Refine
Year of publication
Document Type
- Part of a Book (40)
- Conference Proceeding (24)
- Article (12)
- Other (2)
- Book (1)
Keywords
- Korpus <Linguistik> (72)
- Deutsch (23)
- Deutsches Referenzkorpus (DeReKo) (18)
- Institut für Deutsche Sprache <Mannheim> (11)
- Kontrastive Linguistik (11)
- Textkorpus (7)
- Korpusanalyseplattform (KorAP) (6)
- Rumänisch (6)
- Sprachdaten (6)
- Forschungsdaten (5)
Publicationstate
- Veröffentlichungsversion (46)
- Zweitveröffentlichung (17)
- Postprint (4)
Reviewstate
- Peer-Review (32)
- (Verlags)-Lektorat (29)
- Verlags-Lektorat (2)
Publisher
Für die spezifischen Bedürfnisse der Schreibbeobachtung wurde das Orthografische Kernkorpus (OKK) als virtuelles Korpus in DeReKo entwickelt. Mit derzeit rund 14 Mrd. Token deckt es den Schriftsprachgebrauch in den deutschsprachigen Ländern im Zeitraum von 1995 bis in die Gegenwart ab. Der Zugriff über die Korpusanalyseplattform KorAP erlaubt nicht nur die Nutzung verschiedener Annotationen, sondern über die API-Schnittstellen auch die Einbindung in diverse Auswertungsumgebungen wie RStudio über den RKorAPClient und macht es so für zahlreiche Analyse- und Visualisierungsmöglichkeiten zugänglich.
Der Beitrag betrachtet das Deutsche Referenzkorpus DeReKo in Bezug auf Strategien für seinen Ausbau, den Zugriff über die Korpusanalyseplattform KorAP und seine Einbettung in Forschungsinfrastrukturen und in die deutschsprachige und europäische Korpuslandschaft. Ausgehend von dieser Bestandsaufnahme werden Perspektiven zu seiner Weiterentwicklung aufgezeigt. Zu den Zukunftsvisionen gehören die Verteilung von Korpussressourcen und die Konstruktion multilingualer vergleichbarer Korpora anhand der Bestände der National- und Referenzkorpora, eine Plattform zur Abgabe und Aufbereitung von Sprachspenden als eine Anwendung von Citizen Science sowie eine Komponente zur automatischen Identifikation von übersetzten bzw. maschinenverfassten Texten.
We introduce DeReKoGram, a novel frequency dataset containing lemma and part-of-speech (POS) information for 1-, 2-, and 3-grams from the German Reference Corpus. The dataset contains information based on a corpus of 43.2 billion tokens and is divided into 16 parts based on 16 corpus folds. We describe how the dataset was created and structured. By evaluating the distribution over the 16 folds, we show that it is possible to work with a subset of the folds in many use cases (e.g., to save computational resources). In a case study, we investigate the growth of vocabulary (as well as the number of hapax legomena) as an increasing number of folds are included in the analysis. We cross-combine this with the various cleaning stages of the dataset. We also give some guidance in the form of Python, R, and Stata markdown scripts on how to work with the resource.
Einleitung
(2023)
Einleitung
(2023)
Das Deutsche Referenzkorpus DeReKo dient als eine empirische Grundlage für die germanistische Linguistik. In diesem Beitrag geben wir einen Überblick über Grundlagen und Neuigkeiten zu DeReKo und seine Verwendungsmöglichkeiten sowie einen Einblick in seine strategische Gesamtkonzeption, die zum Ziel hat, DeReKo trotz begrenzter Ressourcen für einerseits möglichst viele und andererseits auch für innovative und anspruchsvolle Anwendungen nutzbar zu machen. Insbesondere erläutern wir dabei Strategien zur Aufbereitung sehr großer Korpora mit notwendigerweise heuristischen Verfahren und Herausforderungen, die sich auf dem Weg zur linguistischen Erschließung solcher Korpora stellen.
The International Comparable Corpus (ICC) (Kirk/Čermáková 2017; Čermáková et al. 2021) is an open initiative which aims to improve the empirical basis for contrastive linguistics by compiling comparable corpora for many languages and making them as freely available as possible as well as providing tools with which they can easily be queried and analysed. In this contribution we present the first release of written language parts of the ICC which includes corpora for Chinese, Czech, English, German, Irish (partly), and Norwegian. Each of the released corpora contains 400k words distributed over 14 different text categories according to the ICC specifications. Our poster covers the design basics of the ICC, its TEI encoding, a demonstration of using the ICC via different query tools, and an outlook on future plans.
