“A Buster Keaton of Linguistics”: First Automated Approaches for the Extraction of Vossian Antonomasia

Abstract

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.

“The Rodney Dangerfield of Stylistic Devices”: End-to-End Detection and Extraction of Vossian Antonomasia Using Neural Networks

“…After using Wikidata for distant supervision linking all candidates to Wikidata “human” entities, they manually created a list to exclude common false positive candidates like “the House of” or “the Prince of”, as those names matched Wikidata entities. Schwab et al ( 2019 ) presented the first fully automated approach by extending the idea of Fischer and Jäschke ( 2019 ). They introduced three different methods.…”

“…Next, we describe the annotation of the target ( ) and the modifier ( ), as the source ( ) has already been identified and annotated by Schwab et al ( 2019 ). If multiple VA appear inside one sentence, we annotate all chunks independently.…”

“…This is an extension of the dataset from Schwab et al ( 2019 ) which was the result of a semi-automated VA extraction method on The New York Times Annotated Corpus (Sandhaus, 2008 ) which contains around 1.8 million articles that were published between 1987 and 2007. The dataset creation is based on a syntax-based approach focusing on nine syntactic patterns around the source (“the/a/an source of/for/among”, cf.…”

“…Combined with the need for background knowledge to differentiate between VA and non-VA phrases, it is a non-trivial task. Its detection can help to improve other NLP tasks, such as fact extraction, machine translation, or entity disambiguation/co-reference resolution (Schwab et al, 2019 ). It can also provide new interesting question answering tasks or support creative natural language generation, especially in news and blog articles to generate fruitful content.…”

“…In addition, we analyze the impact of training data and conduct two robustness studies. One shows the performance on out-of-corpus data and the other one shows the ability to generalize to new types of VA. A second contribution is a fully annotated dataset on the word level which emerged from the dataset in Schwab et al ( 2019 ). The dataset is explained in detail in Section 3.2.…”