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
cs
2017
EVALD 2.0
EVALD 2.0
Evaluator of Discourse (EVALD 2.0) is a software that serves for automatic evaluation of surface coherence (cohesion) in Czech texts written by native speakers of Czech. Software evaluates the level of the given text from the perspective of its surface coherence (i.e. it takes into consideration, e.g., the frequency and diversity of connectives, richness of vocabulary, coreference realtions etc.).
Počítačový nástroj Evaluátor diskurzu (EVALD 2.0) slouží k automatickému hodnocení povrchové koherence (koheze) textů v češtině psaných rodilými mluvčími češtiny. Software tedy hodnotí (známkuje) úroveň předloženého textu z hlediska jeho povrchové výstavby (zohledňuje např. frekvenci a rozmanitost spojovacích prostředků, bohatost slovní zásoby, koreferenční vztahy apod.).
[ "Michal Novák", "Kateřina Rysová", "Jiří Mírovský", "Magdaléna Rysová", "Eva Hajičová" ]
https://www.semanticscholar.org/paper/e975139da766c2db5f4b2f449ae43abb523a0863/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2017
EVALD 2.0 for Foreigners
EVALD 2.0 pro cizince
Evaluator of Discourse for Foreigners (EVALD 2.0 for Foreigners) is a software that serves for automatic evaluation of surface coherence (cohesion) in Czech texts written by non-native speakers of Czech. Software evaluates the level of the given text from the perspective of its surface coherence (i.e. it takes into consideration, e.g., the frequency and diversity of connectives, richness of vocabulary, coreference realtions etc.).
Počítačový nástroj Evaluátor diskurzu pro cizince (EVALD 2.0 pro cizince) slouží k automatickému hodnocení povrchové koherence (koheze) textů v češtině psaných nerodilými mluvčími češtiny. Software hodnotí úroveň předloženého textu z hlediska jeho povrchové výstavby (zohledňuje např. frekvenci a rozmanitost užitých spojovacích prostředků, bohatost slovní zásoby, koreferenční vztahy apod.).
[ "Michal Novák", "Kateřina Rysová", "Jiří Mírovský", "Magdaléna Rysová", "Eva Hajičová" ]
https://www.semanticscholar.org/paper/7ac215cde9944ad1af3547e91c0dfc5ef0bcb510/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Incorporating Coreference to Automatic Evaluation of Coherence in Essays
Začlenění koreference do automatického hodnocení koherence v slohových pracích
The paper contributes to the task of automated evaluation of surface coherence. It introduces a coreference-related extension to the EVALD applications, which aim at evaluating essays produced by native and non-native students learning Czech. Having successfully employed the coreference resolver and coreference-related features, our system outperforms the original EVALD approaches by up to 8 percentage points. The paper also introduces a dataset for non-native speakers' evaluation, which was collected from multiple corpora and the parts with missing annotation of coherence grade were manually judged. The resulting corpora contains sufficient number of examples for each of the grading levels.
Článek popisuje rozšíření systémů EVALD, jejichž cílem je hodnocení úrovně koherence v slohových pracích, o rysy využívající informace o koreferenci a zájmenách.
[ "Michal Novák", "Kateřina Rysová", "Magdaléna Rysová", "Jiří Mírovský" ]
https://www.semanticscholar.org/paper/3e1c2c7ff9fac6aeb29e40bc7cf1d9046d8433f0/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Projection-based Coreference Resolution Using Deep Syntax
Rozpoznávaní koreference založené na projekci s použitím tektogramatiky
The paper describes the system for coreference resolution in German and Russian, trained exclusively on coreference relations projected through a parallel corpus from English. The resolver operates on the level of deep syntax and makes use of multiple specialized models. It achieves 32 and 22 points in terms of CoNLL score for Russian and German, respectively. Analysis of the evaluation results show that the resolver for Russian is able to preserve 66\% of the English resolver's quality in terms of CoNLL score. The system was submitted to the Closed track of the CORBON 2017 Shared task.
Článek popisuje systém pro rozpoznávání koreference v němčině a ruštině, trénovaný výlučně na koreferenčních vztazích projektovaných skrz paralelní korpus. Rozpoznávač operuje na tektogramatické vrstvě a používá vícero specializovných modelů. Měřeno metrikou CoNLL systém dosahuje 32 bodů pro Ruštinu a 22 bodov pro němčinu. Analýza výsledků ukazuje, že rozpoznávač pro ruštinu je schopen při projekci z angličtiny dosáhnout 66% kvality dosažené na angličtině. Systém byl poslán na CORBON 2017 Shared task.
[ "Michal Novák", "Anna Nedoluzhko", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/6527c0b24ee8cc6c8501c79f04df3b7aa8571f7e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
CoNLL 2017 Shared Task - Automatically Annotated Raw Texts and Word Embeddings
Společná úloha CoNLL 2017 – automaticky anotované texty a slovní embeddingy
Automatic segmentation, tokenization and morphological and syntactic annotations of raw texts in 45 languages, generated by UDPipe (http://ufal.mff.cuni.cz/udpipe), together with word embeddings of dimension 100 computed from lowercased texts by word2vec (https://code.google.com/archive/p/word2vec/).
Automatická segmentace, tokenizace a morfologická a syntaktická anotace textů ve 45 jazycích, vygenerovaná pomocí UDPipe (http://ufal.mff.cuni.cz/udpipe), spolu se 100rozměrnými slovními embeddingy vypočítanými nad textem převedeným na malá písmena nástrojem word2vec (https://code.google.com/archive/p/word2vec/).
[ "Filip Ginter", "Jan Hajič", "Juhani Luotolahti", "Milan Straka", "Daniel Zeman" ]
https://www.semanticscholar.org/paper/3ee4ea72679b2ec3814053a0ec5b5a16188b3948/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Findings of the WMT 2017 Biomedical Translation Shared Task
Výsledky soutěže v překladu biomedicínských textů WMT2017
Automatic translation of documents is an important task in many domains, in- cluding the biological and clinical do- mains. The second edition of the Biomed- ical Translation task in the Conference of Machine Translation focused on the au- tomatic translation of biomedical-related documents between English and various European languages. This year, we ad- dressed ten languages: Czech, German, English, French, Hungarian, Polish, Por- tuguese, Spanish, Romanian and Swedish. Test sets included both scientific publica- tions (from the Scielo and EDP Sciences databases) and health-related news (from the Cochrane and UK National Health Ser- vice web sites). Seven teams participated in the task, submitting a total of 82 runs. Herein we describe the test sets, participat- ing systems and results of both the auto- matic and manual evaluation of the trans- lations.
Článek shrnuje výsledky druhého ročníku Biomedical Translation Task konference WMT 2017, tj. strojového překladu biomedicínských textů.
[ "Antonio Jimeno Yepes", "Aurelie Névéol", "Mariana Neves", "Karin Verspoor", "Ondřej Bojar", "Arthur Boyer", "Cristian Grozea", "Barry Haddow", "Madeleine Kittner", "Yvonne Lichtblau", "Pavel Pecina", "Roland Roller", "Rudolf Rosa", "Amy Siu", "Philippe Thomas", "Saskia Trescher" ]
https://www.semanticscholar.org/paper/bcb0b58fe165a593ef5fca30fb5c3463f682c8b7/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Producing Unseen Morphological Variants in Statistical Machine Translation
Generování neviděných tvaroslovných variant ve statistickém strojovém překladu
Translating into morphologically rich languages is difficult. Although the coverage of lemmas may be reasonable, many morphological variants cannot be learned from the training data. We present a statistical translation system that is able to produce these inflected word forms. Different from most previous work, we do not separate morphological prediction from lexical choice into two consecutive steps. Our approach is novel in that it is integrated in decoding and takes advantage of context information from both the source language and the target language sides.
Strojový překlad do tvaroslovně bohatých jazyků představuje složitý problém. Prestože pokrytí na úrovni lemmat může být dostatečné, řada jejich tvaroslovných variant se z trénovacích dat nedá získat. Představujeme statistický překladový systém, který tyto tvaroslovné varinaty dokáže generovat. Na rozdíl od dřívějších prací nerozdělujeme modelování tvarosloví a lexikální volbu na dva navazující kroky. Náš postup je integrován přímo v dekódování a využívá informace z konextu jak na zdrojové, tak na cílové straně.
[ "Matthias Huck", "Aleš Tamchyna", "Ondřej Bojar", "Alexander Fraser" ]
https://www.semanticscholar.org/paper/a28a859693350ae3508b3491332a2dc9b6e6f1eb/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
A Universal Dependencies Treebank for Marathi
Závislostní korpus maráthštiny v Universal Dependencies
This paper describes the creation of a free and open-source dependency treebank for Marathi, the first open-source treebank for Marathi following the Universal Dependencies (UD) syntactic annotation scheme. In the paper, we describe some of the syntactic and morphological phenomena in the language that required special analysis, and how they fit into the UD guidelines. We also evaluate the parsing results for three popular dependency parsers on our treebank.
Tento článek popisuje vytváření volně dostupného závislostního korpusu maráthštiny, který odpovídá anotačnímu schématu Universal Dependencies (UD).
[ "Vinit Ravishankar" ]
https://www.semanticscholar.org/paper/f77ec4c85a8505b9c9502ccbd94f79082a4689d5/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Variable Mini-Batch Sizing and Pre-Trained Embeddings
Proměnné velikosti dávky a předtrénované embedingy
This paper describes our submission to the WMT 2017 Neural MT Training Task. We modified the provided NMT system in order to allow for interrupting and con- tinuing the training of models. This al- lowed mid-training batch size decremen- tation and incrementation at variable rates. In addition to the models with variable batch size, we tried different setups with pre-trained word2vec embeddings. Aside from batch size incrementation, all our ex- periments performed below the baseline
Článek popisuje náš příspěvek do společné úlohy trénování systémů neuronového překladu (WMT2017 Neural MT Training Task).
[ "Mostafa Abdou", "Vladan Glončák", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/3cd3f141df1dd82449cd7b2e4b6fe0206fd3c7df/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Deep Architectures for Neural Machine Translation
Hluboké architektury pro neuronový strojový překlad
It has been shown that increasing model depth improves the quality of neural machine translation. However, different architectural variants to increase model depth have been proposed, and so far, there has been no thorough comparative study. In this work, we describe and evaluate several existing approaches to introduce depth in neural machine translation. Additionally, we explore novel architectural variants, including deep transition RNNs, and we vary how attention is used in the deep decoder. We introduce a novel "BiDeep" RNN architecture that combines deep transition RNNs and stacked RNNs. Our evaluation is carried out on the English to German WMT news translation dataset, using a single-GPU machine for both training and inference. We find that several of our proposed architectures improve upon existing approaches in terms of speed and translation quality. We obtain best improvements with a BiDeep RNN of combined depth 8, obtaining an average improvement of 1.5 BLEU over a strong shallow baseline. We release our code for ease of adoption.
Bylo prokázáno, že zvyšující se hloubka modelu zlepšuje kvalitu neuronového strojového překladu. Přes mnoho návrhů různých variant arhitektur pro zvýšení hloubky modelu doposud nebyla provedena žádná důkladná srovnávací studie. V této práci popisujeme a vyhodnocujeme několik stávajících přístupů k zavedení hloubky v neuronovém strojovém překladu. Navíc prozkoumáváme nové varianty architektur včetně hlubokých přechodových RNN a měníme, jak je hlubokém dekodéru použit mechanismus pozornosti ("attention"). Představujeme novou architekturu "BiDeep" RNN, která kombinuje hluboké přechodové RNN a skládané RNN. Hodnocení provádíme na anglicko-německém datovém souboru WMT pro překlady novinových článků s využitím stroje s jednou GPU pro trénování i inferenci. Zjistili jsme, že několik našich navrhovaných architektur zlepšuje stávající přístupy z hlediska rychlosti a kvality překladu. Nejlepších výsledků jsme získali s BiDeep RNN kombinované hloubky 8, získáním průměrného zlepšení 1,5 BLEU nad silnou baseline. Náš kód je pro snadný přístup zveřejněn.
[ "Antonio Valerio Miceli Barone", "Jindřich Helcl", "Rico Sennrich", "Barry Haddow", "Alexandra Birch" ]
https://www.semanticscholar.org/paper/0b2c790fc30e39d275856cee1aa54002fc1ec0c9/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Tamil Dependency Treebank
Závislostní korpus tamilštiny
License agreement for Google Inc. (Mountain View, California 94043, USA) for using the Tamil Dependency Treebank.
Licenční smlouva pro Google Inc. (Mountain View, California 94043, USA) na užívání Tamilského závislostního korpusu.
[ "Loganathan Ramasamy", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/19c99b4482795c56c50dd63feeeefe7d4f5dac6e/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2017
MUSCIMA++
MUSCIMA++
Optical Music Recognition (OMR) is a sub-field of document analysis and recognition focused on extracting musical content from the image (e.g., scan or photo) of a musical score. Given that more compositions probably exist in written form than have been recorded, automatically transcribing written music can substantially diversify the sources available to digital musicology. The greatest potential of OMR lies in making accessible massive musical archives, such as the Lichtenstein-Castelcorn collection in Kroměříž, to further investigation using methods of digital musicology: automated duplicate or part detection, melody-based search, or tracking the genealogy of musical motifs. Within document analysis, OMR occupies a particularly difficult niche, because modern music notation is one of the most complicated writing systems overall. Especially for manuscripts, where this difficulty compounds with the variability in handwriting, no remotely satisfactory solutions are available today. The problem also only has a small community dedicated to solving it. The talk will give an overview of the current state of the art and the developments in the field, with focus on potential applications for digital musicology.
Rozpoznávání notopisu je specializovaný pod-obor automatického zpracování dokumentů, který se zaměřuje na extrakci hudebního obsahu z obrazové informace (skenu či fotografie) notového zápisu. Vzhledem k tomu, že pravděpodobně existuje více hudebních děl v psané podobě než v podobě nahrávky či jiného záznamu, může automatická transkripce not přinést digitální muzikologii výrazně rozmanitější zdroje. Největší potenciál spočívá v masovém zpřístupnění rozsáhlých notových archivů, jako např. kroměřížská sbírka rodu Lichtenštejn-Kastelkornů, které pak bude možné zkoumat pomocí dalších metod digitální muzikologie, např. automatická detekce duplikátů či partů, vyhledávání pomocí melodií, či sledování genealogie hudebních motivů. V rámci zpracování dokumentů se jedná o úlohu výjimečně obtížnou, neboť moderní hudební notace je jedním z nejkomplikovanějších systémů psaní. Obzvlášť pro rukopisy, kde se její složitost násobí s variabilitou realizace zápisu, neexistují v současnosti ani vzdáleně uspokojivá řešení. Na problému také pracuje relativně málo lidí. Příspěvek nabídne vhled do současného stavu poznání a vývoje oboru, s důrazem na možné aplikace v digitální muzikologii.