Similar to the European Reference Corpus EuReCo (Kupietz et al. 2020), ICC follows the approach of reusing existing linguistic resources wherever possible in order to cover as many languages as possible with realistic effort in as short a time as possible. In contrast to EuReCo, however, comparable corpus pairs are not defined dynamically in the usage phase, but the compositions of the corpora are fixed in the ICC design. The approaches are thus complementary in this respect. The design principles and composition of the ICC are based on those of the International Corpus of English (ICE) (Greenbaum (ed.) 1996), with the deviation that the ICC includes the additional text category blog post and excludes spoken legal texts (see Čermáková et al. 2021 for details). ICC’s fixed-design approach has the advantage that all single-language corpora in the ICC have the same composition with respect to the selected text types and that this guarantees that the selected broad spectrum of potential influencing variables for linguistic variation is always represented. The disadvantage, however, is that this can only be achieved for quite small corpora and that the generalisability of comparative findings based on the ICC corpora will often need to be checked on larger monolingual corpora or translation corpora (Čermáková/Ebeling/Oksefjell Ebeling forthcoming). Arguing that such issues with comparability and representativeness are inevitable, in one way or the other, and need to be dealt with, our poster will discuss and exemplify the text selections in more detail.
It is well known that the distribution of lexical and grammatical patterns is size- and register-sensitive (Biber 1986, and later publications). This fact alone presents a challenge to many corpus-oriented linguistic studies focusing on a single language. When it comes to cross-linguistic studies using corpora, the challenge becomes even greater due to the lack of high-quality multilingual corpora (Kupietz et al. 2020; Kupietz/Trawiński 2022), which are comparable with respect to the size and the register. That was the motivation for the creation of the European Reference Corpus EuReCo, an initiative started in 2013 at the Leibniz Institute for the German Language (IDS) together with several European partners (Kupietz et al. 2020). EuReCo is an emerging federated corpus, with large virtual comparable corpora across various languages and with an infrastructure supporting contrastive research. The core of the infrastructure is KorAP (Diewald et al. 2016), a scalable open-source platform supporting the analysis and visualisation of properties of texts annotated by multiple and potentially conflicting information layers, and supporting several corpus query languages. Until recently, EuReCo consisted of three monolingual subparts: the German Reference Corpus DeReKo (Kupietz et al. 2018), the Reference Corpus of Contemporary Romanian Language (Barbu Mititelu/Tufiş/Irimia 2018), and the Hungarian National Corpus (Váradi 2002). The goal of the present submission is twofold. On the one hand, it reports about the new component of EuReCo: a sample of the National Corpus of Polish (Przepiórkowski et al. 2010). On the other hand, it presents the results of a new pilot study using the newly extended EuReCo. This pilot study investigates selected Polish collocations involving light verbs and their prepositional / nominal complements (Fig. 1) and extends the collocation analyses of German, Romanian and Hungarian (Fig. 2) discussed in Kupietz/Trawiński (2022).
When comparing different tools in the field of natural language processing (NLP), the quality of their results usually has first priority. This is also true for tokenization. In the context of large and diverse corpora for linguistic research purposes, however, other criteria also play a role – not least sufficient speed to process the data in an acceptable amount of time. In this paper we evaluate several state of the art tokenization tools for German – including our own – with regard to theses criteria. We conclude that while not all tools are applicable in this setting, no compromises regarding quality need to be made.
Dieser Beitrag beschreibt die Motivation und Ziele hinter der Initiative Europäisches Referenzkorpus EuReCo. Ausgehend von den Desiderata, die sich aufgrund der Defizite verfügbarer Forschungsdaten wie monolinguale Korpora, Parallelkorpora und Vergleichskorpora für den Sprachvergleich ergeben, werden die bisherigen und die laufenden Arbeiten im Rahmen von EuReCo präsentiert und anhand vergleichender deutsch-rumänischer Kookkurrenzanalysen neue Perspektiven für kontrastive Korpuslinguistik, die die EuReCo-Initiative öffnet, skizziert.
Enabling appropriate access to linguistic research data, both for many researchers and for innovative research applications, is a challenging task. In this chapter, we describe how we address this challenge in the context of the German Reference Corpus DeReKo and the corpus analysis platform KorAP. The core of our approach, which is based on and tightly integrated into the CLARIN infrastructure, is to offer access at different levels. The graduated access levels make it possible to find a low-loss compromise between the possibilities opened up and the costs incurred by users and providers for each individual use case, so that, viewed over many applications, the ratio between effort and results achieved can be effectively optimized. We also report on experiences with the current state of this approach.
