Italian Trulli

“Fear in your Eyes”: Analyzing Threat Perception and Its Influence on Deadly Use of Force by Police Officers against Civilians Using Hebrew NLP Tools

Renana Keydar (renana.keydar@mail.huji.ac.il), Faculty of Law, Hebrew University, Israel

Overview : This study seeks to explore and empirically substantiate the effects of the gap between objective and subjective threat perceptions of police officers in cases of lethal use of force against civilians. It focuses on a newly accessible corpus of digitized transcripts of police officers who testified before an Israeli State Commission of Inquiry in Hebrew. Combining topic modeling and sentiment analysis to explore subjective threat perceptions of duty holders, we aim to establish a deeper, empirically validated, understanding of serious human rights violations. The study makes an important methodological contribution to the field of NLP for non-English languages and non-Latin characters, by applying, for the first time, newly developed NLP tools for Hebrew on a Modern spoken Hebrew corpus. As such, it also contributes to the diversification of the DH community be opening it to minor languages.

Background : Police has extensive powers to use force against civilians. The killing of a civilian is also within the legal and legitimate range of force as long as the police officer acts to protect human life or is in mortal danger himself. Justifying the decision to use lethal force should meet the test of proportionality of the “reasonable police officer.” While the emphasis in previous studies was on the extent of the objective threat posed to the security forces, mainly by analyzing the characteristics of the event, this study takes a broader approach, both theoretically and methodologically, seeking to understand the subjective threat perception of the officers; feelings - of danger, fear or distress – which are crucial in justifying the decision to use of lethal force against civilians.

In order to explore, substantiate and understand the gap between objective and subjective threat perceptions, we aim to examine officers testimonies from the protocols of The Israeli State Commission of Inquiry to investigate clashes between security forces and Israeli civilians during the events of October 2000, in which 13 Arab citizens and a resident of Gaza were killed (henceforth: “The Or Commission”). The Or Commission held 92 public hearings and interviewed security forces involved in the specific incidents, eye-witnesses, politicians, senior police officers, public figures, and journalists. Therefore, examining the totality of the transcripts of the Commission’s hearings will enable us to understand in depth the various aspects of the threat perception that lead to the use of lethal force against civilians. Through an empirical semantic examination of the corpus, we intend to point out the ways in which social constructs that shape subjective threat perceptions and influence decision-making regarding the use of force against civilians.

Corpus and Methodology : Our corpus consists of the protocols of the Or Commission deliberations (comprising more than 2 million words) and the Commission’s final report (holding more than 700 pages). While the Commission’s report is available for online review, the protocols and testimonies are not easily accessible.

The methodological contribution of the study : First, it develops and applies tools and algorithms of topic modeling and sentiment analysis to a corpus of Modern Hebrew legal transcripts.

Recent years have seen a modest rise in studies using computational tools to analyze legal corpora such as constitutions , US Supreme Court decisions or constitutional debates in the US Congress . The scholarly potential, however, is far from exhaustion. In the Israeli field, such research has to overcome crucial language barriers, as in the current state of the field, there are no ready-made or sufficiently-tested NLP tools for modern Hebrew . Our study is first of its kind to employ NLP methods for empirical legal research in Hebrew. Second, it combines topic modeling and sentiment analysis in order to examine subjective perceptions of threat of duty holders. Applied together, these computational methods make it possible to uncover the hidden semantic themes and to identify the emotional and subjective affects expressed by the various voices in the corpus, thus providing an empirical-computational basis for examining the construction of threat perceptions of the security forces.

As such, the research presented here is innovative both in the corpus it examines, and in the methodologies that it implements, in order to establish a deeper understanding of serious human rights violations.