Identifying Deception in Indonesian Transcribed Interviews through Lexical-based Approach

31st Pacific Asia Conference on Language, Information and Computation (PACLIC 31), pages 148–154 Cebu City, Philippines, November 16-18, 2017 Copyright c⃝2017 Tifani Warnita and Dessi Puji Lestari Identifying Deception in Indonesian Transcribed Interviews through Lexical-based Approach Tifani Warnita Dessi Puji Lestari School of Electrical Engineering and Informatics School of Electrical Engineering and Informatics Institut Teknologi Bandung Institut Teknologi Bandung tifaniwarnita@gmail.com dessipuji@gmail.com Abstract This paper aims to present a lexical-based approach in order to identify deception in Indonesian transcribed interviews. Using word calculation from the psychological point of view, we classify each subject utterance into two classes, namely lie and truth. We find that the intentions of the people in both telling the truth and hiding the fact can affect the words used in their utterances. We also find that there is an interesting pattern for Indonesian people when they are answering questions with lies. Despite the promising result of lexical- based approach for detecting deception in the Indonesian language, there are also some cases which cannot be handled by only using the lexical features. Hence, we also present an additional experiment of combining the lexical features with acoustic/prosodic features using the recorded sound data. From the experiment, we find that the combination of lexical features with other features such as acoustic/prosodic can be used as the initial step in order to get better results in identifying deception in Indonesian. 1 Introduction Human social behavior has successfully led to the ubiquitous human communication. In this regard, it is also very possible for people to commit lies when communicating with others. Deceit or commonly referred to as lie is any actions of making others believe what we perceived as false, without the receivers know that they are being fooled (Ekman, 1992; Vrij, 2008). A lie can be divided into a variety of classes when viewed from various aspects involved in such actions. For example, when viewed from how bad a lie is, a lie can be classified into a white lie, gray lie, and real lie (Bryant, 2008). Various motivations may underlie a lie. Based on interviews with children and questionnaire survey results from adults by Ekman (1989), according to most of the children and the adults, someone might lie in order to avoid punishment. Referring to this phenomenon, especially if we focus on the realm of interrogation for solving crimes, it is a compelling matter when people are challenged to be able to tell which utterances contain lies. However, for many people, it seems difficult to recognize any deception, considering that the cues to deception can be reflected from diverse aspects (DePaulo et al., 2003) as well as the need for specific experience in related scientific fields. As in other computational linguistic studies, in order to obtain the best result, sometimes the geographic location of the speakers have to be taken into account when finding the salient features. The location of the speakers can affect their way of thinking, and also their way of speaking. A feature might be very dominant …

Content truncated to 500 words