[ "Jan Hajič, jr." ]
https://www.semanticscholar.org/paper/87a4ed475154212d32e22b349e01cf0beb7b18cd/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Optical Music Recognition: Because music notation is text, too!
Rozpoznávání notopisu: Hudební notace je také text.
Most compositions created throughout history exist today only in written form, usually residing in archive collections; specifically in the Czech Republic, there are many more than 10 000 such manuscripts. To preserve and disseminate this portion of cultural heritage, it is advantageous to digitize it; further usability would be brought by also digitizing the contents of these documents. However, transcribing music with notation editors such as Sibelius or MuseScore is too time-consuming. Optical Music Recognition (OMR), the equivalent of OCR for music notation, can be a key tool for opening the contents of musical archives to large-scale musicological research, better curation (e.g., duplicate search), and for making the way from an archival score to performance easier.
Většina hudebních skladeb, které byly kdy vytvořeny, dnes existuje pouze v psané podobě, většinou v rámci archivů; konkrétně v ČR je více než 10 000 takových rukopisů. Pro zachování a šíření této části kulturního dědictví je vhodné jej digitalizovat, a další výhody přináší digitalizace hudebního obsahu těchto dokumentů. Ruční přepisování not v editorech jako Sibelius či MuseScore je však v tomto rozsahu příliš pomalé. Rozpoznávání notopisu (Optical Music Recognition, OMR), ekvivalent OCR pro hudební notaci, může být klíčovým nástrojem pro zpřístupnění obsahu hudebních archivů - pro široký muzikologický výzkum, pro lepší správu (např. detekce opisů), a pro zkrácení cesty, kterou skladby musí ujít od archiválie k živému provedení.
[ "Jan Hajič, jr." ]
https://www.semanticscholar.org/paper/6acddec324021910ac23099c128a6165fa74c5eb/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Prototyping Full-Pipeline OMR With MUSCIMarker
Prototypy kompletních OMR systémů s MUSCIMarkerem
We present MUSCIMarker, an open-source workbench for developing Optical Music Recognition (OMR) systems from image preprocessing to MIDI export. It is built around the notation graph data model of the MUSCIMA++ dataset for full-pipeline OMR. The system is transparent and interactive, enabling the user to visualize, validate and edit results of individual OMR stages. It is platform-independent, written purely in Python, and can work offline. We demonstrate its value with a prototype OMR system for musical manuscripts that implements the recognition pipeline, up to playing the recognition outputs through MIDI. The audience will interact with the program and can test an integrated OMR system prototype.
Představujeme MUSCIMarker, open-source nástroj pro vývoj systémů rozpoznávání notopisu (Optical Music Recognition, OMR). Nástroj je postavený okolo reprezentace notopisu jakožto grafu, definovaného v datasetu MUSCIMA++. Nástroj je transparentní a interaktivní, umožňuje uživateli vizualizovat, ověřit a upravovat výsledky jednotlivých kroků OMR. Navíc je díky čistě Pythonové implementaci přenosný mezi operačními systémy, a umožňuje pracovat offline. Dokládáme hodnotu MUSCIMarkeru skrze prototyp systému na plnohodnotné rozpoznávání notopisu, od předzpracování obrazu po přehrání výstupu a export do formátu MIDI. Publikum prezentace bude mít příležitost MUSCIMarker i rozpoznávací prototyp vyzkoušet.
[ "Jan Hajič, jr.", "Matthias Dorfer" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Detecting Noteheads in Handwritten Scores with ConvNets and Bounding Box Regression
Detekce notových hlaviček v rukopisech pomocí konvolučních sítí a regrese na ohraničení
Noteheads are the interface between the written score and music. Each notehead on the page signifies one note to be played, and detecting noteheads is thus an unavoidable step for Optical Music Recognition. In printed notation, noteheads are clearly distinct objects; however, the variety of music notation handwriting makes noteheads harder to identify, and while handwritten music notation symbol classification is a well-studied task, symbol detection has usually been limited to heuristics and rule-based systems instead of machine learning methods better suited to deal with the uncertainties in handwriting. We present ongoing work on a simple notehead detector using convolutional neural networks for pixel classification and bounding box regression that achieves a detection f-score of 0.97 on binary score images in the MUSCIMA++ dataset, does not require staff removal, and is applicable to a variety of handwriting styles and levels of musical complexity.
Notové hlavičky představují rozhraní mezi zápisem hudby a hudbou samotnou. Každá hraná nota je kódována pomocí notové hlavičky, a detekovat hlavičky je tím pádem pro rozpoznávání not nevyhnutelné. V tištěné notaci jsou hlavičky jasně rozlišitelné, avšak různorodost rukopisů činí jejich identifikaci obtížnější. Představujeme jednoduchý detektor notových hlaviček používající konvoluční neuronové sítě pro klasifikaci pixelů a regresi na ohraničení, který dosahuje na detekci f-score 0.97 nad datasetem MUSCIMA++, nepotřebuje odstraňování osnov, a lze jej použít na různorodé rukopisné styly a úrovně složitosti zapsané hudby.
[ "Jan Hajič, jr.", "Pavel Pecina" ]
https://www.semanticscholar.org/paper/5c45a6d637cdb44bde65c9ef9cf43c9ab478cc2b/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Groundtruthing (not only) Music Notation with MUSCIMarker: a Practical Overview
Anotace (nejen) hudební notace s MUSCIMarkerem: praktický přehled
Dataset creation for graphics recognition, especially for hand-drawn inputs, is often an expensive and time-consuming undertaking. The MUSCIMarker tool used for creating the MUSCIMA++ dataset for Optical Music Recognition (OMR) led to efficient use of annotation resources, and it provides enough flexibility to be applicable to creating datasets for other graphics recognition tasks where the ground truth can be represented similarly. First, we describe the MUSCIMA++ ground truth to define the range of tasks for which using MUSCIMarker to annotate ground truth is applicable. We then describe the MUSCIMarker tool itself, discuss its strong and weak points, and share practical experience with the tool from creating the MUSCIMA++ dataset.
Tvorba datasetů pro ropzoznávání grafických znaků, především ručně psaných, je často drahá a časově náročná záležitost. Nástroj MUSCIMarker, kterým jsme vytvořili dataset MUSCIMA++ pro rozpoznávání not, pomohl omezené zdroje využít efektivně, a je dostatečně flexibilní na to, aby jej bylo možné použít na tvorbu datasetů pro další úlohy s podobnou reprezentací anotované "pravdy". Nejprve popíšeme tuto reprezentaci, aby byla zřejmé, na jaké úlohy je MUSCIMarker aplikovatelný; následně popisujeme samotný nástroj MUSCIMarker, jeho silné a slabé stránky, a praktické zkušenosti s tvorbou datasetu MUSCIMA++.
[ "Jan Hajič, jr.", "Pavel Pecina" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
How to Exploit Music Notation Syntax for OMR?
Jak využít syntax hudební notace pro OMR?
A major roadblock for Optical Music Recognition, especially for handwritten music notation, is symbol detection: recovering the locations of musical symbols from the input page. This has been attempted both with bottom-up approaches exploiting visual features, and top-down approaches based on the strong constraints that music notation syntax imposes on possible symbol configurations; sometimes joined together at appropriate points in the recognition process. The bottom-up approach has recently greatly improved with the boom of neural networks. However, the reduction in uncertainty that music notation syntax can provide has not yet been married to the power of these neural network models. This extended abstract brainstorms ways in which this can be done, and analyzes the difficulties the various combined approaches will have to address. We hope our work will foster further discussion to clarify the issues involed, provoke OMR researchers to try some of these approaches experimentally, and entice researchers from other parts of the graphics recognition community to share relevant experience.
Podstatným problémem pro rozpoznávání not, především ručně psaných, je lokalizace symbolů ve vstupním obrázku. Řešení jsou stavěna jak odspodu, využívajíce nízkoúrovňové vizuální rysy, tak shora, kde se využívá skutečnost, že se hudební notace řídí silnými omezeními na syntakticky správné konfigurace symbolů. Oba přístupy se občas kombinují. V nedávné době se přístup odspodu výrazně zlepšil pomocí konvolučních neuronových sítí. Snížení nejistoty, které může notační syntax poskytnout, však ještě s těmito modely zkombinována nebyla. Tento rozšířený abstrakt diskutuje způsoby, jak neuronové sítě a notační syntax propojit, a analyzuje obtíže, se kterými by se jednotlivé přístupy měly potýkat. Doufáme, že náš příspěvek podnítí další diskusi o těchto možnostech, vyprovokuje výzkumníky v oboru k experimentálnímu prozkoumání navržených přístupů, a podnítí výzkumníky z příbuzných oblastí sdílet své zkušenosti.
[ "Jan Hajič, jr.", "Pavel Pecina" ]
https://www.semanticscholar.org/paper/b8d08d4d999996f79a6ddb1c9419ee98c1027a40/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
The MUSCIMA++ Dataset for Handwritten Optical Music Recognition
Dataset MUSCIMA++ pro rozpoznávání ručního notopisu
Optical Music Recognition (OMR) promises to make accessible the content of large amounts of musical documents, an important component of cultural heritage. However, the field does not have an adequate dataset and ground truth for benchmarking OMR systems, which has been a major obstacle to measurable progress. Furthermore, machine learn- ing methods for OMR require training data. We design and collect MUSCIMA++, a new dataset for OMR. Ground truth in MUSCIMA++ is a notation graph, which our analysis shows to be a necessary and sufficient representation of music notation. Building on the CVC-MUSCIMA dataset for staffline removal, the MUSCIMA++ dataset v1.0 consists of 140 pages of hand- written music, with 91254 manually annotated notation symbols and 82247 explicitly marked relationships between symbol pairs. The dataset allows training and directly evaluating models for symbol classification, symbol localization, and notation graph assembly, and musical content extraction, both in isolation and jointly. Open-source tools are provided for manipulating the dataset, visualizing the data and annotating more, and the data is made available under an open license.
Od rozpoznávání notopisu (Optical Music Recognition, OMR) si lze slibovat zpřístupnění mnoha hudebních dokumentů, které jsou podstatnou součástí kulturního dědictví. OMR však nemá adekvátní data a příslušný formát anotace, který by umožnil porovnávání systémů OMR, což představuje výraznou překážku pro měřitelný pokrok. Řešení OMR využívající strojové učení navíc potřebují trénovací data. Navrhli jsme a sesbírali jsme nový OMR dataset MUSCIMA++. Poskytované anotace tvoří notační graf, jejž naše analýza odhalila jako nutný a postačující popis hudební notace. Stavíme nad daty CVC-MUSCIMA pro odstraňování notových osnov. MUSCIMA++ v1.0 obsahuje 140 stran hudebního rukopisu, s 91245 ručně vyznačenými symboly a 82247 vztahy mezi nimi. Dataset umožňuje trénovat a přímo evaluovat modely pro klasifikaci a lokalizaci symoblů, rekonstrukce logické struktury notace, a extrakce hudebního obsahu. Jsou poskytnuty open-source nástroje pro manipulaci s datasetem, vizualizaci a rozšiřování anotací, a data samotná jsou poskytnuta pod otevřenou licencí.
[ "Jan Hajič, jr.", "Pavel Pecina" ]
https://www.semanticscholar.org/paper/87a4ed475154212d32e22b349e01cf0beb7b18cd/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Česílko Goes Open-source
Česílko je Open-source
The Macine translation system Česílko has been developed as an answer to a growing need of translation and localization from one source language to many target languages. The system belongs to the shallow parse, shallow transfer RBMT paradigm and it is designed primarily for translation of related languages. The paper presents the architecture, the development design and the basic installation instructions of the translation system.
Článek popisuje přeměnu překladového systému pro příbuzné jazyky Česílko do formátu Open-source. Tento systém byl vytvořen pro účely rychlého a kvalitního překladu z jednoho zdrojového do více příbuzných cílových jazyků. Jeho architektura je založena na mělké analýze a transferu pomocí pravidel. Článek představuje architekturu systému a instalační instrukce.
[ "Jernej Vičič", "Vladislav Kuboň", "Petr Homola" ]
https://www.semanticscholar.org/paper/d174e1623ca3aed017732932c635b467766474ef/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Communication with Robots using Multilayer Recurrent Networks
Komunikace s roboty pomocí vícevrstvých neuronových sítí
In this paper, we describe an improvement on the task of giving instructions to robots in a simulated block world using unrestricted natural language commands
V tomto článku popisujeme vylepšení úlohy pro řízení robota neomezeným přirozeným jazykem.
[ "Bedřich Pišl", "David Mareček" ]
https://www.semanticscholar.org/paper/6a8e2ecb7dfed27c3513aff99fe710a364332907/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
MorphoRuEval-2017: an evaluation track for the automatic morphological analysis methods for Russian
MorphoRuEval-2017: vyhodnocovací stopa pro metody automatické morfologické analýzy pro ruštinu
MorphoRuEval-2017 is an evaluation campaign designed to stimulate the development of the automatic morphological processing technologies for Russian, both for normative texts (news, fiction, nonfiction) and those of less formal nature (blogs and other social media). This article compares the methods participants used to solve the task of morphological analysis. It also discusses the problem of unification of various existing training collections for Russian language.
MorphoRuEval-2017 je hodnotící kampaň určená k povzbuzení rozvoje technologií automatického morfologického zpracování pro ruštinu, a to jak pro normativní texty (novinky, beletrie, fakta), tak pro méně formální povahu (blogy a další sociální média). Tento článek porovnává metody, které účastníci použili při řešení úlohy morfologické analýzy. Rovněž se zabývá problémem sjednocení různých stávajících výcvikových sbírek ruského jazyka.
[ "Alexey Sorokin", "Tatiana Shavrina", "Olga Lyashevskaya", "Victor Bocharov", "Svetlana Alexeeva", "Kira Droganova", "Alena Fenogenova", "Dmitry Granovsky" ]
https://www.semanticscholar.org/paper/b459b460750948463f74be9daed60c2ea0b4aa39/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Beyond simple usage of handles - PIDs in DSpace
Složitější využití handlů - persistentních identifikátorů v DSpace
What else can be achieved with handles in addition to "basic" separation of resource id from its location. Handle metadata. Handles and content negotiation. Template handles.
Co dál je možné dělat s handlem, kromě "základního" oddělení id od lokace. Metadata handlu. Handle a content negotiation. Template handle.
[ "Ondřej Košarko" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
CLARIN DSpace
CLARIN DSpace
History, development and a introduction to usage of clarin-dspace.
Historie vzniku, vývoj a úvod do používání repozitáře clarin-dspace.
