From Theory to Practice: Recent Advances and Future Prospects in Computational Politeness in Natural Language Processing

Abstract

The computational approach to politeness involves the automatic prediction and generation of politeness in text, a crucial task for conversational analysis due to the pervasiveness and complexity of politeness in interactions. This field has garnered significant interest from the conversational analysis community. This tutorial will explore the milestones in computational politeness research, covering supervised and weakly supervised feature extraction, context incorporation, politeness across social factors, and the relationship between politeness and socio-linguistic cues. We will describe the datasets, methodologies, trends, and challenges in this area, discuss performance metrics, and provide directions for future research. The tutorial offers valuable resources along dimensions like feature types, annotation techniques, datasets used, and methodologies proposed.

Outline

The tutorial is organized as follows:

• Introduction (10 minutes)

  In the introduction to this tutorial, we will begin with an overview of the definitions of politeness from authoritative sources such as The Free Dictionary, Oxford Learner’s Dictionary, and Merriam-Webster and politeness literature, emphasizing that politeness encompasses socially correct and acceptable behavior. Next, we will discuss the significance of politeness in human communication, highlighting its critical impact on social interactions and goals. Following this, we will address the relevance of politeness to AI and NLP, underscoring the necessity of emulating human politeness in human-computer interactions [10, 22 ]. Finally, we will provide an overview of the research scope in computational politeness, detailing the computational approaches, methodologies, and advancements in this field.

• Politeness Theory in Linguistics (10 minutes)

  In this part of the tutorial, we provide a concise overview of linguistic studies on politeness. Politeness is pivotal in human communication, influencing the outcomes of interactions significantly. These studies, central to pragmatics research in recent decades, primarily explore communicative strategies for maintaining social harmony. Various theoreticians have offered diverse definitions of politeness, shaping its understanding in scholarly discourse. Further, we will discuss factors emphasizing the role of gender, age, and culture in politeness usage in interaction and its association with other forms of social language and behavior, namely Etiquette, Formality, Emotion, and Offensiveness.

• Computational Politeness in Natural Language Processing (60 minutes)

  In this part of the tutorial, we will focus on two primary research categories within computational politeness: politeness as a natural language understanding task and politeness as a natural language generation task. We will delve into how machine learning and deep learning techniques facilitate the development of computational models that simulate politeness-oriented interactions. Specifically, we will discuss the various problem definitions, benchmark datasets, state-of-the-art machine learning and deep learning-based methodologies, performance evaluation metrics, and recent trends central to computational politeness in NLP. Finally, we will discuss the critical issues that appear in different computational politeness works, such as the acquisition of high-quality data and modeling politeness across diverse contexts.

• Relevance of Politeness for Social Good Applications (20 minutes)

  In this section, we will delve into the role of politeness in dialogue systems for social good applications such as persuasion and mental health support. Recent studies have highlighted the benefits of integrating politeness and complementary cues like emotion and empathy into these systems. We will discuss some of the prominent works that investigated politeness and other related cues, such as emotion, sentiment, and empathy, to enhance user experience and effectiveness in social good applications, highlighting ongoing advancements and methodologies in this field.

• Politeness and Large Language Models (10 minutes)

  In this section, we will present the impact of Large Language Models (LLMs) like GPT-3, LLaMA2, and ChatGPT on computational politeness. LLMs have significantly advanced NLP tasks such as emotion recognition, summarization, and dialogue generation. We will discuss their ability to align with politeness norms and their social competence in recognizing, interpreting, and understanding politeness for effective communication. Recent studies have shown that models like ChatGPT can predict politeness reasonably well, even in zero-shot settings. We will discuss the implications of these findings for applications in fields like healthcare and education, emphasizing the importance of LLMs adhering to social norms to foster trust and collaboration in human-machine interactions. This section aims to provide a comprehensive understanding of the role of LLMs in computational politeness and their potential to enhance human-computer interactions.

• Conclusion and Future Directions (10 minutes)

  Overall, in this tutorial, we will provide an overview of the extensive research conducted on computational approaches to politeness. We will review key milestones in the field, including supervised and weakly supervised feature extraction, context incorporation, and the study of politeness across various social factors such as culture, age, and gender. Additionally, we will discuss the role of socio-linguistic indicators in understanding politeness. The review covers methodologies, datasets, and performance values, highlighting the increasing use of Large Language Models (LLMs) in politeness research. Despite significant advancements, challenges persist in computational politeness research, which will guide the future of this research area.

Materials

Slides

Tutorial Proposal

Computational Politeness in Natural Language Processing: A Survey