Artificial Intelligence And the Challenges of Detecting Rude Conversational Behaviour

On the Challenges of Detecting Rude Conversational Behaviour. Karan Grewal, Khai N. Truong. arXiv.org Computer Science > Human-Computer Interaction, arXiv:1712.09929 [cs.HC], https://arxiv.org/abs/1712.09929

Abstract: In this study, we aim to identify moments of rudeness between two individuals. In particular, we segment all occurrences of rudeness in conversations into three broad, distinct categories and try to identify each. We show how machine learning algorithms can be used to identify rudeness based on acoustic and semantic signals extracted from conversations. Furthermore, we make note of our shortcomings in identifying rudeness in conversations.

Introduction

One-on-one interactions are important in everyday social settings. For instance, in order to attract a potential partner, it is imperative that an individual behave in an appropriate manner. Unfortunately, one-on-one interactions can often result in one party exhibiting rude or inappropriate conversational behaviour. In many cases, the offending party is not aware of the severity of their actions and does not intend to offend the other party. For example, certain individuals may be socially unaware of how others perceive their behaviour. Individuals with learning disabilities, such as autism, may follow this trend. Likewise, young children often lack awareness of their behaviour { a possible explanation for the presence of bullying in elementary schools and why children are generally regarded as immature. In both cases, monitoring a user's conversational behaviour and making them aware of it via active feedback while they are engaged in a one-on-one interaction would be helpful towards correcting their behaviour in such scenarios.

In the last century, there has been a lot of work in the linguistics and psychology domains which attempt to define politeness and acceptable behaviour pertaining to two-person interactions. The most popular of these is Penelope Brown and Steven Levinson's Politeness theory [2]. This theory states that all individuals have two faces: a positive self-image which is the desire to be approved by others, and a negative self-image which is the desire of actions to be unimpeded by others. According to Politeness theory, any external actions which threaten one or more of an individual's faces, such disrespectful gestures, constitute impoliteness. Also, Geoffrey Leech's principle of politeness states that if two individuals are interacting, then there will be some form of disagreement or tension if both individuals are pursuing mutually-incompatible goals -- likening the chance of rude behaviour [8]. Here, goals refers to a psychological state of being. In contrast, Bruce Fraser argues against the theories formulated by Leech, Brown, and Levinson by pointing out that each culture has its own set of social norms which define acceptable behaviour [6]. Therefore, as Fraser argues, the question of whether an individual is behaving in an inappropriate manner is entirely dependent on the context of his/her actions. This view aligns with Robin Lako's notable example of the speaking style in New York [7].  As she states, New Yorkers often use profanity in a casual sense without any intent to offend or be impolite.  However, their conversational behaviour is likely to be interpreted as rude in other cultures.

Is there a grounded definition of rudeness with respect to speech which can be derived from classical theories of politeness? In this study, we define define the notion of rude conversational behaviour and explore methods to identify this type of behaviour in two-person interactions. We do this by extracting acoustic and semantic information from an individual's speech and develop methods which attempt to pinpoint exact instances of rude conversational behaviour. Also, we highlight some existing problems which make the task at hand dicult through our findings. Note that we only focus on signals extracted speech data.