[ "Ondřej Košarko" ]
https://www.semanticscholar.org/paper/47cd6f42a6fc1b1baabb6e18799b9ed2fac37c81/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Annotation of the Evaluative Language in a Dependency Treebank
Anotace evaluativního jazyka v závislostním korpusu
In the paper, we present our efforts to annotate evaluative language in the Prague Dependency Treebank 2.0. The project is a follow-up of the series of annotations of small plaintext corpora. It uses automatic identification of potentially evaluative nodes through mapping a Czech subjectivity lexicon to syntactically annotated data. The annotations unveiled several advantages and disadvantages of the chosen framework.
V příspěvku představujeme projekt anotace evaluativního významu v Pražském závislostním korpusu 2.0. Projekt navazuje na sérii anotací malých korpusů prostého textu. V projektu byla použita automatická identifikace potenciálně evaluativních uzlů prostřednictvím českého slovníku hodnotících výrazů Czech SubLex 1.0. V rámci anotačních prací byly odhaleny výhody i nevýhody zvoleného anotačního schématu.
[ "Jana Šindlerová" ]
https://www.semanticscholar.org/paper/c084d671bbb03c73a4f26811ebf7a354862deddf/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2017
The radio said... or the mutual alignment of location and actor in a bilingual treebank
Hlásili to v rozhlase aneb vzájemné mapování lokace a aktoru v dvojjazyčném závislostním korpusu
In the paper we address one type of structural difference between Czech and English parallel sentences in a bilingual PCEDT treebank, the mutual mapping of actors and locations. Within the analysis of treebank examples, we show the effect of language-specific syntactic and semantic preferences of individual words in the positions of the subject and predicate on the differences in valency structure, as well as the effect of diathesis and the principal of agent demoting.
V příspěvku se zabýváme jedním typem strukturní odlišnosti mezi českými a anglickými paralelními větami v dvojjazyčném korpusu PCEDT, konkrétně vzájemným mapováním valenčních doplnění aktorového typu a (nevalenčních) lokačních doplnění. Analýzou korpusových příkladů ukazujeme, do jaké míry se na rozdílech ve valenci sloves v překladově ekvivalentních větách podílejí jazykově specifické syntakticko-sémantické preference konkrétních slov na pozicích subjektu a predikátu a jak je toto mapování ovlivněno vzájemnou souhrou těchto preferencí, větné diateze a sémantického principu potlačení agentu ve větné struktuře.
[ "Jana Šindlerová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Emotions Translated: Enhancing a Subjectivity Lexicon Using a Parallel Valency Lexicon
Překlad emocí: rozšíření lexikonu hodnotících výrazů (???) pomocí paraleního valenčního slovníku
This paper documents the behavior of verb valency complementations regarding the position of the target of evaluation within the valency frame. We classify the types of evaluative meaning expressed by the verbs and identify shared characteristic features considering the valency patterns of the verbs. In the analysis, we comment on three major issues of interest: the semantic classification of evaluative verbs and its relation to the propagation of sentiment value to the participants, the possible non-matching structural positions of the target of evaluation in the valency frame of a verb and its translation, i.e., the possible shift in evaluative focus and scope, and the possible loss of evaluative stance in the process of translation.
Příspěvek sleduje chování valenčních doplnění sloves vzhledem k pozici evaluačního cíle mezi participanty valenčního rámce. Předkládáme klasifikaci typů evaluativnío významu vyjádřeného slovesy a popisujeme společné charakteristické rysy valenčních rámců evaluačních sloves. V analýze se zabýváme třemi problémy: sémantickou klasifikací evaluačních sloves a jejím vztahem k propagaci evaluačního významu jednotlivým participantům, různým valenčním pozicím evaluačního cíle u vzýjemně překladových sloves, tj. možným posunům evaluativního fokusu a dosahu a ztrátě evaluativního významu během překladu.
[ "Jana Šindlerová", "Aleš Tamchyna" ]
https://www.semanticscholar.org/paper/0527afc1506a153ed91dfb151c3229ba3396d3a8/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2017
Linguistic reasons for choosing constructivist approach to teaching word class categorization
Lingvistické opodstatnění konstruktivistické didaktiky při výuce slovních druhů
The paper addresses the advantages of teaching word class categorization in Czech in a constructivist manner. The problem of vagueness and blurriness of the linguistic category of word class is analyzed from the linguistic point of view, the lexicographer's point of view and the pedagogical point of view. We argue that because of the overall change of the aims of the teaching process, constructivism is more in compliance with the principles of modern linguistic approaches than the transmissive teaching. It allows pupils to build their own linguistic concepts by their logical thinking and direct manipulation of language, shifting the attention from the word class categorization training to the analytical thinking.
V příspěvku se zabýváme možnostmi pojetí výuky slovních druhů v konstruktivistickém vyučování. Problematiku vágnosti a vzájemné prostupnosti jazykových kategorií analyzujeme z hlediska lingvistické teorie, která usiluje o konzistentnost a přístupnost pro školní praxi, z hlediska lexikografické práce, jejímž cílem je připravit jasnou oporu pro běžného uživatele i pedagoga, i z hlediska moderní pedagogiky, jejímž úkolem je zpřístupňovat žákům vědecké koncepty a teoretické konstrukty důvěryhodným a smysluplným způsobem. Důraz klademe zejména na proměnu cílů jazykového vyučování i na proměnu vnímání role učitele. Tvrdíme, že konstruktivistické vyučování je více v souladu s principy současné jazykové deskripce než vyučování transmisivní, neboť umožňuje žákovské budování pojmů vlastní manipulací s materiálem a vlastní logickou úvahou a přechod od nácviku jednoznačné kategorizace k prohlubování obecných analytických schopností.
[ "Jana Šindlerová", "Barbora Štěpánková" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Universal Dependencies for Arabic
Universal Dependencies pro arabštinu
We describe the process of creating NUDAR, a Universal Dependency treebank for Arabic. We present the conversion from the Penn Arabic Treebank to the Universal Dependency syntactic representation through an intermediate dependency representation. We discuss the challenges faced in the conversion of the trees, the decisions we made to solve them, and the validation of our conversion. We also present initial parsing results on NUDAR.
Popisujeme proces vzniku NUDAR, arabského treebanku ve stylu Universal Dependencies. Představujeme převod z Penn Arabic Treebanku do syntaktické reprezentace Universal Dependencies přes mezilehlou závislostní reprezentaci. Probíráme obtíže, se kterými je převod závislostních stromů spojen, řešení, která jsme použili, a hodnocení námi převedených dat. Dále představujeme prvotní výsledky parsingu na NUDARu.
[ "Dima Taji", "Nizar Habash", "Daniel Zeman" ]
https://www.semanticscholar.org/paper/a16785a57f0055527d8969a72b88e650d3d613ee/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Paying Attention to Multi-Word Expressions in Neural Machine Translation
Pozornost víceslovným výrazům v neuronovém strojovém překladu
Processing of multi-word expressions (MWEs) is a known problem for any natural language processing task. Even neural machine translation (NMT) struggles to overcome it. This paper presents results of experiments on investigating NMT attention allocation to the MWEs and improving automated translation of sentences that contain MWEs in English→Latvian and English→Czech NMT systems. Two improvement strategies were explored—(1) bilingual pairs of automatically extracted MWE candidates were added to the parallel corpus used to train the NMT system, and (2) full sentences containing the automatically extracted MWE candidates were added to the parallel corpus. Both approaches allowed to increase automated evaluation results. The best result—0.99 BLEU point increase—has been reached with the first approach, while with the second approach minimal improvements achieved. We also provide open-source software and tools used for MWE extraction and alignment inspection.
Víceslovné výrazy představují problém pro všechny aplikace zpracování přirozeného jazyka. V článku představujeme výsledky experimentů s neuronovým strojovým překladem z angličtiny do češtiny a lotyšštiny s důrazem na víceslovné výrazy.
[ "Matiss Rikters", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/13b79cb62747e817358ff2d6746c1eadae31ffa9/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Visualizing Neural Machine Translation Attention and Confidence
Zobrazování pozornosti a spolehlivosti neuronového strojového překladu
In this article, we describe a tool for visualizing the output and attention weights of neural machine translation systems and for estimating confidence about the output based on the attention. Our aim is to help researchers and developers better understand the behaviour of their NMT systems without the need for any reference translations. Our tool includes command line and web-based interfaces that allow to systematically evaluate translation outputs from various engines and experiments. We also present a web demo of our tool with examples of good and bad translations: http://ej.uz/nmt-attention
V článku popisuje nástroj pro vykreslování výstupu a pozornostních vah neuronového překladu a pro odhad spolehlivosti překladu vypočtený na základě pozornosti.
[ "Matiss Rikters", "Mark Fishel", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/af65974aa5e3df6e8b5c30b2771d63661918004c/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Understanding text coherence: Implicit discourse relations
Porozumění textové koherenci: Implicitní diskurzní vztahy
The paper deals with secondary language signals that can express the semantics of implicit discourse relations. These include, for example, evaluative expressions in focus, usually followed by an argument with the meaning of explication or specification.
Příspěvek se zabývá vedlejšími jazykovými signály, které mohou vyjadřovat sémantiku implicitních vztahů. K nim patří např. hodnotící výrazy ve fokusu, za nimiž následuje obvykle argument s významem explikace či specifikace.
[ "Šárka Zikánová" ]
https://www.semanticscholar.org/paper/9e689631885fc1a9204a920b7700a15f39c1eac8/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2017
Prague workshop Discourse Relations in Multilingual Context (Kontakt II)
Pražský workshop Discourse Relations in Multilingual Context (Kontakt II)
A report on the contents of an international research project on the structure of discourse and on the means of its connectivity and on a workshop organized in Prague.
Zpráva o vědecké náplni mezinárodního projektu týkajícího se výstavby diskurzu a prostředků jeho konektivity a o workshopu pořádaném v Praze v rámci tohoto projektu.
[ "Šárka Zikánová", "Eva Hajičová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Discourse Connectives: From Historical Origin to Present-Day Development
Diskurzní konektory: Od původu k současnosti
The paper focuses on the description and delimitation of discourse connectives, i.e. linguistic expressions significantly contributing to text coherence and generally helping the reader to better understand semantic relations within a text. The paper discusses the historical origin of discourse connectives viewed from the perspective of present-day linguistics. Its aim is to define present-day discourse connectives according to their historical origin through which we see what is happening in discourse in contemporary language. The paper analyzes the historical origin of the most frequent connectives in Czech, English and German (which could be useful for more accurate translations of connectives in these languages) and point out that they underwent a similar process to gain a status of present-day discourse connectives. The paper argues that this historical origin or process of rising discourse connectives might be language universal. Finally, the paper demonstrates how these observations may be helpful for annotations of discourse in large corpora.
Kapitola se zaměřuje na popis a delimitaci diskurzních konektorů, tj. výrazů přispívajících ke kohereci textu a pomáhajících čtenáři lépe porozumět sémantickým vztahům v textu. V příspěvku je představen etymologický původ diskurzních konektorů z hlediska současné lingvistiky. Cílem je podat definici diskurzních konektorů s ohledem na jejich historický vývoj, díky němuž můžeme lépe nahlížet na diskurzní vztahy v současném jazyce. V kapitole analyzujeme etymologický původ a vývoj deseti nejfrekventovanějších konektorů v češtině, angličtině a němčině (mezijazykový pohled může napomoci také přesnějšímu překladu konektorů) a poukazujeme na fakt, že tyto konektory prošly ve všech sledovaných jazycích velmi podobným vývojem, než se ustálily v roli současných diskurzních konektorů. Domníváme se proto, že způsob vzniku diskurzních konektorů může být jazykovou univerzálií. V závěru kapitoly ukazujeme, jak naše zjištění mohou pomoci při anotacích diskurzu ve velkých korpusech.
[ "Magdaléna Rysová" ]
https://www.semanticscholar.org/paper/b661b42cf6d5a086693f826f9e060a62b5e92824/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2017
Installation and user settings for Czech-English and English-Czech translation of Czech technical terms of the INSPIRE Directive and thesaurea GEMET and AgroVoc (spatial data) for exclusive use in a research project
Instalace a uživatelské nastavení pro česko-anglický a anglicko-český překlad českých odborných termínů směrnice INSPIRE a thesaurů GEMET a AgroVoc (prostorová data) pro výhradní použití ve výzkumném projektu
Installation and user setting for Czech-English and English-Czech translation of Czech technical terms of the INSPIRE Directive and thesauri GEMET and AgroVoc (spatial data). For the project a package containing trained Moses models and commands for translation was prepared.
Instalace a uživatelské nastavení odborných překladů z češtiny do angličtiny a z angličtiny do češtiny. Překládají se české odborné termíny směrnice INSPIRE a thesaury GEMET a AgroVoc (prostorová data). Pro účely projektu byl připraven balíček obsahující připravené modely v systému Moses a seznam příkazů pro provádění překladu.
[ "Roman Sudarikov", "Ondřej Bojar" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Tokenizing, POS Tagging, Lemmatizing and Parsing UD 2.0 with UDPipe
Tokenizace, tagging, lemmatizace a syntaktická analýza UD 2.0 pomocí UDPipe
We present an update to UDPipe 1.0, a trainable pipeline which performs sentence segmentation, tokenization, POS tagging, lemmatization and dependency parsing. We provide models for all 50 languages of UD 2.0, and furthermore, the pipeline can be trained easily using data in CoNLL-U format. For the purpose of the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies, the updated UDPipe 1.1 was used as one of the baseline systems, finishing as the 13th system of 33 participants. A further improved UDPipe 1.2 participated in the shared task, placing as the 8th best system, while achieving low running times and moderately sized models. The tool is available under open-source Mozilla Public Licence (MPL) and provides bindings for C++, Python (through ufal.udpipe PyPI package), Perl (through UFAL::UDPipe CPAN package), Java and C#.
Představujeme novou verzi UDPipe 1.0, což je trénovatelný nástroj provádějící větnou segmentaci, tokenizaci, morfologické značkování, lemmatizaci a syntaktickou analýzu. Poskytujeme modely pro všech 50 jazyků UD 2.0, a navíc lze jednoduše UDPipe natrénovat pomocí vlastních dat v CoNLL-U formátu. Pro potřeby CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies, upravená verze UDPipe 1.1 byla použita jako základový systém a umístila se na 13. místě z 33 účastníků. Nejnovější verze UDPipe 1.2, která se také účastnila, dosáhla na 8. místo, přičemž potřebuje jen malý čas na běh a středné velké modely. Nástroj je k dispozici pod open-source licencí MPL a poskytuje rozhraní pro C++, Python (pomocí ufal.udpipe balíčku PyPI), Perl (pomocí UFAL::UDPipe balíčku CPAN), Javu a C#.
