Attributing a particular property to a person by naming another person, who is typically wellknown for the respective property, is called a Vossian Antonomasia (VA). This subtpye of metonymy, which overlaps with metaphor, has a specific syntax and is especially frequent in journalistic texts. While identifying Vossian Antonomasia is of particular interest in the study of stylistics, it is also a source of errors in relation and fact extraction as an explicitly mentioned entity occurs only metaphorically and should not be associated with respective contexts. Despite rather simple syntactic variations, the automatic extraction of VA was never addressed as yet since it requires a deeper semantic understanding of mentioned entities and underlying relations. In this paper, we propose a first method for the extraction of VAs that works completely automatically. Our approaches use named entity recognition, distant supervision based on Wikidata, and a bi-directional LSTM for postprocessing. The evaluation on 1.8 million articles of the New York Times corpus shows that our approach significantly outperforms the only existing semi-automatic approach for VA identification by more than 30 percentage points in precision.

No abstract

No abstract

Vossian Antonomasia (VA) is a well-known stylistic device based on attributing a certain property to a person by relating them to another person who is famous for this property. Although the morphological and semantic characteristics of this phenomenon have long been the subject of linguistic research, little is known about its distribution. In this paper, we describe end-to-end approaches for detecting and extracting VA expressions from large news corpora in order to study VA more broadly. We present two types of approaches: binary sentence classifiers that detect whether or not a sentence contains a VA expression, and sequence tagging of all parts of a VA on the word level, enabling their extraction. All models are based on neural networks and outperform previous approaches, best results are obtained with a fine-tuned BERT model. Furthermore, we study the impact of training data size and class imbalance by adding negative (and possibly noisy) instances to the training data. We also evaluate the models' performance on out-of-corpus and real-world data and analyze the ability of the sequence tagging model to generalize in terms of new entity types and syntactic patterns.