Crosslingual Story Generation with Large Language Models

Khái niệm cốt lõi

Planning in story generation across languages enhances coherence and control, particularly with a three-act structure.

Tóm tắt

The article explores the effectiveness of planning in crosslingual story generation using large language models. It introduces a new task of crosslingual story generation, presents a dataset, and conducts a study on different plans. Results show that the three-act structure leads to more coherent narratives.
Introduction

Automated story generation history.
Importance of planning in narrative coherence.
Shift from symbolic planning to large pre-trained language models.
Data Extraction

"Previous work has demonstrated the effectiveness of planning for story generation exclusively in a monolingual setting focusing primarily on English."
"Our results demonstrate that plans which structure stories into three acts lead to more coherent and interesting narratives."
Related Work

Plans improve coherence in automated story generation.
Previous approaches relied on planning before generating full stories.
Inquiry and Critical Thinking:
How can the three-act structure enhance narrative coherence?
What are the implications of using large language models for crosslingual story generation?
How might cultural differences impact the effectiveness of story plans across languages?

Thống kê

"Previous work has demonstrated the effectiveness of planning for story generation exclusively in a monolingual setting focusing primarily on English."
"Our results demonstrate that plans which structure stories into three acts lead to more coherent and interesting narratives."

Trích dẫn

"Plans have been widely used to improve coherence in automated story generation."
"We propose a new task of crosslingual story generation with planning and present a new dataset for this task."

Thông tin chi tiết chính được chắt lọc từ

Little Red Riding Hood Goes Around the Globe

by Evgeniia Raz... lúc arxiv.org 03-26-2024

https://arxiv.org/pdf/2212.10471.pdf

Yêu cầu sâu hơn

How can cultural nuances be incorporated into crosslingual story generation?

Incorporating cultural nuances into crosslingual story generation is crucial for creating authentic and engaging narratives. One way to achieve this is by conducting thorough research on the cultures and languages involved in the storytelling process. Understanding the values, beliefs, traditions, and storytelling conventions of different cultures can help in crafting stories that resonate with diverse audiences.
Additionally, working with local experts or consultants from various cultural backgrounds can provide valuable insights into how to infuse cultural elements into the stories. This collaboration ensures that the stories are respectful, accurate, and reflective of the diversity present in different communities.
Using language models trained on multilingual datasets that include a wide range of cultural contexts can also help capture linguistic nuances specific to different languages. By exposing these models to a variety of cultural texts during training, they become more adept at generating culturally relevant content during crosslingual story generation.
Moreover, incorporating folklore, myths, legends, and other traditional narratives from various cultures as inspiration for storytelling can add richness and depth to crosslingual stories. By drawing upon these sources, storytellers can create narratives that not only entertain but also educate audiences about different cultures around the world.

What ethical considerations should be taken into account when using large language models for storytelling?

When utilizing large language models (LLMs) for storytelling purposes, several ethical considerations must be carefully addressed:

Bias Mitigation: LLMs have been known to amplify biases present in their training data. It's essential to mitigate bias by ensuring diverse representation in training data and implementing bias detection mechanisms during model development.

Cultural Sensitivity: Stories generated by LLMs should respect diverse cultures and avoid perpetuating stereotypes or misrepresentations of marginalized groups.

Consent & Privacy: If personal data or sensitive information is used as input for generating stories (e.g., user-generated prompts), obtaining consent from individuals is crucial to uphold privacy rights.

Transparency & Accountability: Storytellers should disclose when AI systems are involved in generating content so users are aware they are interacting with machine-generated text rather than human-authored material.

Fairness & Accessibility: Ensuring fair access to AI-generated stories regardless of factors like socioeconomic status or disability is important for promoting inclusivity within storytelling platforms.

How might interactive storytelling tools benefit from incorporating structured plans like the three-act structure?

Incorporating structured plans like the three-act structure into interactive storytelling tools offers several benefits:

Enhanced Coherence: The three-act structure provides a clear framework for organizing plot events cohesively throughout an interactive narrative experience.

Improved Engagement: Structured plans help maintain audience interest by guiding the progression of events logically and building suspense effectively.

Narrative Control: Interactive tools utilizing structured plans allow creators greater control over story arcs while still providing flexibility for user choices within each act.

4 .User Guidance: The three-act structure serves as a roadmap for both storytellers and users navigating through interactive experiences—helping them anticipate key plot points.
5 .Personalization Opportunities: By integrating user choices within each act based on pre-defined structures such as character goals or conflicts set up in earlier acts allows tailored experiences without sacrificing overall coherence.