[ "Milan Straka", "Jana Straková" ]
https://www.semanticscholar.org/paper/ea40fba2dc52bde32dedd45b05f83d27dc2ea6c7/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Prague at EPE 2017: The UDPipe System
Praha na EPE 2017: Systém UDPipe
We present our contribution to The First Shared Task on Extrinsic Parser Evaluation (EPE 2017). Our participant system, the UDPipe, is an open-source pipeline performing tokenization, morphological analysis, part-of-speech tagging, lemmatization and dependency parsing. It is trained in a language agnostic manner for 50 languages of the UD version 2. With a relatively limited amount of training data (200k tokens of English UD) and without any English specific tuning, the system achieves overall score 56.05, placing as the 7th participant system.
Představujeme náš příspěvek do First Shared Task on Extrinsic Parser Evaluation (EPE 2017). Náš systém, UDPipe, je trénovatelný nástroj provádějící tokenizaci, morfologickou analýzu, morfologické značkování, lemmatizaci a syntaktickou analýzu. Je nezávislý na jazyku a k dispozici jsou modely pro všech 50 jazyků UD 2.0. Použitím relativně omezeného množství trénovacích dat (200 tisíc tokenů z anglického korpusu UD) a bez nastavení specifického pro angličtinu získal systém celkové hodnocení 56.05 a umístil se mezi soutěžícími systémy jako 7.
[ "Milan Straka", "Jana Straková", "Jan Hajič" ]
https://www.semanticscholar.org/paper/a8ff0160ea28c02573acbcfdd9083d3d9590d09a/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
clarin-dspace Architecture and RDA-DFT/FAIR/OAIS
Architektura clarin-dspace a RDA-DFT/FAIR/OAIS
Describing clarin-dspace architecture and the compliance to RDA-DFT/FAIR/OAIS.
Popis architektury projektu clarin-dspace a dodrziavanie standardov a doporuceni RDA-DFT/FAIR/OAIS.
[ "Jozef Mišutka" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
The QT21 Combined Machine Translation System for English to Latvian
Společný kombinovaný systém projektu QT21 pro překlad z angličtiny do lotyšštiny
This paper describes the joint submis- sion of the QT21 projects for the English → Latvian translation task of the EMNLP 2017 Second Conference on Ma- chine Translation (WMT 2017). The sub- mission is a system combination which combines seven different statistical ma- chine translation systems provided by the different groups. The systems are combined using either RWTH’s system combination approach, or USFD’s consensus-based system- selection approach. The final submission shows an improvement of 0.5 B LEU compared to the best single system on newstest2017.
Článek popisuje společný systém projektu QT21 pro překlad z angličtiny do lotyšštiny na druhé konferenci WMT 2017.
[ "Jan-Thorsten Peter", "Hermann Ney", "Ondřej Bojar", "Ngoc-Quam Pham", "Jan Niehues", "Alex Waibel", "Franck Burlot", "François Yvon", "Marcis Pinnis", "Valters Sics", "Joost Bastings", "Miguel Rios", "Wilker Aziz", "Phil Williams", "Frédéric Blain", "Lucia Specia" ]
https://www.semanticscholar.org/paper/a3c5faa8257f784ffdf4b516f52a40e85bccac27/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Constrained Deep Answer Sentence Selection
Omezené Hluboký Výběr Věty
We propose Constrained Deep Neural Network (CDNN) a deep neural model for answer sentence selection in the context of Question Answering (QA) systems. To produce the best predictions, CDNN combines neural reasoning with a kind of symbolic constraint. It integrates pattern matching technique into sentence vector learning. When trained using enough samples, CDNN outperforms the other best models for sentence selection. We show how the use of other sources of training can enhance the performance of CDNN. In a well-studied dataset for answer sentence selection, our model improves the state-of-the-art significantly
Navrhujeme propuštěnou hlubokou neuronovou síť (CDNN) hluboký neuronový model pro výběr věty odpovědi v kontextu systémů QA (Question Answering). Pro vytvoření nejlepších předpovědí kombinuje CDNN neurální uvažování s určitým symbolickým omezením. Integruje techniku přizpůsobení vzoru do vektoru vět učení se. Při výcviku za použití dostatečných vzorků převyšuje CDNN ostatní nejlepší modely pro výběr vět. Ukazujeme, jak se využívají další zdroje Školení může zvýšit výkon CDNN. V dobře studovaném datovém souboru pro výběr věty odpovědi náš model výrazně zlepšuje nejmodernější technologii.
[ "Ahmad Aghaebrahimian" ]
https://www.semanticscholar.org/paper/da5086b1cc9f9a02a5bbf8cefc15f40a1bf0a177/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Hybrid Deep Open-Domain Question Answering
Hybridní Hluboký Otevřená Doména Otázka Odpovědět
In this short paper, we report a progressing work on a hybrid deep neural system for answering open-domain factoid and non-factoid questions. This system supplements knowledge graph based Question Answering with free-texts searching techniques to address the sparsity issues in knowledge graphs. We justify the efficiency of the proposed system based on the results of a pilot experiment. We also describe the settings of an on-going project in the context of this system.
V tomto krátkém časopise oznamujeme postupnou práci na hybridním hlubokém nervovém systému, který odpovídá na faktoidní a nepravotní otázky otevřené oblasti. Tento systém doplňuje znalostní graf založený na zodpovězení dotazu s vyhledávacími technikami pro volné texty, které řeší problematiku sparsity ve znalostních grafech. Ospravedlňujeme účinnost navrhovaného systému na základě výsledků pilotního experimentu. Rovněž popisujeme nastavení probíhajícího projektu v kontextu tohoto systému.
[ "Ahmad Aghaebrahimian" ]
https://www.semanticscholar.org/paper/890318caeda7d76dfbb5f0b1702f64d17b1ea345/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Quora Question Answer Dataset
Quora Otázka Odpovědět Datových sad
We report on a progressing work for compiling Quora question answer dataset. Quora dataset is composed of the questions which are posed in Quora question answering site. It is the only dataset which provides answers in sentence level and word level at the same time. Moreover, the questions in the dataset are authentic which is much more realistic for question answering systems. We test the performance of a state-of-the-art question answering system on the dataset and compare it with human performance to establish an upper bound for the dataset
Uvádíme zprávu o pokročilejších úkolech při sestavování datových sad otázka odpovědět Quora. Datová sada Quora se skládá z otázek, které jsou položeny na stránkách Quora answering site. Je to jediná datová sada, která poskytuje odpovědi současně na úrovni vět a slov. Otázky v datové sadě jsou navíc autentické, což je mnohem realističtější pro systémy odpovědí na otázky. Testujeme výkonnost nejmodernějšího záznamníku otázek na datové množině a porovnáváme ji s lidskou výkonností, abychom vytvořili horní hranici datové sady.
[ "Ahmad Aghaebrahimian" ]
https://www.semanticscholar.org/paper/fe31a7dc3c744ef35e70810d253a03c98eda741d/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
CzeDLex and other Lexicons of Discourse Connectives
CzeDLex a další slovníky textových konektorů
The presentation offers an overview and comparison of present-day electronic lexicons of discourse connectives in different languages, with a special regard on the recently developed lexicon of Czech connectives – CzeDLex.
Prezentace nabízí přehled a srovnání současných elektronických lexikonů textových konektorů v různých jazycích se zvláštním zřetelem na nový lexikon českých konektorů - CzeDLex.
[ "Lucie Poláková" ]
https://www.semanticscholar.org/paper/1ae29c65f7cc846e639859635687d0276be2ea00/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Signalling Implicit Relations: A PDTB - RST Comparison
Signály implicitních vzahů: Srovnání PDTB a RST
Describing implicit phenomena in discourse is known to be a problematic task, both from the theoretical and from the empirical perspective. The present article contributes to this topic by a novel comparative analysis of two prominent annotation approaches to discourse (coherence) relations that were carried out on the same texts. We compare the annotation of implicit relations in the Penn Discourse Treebank 2.0, i.e. discourse relations not signalled by an explicit discourse connective, to the recently released analysis of signals of rhetorical relations in the RST Signalling Corpus. Our data transformation allows for a simultaneous depiction and detailed study of these two resources.
Článek předkládá podrobné srovnání dvou proslulých přístupů k analýze imlicitních diskurzních vztahů, Penn Discourse Treebanku a databáze Rhetorical Structure Theory Signalling Corpus.
[ "Lucie Poláková", "Jiří Mírovský", "Pavlína Synková" ]
https://www.semanticscholar.org/paper/e6f07af23f136893fc2a31a38e31c8a3a0c6cc56/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Deep Learning in MT / NMT
Hluboké učení v (neuronovém) strojovém překladu
Neural machine translation (NMT) has become a widely-adopted approach to machine translation in the past few years. In our tutorial, we will start with the introduction to the basics of the deep learning methods used in NMT, such as recurrent neural networks and their advanced variants (GRU or LSTM networks), or the algorithms for their optimization. We introduce the NMT-specific models, such as the attention mechanism, and describe the methods used for decoding the target sentences, including model ensembling and beam search. We will go through the recent advancements in the field and discuss their impact on the state-of-the-art methods used in this year's WMT competition (http://www.statmt.org/wmt17/).
Neuronový strojový překlad (NMT) se stal v posledních letech široce využívaným přístupem ke strojovému překladu. V našem tutoriálu začneme úvodem do základů metod hloubkového učení používaných v NMT jako jsou rekurentní neuronové sítě a jejich pokročilé varianty (GRU nebo LSTM sítě) nebo algoritmy pro jejich optimalizaci. Představujeme modely specifické pro NMT, jako je mechanizmus pozornosti, a popisujeme metody použité pro dekódování cílových vět, včetně ensemblování modelů a paprskového prohledávání. Projdeme nedávné pokroky v této oblasti a budeme diskutovat o jejich dopadu na nejmodernější metody používané na letošní soutěži WMT (http://www.statmt.org/wmt17/).
[ "Jindřich Helcl", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/a21d11a1932ae0a66329fb1b0c8371cce6059e13/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
CUNI System for the WMT17 Multimodal Translation Task
Systém Univerzity Karlovy pro úlohu multimodálního strojového překladu na WMT17
In this paper, we describe our submissions to the WMT17 Multimodal Translation Task. For Task 1 (multimodal translation), our best scoring system is a purely textual neural translation of the source image caption to the target language. The main feature of the system is the use of additional data that was acquired by selecting similar sentences from parallel corpora and by data synthesis with back-translation. For Task 2 (cross-lingual image captioning), our best submitted system generates an English caption which is then translated by the best system used in Task 1. We also present negative results, which are based on ideas that we believe have potential of making improvements, but did not prove to be useful in our particular setup.
V tomto článku popisujeme naše příspěvky do multimodální překladové úlohy na WMT17. Pro úlohu 1 (multimodální překlad) byl nejlepším systémem čistě textový neuronový překlad titulku zdrojového obrázku do cílového jazyka. Hlavním rysem našeho systému je využití dalších dat získaných výběrem podobných vět z paralelních korpusů a syntézou dat zpětným překladem. Pro úlohu 2 (vícejazyčné generování popisu obrázků) náš nejlepší systém generuje anglický popis obrázku, který je poté přeložen podle nejlepšího systému používaného v úloze č. 1. Také předkládáme negativní výsledky, které jsou založeny na myšlenkách, o kterých se domníváme, že mají potenciál překlad zlepšit, ale v našem konkrétním uspořádání neprokázaly býti prospěšnými.
[ "Jindřich Helcl", "Jindřich Libovický" ]
https://www.semanticscholar.org/paper/35c5b84fa47e5b6cee201f831507f62482ccebcc/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Neural Monkey: An Open-source Tool for Sequence Learning
Neural Monkey: Open source nástroj pro sekvenční učení
In this paper, we announce development of Neural Monkey — an open-source neural machine translation (NMT) and general sequence-to-sequence learning system built over TensorFlow machine learning library. The system provides a high-level API with support for fast prototyping of complex architectures with multiple sequence encoders and decoders. These models’ overall architecture is specified in easy-to-read configuration files. The long-term goal of Neural Monkey project is to create and maintain a growing collection of implementations of recently proposed components or methods, and therefore it is designed to be easily extensible. The trained models can be deployed either for batch data processing or as a web service. In the presented paper, we describe the design of the system and introduce the reader to running experiments using Neural Monkey.
V tomto článku představujeme vývoj Neural Monkey — open source nástroj pro neuronový strojový překlad a sekvenční učení obecně, který je postavený na knihovně TensorFlow. Náš nástroj poskytuje svým uživatelům vysokoúrovňové rozhraní, které umožňuje rychlé vytváření prototypů komplexních modelů s několika enkodéry a dekodéry. Architektura modelů se vytváří pomocí snadno čitelných konfiguračních souborů. Dlouhodobým cílem Neural Monkey je vytvořit a udržovat kolekci moderních metod pro sekvenční učení. Tomu odpovídá i modulární snadno rozšiřitelný design. Natrénované modely je možné použít pro dávkové zpracování dat nebo spustit jako webovou službu. V předkládaném článku popisujeme základní design nástroje a postup, jak spustit trénování jednoduchého překladače.
[ "Jindřich Helcl", "Jindřich Libovický" ]
https://www.semanticscholar.org/paper/0267e8dab6669a6c10be2e29bea44ea32c41e990/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Neural Monkey lab
Lab o Neural Monkey
The aim of this lab is to familiarize the users with the Neural Monkey toolkit for sequence learning experiments.
Účel tohoto labu je seznámit uživatele s toolkitem Neural Monkey, využívaným pro experimenty se sekvenčním učením.
[ "Jindřich Helcl", "Dušan Variš" ]
https://www.semanticscholar.org/paper/f681b2b8746d8f702a0dc85f9c5d8b8979022bd2/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Attention Strategies for Multi-Source Sequence-to-Sequence Learning
Strategie používání attention modelů ve vícezdrojovém sekvenčním učení
Modeling attention in neural multi-source sequence-to-sequence learning remains a relatively unexplored area, despite its usefulness in tasks that incorporate multiple source languages or modalities. We propose two novel approaches to combine the outputs of attention mechanisms over each source sequence, flat and hierarchical. We compare the proposed methods with existing techniques and present results of systematic evaluation of those methods on the WMT16 Multimodal Translation and Automatic Post-editing tasks. We show the proposed methods achieve competitive results on both tasks.
Attention modely ve vícezdrojovém neuronovém sekvenčním učení zůstávají poměrně neprobádanou oblastí, a to navzdory jeho užitečnosti v úkolech, které které využívají více zdrojových jazyků či modalit. Navrhujeme dvě nové strategie jak kombinovat výstupy attentiion modelu z různých vstupů, plochou a hierarchickou. Navrhované metody porovnáváme se stávajícími a výsledky vyhodnocujeme na datech pro multimodální překlad a automatické post-editování překladu z WMT16. Navrhované metody dosažení konkurenceschopných výsledků na obou úlohách.
