a argumenty.
[ "Pavlína Jínová", "Jiří Mírovský", "Lucie Poláková" ]
https://www.semanticscholar.org/paper/3b443a525e73170ea944c1d98ff8f01a2ada9683/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
A Richly Annotated, Multilingual Parallel Corpus for Hybrid Machine Translation
Bohatě anotovaný mnohojazyčný paralelní korpus pro hybridní strojový překlad
In recent years, machine translation (MT) research has focused on investigating how hybrid machine translation as well as system combination approaches can be designed so that the resulting hybrid translations show an improvement over the individual “component” translations. As a first step towards achieving this objective we have developed a parallel corpus with source text and the corresponding translation output from a number of machine translation engines, annotated with metadata information, capturing aspects of the translation process performed by the different MT systems. This corpus aims to serve as a basic resource for further research on whether hybrid machine translation algorithms and system combination techniques can benefit from additional (linguistically motivated, decoding, and runtime) information provided by the different systems involved. In this paper, we describe the annotated corpus we have created. We provide an overview on the component MT systems and the XLIFF-based annotation format we have developed. We also report on first experiments with the ML4HMT corpus data.
V posledních letech se výzkum v oblasti strojového překladu zaměřil na hybridní strojový překlad a kombinaci překladových systémů za účelem vylepšení překladové kvality. Jako první krok ke splnění tohoto cíle jsme vytvořili paralelní korpus vět přeložených několika různými systémy opatřenými různými metadaty a anotacemi získanými při překladu.
[ "Eleftherios Avramidis", "Marta R. Costa-Jussà", "Christian Federmann", "Pavel Pecina", "Josef van Genabith" ]
https://www.semanticscholar.org/paper/91c89af3c6c94c1769044ff18a862e3fd4344e10/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Data-Driven Methods for Spoken Language Understanding
Metody porozumění mluvené řeči založené na učení z dat
This chapter described data-driven methods suitable for development of spoken language understanding components in dialogue systems.
Tato kapitola popisuje metody učení z dat vhodné pro vývoj systémů porozumění mluvené řeči pro dialogové systémy.
[ "James Henderson", "Filip Jurčíček" ]
https://www.semanticscholar.org/paper/99c8008e513175fa483397e69728bd301d42d79f/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Automatic polarity classification systems for Czech
Systémy pro automatickou klasifikaci polarity v češtině
In this talk I present work towards building a reliable polarity classifiers based on Czech data. I will describe a method for annotating Czech evaluative structures and introduce a standard unigram-based Naive Bayes classifier together with the lexicon-based classifier, both trained on three different types of data. Also, I will analyze existing results for both manual and automatic annotation. Joint work with Jan Hajič, jr. and Jana Šindlerová.
V příspěvku představím kroky vedoucí k spolehlivým klasifikátorům polarity založeným na českých datech. Popíšu metodologické základy anotování českých struktur a unigramové klasifikátory trénované na třech druzích dat. Rovněž zanalyzuji existující výsledky pro manuální i automatickou anotaci.
[ "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/0d29b423d5e806f4668730ee745167a4a0d0c185/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2012
International conference on dependency linguistics Depling 2011 in Barcelona
Mezinárodní konference závislostní lingvistiky Depling 2011 v Barceloně
Report on the first international conference in dependency grammar and reflexion of its current state.
Reportáž z první mezinárodní konference v závislostní lingvistice spojená se zamyšlením nad jejím současným stavem.
[ "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/2514baaf8c253158693cbe7819181b5ee0d2c957/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Sentence-Level Sentiment Analysis in Czech
Postojová analýza na úrovni věty v češtině
This paper presents initial research in the area of sentiment analysis in Czech. We will describe a method for annotating Czech evaluative structures and analyze existing results for both manual and automatic annotation of the plain text data which represents the basis for further subjectivity clues learning.
V článku je prezentována počáteční fáze výzkumu postojové analýzy v češtině: metoda anotování evaluativních struktur a první výsledky manuální i automatické anotace.
