Identification of toxicity in speech

Abstract

Communication of information is apparently instantaneous nowadays than ever before. With the advent of social media platforms, such as, online forums, and networking websites, interactions between people today is at an all-time high. In these interactions, even though the primary form of passing information is text, rapid technological developments have allowed people to communicate instantly over the internet using audio and visual modes as well. With the increase in interactions, there has also been an increase in the form of violence that are incited due to the spread of hatred through speech, and has many times proved dangerous if not prevented at the source. Over the years, toxicity detection has been a popular field of interest for research work, however toxicity detection in speech remains relatively unexplored. Since speech is the one mode of communication that is, by far, unique to mostly humans, the recognition of vicious and spiteful elements that are spread through speech is an important step in preventing violent, hateful crimes. An apparent solution is the application of Natural Language Processing. Natural Language Processing is a technique that is popularly used to work with language in the way it is written or spoken by humans. Combining linguistics and artificial intelligence, it paves the way for a sufficiently reliable solution. In this thesis, we have attempted to investigate the identification of offensive elements in speech using some of the methods from Natural Language Processing, and presented our findings.