Automated Detection of Voice in News Texts

Evaluating Tools for Reported Speech and Speaker Recognition

Abstract

The automated content analysis of text has become integral to today's communication science. Automated possibilities to analyze reported voice in such text have, however, been seldomly utilized. Yet, approaches to automatically detect reported speech from persons and organizations in textual data would offer valuable insights into media and communication practices. Bridging the fields of communication science and computational linguistics, this manuscript reviews and evaluates available tools for automated voice detection (of direct/indirect speech and of the speakers) with respect to user experience and validity. A manually annotated English news article corpus served as baseline for the evaluation of the automated detection of voice. Findings indicate that the evaluated tools offer highly satisfactory user experience and provide promising solutions for detecting direct speech automatically, encouraging fellow researchers to utilize automated detection for direct quotations. However, the recognition of indirect speech and of speakers needs considerable improvement. Furthermore, this manuscript emphasizes the need for researchers to perform quality checks when external tools are used as part of their text-analysis pipelines.