Refine
Document Type
- Conference Proceeding (6)
- Part of a Book (3)
- Article (2)
Language
- English (11) (remove)
Has Fulltext
- yes (11)
Keywords
- Deutsch (8)
- Korpus <Linguistik> (5)
- Direkte Rede (3)
- Annotation (2)
- Automatic recognition of speech (2)
- Automatische Sprachanalyse (2)
- German (2)
- Indirect speech (2)
- Indirekte Rede (2)
- Prosa (2)
Publicationstate
- Veröffentlichungsversion (7)
- Zweitveröffentlichung (2)
- Postprint (1)
Reviewstate
- Peer-Review (5)
- (Verlags)-Lektorat (3)
- Verlags-Lektorat (1)
We present recognizers for four very different types of speech, thought and writing representation (STWR) for German texts. The implementation is based on deep learning with two different customized contextual embeddings, namely FLAIR embeddings and BERT embeddings. This paper gives an evaluation of our recognizers with a particular focus on the differences in performance we observed between those two embeddings. FLAIR performed best for direct STWR (F1=0.85), BERT for indirect (F1=0.76) and free indirect (F1=0.59) STWR. For reported STWR, the comparison was inconclusive, but BERT gave the best average results and best individual model (F1=0.60). Our best recognizers, our customized language embeddings and most of our test and training data are freely available and can be found via www.redewiedergabe.de or at github.com/redewiedergabe.
Corpus REDEWIEDERGABE
(2020)
This article presents the corpus REDEWIEDERGABE, a German-language historical corpus with detailed annotations for speech, thought and writing representation (ST&WR). With approximately 490,000 tokens, it is the largest resource of its kind. It can be used to answer literary and linguistic research questions and serve as training material for machine learning. This paper describes the composition of the corpus and the annotation structure, discusses some methodological decisions and gives basic statistics about the forms of ST&WR found in this corpus.
This paper describes a rule-based approach to detect direct speech without the help of any quotation markers. As datasets fictional and non-fictional texts were used. Our evaluation shows that the results appear stable throughout different datasets in the fictional domain and are comparable to the results achieved in related work.
In this paper, we present our work-inprogress to automatically identify free indirect representation (FI), a type of thought representation used in literary texts. With a deep learning approach using contextual string embeddings, we achieve f1 scores between 0.45 and 0.5 (sentence-based evaluation for the FI category) on two very different German corpora, a clear improvement on earlier attempts for this task. We show how consistently marked direct speech can help in this task. In our evaluation, we also consider human inter-annotator scores and thus address measures of certainty for this difficult phenomenon.
In this paper we outline our corpus-driven approach to detecting, describing and presenting multi- word expressions (MWEs). Our goal is to treat MWEs in a way that gives credit to their flexible nature and their role in language use. The bases of our research are a very large corpus and a Statistical method of collocation analysis. The rich empirical data is interpreted linguistically in a structured way which captures the interrelations, patterns and types of variances of MWEs. Several levels of abstraction build on each other: surface patterns, lexical realizations (LRs), MWEs and MWE patterns. Generalizations are made in a controlled way and in adherence to corpus evidence. The results are published online in a hypertext format.
We present a corpus-driven approach to the study of multi-word expressions, which constitute a significant part of. As a data basis, we use collocation profiles computed from DeReKo (Deutsches Referenzkorpus), the largest available collection of written German which has approximately two billion word tokens and is located at the Institute for the German Language (IDS). We employ a strongly usage-based approach to multi-word expressions, which we think of as conventionalised patterns in language use that manifest themselves in recurrent syntagmatic patterns of words. They are defined by their distinct function in language. To find multi-word expressions, we allow ourselves to be guided by corpus data and statistical evidence as much as possible, making interpretative steps carefully and in a monitored fashion. We develop a procedure of interpretation that leads us from the evidence of collocation profiles to a collection of recurrent word patterns and finally to multi-word expressions. When building up a collection of multi-word expressions in this fashion, it becomes clear that the expressions can be defined on different levels of generalisation and are interrelated in various ways. This will be reflected in the documentation and presentation of the findings. We are planning to add annotation in a way that allows grouping the multi-word expressions according to different features and to add links between them to reflect their relationships, thus constructing a network of multi-word expressions.
Automatic recognition of speech, thought, and writing representation in German narrative texts
(2013)
This article presents the main results of a project, which explored ways to recognize and classify a narrative feature—speech, thought, and writing representation (ST&WR)—automatically, using surface information and methods of computational linguistics. The task was to detect and distinguish four types—direct, free indirect, indirect, and reported ST&WR—in a corpus of manually annotated German narrative texts. Rule-based as well as machine-learning methods were tested and compared. The results were best for recognizing direct ST&WR (best F1 score: 0.87), followed by indirect (0.71), reported (0.58), and finally free indirect ST&WR (0.40). The rule-based approach worked best for ST&WR types with clear patterns, like indirect and marked direct ST&WR, and often gave the most accurate results. Machine learning was most successful for types without clear indicators, like free indirect ST&WR, and proved more stable. When looking at the percentage of ST&WR in a text, the results of machine-learning methods always correlated best with the results of manual annotation. Creating a union or intersection of the results of the two approaches did not lead to striking improvements. A stricter definition of ST&WR, which excluded borderline cases, made the task harder and led to worse results for both approaches.
This contribution presents the newest version of our ’Wortverbindungsfelder’ (fields of multi-word expressions), an experimental lexicographic resource that focusses on aspects of MWEs that are rarely addressed in traditional descriptions: Contexts, patterns and interrelations. The MWE fields use data from a very large corpus of written German (over 6 billion word forms) and are created in a strictly corpus-based way. In addition to traditional lexicographic descriptions, they include quantitative corpus data which is structured in new ways in order to show the usage specifics. This way of looking at MWEs gives insight in the structure of language and is especially interesting for foreign language learners.
This contribution presents an XML Schema for annotating a high level narratological category: speech, thought and writing representation (ST&WR). It focusses on two aspects: Firstly, the original Schema is presented as an example for the challenge to encode a narrative feature in a structured and flexible way and secondly, ways of adapting this Schema to TEI are considered, in Order to make it usable for other, TEI-based projects.