University of the Aegean
Text analysis and machine learning classification of defendants’ testimonies in greek courtroom in order to create their linguistic profile
Abstract
dc:descriptionForensic Linguistics attempts to analyze the language that relates to the law, either as evidence or as legal discourse. Language as legal discourse includes, among others, the discourse inside the court room. Crime profiling, or offender profiling, is one of the most important areas of research in Forensic Linguistics and should be its fundamental task, since by examining a criminal behavior one can evaluate or even predict future criminal actions. The identification of specific characteristics of an individual committing a crime is achieved by a thorough systematic observational process and an analysis of the crime scene, the victim, the forensic evidence, and the known facts of the crime. In this dissertation, using natural language analysis techniques from the field of author profiling, where one can extract information about the age, education, sex, etc., of the author of a given text, we attempt to define the linguistic profile of a criminals’ category (that of the murderers) and, at a later stage, to develop a machine learning classifier which would predict whether a text belongs to that category, i.e., it has been written or said by a murderer or not. First, we created three corpora from text data that we derived from real trial briefs of a Greek court. The first one concerned testimonies of defendants accused of murder, the second one was constructed from testimonies of witnesses and the last one consisted of testimonies of the defendants in their interrogation phase before their trial. It is obvious that the creation of this research would not have been possible without the possession of these trial briefs, which were difficult to get access to and required a time-consuming procedure. Τhe latter is the main reason that no corresponding research has been done so far in Greece. Having created these corpora, we quantified the way defendants of murder speak inside a Greek courtroom during their testimony, by studying several stylometric features of their language and comparing them with both the general language and the language of the witnesses. As a result, we have been able to extract some linguistic patterns used by murders in their testimonies. Moreover, some of these features proved to be more crucial, than others, in being able to describe the language profile of the speaker of a testimony. The results we extracted of the quantitative analysis and knowing that the court proceedings and police investigations in Greece do not have appropriate and easy-to-use tools that can provide additional assistance in evaluating the statements of the accused, led us to the idea of constructing an automated text classifier using as training data the most useful stylometric features of the defendants’ testimonies. Automated text classification has been considered as a vital method to manage a vast number of documents in digital form since its goal is the construction of a classification model (classifier) that is able to automatically assign labels to electronic texts by learning specific features of each category. In any case, statistics has been more concerned with testing hypotheses, whereas machine learning has been more concerned with formulating the process of generalization as a search through possible hypotheses. Hence, we present a text classification machine learning model, the GDCT classifier, which was trained using the appropriate stylometric features, as demonstrated in our study. The experimental results of our corpora, covering the testimonies of 269 defendants and witnesses in total, verify the effectiveness of our method. Specifically, we prove that GCDT classifier can characterize a person who testifies, as guilty or not, with 93% accuracy. Our model does not seek to replace any judge or investigator but can offer to the trial procedure an additional tool in evaluating a murderer’s testimony. This research is a pioneering method both in Greek Forensic Linguistics and in the Greek judicial process.
Degree
thesis:*- Grantor dc:publisher
- University of the Aegean
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Katranidou, Anastasia
- Κατρανίδου, Αναστασία
Subjects
dc:subject × 28- Δικανική Γλωσσολογία
- Υπολογιστική γλωσσολογία
- Μηχανική μάθηση
- Ανάλυση κειμένου
- Ταξινόμηση κειμένων
- Σώματα κειμένων
- Υφομετρία
- Καταθέσεις
- Forensic Linguistics
- Computational linguistics
- Machine learning
- Text analysis
- Text classification
- Corpora
- Stylometry
- Testimonies
- Γλώσσα και Λογοτεχνία
- Ανθρωπιστικές Επιστήμες και Τέχνες
- Γλώσσα και Γλωσσολογία
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Διεπιστημονικές εφαρμογές επιστήμης ηλεκτρονικών υπολογιστών
- Languages and Literature
- Humanities and the Arts
- Language and Linguistics
- Computer and Information Sciences
- Natural Sciences
- Computer Science Interdisciplinary Applications
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/50840