[ "Jindřich Libovický", "Jindřich Helcl" ]
https://www.semanticscholar.org/paper/e7dc0d6669d411849010a32ee75b49cf4c853f55/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Querying Multiword Expressions Annotation with NoSke
Vyhledávání v datech s anotací víceslovných jednotek pomocí CQL
This paper demonstrates one of the possible ways on how to represent and query corpora with multiword expression (MWE) annotation. We exploit the multilingual corpus of 18 languages created under the PARSEME project with verbal multiword expression (VMWE) annotation. VMWEs include categories such as idioms, light verb constructions, verb-particle constructions, inherently reflexive verbs, and others. The corpus was mainly used for the purposes of training predictive models, yet not much linguistic research was conducted based on this data. We discuss how to allow linguists to query for MWEs in a simple user interface using the Corpus Query Language (CQL) within the NoSke corpus management and concordance system. Despite its limited abilities to represent challenging cases such as discontinuous, coordinated or embedded VMWEs, CQL can be sufficient to make basic analysis of the MWE-annotated data in corpus-based studies.
Článek popisuje jeden z možných způsobů, jak zobrazit a prohledávat data s anotací víceslovných jednotek. Využíváme mnohojazyčný korpus PARSEME s anotací verbálních víceslovných jednotek v 18 jazycích. Anotované jednotky zahrnují různé typy, jako např. idiomy, konstrukce s lehkými slovesy, inherentně reflexivní slovesa nebo konstrukce se slovesem a částicí. Korpus byl dosud využíván zejména pro trénování prediktivních modelů, ale nikoli k lingvistickému výzkumu per se. Článek nabízí způsob, jak data zpřístupnit lingvistům skrze jednoduché vyhledávací prostředí a jazyk Corpus Query Language (CQL) známy například z často užívané platformy NoSke. I přes omezené možnosti k zachycení komplexních jevů jakými jsou nespojité, koordinované nebo vnořené víceslovné predikáty, CQL může postačovat k základním vyhledávkám víceslovných jednotek pro korpusově založený výzkum.
[ "Natalia Klyueva", "Anna Vernerová", "Behrang QasemiZadeh" ]
https://www.semanticscholar.org/paper/ea12b18fe8c55c04bebc4d8eed8f86fcb906d430/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Neural Networks for Multi-Word Expression Detection
Umělé neuronové sítě pro rozpoznávání víceslovných výrazů
In this paper we describe the MUMULS system that participated to the 2017 shared task on automatic identification of verbal multiword expressions (VMWEs). The MUMULS system was implemented using a supervised approach based on recurrent neural networks using the open source library TensorFlow. The model was trained on a data set containing annotated VMWEs as well as morphological and syntactic information. The MUMULS system performed the identification of VMWEs in 15 languages, it was one of few systems that could categorize VMWEs type in nearly all languages.
Tento článek popisuje systém MUMULS, který se účastnil 2017 Shared Task on Automatic Identification of Verbal Multiword expressions (VMWEs). Systém MUMULS byl implementován přístupem učení s učitelem pomocí rekurentních neuronových sítí v open-source knihovně TensorFlow. Model byl trénován na poskytnutých datech s VMWEs a také na morfologických a syntaktických anotacích. MUMULS provádí identifikaci VMWEs v patnácti jazycích, byl to jeden z mála systémů který dokázal kategorizovat VMWEs ve skoro všech jazycích.
[ "Natalia Klyueva", "Antoine Doucet", "Milan Straka" ]
https://www.semanticscholar.org/paper/10c864539f27e591e2821ce856d361aea6d63d38/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Denotation Extraction for Interactive Learning in Dialogue Systems
Extrakce denotací pro interaktivní učení v dialogových systémech
This paper presents a novel task using real user data obtained in human-machine conversation. The task concerns with denotation extraction from answer hints collected interactively in a dialogue. The task is motivated by the need for large amounts of training data for question answering dialogue system development, where the data is often expensive and hard to collect. Being able to collect denotation interactively and directly from users, one could improve, for example, natural understanding components online and ease the collection of the training data. This paper also presents introductory results of evaluation of several denotation extraction models including attention-based neural network approaches.
Tento článek představuje novou úlohu na využití dat od uživatelů dialogových systémů v konverzacích člověk-stroj. Tato úloha se zaměřuje na sběr denotací extrahováním z přirozených vět získaných v průběhu dialogu. Motivace spočívá v potřebě velkého monžství trénovacích dat pro vývoj Q&A dialogových systémů, přičemž získání těchto dat je obvykle těžké a nákladné. Získávání denotací při interakcích s uživateli například umožňuje online vylepšování komponent pro porozumění přirozenému jazyku a zjednodušit sběr trénovacích dat. Tento článek také prezentuje výsledky evaluace několika přístupů k extrakci denotací zahrnující modely založené na neuronových sítích s attention architekturou.
[ "Miroslav Vodolán", "Filip Jurčíček" ]
https://www.semanticscholar.org/paper/9162fa21f83b1bdbd302dffcbdce7a1c6d54dc3f/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Hybrid Dialog State Tracker with ASR Features
Hybridní State Tracker pro Dialog s ASR Featurami
This paper presents a hybrid dialog state tracker enhanced by trainable Spoken Language Understanding (SLU) for slot-filling dialog systems. Our architecture is inspired by previously proposed neural-network-based belief-tracking systems. In addition, we extended some parts of our modular architecture with differentiable rules to allow end-to-end training. We hypothesize that these rules allow our tracker to generalize better than pure machine-learning based systems. For evaluation, we used the Dialog State Tracking Challenge (DSTC) 2 dataset - a popular belief tracking testbed with dialogs from restaurant information system. To our knowledge, our hybrid tracker sets a new state-of-the-art result in three out of four categories within the DSTC2.
Tento článek představuje hybridní dialog state tracker, rozšířený o trénovatelnou jednotku zpracování přirozeného jazyka (SLU). Naše architektura je inspirována belief trackery založenými na neuronových sítích. Tento přístup navíc rozšiřujeme o derivovatelná prvidla, která umožní end-to-end trénink. Tato pravidla umožní našemu trackeru lépe generalizovat v porovnání s trackery založenými pouze na strojovém učení. Pro evaluaci používáme Dialog State Tracking Challenge (DSTC) 2 - populární dataset využívaný pro srovnání výkonnosti belief trackerů. Podle informací, které máme, náš tracker dosahuje state-of-the-art výsledků ve třech ze čtyř kategorií datasetu DSTC2.
[ "Miroslav Vodolán", "Rudolf Kadlec", "Jan Kleindienst" ]
https://www.semanticscholar.org/paper/049ecf2717e8346cdc03bba55cd3e76dd09b1e18/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
Udapi: Universal API for Universal Dependencies
Udapi: univerzální API pro Universal Dependencies
Udapi is an open-source framework providing an application programming interface (API) for processing Universal Dependencies data. Udapi is available in Python, Perl and Java. It is suitable both for full-fledged applications and fast prototyping: visualization of dependency trees, format conversions, querying, editing and transformations, validity tests, dependency parsing, evaluation etc.
Udapi je open-source framework poskytující APO pro zpracovávání dat z projektu Universal Dependencies. Implementace Udapi je dostupné pro programovací jazyky: Python, Perl a Java. Udapi je vhodné jak pro plnohodnotné aplikace, tak pro rychlé vyváření prototypů: vizualizace stromů, konverze formátu, dotazování, editace, transformace, testy validity, závislostní parsing, vyhodnocování, atd.
[ "Martin Popel", "Zdeněk Žabokrtský", "Martin Vojtek" ]
https://www.semanticscholar.org/paper/90b4e9cef5fcf6194b666240c80fb0cb1719b787/
null
null
null
null
null
null
null
null
null
null
null
null
en
2017
CUNI System for WMT17 Automatic Post-Editing Task
Systém UK pro úlohu automatické posteditace strojového překladu WTM17
Following upon the last year's CUNI system for automatic post-editing of machine translation output, we focus on exploiting the potential of sequence-to-sequence neural models for this task. In this system description paper, we compare several encoder-decoder architectures on a smaller-scale models and present the system we submitted to WMT 2017 Automatic Post-Editing shared task based on this preliminary comparison. We also show how simple inclusion of synthetic data can improve the overall performance as measured by an automatic evaluation metric. Lastly, we list few example outputs generated by our post-editing system.
V návaznosti na loňský systém pro automatickou post-editaci strojového překladu Karlovy Univerzity se soustředíme na využití potenciálu sequence-to-sequence neuronových modelů pro danou úlohu. V článku nejprve porovnáváme několik architektur typu enkodér-dekodér na modelech menšího měřítka a představujeme systém, který byl vybrán na základě těchto předběžných výsledků a odeslán na WMT 2017 Automatic Post-Editing shared task. V článku také ukazujeme jak jednoduchá inkluze umělých dat dokáže vylepšit úspěšnost modelu na základě automatických evaluačních metrik. V závěru uvádíme několik příkladů výstupů vygenerových našim post-editačním systémem.
[ "Dušan Variš", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/6dc33afdf5a90e44e3fa20086fc270ddc88681a2/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Edinburgh’s Statistical Machine Translation Systems for WMT16
Edinburské statistické překladové systémy pro WMT16
This paper describes the University of Edinburgh’s phrase-based and syntax-based submissions to the shared translation tasks of the ACL 2016 First Conference on Machine Translation (WMT16). We submitted five phrase-based and five syntaxbased systems for the news task, plus one phrase-based system for the biomedical task.
Článek popisuje frázové a syntaktické překladové systémy Univerzity v Edinburku účastnící se soutěžní překladového úlohy WMT16.
[ "Phil Williams", "Rico Sennrich", "Maria Nadejde", "Matthias Huck", "Barry Haddow", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/8d4cb41d377569249895dde9b809cdb02cd6861e/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2016
42th Year of the Olympiad in Czech Language
42. ročník Olympiády v českém jazyce
Analysis of competition tasks of the Olympiad in Czech language, evaluation of students' language skills.
Analýza soutěžních úkolů z Olympiády v českém jazyce, hodnocení jazykových dovedností studentů.
[ "Kateřina Rysová", "Karel Oliva" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
EVALD 1.0
EVALD 1.0
Evaluator of Discourse (EVALD 1.0) is a software that serves for automatic evaluation of surface coherence (cohesion) in Czech texts written by native speakers of Czech. Software evaluates the level of the given text from the perspective of its surface coherence (i.e. it takes into consideration, e.g., the frequency and diversity of connectives, richness of vocabulary etc.).
Počítačový nástroj Evaluátor diskurzu (EVALD 1.0) slouží k automatickému hodnocení povrchové koherence (koheze) textů v češtině psaných rodilými mluvčími češtiny. Software tedy hodnotí (známkuje) úroveň předloženého textu z hlediska jeho povrchové výstavby (zohledňuje např. frekvenci a rozmanitost spojovacích prostředků, bohatost slovní zásoby apod.).
[ "Kateřina Rysová", "Jiří Mírovský", "Michal Novák", "Magdaléna Rysová" ]
https://www.semanticscholar.org/paper/61fb1a8bdc94d6b2559b8883744deed9d6e99545/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
EVALD 1.0 for Foreigners
EVALD 1.0 pro cizince
Evaluator of Discourse for Foreigners (EVALD 1.0 for Foreigners) is a software that serves for automatic evaluation of surface coherence (cohesion) in Czech texts written by non-native speakers of Czech. Software evaluates the level of the given text from the perspective of its surface coherence (i.e. it takes into consideration, e.g., the frequency and diversity of connectives, richness of vocabulary etc.).
Počítačový nástroj Evaluátor diskurzu pro cizince (EVALD 1.0 pro cizince) slouží k automatickému hodnocení povrchové koherence (koheze) textů v češtině psaných nerodilými mluvčími češtiny. Software hodnotí úroveň předloženého textu z hlediska jeho povrchové výstavby (zohledňuje např. frekvenci a rozmanitost užitých spojovacích prostředků, bohatost slovní zásoby apod.).
[ "Kateřina Rysová", "Jiří Mírovský", "Michal Novák", "Magdaléna Rysová" ]
https://www.semanticscholar.org/paper/61fb1a8bdc94d6b2559b8883744deed9d6e99545/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2016
Coreference and Ellipsis in Prague Dependency Treebank
Koreference a elipsa v Pražském závislostním korpusu
The paper focuses on which role is given to elided sentence participants in coreference chains in the text, i.e. whether (and to which degree) the participants that are present only implicitly in the surface layer are involved in relations of textual and grammatical coreference. Generally, the paper introduces he methods how it is possible to examine the interplays of different language phenomena in corpus data of the Prague Dependency Treebank containing multilayer annotation.
Článek popisuje, jakou roli hrají elidované větné participanty v koreferenčních řetězcích v textu, tj. zda a do jaké míry se participanty, které jsou v povrchové větné struktuře přítomné pouze implicitně, zapojují do vztahů textové a gramatické koreference. Článek zároveň představuje metody, jimiž je možné zkoumat vzájemné vztahy mezi různými jazykovými jevy, a to na datech Pražského závislostního korpusu, který obsahuje anotaci lingvistických jevů na několika jazykových rovinách.
[ "Kateřina Rysová", "Magdaléna Rysová" ]
https://www.semanticscholar.org/paper/334db7cbffc537bb4d9b25e0dd128e6434691af6/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Automatic evaluation of surface coherence in L2 texts in Czech
Automatická evaluace povrchové koherence v textech nerodilých mluvčích češtiny
We introduce possibilities of automatic evaluation of surface text coherence (cohesion) in texts written by learners of Czech during certified exams for non-native speakers. On the basis of a corpus analysis, we focus on finding and describing relevant distinctive features for automatic detection of A1–C1 levels (established by CEFR – the Common European Framework of Reference for Languages) in terms of surface text coherence. The CEFR levels are evaluated by human assessors and we try to reach this assessment automatically by using several discourse features like frequency and diversity of discourse connectives, density of discourse relations etc. We present experiments with various features using two machine learning algorithms. Our results of automatic evaluation of CEFR coherence/cohesion marks (compared to human assessment) achieved 73.2% success rate for the detection of A1–C1 levels and 74.9% for the detection of A2–B2 levels.
V článku jsou představeny možnosti automatické evaluace povrchové koherence (koheze) textů psaných nerodilými mluvčími češtiny během certifikovaných zkoušek. Na základě korpusové analýzy jsou vyhledávány a popisovány relevantní rozlišovací rysy (týkající se povrchové koherence textu) pro automatickou detekci úrovní textů nerodilých mluvčích A1–C1 (úrovně jsou ustanoveny Společným evropským referenčním rámcem pro jazyky). Úrovně A1–C1 byly hodnoceny nejprve lidmi (anotátory) – poté byly dělány strojové experimenty s cílem přiblížit se lidskému hodnocení automaticky, a to sledováním vybraných textových rysů, např. frekvence a různorodosti diskurzních konektorů nebo hustoty diskurzních vztahů v daném textu ap. V článku jsou představeny experimenty sledující vždy různé textové rysy při použití dvou algoritmů strojového učení. Úspěšnost automatického měření povrchové koherence (koheze) textu podle Společného evropského referenčního rámce pro jazyky je 73,2 % pro rozpoznávání úrovní A1–C1 a 74,9 % pro rozpoznávání úrovní A2–B2.