[ "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/d7e4713826eab41bf07daf3d94bb21837f755744/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Using Czech-English Parallel Corpora in Automatic Identification of It
Využití česko-anglických paralelních dat při automatické identifikaci "it"
In this paper we have two goals. First, we want to present a part of the annotation scheme of the recently released Prague Czech-English Dependency Treebank 2.0 related to the annotation of personal pronoun it on the tectogrammatical layer of sentence representation. Second, we introduce experiments with the automatic identification of English personal pronoun it and its Czech counterpart. We design sets of tree-oriented rules and on the English side we combine them with the state-of-the-art statistical system that altogether results in an improvement of the identification. Furthermore, we design and successfully apply rules, which exploit information from the other language.
Tento článek má dvojí cíl: jednak chceme prezentovat tu část anotačního schématu nedávno vydaného korpusu PCEDT2.0, která je spjata s anotací anglického zájmene “it” na tektogramatické rovině, jednak představujeme experimenty týkající se automatické identifikace anglického “it” a jeho českého protějšku. Navrhli jsme soubor pravidel pro stromové struktury, která v rámci anglické části korpusu kombinujeme se současnými statistickými systémy, což v důsledku vede ke zlepšení automatické detekce. Mimoto jsme také navrhli a úspěšně aplikovali pravidla, která využívají informace z paralelních českých struktur.
[ "Kateřina Veselovská", "Giang Linh Nguy", "Michal Novák" ]
https://www.semanticscholar.org/paper/01bf7708b39f064a926656b71fc633615c64d0c1/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Creating annotated resources for polarity classification in Czech
Vytváření anotovaných zdrojů pro klasifikaci polarity v češtině
This paper presents the first steps towards reliable polarity classification based on Czech data. We describe a method for annotating Czech evaluative structures and build a standard unigram-based Naive Bayes classifier on three different types of annotated texts. Furthermore, we analyze existing results for both manual and automatic annotation, some of which are promising and close to the state-of-the-art performance, see Cui(2006).
Článek popisuje první kroky vedoucí ke spolehlivé automatické klasifikaci větné polarity v češtině. Objasňujeme metodu anotace českých evaluativních struktur a trénování základního bayesovského klasifikátoru na třech typech dat, včetně analýzy dosažených výsledků, z nichž některé jsou srovnatelné s výsledky, jichž bylo dosud docíleno v oblasti postojové analýzy jako takové (viz např. Cui, 2006).
[ "Kateřina Veselovská", "Jan Hajič, jr.", "Jana Šindlerová" ]
https://www.semanticscholar.org/paper/1ee0ce86645f02ca5f9b8f91cc3d009f7b79f861/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Unsupervised Dependency Parsing
Neřízená závislostní analýza
Unsupervised dependency parsing is an alternative approach to identifying relations between words in a sentence. It does not require any annotated treebank, it is independent of language theory and universal across languages. However, so far quite low parsing quality is its main disadvantage. This thesis discusses some previous works and introduces a novel approach to unsupervised parsing. Our dependency model consists of four submodels: (i) edge model, which controls the distribution of governor-dependent pairs, (ii) fertility model, which controls the number of node's dependents, (iii) distance model, which controls the length of the dependency edges, and (iv) reducibility model. The reducibility model is based on a hypothesis that words that can be removed from a sentence without violating its grammaticality are leaves in the dependency tree. Induction of the dependency structures is done using Gibbs sampling method. We introduce a sampling algorithm that keeps the dependency trees projective, which is a very valuable constraint. In our experiments across 30 languages, we discuss the results of various settings of our models. Our method outperforms the previously reported results on a majority of the test languages.