We present the use of count-based and predictive language models for exploring language use in the German Reference Corpus DeReKo. For collocation analysis along the syntagmatic axis we employ traditional association measures based on co-occurrence counts as well as predictive association measures derived from the output weights of skipgram word embeddings. For inspecting the semantic neighbourhood of words along the paradigmatic axis we visualize the high dimensional word embeddings in two dimensions using t-stochastic neighbourhood embeddings. Together, these visualizations provide a complementary, explorative approach to analysing very large corpora in addition to corpus querying. Moreover, we discuss count-based and predictive models w.r.t. scalability and maintainability in very large corpora.
In a recent article, Meylan and Griffiths (Meylan & Griffiths, 2021, henceforth, M&G) focus their attention on the significant methodological challenges that can arise when using large-scale linguistic corpora. To this end, M&G revisit a well-known result of Piantadosi, Tily, and Gibson (2011, henceforth, PT&G) who argue that average information content is a better predictor of word length than word frequency. We applaud M&G who conducted a very important study that should be read by any researcher interested in working with large-scale corpora. The fact that M&G mostly failed to find clear evidence in favor of PT&G's main finding motivated us to test PT&G's idea on a subset of the largest archive of German language texts designed for linguistic research, the German Reference Corpus consisting of ∼43 billion words. We only find very little support for the primary data point reported by PT&G.
When comparing different tools in the field of natural language processing (NLP), the quality of their results usually has first priority. This is also true for tokenization. In the context of large and diverse corpora for linguistic research purposes, however, other criteria also play a role – not least sufficient speed to process the data in an acceptable amount of time. In this paper we evaluate several state-ofthe-art tokenization tools for German – including our own – with regard to theses criteria. We conclude that while not all tools are applicable in this setting, no compromises regarding quality need to be made.
This paper reports on the efforts of twelve national teams in building the International Comparable Corpus (ICC; https://korpus.cz/icc) that will contain highly comparable datasets of spoken, written and electronic registers. The languages currently covered are Czech, Finnish, French, German, Irish, Italian, Norwegian, Polish, Slovak, Swedish and, more recently, Chinese, as well as English, which is considered to be the pivot language. The goal of the project is to provide much-needed data for contrastive corpus-based linguistics. The ICC corpus is committed to the idea of re-using existing multilingual resources as much as possible and the design is modelled, with various adjustments, on the International Corpus of English (ICE). As such, ICC will contain approximately the same balance of forty percent of written language and 60 percent of spoken language distributed across 27 different text types and contexts. A number of issues encountered by the project teams are discussed, ranging from copyright and data sustainability to technical advances in data distribution.
Die Korpusanalyseplattform KorAP ist von Grund auf sprachenunabhängig konzipiert. Dies gilt sowohl in Bezug auf die Lokalisierung der Benutzeroberfläche als auch hinsichtlich unterschiedlicher Anfragesprachen und der Unterstützung fremdsprachiger Korpora und ihren Annotationen. Diese Eigenschaften dienen im Rahmen der EuReCo Initiative aktuell besonders der Bereitstellung weiterer National- und Referenzkorpora neben DeReKo. EuReCo versucht, Kompetenzen beim Aufbau großer Korpora zu bündeln und durch die Verfügbarmachung vergleichbarer Korpora quantitative Sprachvergleichsforschung zu erleichtern. Hierzu bietet KorAP inzwischen, neben dem Zugang durch die Benutzeroberfläche, einen Web API Client an, der statistische Erhebungen, auch korpusübergreifend, vereinfacht.
The automatic recognition of idioms poses a challenging problem for NLP applications. Whereas native speakers can intuitively handle multiword expressions whose compositional meanings are hard to trace back to individual word semantics, there is still ample scope for improvement regarding computational approaches. We assume that idiomatic constructions can be characterized by gradual intensities of semantic non-compositionality, formal fixedness, and unusual usage context, and introduce a number of measures for these characteristics, comprising count-based and predictive collocation measures together with measures of context (un)similarity. We evaluate our approach on a manually labelled gold standard, derived from a corpus of German pop lyrics. To this end, we apply a Random Forest classifier to analyze the individual contribution of features for automatically detecting idioms, and study the trade-off between recall and precision. Finally, we evaluate the classifier on an independent dataset of idioms extracted from a list of Wikipedia idioms, achieving state-of-the art accuracy.