Towards Generating Stylistic Dialogues for Narratives Using Data-Driven Approaches

Abstract

Recently, there has been a renewed interest in generating dialogues for narratives. Within narrative dialogues, their structure and content are essential, though style holds an important role as a mean to express narrative dialogue through telling stories. Most existing approaches of narrative dialogue generation tend to leverage hand-crafted rules and linguistic-level styles, which lead to limitations in their expressivity and issues with scalability. We aim to investigate the potential of generating more stylistic dialogues within the context of narratives. To reach this, we propose a new approach and demonstrate its feasibility through the support of deep learning. We also describe this approach using examples, where story-level features are analysed and modelled based on a classification of characters and genres.

CloChat: Understanding How People Customize, Interact, and Experience Personas in Large Language Models

“…Various research efforts have focused on how users can effectively tailor and configure the personas of conversational agents to align with their individual preferences and needs. Previous works have employed frameworks categorizing agent personas based on their characteristics [10,43,49,72,87,94]. The Big Five model, for example, encapsulates five core personality traits: extroversion, agreeableness, conscientiousness, neuroticism, and openness [55].…”

“…Despite its recognized importance, the processes of how people customize, experience, and interact with personas in LLMs, and how these experiences differ from those with generic and universal conversational agents, remain relatively unexplored. Past research has predominantly focused on categorizing personality types for crafting personas [9,23,43,49,72,87,94], often prioritizing the convenience of designers or developers [37], while overlooking a broader range of diverse personality types [17,67]. There have been few studies that delve into persona designs tailored to individual user preferences or interaction histories [67].…”

On the Need for Thoughtful Data Collection for Multi-Party Dialogue: A Survey of Available Corpora and Collection Methods

“…The Character Style From Film Corpus (Walker et al, 2012a) is another resource contributing towards guided text generation by providing character styles, created from the archive IMSDB. It has been used to generate stylistic dialogue for narratives (Xu et al, 2018).…”