Nerízená závislostní analýza je alternativní zpusob urcování vztahu mezi slovy ve vete. Nepotrebuje žádný anotovaný závislostní korpus, je nezávislý na jazykové teorii a univerzální pro velké množství jazyku. Jeho nevýhodou je ale zatím relativne nízká úspešnost. V této práci diskutujeme nekteré predchozí práce a predstavujeme novou metodu nerízenéhé analýzy. Náš závislostní model se skládá ze ctyr podmodelu: (i) hranový model, který rídí rozdelení dvojic rídících a závislých clenu, (ii) model plodnosti, který rídí pocet clenu závislých na uzlu, (iii) model vzdálenosti, který rídí délku závislostních hran a (iv) model vypustitelnosti. Tento model je založen na predpokladu, že slovau která se mohou z vety vypustit, aniž by se porušila její gramaticnost jsou v závislostním slove listy. Odvození závislostních struktur provádíme pomocí Gibbsova vzorkovace. Predstavujeme vzorkovací algoritmus, který zachovovává projektivitu závislostních stromu, cože je velmi užitecnou vlastností. V našich experimentech na 30 jazycích srovnáváme výsledky pro ruzné parametry modelu. Naše metoda prekonávvá dríve publikované výsledky pro vetšinu zkoumaných jazyku.
[ "David Mareček" ]
https://www.semanticscholar.org/paper/16160fbefc830117a8395a4b665c39feae834e48/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Exploiting Reducibility in Unsupervised Dependency Parsing
Využití vypustitelnosti v neřízeném závislostním parsingu
Predstavujeme novy pristup k nerizenemu zavislostnimu parsingu zalozenemu na mire vypustitelnosti jednotlivych sekvenci slov.
The possibility of deleting a word from a sentence without violating its syntactic correctness belongs to traditionally known manifes tations of syntactic dependency. We introduce a novel unsupervised parsing approach that is based on a new n-gram reducibility measure. We perform experiments across 18 languages available in CoNLL data and we show that our approach achieves better accuracy for the majority of the languages then previously reported results.
[ "David Mareček", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/d25d0426c5d39d3507e443e3b6f0bcb0427b45f6/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Unsupervised Dependency Parsing using Reducibility and Fertility features
Neřízený závislostní parsing využívající vlastnost vypustitelnosti a fertility
Popis systemu nerizeneho zavislostniho parsingu zalozeneho na Gibbsove samplingu. Novy pristup predstavuje vlastnosti vypusttelnsti a fertility.
This paper describes a system for unsuper- vised dependency parsing based on Gibbs sampling algorithm. The novel approach in- troduces a fertility model and reducibility model, which assumes that dependent words can be removed from a sentence without vio- lating its syntactic correctness.
[ "David Mareček", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/04f44c582c54653c0516e09bc0ea24cf4e7fcc39/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Formemes in English-Czech Deep Syntactic MT
Formémy v anglicko-českém hloubkově syntaktickém strojovém překladu
One of the most notable recent improvements of the TectoMT English-to-Czech translation is a systematic and theoretically supported revision of formemes—the annotation of morpho-syntactic features of content words in deep dependency syntactic structures based on the Prague tectogrammatics theory. Our modifications aim at reducing data sparsity, increasing consistency across languages and widening the usage area of this markup. Formemes can be used not only in MT, but in various other NLP tasks.
Jedním z nejvýznamnějších nedávných vylepšení anglicko-českého překladače TectoMT je systematická a teoreticky podložená revize formémů. Formémy jsou anotací mofo-synaktických rysů plnovýznamových slov v hloubkově syntaktické závislostní syntaxi založené na tektogramatice. Naše úpravy měly za cíl snížit řídkost dat, zvýšit konzistenci anotace napříč jazyky a otevřít nové oblasti využití formémů. Formémy jsou využívány ve strojovém překladu i v dalších úlohách zpracování přirozeného jazyka.
