Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation

Judith van Stegeren; Jakub Myśliwiec

doi:10.1145/3472538.3472595

Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation

Judith van Stegeren, Jakub Myśliwiec

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

21 Citations (Scopus)

188 Downloads (Pure)

Abstract

GPT-2, a neural language model trained on a large dataset of English web text, has been used in a variety of natural language generation tasks because of the language quality and coherence of its outputs. In order to investigate the usability of GPT-2 for text generation for video games, we fine-tuned GPT-2 on a corpus of video game quests and used this model to generate dialogue lines for quest-giver NPCs in a role-playing game. We show that the model learned the structure of quests and NPC dialogue, and investigate how the temperature parameter influences the language quality and creativity of the output artifacts. We evaluated our approach with a crowdsource experiment in which human judges were asked to rate hand-written and generated quest texts on language quality, coherence and creativity.

Original language	English
Title of host publication	FDG'21: The 16th International Conference on the Foundations of Digital Games (FDG) 2021
Editors	Allan Fowler, Johanna Pirker, Alesandro Alessandro Canossa, Ali Ali Arya, Casper Harteveld
Place of Publication	Montreal, QC, Canada
Publisher	Association for Computing Machinery
Pages	1
Number of pages	8
ISBN (Electronic)	9781450384223
DOIs	https://doi.org/10.1145/3472538.3472595
Publication status	Published - 3 Aug 2021
Event	16th International Conference on the Foundations of Digital Games, FDG 2021 - Online Conference Duration: 2 Aug 2021 → 6 Aug 2021 Conference number: 16 http://fdg2021.org/

Conference

Conference	16th International Conference on the Foundations of Digital Games, FDG 2021
Abbreviated title	FDG 2021
City	Online Conference
Period	2/08/21 → 6/08/21
Internet address	http://fdg2021.org/

Keywords

GPT-2
Transformers
World of Warcraft
English
NPC dialogue
Natural language generation
Procedural content generation for games
Quest generation
MMORPG

Access to Document

10.1145/3472538.3472595Licence: CC BY

VanStegeren2021fine-tuningFinal published version, 491 KBLicence: CC BY

Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation
van Stegeren, J. E. (Speaker)
3 Aug 2021
Activity: Talk or presentation › Oral presentation

Cite this

van Stegeren, J., & Myśliwiec, J. (2021). Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation. In A. Fowler, J. Pirker, A. A. Canossa, A. A. Arya, & C. Harteveld (Eds.), FDG'21: The 16th International Conference on the Foundations of Digital Games (FDG) 2021 (pp. 1). Article 3472595 Association for Computing Machinery. https://doi.org/10.1145/3472538.3472595

van Stegeren, Judith ; Myśliwiec, Jakub. / Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation. FDG'21: The 16th International Conference on the Foundations of Digital Games (FDG) 2021. editor / Allan Fowler ; Johanna Pirker ; Alesandro Alessandro Canossa ; Ali Ali Arya ; Casper Harteveld. Montreal, QC, Canada : Association for Computing Machinery, 2021. pp. 1

@inproceedings{e620f864167e45378406e089cb6f31fe,

title = "Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation",

abstract = "GPT-2, a neural language model trained on a large dataset of English web text, has been used in a variety of natural language generation tasks because of the language quality and coherence of its outputs. In order to investigate the usability of GPT-2 for text generation for video games, we fine-tuned GPT-2 on a corpus of video game quests and used this model to generate dialogue lines for quest-giver NPCs in a role-playing game. We show that the model learned the structure of quests and NPC dialogue, and investigate how the temperature parameter influences the language quality and creativity of the output artifacts. We evaluated our approach with a crowdsource experiment in which human judges were asked to rate hand-written and generated quest texts on language quality, coherence and creativity.",

keywords = "GPT-2, Transformers, World of Warcraft, English, NPC dialogue, Natural language generation, Procedural content generation for games, Quest generation, MMORPG",