[ "Kateřina Rysová", "Magdaléna Rysová", "Jiří Mírovský" ]
https://www.semanticscholar.org/paper/26e229385bec7dae3cea8f3664a3c780c1b0ad5d/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Logical vs. Natural Language Conjunctions in Czech: A Comparative Study
Jazykové spojky v češtině: komparativní studie
This paper studies the relationship between conjunctions in a natural language (Czech) and their logical counterparts. It shows that the process of transformation of a natural language expression into its logical representation is not straightforward. The paper concentrates on the most frequently used logical conjunctions, AND and OR, and it analyzes the natural language phenomena which influence their transformation into logical conjunction and disjunction. The phenomena discussed in the paper are temporal sequence, expressions describing mutual relationship and the consequences of using plural.
Názvy logických spojek jsou často motivovány názvy spojek z přirozeného jazyka. V případě, že však chceme výroku v přirozeném jazyce (konkrétně češtině) přiřadit jednoznačně logické schéma, zjišt’ujeme, že to není možné dělat automaticky. V tomto článku se zabýváme tím, kdy je možné české spojky „a“ a „nebo“ zapsat pomocí konjunkce a disjunkce a kde je nutné zvolit jinou interpretaci – s ohledem na problematiku množného čísla a vyjádření vztahů.
[ "Katrin Přikrylová", "Vladislav Kuboň", "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/c65cf88ffa6cd7316d8b3c3bb57c09480f97ce51/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
The Role of Conjunctions in Adjective Polarity Analysis in Czech
Úloha spojek v při určování polarity českých adjektiv
Adjectives very often determine the polarity of an utterance. This paper studies the role of conjunctions in the analysis of adjective polarity. The study is being performed on Czech language on the basis of an existing algorithm for English. The algorithm has been modified in order to reflect the differences between the two typologically different languages. The results of the original and modified algorithm are being compared and discussed. The paper also contains a thorough discussion of exceptions and special cases supported by a number of examples from a large corpus of Czech.
Adjektiva velmi často určují polaritu věty. Tento článek studuje roli spojek v analýze polarity adjektiv. Celá studie je prováděna pro češtinu modifikací algoritmu vyvinutého původně pro angličtinu. Modifikace berou v úvahu typologické rozdíly mezi oběma jazyky. Výsledky dosažené pro oba jazyky jsou porovnány a doplněny rozsáhlým rozborem výjimečných případů.
[ "Katrin Přikrylová", "Vladislav Kuboň", "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/3e17fc68761629f33b52f657ca281ee998ec6ae3/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
If You Even Don't Have a Bit of Bible: Learning Delexicalized POS Taggers
Když nemáte ani kus Bible: učení delexikalizovaných POS taggerů
Part-of-speech (POS) tagging is sometimes considered an almost solved problem in NLP. Standard supervised approaches often reach accuracy above 95% if sufficiently large hand-labeled training data are available (typically several hundred thousand tokens or more). However, we still believe that it makes sense to study semi-supervised and unsupervised approaches.
Značkování slovními druhy (POS tagging) se v počítačovém zpracování přirozeného jazyka někdy považuje za téměř vyřešený problém. Standardní řízené přístupy často dosahují úspěšnosti přes 95 %, pokud je k dispozici dostatek ručně anotovaných trénovacích dat (typicky několik set tisíc tokenů nebo více). My si nicméně myslíme, že je stále užitečné studovat polořízené a neřízené přístupy.
[ "Zhiwei Yu", "David Mareček", "Zdeněk Žabokrtský", "Daniel Zeman" ]
https://www.semanticscholar.org/paper/f3f5e44aa87b564a6ba986d2f3dc46622338bee4/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Deeper Machine Translation and Evaluation for German
Hlubší strojový překlad a evaluace pro němčinu
This paper describes a hybrid Machine Translation (MT) system built for translating from English to German in the domain of technical documentation. The system is based on three different MT engines (phrase-based SMT, RBMT, neural) that are joined by a selection mechanism that uses deep linguistic features within a machine learning process. It also presents a detailed source-driven manual error analysis we have performed using a dedicated “test suite” that contains selected examples of relevant phenomena. While automatic scores show huge differences between the engines, the overall average number or errors they (do not) make is very similar for all systems. However, the detailed error breakdown shows that the systems behave very differently concerning the various phenomena.
Tento článek popisuje systém hybridního strojového překladu (MT), který byl vytvořen pro překlad z angličtiny do němčiny v oblasti technické dokumentace. Systém je založen na třech různých systémech MT (frázový, pravidlový a neuronový), které jsou spojeny výběrovým mechanismem, který používá hluboké jazykové rysy v procesu strojového učení. Součástí je také podrobná manuální analýza chyb, kterou jsme provedli pomocí specializované "zkušební sady", která obsahuje vybrané příklady relevantních jevů. Zatímco automatické výsledky ukazují obrovské rozdíly mezi systémy, celkový průměrný počet chyb, které (ne) dělají, je pro všechny systémy velmi podobný. Podrobné rozdělení chyb však ukazuje, že systémy se chovají velmi odlišně, pokud jde o různé jevy.
[ "Eleftherios Avramidis", "Vivien Macketanz", "Aljoscha Burchardt", "Jindřich Helcl", "Hans Uszkoreit" ]
https://www.semanticscholar.org/paper/c3c804b5cb59f6faa21c0c327ad68bc040423a14/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
VPS-GradeUp: Graded Decisions on Usage Patterns
VPS-GradeUp: Škálované posouzení vzorů užití
We present VPS-GradeUp ― a set of 11,400 graded human decisions on usage patterns of 29 English lexical verbs from the Pattern Dictionary of English Verbs by Patrick Hanks. The annotation contains, for each verb lemma, a batch of 50 concordances with the given lemma as KWIC, and for each of these concordances we provide a graded human decision on how well the individual PDEV patterns for this particular lemma illustrate the given concordance, indicated on a 7-point Likert scale for each PDEV pattern. With our annotation, we were pursuing a pilot investigation of the foundations of human clustering and disambiguation decisions with respect to usage patterns of verbs in context. The data set is publicly available at http://hdl.handle.net/11234/1-1585.
Článek představuje VPS-GradeUp, sadu 11 400 manuálně anotovaných ohodnocení vzorů užití 29 anglických sloves z Pattern Dictionary of English Verbs Patricka Hankse.
[ "Vít Baisa", "Silvie Cinková", "Ema Krejčová", "Anna Vernerová" ]
https://www.semanticscholar.org/paper/916a51b4269359f728584011ca9f7e75654f4fea/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Enriching a Valency Lexicon by Deverbative Nouns
Rozšíření valenčního slovníku o deverbativní substantiva
In this paper, we present an attempt to automatically identify Czech deverbative nouns using several methods that use large corpora as well as existing lexical resources. The motivation for the task is to extend a verbal valency (i.e., predicate-argument) lexicon by adding nouns that share the valency properties with the base verb, assuming their properties can be derived (even if not trivially) from the underlying verb by deterministic grammatical rules. At the same time, even in inflective languages, not all deverbatives are simply created from their underlying base verb by regular lexical derivation processes. We have thus developed hybrid techniques that use both large parallel corpora and several standard lexical resources. Thanks to the use of parallel corpora, the resulting sets contain also synonyms, which the lexical derivation rules cannot get. For evaluation, we have manually created a gold dataset of deverbative nouns linked to 100 frequent Czech verbs since no such dataset was initially available for Czech.
Článek popisuje metodu automatické identifikace deverbativ na základě slovníků a manuálně a automaticky anotovaných korpusů. Metoda je evaluována na novém testovacím korpusu, který je rovněž zveřejněn.
[ "Eva Fučíková", "Jan Hajič", "Zdeňka Urešová" ]
https://www.semanticscholar.org/paper/6c1c4b767c15655f1c085bd1ba87872f0597077f/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Joint search in a bilingual valency lexicon and an annotated corpus
Vyhledávání ve dvojjazyčném slovníku a anotovaném korpusu
In this paper and the associated system demo, we present an advanced search system that allows to perform a joint search over a (bilingual) valency lexicon and a correspondingly annotated linked parallel corpus. This search tool has been developed on the basis of the Prague Czech-English Dependency Treebank, but its ideas are applicable in principle to any bilingual parallel corpus that is annotated for ependencies and valency (i.e., predicate-argument structure), and where verbs are linked to appropriate entries in an associated valency lexicon. Our online search tool consolidates more search interfaces into one, providing expanded structured search capability and a more efficient advanced way to search, allowing users to search for verb pairs, verbal argument pairs, their surface realization as recorded in the lexicon, or for their surface form actually appearing in the linked parallel corpus. The search system is currently under development, and is replacing our current search tool available at http://lindat.mff.cuni.cz/services/CzEngVallex, which could search the lexicon but the queries cannot take advantage of the underlying corpus nor use the additional surface form information from the lexicon(s). The system is available as open source.
Článek popisuje vyhledávací systém pro spojené vyhledávání ve dvojjazyčném slovníku CzEngVallex a v paralelním anglicko-českém korpusu PCEDT.
[ "Eva Fučíková", "Jan Hajič", "Zdeňka Urešová" ]
https://www.semanticscholar.org/paper/23af8bf92ce4993f22bd5754286e0214d1e7cb46/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Word2vec Based System for Recognizing Partial Textual Entailment
Word2vec systém pro rozpoznávání Partial Textual Entailment
Recognizing textual entailment is typically considered as a binary decision task – whether a text T entails a hypothesis H. Thus, in case of a negative answer, it is not possible to express that H is “almost entailed” by T. Partial textual entailment provides one possible approach to this issue. This paper presents an attempt to use word2vec model for recognizing partial (faceted) textual entailment. The proposed approach does not rely on language dependent NLP tools and other linguistic resources, therefore it can be easily implemented in different language environments where word2vec models are available.
Recognizing textual entailment is typically considered as a binary decision task – whether a text T entails a hypothesis H. Thus, in case of a negative answer, it is not possible to express that H is almost entailed by T. Partial textual entailment provides one possible approach to this issue. This paper presents an attempt to use word2vec model for recognizing partial (faceted) textual entailment. The proposed approach does not rely on language dependent NLP tools and other linguistic resources, therefore it can be easily implemented in different language environments where word2vec models are available.
[ "Martin Víta", "Vincent Kríž" ]
https://www.semanticscholar.org/paper/ed5dff0498bda2f8ae012e0a64b5dc27d0b14d87/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2016
Sentiment analysis of a written text
Analýza sentimentu textu
Sentiment analysis or automatic extraction of subjective opinions from a written texts is currently one of the major topics in NLP. In this talk, I will introduce the state-of-the-art methods and approaches to sentiment analysis on Czech data.
Sentiment analysis neboli automatická extrakce subjektivních názorů z textu je v současnosti jedním z hlavních témat počítačového zpracovávání přirozeného jazyka. V příspěvku bude shrnut současný stav bádání v této oblasti na českých datech. Obšírněji představím základní metody vytěžování emocí z textu, datové zdroje pro češtinu i specifika českých dat a přiblížím konkrétní aplikace a use cases např. z oblasti internetového marketingu, monitoringu sociálních sítí apod.
[ "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/f3c0bb77c87769e84915cb80f64894ba09f39460/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2016
Automatic sentiment or Where does the World go
Automatický sentiment aneb Kam spěje svět
Ignite talk on automatic detection of emotions, basic approaches, data resources overview.
Ignite talk na téma automatické detekce emocí, základní přístupy a datové zdroje.
[ "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/1f6fa7ece20b2e02d81e6c02038b8555a0c8b2a2/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2016
How do we count emotions in languages
Jak spočítat emoce v jazyce
If well trained, computers can count emotions in texts. Is that so? And what about irony?
Pokud mají dostatek trénovacích dat, umějí stroje rozpoznat emoce v psaném textu. Je tomu skutečně tak? A jak se řeší ironie?
[ "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/8bfdd9898c1e1f1f821ca58933b0133935322d89/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Machine Learning Approaches to Sentiment Analysis
Strojové učení a postojová analýza
Current computational linguistics witnesses a massive increase of interest in sentiment analysis, as it is a powerful means of public opinion mining. However, is it even possible to distinguish between good and evil using statistical methods? How would you treat irony, idioms, innovative vulgarisms and other inherent elements of natural language? Are emotions language-independent? I will share my thoughts and opinions towards these issues, describing the state-of-the art approaches employed in sentiment analysis.
Současná počítačová lingvistika zažívá masový zájem o postojovou analýzu jakožto nástroj zjišťování veřejného mínění. Je ale možné rozlišit dobro a zlo pomocí statistických metod? Jak zacházet s ironií, idiomy či vulgarismy? Jsou emoce jazykově nezávislé? V příspěvku zodpovím tyto otázky a popíšu současné přístupy k postojové analýze.
[ "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/217b5a97b071bd8ddbd67160fee431e263f6f18b/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2016
Computer processing of emotions in text
Počítačové zpracování emocí v textu
The volume of text data is increasing constantly, the rise of Web 2.0 brings lots of data created by the Internet users themselves. How do we categorize these data automatically with respect to emotions they express? In this talk, I will introduce basic means of emotional language and methods we use to process it automatically.
Objem textových dat stále roste, rozvoj Webu 2.0 přináší množství textů generovaných samotnými uživateli Internetu. Jejich příspěvky nezřídka obsahují subjektivní názory, emoce, hodnocení… K čemu a jak můžeme tato data použít? Je možné emoce v textu spolehlivě strojově třídit? V přednášce si představíme základní prostředky emocionálního vyjadřování v češtině na všech rovinách popisu jazyka a popíšeme metody automatického zpracovávání emocionálně laděných textů i jejich praktické aplikace z oblasti internetového marketingu, monitoringu sociálních sítí apod.
[ "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/5e86d7c72ff85d35f1029f17522ad1256fa1a7e0/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2016
Stance modality and possibilities of automatic extraction of emotions from written text
Postojová modalita a možnosti automatické extrakce emocionality z psaného textu
In this talk, I will introduce basic means of Czech emotional language on all the levels of linguistic description. I will focus on both grammar and lexicon and semantics and pragmatics of evaluative utterances.
V přednášce si představíme základní prostředky emocionálního vyjadřování v češtině na všech rovinách lingvistického popisu: zaměříme se na stránku lexikální (slovník emocionálních výrazů pro češtinu, evaluativní idiomy, vulgarismy), gramatickou (význam jednotlivých slovních druhů, typické syntaktické vzorce hodnotících vět) a především sémantickou a pragmatickou (ironie, sarkasmus). Popíšeme také základní metody automatického zpracovávání emocionálně laděných textů a jejich praktické aplikace.