[ "Ondřej Dušek", "Zdeněk Žabokrtský", "Martin Popel", "Martin Majliš", "Michal Novák", "David Mareček" ]
https://www.semanticscholar.org/paper/8cc6ee1a2251f57e1a7b120da97d5e97518fb6b0/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2012
Lexical-semantic Conversions in a Valency Lexicon
Lexikálně-sémantické konverze ve valenčním slovníku
In this thesis, we provide an adequate lexicographic representation of lexicalsemantic conversion. Under the term lexical-semantic conversion, the relation between semantically similar syntactic structures which are based on separate lexical units of the same verb lexeme is understood. These relations are associated with various changes in valency structure of verbs – they may involve a number of valency complementations, their type, obligatoriness as well as morphemic forms. These changes arise from differences in the mapping of situational participants onto valency complementations. On the basis of semantic and syntactic analysis of two types of Czech lexical-semantic conversions, the locative conversion and the conversion Bearer of action-Location, we propose to represent lexical units creating syntactic variants in the relation of lexical semantic conversion by separate valency frames stored in the data component of the lexicon. The special attribute -conv whose value is a type of lexical-semantic conversion is assigned to relevant valency frames. Then the rule component of the lexicon consists of general rules determining changes in the correspondence between situational participants and valency complementations. This proposal is primarily designed for the valency lexicon of Czech verbs, VALLEX. However, we suppose that this representation can be applied in other lexical resources as well.
V disertační práci podáváme návrh reprezentace lexikálně-sémantických konverzí ve valenčním lexikonu. Lexikálně-sémantické konverze chápeme jako vztahy mezi sémanticky blízkými syntaktickými konstrukcemi tvořenými odlišnými lexikálními jednotkami lexematicky totožného slovesa. Tyto vztahy jsou spojeny se změnami ve valenční struktuře slovesa, které vyplývají ze změn ve vzájemném přiřazení situačních participantů a valenčních doplnění. Tyto změny mohou zasahovat počet, typ valenčních doplnění, jejich obligatornost i morfematickou formu. Na základě sémantické a syntaktické analýzy dvou typů lexikálně-sémantické konverze, lokativní konverze a konverze Nositel děje-Místo, navrhujeme, aby lexikálním jednotkám vytvářejícím syntaktické varianty ve vztahu lexikálně-sémantické konverze v datové části lexikonu odpovídaly odlišné valenční rámce propojené atributem -conv, jehož hodnotou bude typ lexikálně-sémantické konverze. Součástí pravidlové komponenty valenčního lexikonu pak budou pravidla určující změny ve vzájemné korespondenci situačních participantů a valenčních doplnění. Ačkoli reprezentaci lexikálně-sémantické konverze primárně navrhujeme pro účely popisu valence českých sloves ve valenčním slovníku VALLEX, předpokládáme, že poznatky uváděné v této práci mohou být využity při popisu tohoto typu vztahu i v dalších lexikálních zdrojích.
[ "Václava Kettnerová" ]
https://www.semanticscholar.org/paper/52fe17da350999456e9bb9f381d3f6d23bf2f98e/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2012
The Syntactic Constructions Bess are swarming in the garden – The garden is swarming with bees
Syntaktické konstrukce typu Včely se hemží na zahradě – Zahrada se hemží včelami
In this paper, we focus on the relation between Czech syntactic constructions Bees are swarming in the garden –- The garden is swarming with bees. We refer to this relation as a lexical-semantic conversion. This lexical-semantic conversion is associated with changes in the correspondence between two situational participants –- Bearer of action and Location –- and valency complementations. These changes result in a permutation of these participants which affects the prominent surface syntactic position of subject. We demonstrate that whereas the first syntactic variant refers to a simple action, the latter expresses rather a state arising from this simple action. In the latter case, the participant Location is usually holistically interpreted. We assume that the holistic ef- fect of this participant follows from the fact that a state is generally attributed to each participant expressed in the subject position as a whole. Furthermore, syntactic variants with formal subject “to” and subjectless constructions (usually conceived as formal paraphrases of the pairs of the observed constructions) are discussed. Finally, a lexicographic representation of these syntactic structures is presented
V příspěvku jsme se zabývali syntaktickými konstrukcemi typu Včely se hemží na zahradě –- Zahrada se hemží včelami, které chápeme jako specifický typ lexikálně-sémantické konverze, u níž dochází ke změnám v korespondenci dvou situačních participantů (s rolí Nositel děje a Místo) a valenčních doplnění. Tyto změny mají za následek změny v povrchově syntaktické realizaci daných participantů, které zasahují prominentní pozici subjektu, v níž je jednou vyjádřen Nositel děje (Včely se hemží na zahradě) a podruhé Místo (Zahrada se hemží včelami). Zatímco prvně zmíněná syntaktická varianta vyjadřuje neakční elementární děj, druhá varianta označuje procesuálně pojatý stav. Předpokládáme potom, že holistická interpretace participantu Místo ve druhé syntaktické variantě vyplývá přímo ze stavového významu dané konstrukce. Dále příspěvek pojednává o bezpomětových konstrukcích a konstrukcích s formálním subjektem to, které vytvářejí formální paralelu k probíraným konstrukcím. Na závěr předkládáme možnou lexikografickou reprezentaci lexikálně-sémantické konverze daného typu, v níž informaci přerozdělujeme mezi datovou a pravidlovou část lexikonu.