author = "{van Stegeren}, Judith and Jakub My{\'s}liwiec",

note = "Funding Information: This research is supported by the Netherlands Organisation for Scientific Research (NWO) via the Data2Game project (project number 055.16.114). Publisher Copyright: {\textcopyright} 2021 Owner/Author.; 16th International Conference on the Foundations of Digital Games, FDG 2021, FDG 2021 ; Conference date: 02-08-2021 Through 06-08-2021",

year = "2021",

month = aug,

day = "3",

doi = "10.1145/3472538.3472595",

language = "English",

pages = "1",

editor = "Allan Fowler and Johanna Pirker and Canossa, {Alesandro Alessandro} and Arya, {Ali Ali} and Casper Harteveld",

booktitle = "FDG'21: The 16th International Conference on the Foundations of Digital Games (FDG) 2021",

publisher = "Association for Computing Machinery",

address = "United States",

url = "http://fdg2021.org/",

}

van Stegeren, J & Myśliwiec, J 2021, Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation. in A Fowler, J Pirker, AA Canossa, AA Arya & C Harteveld (eds), FDG'21: The 16th International Conference on the Foundations of Digital Games (FDG) 2021., 3472595, Association for Computing Machinery, Montreal, QC, Canada, pp. 1, 16th International Conference on the Foundations of Digital Games, FDG 2021, Online Conference, 2/08/21. https://doi.org/10.1145/3472538.3472595

Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation. / van Stegeren, Judith; Myśliwiec, Jakub.
FDG'21: The 16th International Conference on the Foundations of Digital Games (FDG) 2021. ed. / Allan Fowler; Johanna Pirker; Alesandro Alessandro Canossa; Ali Ali Arya; Casper Harteveld. Montreal, QC, Canada: Association for Computing Machinery, 2021. p. 1 3472595.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation

AU - van Stegeren, Judith

AU - Myśliwiec, Jakub

N1 - Conference code: 16

PY - 2021/8/3

Y1 - 2021/8/3

N2 - GPT-2, a neural language model trained on a large dataset of English web text, has been used in a variety of natural language generation tasks because of the language quality and coherence of its outputs. In order to investigate the usability of GPT-2 for text generation for video games, we fine-tuned GPT-2 on a corpus of video game quests and used this model to generate dialogue lines for quest-giver NPCs in a role-playing game. We show that the model learned the structure of quests and NPC dialogue, and investigate how the temperature parameter influences the language quality and creativity of the output artifacts. We evaluated our approach with a crowdsource experiment in which human judges were asked to rate hand-written and generated quest texts on language quality, coherence and creativity.

AB - GPT-2, a neural language model trained on a large dataset of English web text, has been used in a variety of natural language generation tasks because of the language quality and coherence of its outputs. In order to investigate the usability of GPT-2 for text generation for video games, we fine-tuned GPT-2 on a corpus of video game quests and used this model to generate dialogue lines for quest-giver NPCs in a role-playing game. We show that the model learned the structure of quests and NPC dialogue, and investigate how the temperature parameter influences the language quality and creativity of the output artifacts. We evaluated our approach with a crowdsource experiment in which human judges were asked to rate hand-written and generated quest texts on language quality, coherence and creativity.

KW - GPT-2

KW - Transformers

KW - World of Warcraft

KW - English

KW - NPC dialogue

KW - Natural language generation

KW - Procedural content generation for games

KW - Quest generation

KW - MMORPG

U2 - 10.1145/3472538.3472595

DO - 10.1145/3472538.3472595

M3 - Conference contribution

SP - 1

BT - FDG'21: The 16th International Conference on the Foundations of Digital Games (FDG) 2021

A2 - Fowler, Allan

A2 - Pirker, Johanna

A2 - Canossa, Alesandro Alessandro

A2 - Arya, Ali Ali

A2 - Harteveld, Casper

PB - Association for Computing Machinery

CY - Montreal, QC, Canada

T2 - 16th International Conference on the Foundations of Digital Games, FDG 2021

Y2 - 2 August 2021 through 6 August 2021

ER -

Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation

Abstract

Conference

Keywords

Access to Document

Fingerprint

Activities

Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation

Cite this