1,720,970 research outputs found
Automatic Identification of Generalizing Passages in German Fictional Texts using BERT with Monolingual and Multilingual Training Data
No full textExtended abstract submitted and accepted for the KONVENS 2022 Student Poster Session. The poster and code are available at https://github.com/tschomacker/generalizing-passages-identification-bert
Neither Telling nor Describing. Reflective Passages and Perceived Reflectiveness 1700-1945
Get PDFThe paper analyses within-fiction reflections in 250 years of literary history.
To this end, we formalised the concept of “reflective passage”, demonstrate
how our annotation categories are deduced from literary theory and derive
three subphenomena – comment, generalisation, and non-fictional speech
– that constitute literary reflection. A collaborative annotation serves (a) as
basis for the training of a neural classifier and (b) as dataset for a reception
experiment leading to the calculation of a ”reflection score”, a measurement for
the perceived reflectiveness of a textual passage. The classifier is applied to a
diachronic corpus of German-language literary fictions derived from the KOLIMO
corpus through extensive metadata enrichment and filtering. The results suggest
three boom periods of reflective passages: around 1755, 1835 and 1920 and show
effects of text length, canonisation status and authors’ sex
Simplification of German Narrative Documents with Longformer mBART
Get PDFTransformer-models have become the most prominent method for solving a multitude of natural language processing (NLP) tasks since their introduction in 2017. Natural Language Generation (NLG) is one of these problems. In this thesis we applied modern NLG-techniques to the problem of text simplification. Text simplification can be desribed as an intra-language translation task, where standard language is translated to simple language. Currently there are only a few German datasets available for Text Simplification. Even fewer with larger and aligned Documents, and not a single one with narrative
texts. With this paper we firstly explore to which degree modern NLG-techniques can be applied to our newly proposed German Narrative Text Simplifications dataset. We used Longformer Attention and a pre-trained mBART model. Our findings indicate that currently avaible approach are not able to solve the task properly. We conclude on a few directions for future research to adress this problem.Transformator-Modelle haben sich seit ihrer Einführung im Jahr 2017 zur Lösung einer Vielzahl von Aufgaben der natürlichen Sprachverarbeitung (NLP) durchgesetzt. Natural Language Generation (NLG) ist eines dieser Probleme. In dieser Arbeit haben wir moderne NLG-Techniken auf das Problem der Textvereinfachung angewendet. Textvereinfachung kann als eine innersprachliche Übersetzungsaufgabe beschrieben werden, bei der Standardsprache in einfache Sprache übersetzt wird. Derzeit gibt es nur wenige deutsche Datensätze zur Textvereinfachung. Noch weniger mit größeren und Dokumenten, die in beiden Versionen vorliegen. Und kein einziger mit narrativen Texten. In diesem Beitrag untersuchen wir zunächst, inwieweit sich moderne NLG-Techniken auf unseren neu eingeführten deutschen Datensatz für narrative Textvereinfachungen anwenden lassen. Wir haben Longformer Attention und ein vortrainiertes mBART-Modell verwendet. Unsere Ergebnisse zeigen, dass die derzeit verfügbaren Ansätze nicht in der Lage sind, die Aufgabe richtig zu lösen. Wir schließen mit einigen Hinweisen für die zukünftige Forschung, um dieses Problem zu adressieren
Reflection Tagger Single
No full textThis code and models (in `output/saved_models`) were downloaded from https://github.com/tschomacker/generalizing-passages-identification-bert/releases/tag/v0.3.0.
The code is licensed under MIT
Generalising Tagger Single
No full textThis code and models (in `output/saved_models`) were downloaded from https://github.com/tschomacker/generalizing-passages-identification-bert/releases/tag/v0.3.0.
The code is licensed under MIT
Automatic German Easy Language (Leichte Sprache) Simplification: Data, Requirements and Approaches
No full textWith the rise of the internet, it has become convenient and often free to access an abundance of texts. However, not all people, who have access, can really read and understand the texts. Despite the fact that, they speak the language that the text is written in. Most often this problem originates in the too complex nature of the texts. Text Simplification can help to overcome this barrier. In my dissertation, I want to specially focus on Leichte Sprache (German Easy Language). Which is a simplified version of German, that is tailored to the needs of people with cognitive disabilities
Application of Transformer-based Methods to Latin Text Analysis
Get PDFTextzusammenfassung ist ein etabliertes Problem im NLP-Bereich. Der rasch anwachsende Erfolg von deep learning Algorithmen führte zur Entwicklung des attention Mechanismus, welcher wiederum die Grundlage für die Transformer Architektur bildet. Die Transformer Architektur ist ein transfer learning Ansatz NLP Probleme zu lösen. BERT, ein pre-trained Transformer Modell, hat herausragende Ergebnis beim Lösen verschiedener NLP-Probleme erzielt. In dieser Abschlussarbeit wird BertSum, eine Erweiterung BERTs spezialisiert auf extrahierende Textzusammenfassung, auf neuronale Textzusammenfassung von
lateinischen und deutschen Texten angewandt. Dies stellt eine besondere Herausforderung dar, denn die Texte wurden zu einer Zeit verfasst, in der noch keine festgeschriebene Orthographie, Morphologie und Semantik existierte.Text summarization is an established problem in the field of NLP. The rapidly growing success of deep learning algorithms in solving NLP problems has led to the attention mechanism, which is the foundation for the Transformer architecture, a transfer learning approach for NLP tasks. BERT, a pre-trained Transformer model, has performed exceptionally well on various NLP tasks. This thesis applies BertSum, an enhancement of BERT specialized for extractive text summarization, to the neural text summarization of Latin and German texts. The distinctiveness of the chosen corpus is that it consists of medieval documents. This poses a challenge because the documents were written in a time where orthography, morphology,
and semantics were not well defined
- …