[ "Václava Kettnerová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
The Rule-Based Approach to Czech Grammaticalized Alternations
Pravidlový přístup k českým gramatikalizovaným alternacím
Under the term grammaticalized alternations, we understand changes in valency frames of verbs corresponding to different surface syntactic structures of the same lexical unit of a verb. Czech grammaticalized alternations are expressed either (i) by morphological means (diatheses), or (ii) by syntactic means (reciprocity). These changes are limited to changes in morphemic form(s) of valency complementations; moreover, they are regular enough to be captured by formal syntactic rules. In this paper a representation of Czech grammaticalized alternations and their possible combination is proposed for the valency lexicon of Czech verbs, VALLEX.
Termínem gramatické alternace označujeme změny ve valenční stuktuře sloves, které odpovídají rozdílným povrchově syntaktickým strukturám tvořeným jedinou lexikální jednotkou slovesa. České gramatikalizované alternace jsou vyjadřovány buď (i) morfologickými prostředky (diateze), nebo (ii) prostředky převážně syntaktickými (reciprocita). Změny ve valenčním rámci slovesa jsou v tomto případě omezeny na změny v morfematické realizaci valenčních doplnění, které jsou natolik pravidelné, že mohou být zachyceny formálními syntaktickými pravidly.
[ "Václava Kettnerová", "Markéta Lopatková", "Zdeňka Urešová" ]
https://www.semanticscholar.org/paper/c7394ca22f446d6f00709734169d3eb1f7db80c7/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Mapping Semantic Information from FrameNet onto VALLEX
Obohacení VALLEXu o sémantické informace z FrameNetu
In this article, we introduce a project aimed at enhancing a valency lexicon of Czech verbs with semantic information. For this purpose, we make use of FrameNet, a semantically ori- ented lexical resource. At the present stage, semantic frames from FrameNet have been mapped to eight groups of verbs with various semantic and syntactic properties. The feasibility of this task has been verified by the achieved inter-annotator agreement measured on two semantically and syntactically different groups of verbs –- verbs of communication and exchange (85.9% and 78.5%, respectively). Based on the upper level semantic frames from the relation of ‘Inheritance’ built in FrameNet, the verbs of these eight groups have been classified into more coherent se- mantic classes. Moreover, frame elements from these upper level semantic frames have been assigned to valency complementations of the verbs of the listed groups as semantic roles. As in case of semantic frames, the achieved interannotator agreement concerning assigning frame elements measured on verbs of communication and exchange has been promising (95.6% and 91.2%, respectively). As a result, 1 270 lexical units pertaining to the verbs of communication, mental action, psych verbs, social interaction, verbs of exchange, motion, transport and location (2 129 Czech verbs in total if perfective and imperfective verbs being counted separately) have been classified into syntactically and semantically coherent classes and their valency complementations have been characterized by semantic roles adopted from the FrameNet lexical database.
