A chatbot service for use in video game development
No Thumbnail Available
Date
2015-04-28
Authors
Larsen, Alec John
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Character dialogue writing for modern digital games is a di cult process as many of
them are non-linear and as such the authors have to write many di erent versions
of the characters' dialogues to compensate for di erent orders that the players
may perform the tasks. The proposed solution is to replace game cut-scenes with
interactive dialogue using a chatbot - a computer program that simulates conversation
by responding to user's text inputs in a natural language such as English. This
falls into the eld of interactive ction. By investigating existing interaction ction
systems it was determined that they are di cult to author, use prede ned dialogue
and do not handle mood. The proposed solution allows authors to create di erent non-
playable characters (NPCs) using a single chatbot, based on the Arti cial Linguistic
Internet Computer Entity (A.L.I.C.E.) open source project. This single chatbot acts
as the knowledge-base for all the non-playable characters (NPCs) and provides stock
responses to the player's inputs. These stock responses are then translated to match
the NPC being interacted with. The translation takes the NPCs' dialects and moods
into account, generating emergent dialogue. This approach simpli es the authoring
approach as the knowledge-base is created once, independently of the NPCs and
simple rules are de ned that allow each NPC to convey the relevent persona. The
general chatbot and NPC translation rules are created using a GUI. The hedonic
quality of the GUI was tested by ve people via a questionnaire. This showed that
while the GUI is easy to use, the general chatbot work
ow could be simpli ed. A
test game was created and 35 people rated the system via a questionnaire. This
gave a pleasing qualitative result as the NPCs are e ective at conveying information,
their responses are variable, their personas are evident through their responses and
players noticed a change in mood. Theoretical analysis shows that the simple rules
can produce extremely variable outputs.