[ "Kateřina Veselovská" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2016
Artificial Intelligence with Opinion: Objectivity in computer linguistics
Umělá inteligence s názorem: objektivita v počítačové lingvistice
Can artificial intelligence have an opinion? And how do we distinguish between subjective and objective stances using computer linguistic? This talk introduces basic approaches to these problems.
Může mít umělá inteligence vlastní úsudek? A jak dokážeme počítačově odhalit objektivní či subjektivní názor? Přednáška shrne základní přístupy k objektivitě a subjektivitě v textu.
[ "Kateřina Veselovská" ]
https://www.semanticscholar.org/paper/7b8a202ea4136ae76bf679ef058954600a079c3d/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Using Universal Dependencies for Cross-linguistic Comparison of Syntactic Features in Evaluative Structures
Využití univerzálních závislostí pro mezijazykové srovnání syntaktických rysů evaluativních struktur
This paper is to present a unique project of parallel treebank annotation and a case study of its practical use. We describe the project of Universal Dependencies, explain the methodological approach, illustrate the basic principles of parallel annotation and introduce the universal annotation scheme. Also, we provide a contrastive syntactic analysis of evaluative sentences in typologically different languages.
Příspěvek představuje projekt Universal Dependencie a konkrétní případ jeho využití v oblasti sentiment analysis.
[ "Kateřina Veselovská" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
cs
2016
Expressing of emotions in Czech
Vyjadřování emocí v češtině a možnosti extrakce emocionality z psaného textu
In this talk, we introduce the basic means of expressing emotions in Czech on all levels of the linguistic description.
V přednášce si představíme základní prostředky emocionálního vyjadřování v češtině na všech rovinách lingvistického popisu: zaměříme se na stránku lexikální (slovník emocionálních výrazů pro češtinu, evaluativní idiomy, vulgarismy), gramatickou (význam jednotlivých slovních druhů, typické syntaktické vzorce hodnotících vět) a především sémantickou a pragmatickou (ironie, sarkasmus). Popíšeme také základní metody automatického zpracovávání emocionálně laděných textů a jejich praktické aplikace.
[ "Kateřina Veselovská" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Combining Hand-Crafted Rules and Machine Learning Methods in Opinion Target Identification
Kombinace pravidel a metod strojového učení pro identifikaci cílů hodnocení
This poster shows how we created the rules for opinion target identification based on evaluative structures automatically found in Prague Dependency Treebank and combined them with machine learning methods, namely linear-chain conditional random fields.
Poster představuje tvorbu pravidel pro určení cílů hodnocení založenou na evaluativních strukturách automaticky vyhledaných v Pražském závislostním korpusu. Tato pravidla jsme zkombinovali s metodami strojového učení, konkrétně s conditional random fields.
[ "Kateřina Veselovská", "Aleš Tamchyna" ]
https://www.semanticscholar.org/paper/194221a6c7caa33091f6c5bca71f69b03f29bd50/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Czech-English Manual Word Alignment
Česko-anglické ruční zarovnání slov
Czech-English Manual Word Alignment. Czech-English Manual Word Alignment.
Česko-anglické ruční zarovnání slov. Česko-anglické ruční zarovnání slov.
[ "David Mareček" ]
https://www.semanticscholar.org/paper/b734805fff4fe8c6fe3fe60ab33e5aef743accfd/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Delexicalized and Minimally Supervised Parsing on Universal Dependencies
Delexikalizovaný a minimálně supervizovaný parsing na Universal Dependencies
In this paper, we compare delexicalized transfer and minimally supervised parsing techniques on 32 different languages from Universal Dependencies treebank collection. The minimal supervision is in adding handcrafted universal grammatical rules for POS tags. The rules are incorporated into the unsupervised dependency parser in forms of external prior probabilities. We also experiment with learning this probabilities from other treebanks. The average attachment score of our parser is slightly lower then the delexicalized transfer parser, however, it performs better for languages from less resourced language families (non-Indo-European) and is therefore suitable for those, for which the treebanks often do not exist.
V tomto článku budeme porovnávat delexicalized přenos a minimálně pod dohledem rozebrat techniky na 32 různých jazyků od Universal závislostí korpusu sbírky. Minimální dohled při přidávání pouštět univerzální gramatická pravidla pro POS tagy. Pravidla jsou začleněny do nekontrolované závislost parseru ve formách externích dřívějších pravděpodobnostech. Také jsme experimentovat s učit se toto pravděpodobností z jiných stromových korpusů. Průměrná připevnění skóre našeho parseru je o něco nižší než v delexicalized přenosu analyzátor, nicméně, to funguje lépe pro jazyky z méně zdroji jazykových rodin (non-Indo-Evropan), a proto je vhodný pro ty, pro které stromových korpusů často neexistují.
[ "David Mareček" ]
https://www.semanticscholar.org/paper/d2f972e71d269c2aad1d850afef27dfe208b96ce/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Merged bilingual trees based on Universal Dependencies in Machine Translation
Dvojjazyčné stromy založené na Universal Dependencies ve strojovém překladu
In this paper, we present our new experimental system of merging dependency representations of two parallel sentences into one dependency tree. All the inner nodes in dependency tree represent source-target pairs of words, the extra words are in form of leaf nodes. We use Universal Dependencies annotation style, in which the function words, whose usage often differs between languages, are annotated as leaves. The parallel treebank is parsed in minimally supervised way. Unaligned words are there automatically pushed to leaves. We present a simple translation system trained on such merged trees and evaluate it in WMT 2016 English-to-Czech and Czech-to-English translation task. Even though the model is so far very simple and no language model and word-reordering model were used, the Czech-to-English variant reached similar BLEU score as another established tree-based system.
V tomto příspěvku představujeme náš nový experimentální systém sloučení závislost reprezentace dvou rovnoběžných vět do jednoho strom závislostí. Všechny vnitřní uzly v závislosti stromu představují zdroj-cílové páry slovy, další slova jsou ve formě koncové uzly. Používáme Univerzální Závislosti anotace styl, ve kterém funkční slova, jejichž použití se často liší mezi jazyky, jsou zaznamenány jako listy. Paralelní korpus je analyzován v minimálně dohlíží způsobem. Nezarovnaný slova jsou zde automaticky tlačil na povrch listů. Představujeme jednoduchý systém překladu vyškoleného na takových sloučených stromech a vyhodnocovat jej WMT 2016 anglicko-to-český a česko-to-anglický překlad úloh. I přesto, že model je doposud velmi jednoduché a byl používán žádný jazykový model a model word-li řazení varianta Český k angličtině dosáhl podobného Bleu skóre jako další zavedeného systému stromu bázi.
[ "David Mareček" ]
https://www.semanticscholar.org/paper/491833087ba57eaabad5c762927e57aeaa4baaaf/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Twelve Years of Unsupervised Dependency Parsing
Dvanáct let neřízeného závislostního parsingu
In the last 12 years, there has been a big progress in the field of unsupervised dependency parsing. Different approaches however sometimes differ in motivation and definition of the problem. Some of them allow using resources that are forbidden by others, since they are treated as a kind of supervision. The goal of this paper is to define all the variants of unsupervised dependency parsing problem and show their motivation, progress, and the best results. We also discuss the usefulness of the unsupervised parsing generally, both for the formal linguistics and for the applications.
V posledních 12 letech došlo k velkým pokroku v oblasti bez dozoru závislostní parsování. Různé přístupy však někdy liší v motivaci a ve vymezení problému. Některé z nich umožňují využití zdrojů, které jsou zakázány jinými, neboť se s nimi zachází jako druh dohledu. Cílem tohoto příspěvku je definovat všechny varianty bez dozoru závislost rozebrat problém a ukázat jejich motivace, pokrok, a nejlepší výsledky. Také jsme diskutovali o užitečnosti látky jako součásti bez dozoru analýze obecně, a to jak pro formální lingvistiky a pro aplikace.
[ "David Mareček" ]
https://www.semanticscholar.org/paper/50843155d395d3f93151652220ac618c9efa90e6/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Deltacorpus
Deltacorpus
Texts in 107 languages from the W2C corpus (http://hdl.handle.net/11858/00-097C-0000-0022-6133-9), first 1,000,000 tokens per language, tagged by the delexicalized tagger described in Yu et al. (2016, LREC, Portorož, Slovenia).
Texty ve 107 jazycích z korpusu W2C (http://hdl.handle.net/11858/00-097C-0000-0022-6133-9), první 1000000 tokenů pro každý jazyk, označkované delexikalizovaným taggerem popsaným v Yu et al. (2016, LREC, Portorož, Slovenia).
[ "David Mareček", "Zhiwei Yu", "Daniel Zeman", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/5d607e524716c7dd5869871cc5625d41596662ae/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Gibbs Sampling Segmentation of Parallel Dependency Trees for Tree-Based Machine Translation
Segmentace paralelních závislostních stromů pomocí Gibbsova samplingu pro stromový strojový překlad
We present a work in progress aimed at extracting translation pairs of source and target dependency treelets to be used in a dependency-based machine translation system. We introduce a novel unsupervised method for parallel tree segmentation based on Gibbs sampling. Using the data from a Czech-English parallel treebank, we show that the procedure converges to a dictionary containing reasonably sized treelets; in some cases, the segmentation seems to have interesting linguistic interpretations.
Představujeme nedokončené zaměřená na těžbu překladové páry zdrojové a cílové závislost treelets které mají být použity v systému strojového překládání závislostí na bázi. Představíme nový voze metodu pro paralelní segmentaci stromu na základě vzorků Gibbs. S použitím dat z české, anglické paralelním korpusu, ukážeme, že postup konverguje ke slovníku, který obsahuje poměrně velké treelets; V některých případech se zdá, že segmentace mít zajímavé lingvistické výklady.
[ "David Mareček", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/f73c67cc4818cd83e8a0f2f04f64e8b322c88fde/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
A Context-aware Natural Language Generation Dataset for Dialogue Systems
Kontextová datová sada pro generování jazyka v dialogových systémech
We present a novel dataset for natural language generation (NLG) in spoken dialogue systems which includes preceding context (user utterance) along with each system response to be generated, i.e., each pair of source meaning representation and target natural language paraphrase. We expect this to allow an NLG system to adapt (entrain) to the user’s way of speaking, thus creating more natural and potentially more successful responses. The dataset has been collected using crowdsourcing, with several stages to obtain natural user utterances and corresponding relevant, natural, and contextually bound system responses. The dataset is available for download under the Creative Commons 4.0 BY-SA license.
Představujeme novou datovou sadu pro generování jazyka v hlasových dialogových systémech, která spolu s každou odpovědí systému k vygenerování (pár zdrojová sémantická reprezentace – cílová věta v přirozeném jazyce) uvádí i předcházející kontext (uživatelský dotaz). Očekáváme, že tento kontext dovolí generátorům jazyka adaptovat se na způsob vyjadřování uživatele a tím docílit přirozenějších a potenciálně úspěšnějších odpovědí. Datová sada byla vytvořena za pomoci crowdsourcingu v několika fázích, aby bylo možno získat přirozené uživatelské dotazy a odpovídající přirozené, relevantní a kontextově zapojené odpovědi systému. Datová sada je dostupná online pod otevřenou licencí Creative Commons 4.0 BY-SA.
[ "Ondřej Dušek", "Filip Jurčíček" ]
https://www.semanticscholar.org/paper/5abc7c678ea98fea5d4912cd94979c4220617c73/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
A Context-aware Natural Language Generator for Dialogue Systems
Kontextový generátor přirozeného jazyka pro dialogové systémy
We present a novel natural language generation system for spoken dialogue systems capable of entraining (adapting) to users' way of speaking, providing contextually appropriate responses. The generator is based on recurrent neural networks and the sequence-to-sequence approach. It is fully trainable from data which include preceding context along with responses to be generated. We show that the context-aware generator yields significant improvements over the baseline in both automatic metrics and a human pairwise preference test.
Představujeme nový generátor přirozeného jazyka pro hlasové dialogové systémy, který je schopný přizpůsobit se způsobu, jakým mluví uživatel, a poskytovat odpovědi přiměřené kontextu dialogu. Generátor je založen na neuronových sítích a přístupu sequence-to-sequence. Je plně trénovatelný z dat, která spolu s trénovacími výstupy generátoru obsahují také předchozí kontext. Ukazujeme, že kontextový generátor přináší signifikantní zlepšení oproti základnímu generátoru, a to jak z pohledu automatických metrik, tak v preferenčním testu lidského hodnocení.
[ "Ondřej Dušek", "Filip Jurčíček" ]
https://www.semanticscholar.org/paper/9fb07b2dbb488b8890debf62c1689d5700e12383/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Alex Context NLG Dataset
Kontextová datová sada pro generování jazyka
A dataset intended for fully trainable natural language generation (NLG) systems in task-oriented spoken dialogue systems (SDS), covering the English public transport information domain. It includes preceding context (user utterance) along with each data instance (pair of source meaning representation and target natural language paraphrase to be generated). Taking the form of the previous user utterance into account for generating the system response allows NLG systems trained on this dataset to entrain (adapt) to the preceding utterance, i.e., reuse wording and syntactic structure. This should presumably improve the perceived naturalness of the output, and may even lead to a higher task success rate. Crowdsourcing has been used to obtain natural context user utterances as well as natural system responses to be generated.
Datová sada pro plně trénovatelné generátory jazyka v hlasových dialogových systémech, která pokrývá doménu anglických informací o veřejné dopravě. Spolu s každou datovou položkou (pár zdrojové reprezentace významu a věty v přirozeném jazyce jako cíl generování) obsahuje i předcházející kontext (uživatelův dotaz, který má systém egnerovanou větou zodpovědět). Zohlednění formy předchozího dotazu pro generování umožní generátorům natrénovaným na této datové sadě adaptovat se na předchozí dotazy, tj. používat stejné shodné výrazy a syntaktické konstrukce jako uživatel dialogového systému. Předpokládáme, vygenerované věty tak budou vnímány jako přirozenější, což může vést i k úspěšnějším dialogům. Pro získání přirozených uživatelských dotazů i odpovědí systému byla použita metoda crowdsourcingu.
[ "Ondřej Dušek", "Filip Jurčíček" ]
https://www.semanticscholar.org/paper/f27d79ef8d48fde31b35fd4ab84eb679949149ed/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Sequence-to-Sequence Generation for Spoken Dialogue via Deep Syntax Trees and Strings
Generování hloubkově syntaktických stromů a řetězců pro dialogové systémy s použitím metody sequence-to-sequence
We present a natural language generator based on the sequence-to-sequence approach that can be trained to produce natural language strings as well as deep syntax dependency trees from input dialogue acts, and we use it to directly compare two-step generation with separate sentence planning and surface realization stages to a joint, one-step approach. We were able to train both setups successfully using very little training data. The joint setup offers better performance, surpassing state-of-the-art with regards to n-gram-based scores while providing more relevant outputs.