V tomto článku popisujeme projekt zaměřený na obohacení valenčního lexikonu českých sloves, VALLEX, o sémantické informace z FrameNetu. V tomto projektu bylo celkem osm skupin českých sloves anotováno sémantickými rámci z FrameNetu. Na základě sémantických rámců z horních pater relace dědičnosti byla tato slovesa rozdělena do koherentních sémantických tříd. Valenčním doplněním daných sloves byly dále přiděleny elementy sémantických rámců z horních pater relace dědičnosti jako sémantické role.
[ "Václava Kettnerová", "Markéta Lopatková", "Eduard Bejček" ]
https://www.semanticscholar.org/paper/c67ea7fdf4e8488da7b7d1c66518e8f09980b4c0/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
The Syntax-Semantics Interface of Czech Verbs in the Valency Lexicon
Jazykové jevy na pomezí syntaxe a sémantiky ve valenčním lexikonu
In this paper, alternation based model of the valency lexicon of Czech verbs, VALLEX, is described. Two types of alternations (changes in valency frames of verbs) are distinguished on the basis of used linguistic means: (i) grammaticalized alternations and (ii) lexicalized alternations. Both grammaticalized and lexicalized alternations are either conversive, or non-conversive. While grammaticalized alternations relate different surface syntactic structures of a single lexical unit of a verb, lexicalized alternations relate separate lexical units. For the purpose of the representation of alternations, we divide the lexicon into data and rule components. In the data part, each lexical unit is characterized by a single valency frame and by applicable alternations. In the rule part, two types of rules are contained: (i) syntactic rules describing grammaticalized alternations and (ii) general rules determining changes in the linking of situational participants with valency complementations typical of lexicalized alternations.
V tomto příspěvku přinášíme popis alternačního modelu valenčního lexikonu českých sloves, VALLEX. V práci rozlišujeme celkem dva základní typy alternací (změn ve valenční struktuře sloves): (i) gramatikalizované a (ii) lexikalizované alternace. Jak gramatikalizované, tak lexikalizované alternace mohou být konverzní či nekonverzní povahy. Zatímco gramatikalizované alternace spojují rozdílné povrchové strukturace téže lexikální jednotky, lexikalizované alternace charakterizují syntaktické struktury tvořené rozdílnými lexikálními jednotkami. Pro účely reprezentace alternací rozdělujeme lexikon na datovou a pravidlovou část. V datové části je každá lexikální jednotka reprezentována jedním valenčním rámcem. Pravidlová část obsahuje dva typy pravidel: (i) formální syntaktická pravidla zachycující gramatikalizované alternace a (ii) obecná pravidla určující změny v korespondenci participantů a valenčních doplnění typické pro lexikalizované alternace.
[ "Václava Kettnerová", "Markéta Lopatková", "Eduard Bejček" ]
https://www.semanticscholar.org/paper/89c2078fbc09c2b24d92c1f74e726ebcd6491e04/
null
null
null
null
null
null
null
null
null
null
null
null
en
2012
Optimisation for POMDP-Based Spoken Dialogue Systems
Optimalizace POMDP dialogových systémů
This chapter presents current knowledge about statistical dialogue systems based on POMDP model of interaction with human users. The chapter introduces basic concepts of POMDP and suggest appropriate methods for implementation it the context of dialogue systems.
Tato kapitola prezentuje současné poznatky v oblasti vývoje statistických dialogových systému založených na POMDP modelu řízení komunikace s uživatelem. Postupně jsou vysvětleny základy POMDP a vhodné metody pro implementaci v oblasti dialogových systémů.
[ "Milica Gašić", "Filip Jurčíček", "Blaise Thomson", "Steve Young" ]
https://www.semanticscholar.org/paper/4c500687964f6a8357977169320b440c3269213b/
null
null
null
null
null
null
null
null
null
null
null
null