Prezentujeme generátor přirozeného jazyka založený na metodě sequence-to-sequence, který lze natrénovat tak, aby z dialogových aktů produkoval buď věty v přirozeném jazyce, nebo hloubkově syntaktické závislostní stromy. Na generátoru potom přímo porovnáváme dvoufázové generování, které používá oddělené větné plánování a povrchovou realizaci, s jednofázovým přístupem. Obě nastavení generátoru jsme byli schopni natrénovat s použitím velmi malého množství trénovacích dat. Jednofázové generování dosahuje lepších výsledků; překonává nejlepší předchozí výsledek podle n-gramových automatických metrik a zároveň nabízí relevantnější výstupy.
[ "Ondřej Dušek", "Filip Jurčíček" ]
https://www.semanticscholar.org/paper/663dc7723e1748f1e28ac18f3e1a4c1212f2385e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Diatheses of Complex Predicates in Czech
Diateze komplexních predikátů v češtině
Diatheses represent relations between different surface syntactic structures of verbs. They are conditioned by changes in morphological meaning of a verb and they are associated with specific changes in its valency structure. In this contribution, I discussed changes in the surface syntactic expressions of Czech complex predicates.
Diateze představují vztahy mezi různými povrchověsyntaktickými strukturami sloves. Jsou podmíněny změnou morfologické charakteristiky slovesného rodu a jsou spojeny se změnami v povrchověsyntaktickém vyjádření valenčních doplnění sloves. V příspěvku jsme se zaměřili na specifické změny, kterým podléhají valenční doplnění českých funkčních sloves.
[ "Václava Kettnerová" ]
https://www.semanticscholar.org/paper/472dd385114e533dd412f8999f91448053f6bb92/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Distribution of Valency Complementations in Light Verb Constructions
Distribuce valenčních doplnění v konstrukcích s funkčními slovesy
The contribution discusses the distribution of ACTors in light verb constructions in Czech.
Příspěvek se zabývá distribucí Konatele v konstrukcích komplexních predikátů s kategoriálními slovesy.
[ "Václava Kettnerová" ]
https://www.semanticscholar.org/paper/481398b63e6f53aadff1e00c03ea61388da1e266/
null
null
null
null
null
null
null
null
null
null
null
null
cs
2016
Syntactic Structure of Complex Predicates and its Description in the Valency Lexicon
Syntaktická struktura komplexních predikátů a její popis ve valenčním slovníku
This chapter describes the syntactic structure of Czech complex predicates with light verbs in active voice. Four types of Czech complex predicates are distinguished. Their rules based description and representation in the valency lexicon are proposed.
Kapitola popisuje syntaktickou strukturu českých komplexních predikátů s funkčním slovesem v aktivu. Vymezuje čtyři typy komplexních predikátů. Navrhuje pravidlový popis daných jazykových jevů a jejich zachycení ve slovníku.
[ "Václava Kettnerová" ]
null
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Lexicographic Description of Complex Predicates in Czech: Between Lexicon and Grammar
Lexikografický popis komplexních predikátů v češtině: Mezi lexikonem a gramatikou
In this contribution, we propose a lexicographic representation of Czech complex predicates. Special attention is devoted to the selection of collocations of function verbs and predicative nouns and to the annotation of their syntactic structure.
V příspěvku navrhujeme lexikografickou reprezentaci komplexivních predikátů v češtině. Zvláštní oddíl je věnován výběru kolokací funkčních sloves a predikativních jmen a anotaci jejich syntaktické struktury.
[ "Václava Kettnerová", "Petra Barančíková", "Markéta Lopatková" ]
https://www.semanticscholar.org/paper/c3fb84d83caababfd8607699c75d2254ed72cfe7/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Distribution of Valency Complements in Czech Complex Predicates: Between Verb and Noun
Distribuce valenčních doplnění v českých složených predikátech: Mezi slovesem a substantivem
In this paper, we focus on Czech complex predicates formed by a light verb and a predicative noun expressed as the direct object. Although Czech – as an inflectional language encoding syntactic relations via morphological cases – provides an excellent opportunity to study the distribution of valency complements in the syntactic structure with complex predicates, this distribution has not been described so far. On the basis of a manual analysis of the richly annotated data from the Prague Dependency Treebank, we thus formulate principles governing this distribution. In an automatic experiment, we verify these principles on well-formed syntactic structures from the Prague Dependency Treebank and the Prague Czech-English Dependency Treebank with very satisfactory results: the distribution of 97% of valency complements in the surface structure is governed by the proposed principles. These results corroborate that the surface structure formation of complex predicates is a regular process.
V tomto článku se zaměřujeme na české složené predikáty, které jsou tvořené lehkým slovesem a predikativním jménem vyjádřeným jako přímý objekt. Ačkoli čeština -- jakožto inflekční jazyk -- poskytuje výbornou příležitost pro studium distribuce valenčních doplnění složených predikátů, tato distribuce dosud nebyla poplsána. Na základě manuální analýzy bohatě anotovaných dat PDT formulujeme poučky řídicí tuto distribuci. V automatickém experimentu tyto poučky ověřujeme na korektních syntaktických strukturách PDT a PCEDT s velmi uspokojivými výsledky: distribuce 97% valenčních doplnění se řídí navrženými poučkami. Tyto výsledky dokládají, že vytváření povrchové struktury složených predikátů je pravidelný proces.
[ "Václava Kettnerová", "Eduard Bejček" ]
https://www.semanticscholar.org/paper/e0ad5a5c7ad7ebef4d4372a61757125e94f039fa/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Towards Comparability of Linguistic Graph Banks for Semantic Parsing
Ke srovnatelnosti lingvistických závislostních grafů pro sémantickou analýzu
Increased interest in natural language ‘understanding’ has brought into the focus of much current work a variety of techniques often described as ‘semantic parsing’.
Zvýšený zájem o ‚porozumění‘ přirozeným jazykům způsobil, že do centra pozornosti současného výzkumu se dostávají různé postupy, často popisované jako ‚sémantická analýza‘.
[ "Stephan Oepen", "Marco Kuhlmann", "Yusuke Miyao", "Daniel Zeman", "Silvie Cinková", "Dan Flickinger", "Jan Hajič", "Angelina Ivanova", "Zdeňka Urešová" ]
https://www.semanticscholar.org/paper/518684a4aacf9cd8af5a13ac8b9a4a4706282f96/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Neural Networks for Featureless Named Entity Recognition in Czech
Umělé neuronové sítě pro rozpoznávání pojmenovaných entit v češtině
We present a completely featureless, language agnostic named entity recognition system. Following recent advances in artificial neural network research, the recognizer employs parametric rectified linear units (PReLU), word embeddings and character-level embeddings based on gated linear units (GRU). Without any feature engineering, only with surface forms, lemmas and tags as input, the network achieves excellent results in Czech NER and surpasses the current state of the art of previously published Czech NER systems, which use manually designed rule-based orthographic classification features. Furthermore, the neural network achieves robust results even when only surface forms are available as input. In addition, the proposed neural network can use the manually designed rule-based orthographic classification features and in such combination, it exceeds the current state of the art by a wide margin.
Předkládáme systém pro rozpoznávání pojmenovaných entit, který je jazykově nezávislý a nepotřebuje klasifikační rysy pro strojové učení. Systém využívá současných výsledků v oblasti umělých neuronových sítí, jako jsou parametric rectified linear units (PReLU), embeddingy slov a embeddingy charakterů ve slovech založené na gated linear units (GRU). Systém nepotřebuje vyhledávání vhodné sady klasifikačních rysů (feature engineering) a pouze s využitích povrchových forem, lemmat a slovních druhů na vstupu dosahuje vynikajících výsledků v rozpoznávání pojmenovaných entit v češtině a překonává stávající výsledky dříve publikovaných prací, které využívají ručně vytvořené klasifikační rysy založené na ortografické podobnosti slov. Navíc tato síť podává robustní výkon i v případě, kdy jsou na vstupu pouze povrchové formy. Síť dovede využít navíc i kombinaci ručně vytvořených klasifikačních rysů a v tom případě překonává stávající výsledky s markantním rozdílem.
[ "Jana Straková", "Milan Straka", "Jan Hajič" ]
https://www.semanticscholar.org/paper/5d3ab49a338d24c1495d958f6e49a7aeb5645134/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Creating Hybrid Dependency Parsers for Syntax-Based MT
Hybridní závislostní analyzátory pro strojový překlad založený na syntaxi
Dependency parsers are almost ubiquitously evaluated on their accuracy scores, these scores say nothing of the complexity and usefulness of the resulting structures. As dependency parses are basic structures in which other systems are built upon, it would seem more reasonable to judge these parsers down the NLP pipeline. In this chapter, we will discuss how different forms and different hybrid combinations of dependency parses effect the overall output of Syntax-Based machine translation both through automatic and manual evaluation. We show results from a variety of individual parsers, including dependency and constituent parsers, and describe multiple ensemble parsing techniques with their overall effect on the Machine Translation system. We show that parsers’ UAS scores are more correlated to the NIST evaluation metric than to the BLEU Metric, however we see increases in both metrics. To truly see the effect of hybrid dependency parsers on machine translation, we will describe and evaluate a combined resource we have released, that contains gold standard dependency trees along with gold standard translations.
Závislostní parsery jsou obvykle vyhodnocovány pomocí podílu správně zavěšených uzlů, toto skóre ale neříká nic o užitečnosti daného parseru pro koncové aplikace, v nichž slouží jako komponenta. V této kapitole se zabývá tím, jak rozdílné parsery a jejich různé kombinace přispívají ke kvalitě strojového překladu založeného na syntaxi. Překládáme výsledky pro několik základních typů parserů a pro různé metody jejich kombinace. Ukazujeme korelace se standardními metrikami pro vyhodnocování kvality strojového překladu.
[ "Nathan David Green", "Zdeněk Žabokrtský" ]
https://www.semanticscholar.org/paper/34c8f6b9ccde6282e945fc96fe3c5f938d553c36/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
SubGram: Extending Skip-gram Word Representation with Substrings
SubGram: Rozšíření Skip-gramové slovní reprezentace o podřetězce
Skip-gram (word2vec) is a recent method for creating vector representations of words (“distributed word representations”) using a neural network. The representation gained popularity in various areas of natural language processing, because it seems to capture syntactic and semantic information about words without any explicit supervision in this respect. We propose SubGram, a refinement of the Skip-gram model to consider also the word structure during the training process, achieving large gains on the Skip-gram original test set.
Představujeme SubGram, rozšíření Skip-gram modelu, které používá podřetězce slov během trénování reprezentace slov.
[ "Tom Kocmi", "Ondřej Bojar" ]
https://www.semanticscholar.org/paper/733966ee11461746822b10f0a0af2c0aa04572ca/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
Universal Dependencies for Russian: A New Syntactic Dependencies Tagset
Universal Dependencies pro ruštinu: nová sada závislostních syntaktických značek
We present a new annotation schema of dependency syntax for Russian. The schema is based on the Universal Dependencies formalism. It has been applied to the SynTagRus corpus.
Představujeme nové anotační schéma závislostní syntaxe pro ruštinu, založené na formalismu Universal Dependencies a aplikované na korpus SynTagRus.
[ "Olga Lyashevskaya", "Kira Droganova", "Daniel Zeman", "Maria Alexeeva", "Tatiana Gavrilova", "Nina Mustafina", "Elena Shakurova" ]
https://www.semanticscholar.org/paper/45ccb5a00d0e6a7d9231fefc7f222b4cfa063c4a/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
CUNI at TRECVID 2015 Video Hyperlinking Task
CUNI na benchmarku TRECVID 2015 Video Hyperlinking Task
In this paper, we present our approach used in the TRECVID 2015 Video Hyperlinking Task. Our approach combines text-based similarity calculated on subtitles, visual similarity between keyframes calculated using Feature Signatures, and preference whether the query and retrieved answer come from the same TV series. All experiments were tuned and tested on about 2500 hours of BBC TV programmes. Our Baseline run exploits fixed-length segmentation, text-based retrieval of subtitles, and query expansion which utilizes metadata, context, in-formation about music and artist contained in the query segment and visual concepts. The Series run combines the Baseline run with weighting based on information whether the query and data segment come from the same TV series. The FS run combines the Baseline run with the similarity between query and data keyframes calculated using Feature Signatures. The FSSeriesRerank run is based on the FS run on which we applied reranking which, again, uses information about the TV series. The Series run significantly outperforms the FSSeriesRerank run. Both these runs are significantly inferior to our Baseline run in terms of all our reported measures. The FS run outperforms the Baseline run in terms of all measures but it is significantly better than the Baseline run only in terms of the MAP score. Our test results confirm that employment of visual similarity can improve video retrieval based on information contained in subtitles but information about TV series which was most helpful in our training experiments did not lead to further improvements.
V článku popisujeme náš postup aplikovaný v benchmarku TRECVID 2015 v úlohe Video Hyperlinking. Náš prístup kombinuje podobnosť textov vypočítanú z titulkov, vizuálnu podobnosť medzi zábermi vypočítanú pomocou Feature Signatures a informáciou o tom, či query segment a vyhľadaný segment pochádzajú z toho istého televízneho programu. Všetky experimenty boli odladené a otestované na kolekcii 2500 hodín televíznych programov poskytnutých BBC.
[ "Petra Galuščáková", "Michal Batko", "Martin Kruliš", "Jakub Lokoč", "David Novák", "Pavel Pecina" ]
https://www.semanticscholar.org/paper/21cac67c675ff6e187eec9013b8ce82e2525448e/
null
null
null
null
null
null
null
null
null
null
null
null
en
2016
SHAMUS: UFAL Search and Hyperlinking Multimedia System
SHAMUS: Systém na vyhľadávanie v multimédiách
In this paper, we describe SHAMUS, our system for an easy search and navigation in multimedia archives. The system consists of three components. The Search component provides a text-based search in a multimedia collection, the Anchoring component determines the most important segments of videos, and segments topically related to the anchoring ones are retrieved by the Hyperlinking component. In the paper, we describe each component of the system as well as the online demo interface which currently works with a collection of TED talks.
Článok popisuje systém SHAMUS určený na jednoduché vyhľadávanie a navigáciu v multimediálnych archívoch. Systém pozostáva z troch komponentov. Search poskytuje textové vyhľadávanie v mutlimediálnej kolekcii, Anchoring automaticky detekuje najvýznamnejšie segmenty videa a sémanticky súvisiace segmenty sú vyhľadané v komponente Hyperlinking. V článku popisujeme jednotlivé komponenty systému, ako aj online demo rozhranie, ktoré pracuje s kolekciou videí z konferencie TED.
[ "Petra Galuščáková", "Shadi Saleh", "Pavel Pecina" ]
https://www.semanticscholar.org/paper/f67cd5de50bf774745bb5ad68273049a4c91f2be/
null
null
null
null
null
null
null
null
null
null
null
null