Intertwining World and Narrative Generation for a
                                     Mobile Role-playing Game

                         Joel Jonasson1 , Brenden Lech2 , Sasha Azad2 , Chris Martens2
                                                  1
                                               Blast Bit Enterprises AB, Sweden
                                              2
                                             North Carolina State University, USA
                  joel.jonasson@blastbit.net,{bglech, sasha.azad, mawellni}@ncsu.edu, crmarten@ncsu.edu




   Making a game where the environments, characters, and
story are all procedurally generated is a complex task. We
are a small studio in Sweden trying to develop such a game
for the mobile market (working title “Lance a Lot”). Our
game adopts role-playing game conventions, such as an ex-
plorable world map, non-player characters, and quests. It is
our aim to develop as our ”story engine” a system capable
of procedurally generating quests and narratives to guide the   Figure 1: A root area is an area with the world root as its
gameplay experience. Such a system would encourage re-          parent. When the player enters a new root area all content
playability and ensure players never run out of content to      inside is expected to have been instantiated into the game
experience, while affording players more agency to shape        world.
the narrative and the world through their actions.
   Our approach to knitting together multiple kinds of pro-
cedural generation involves two types of generators: abstract      Our abstract generators receive as input possible player
and concrete. The abstract generators construct representa-     interactions, facts about the world and terrain, details avail-
tions of game content as sentences in first-order logic, and    able in the knowledge base, and hyper-parameters such as
the concrete generators that instantiates the in-game repre-    a series of morphological functions (Propp 2010), pacing or
sentations of those logical sentences.                          story beats (Mateas and Stern 2005), and other thematic vari-
   The abstract generators include generating the history of    ables.
the world, cultures and belief systems, character personality      The abstract generators work in a gradual process of mak-
and backstory. When the player enters a previously unvisited    ing information more concrete until the properties of an area
area, the engine generates the next area (see Fig. 1), begin-   are defined to the degree that the information can be used
ning with the generation of the terrain, including the area’s   as input to the concrete generators. An example of some of
biome, preliminary information about connecting areas, and      the properties associated with regions are locative properties
any other high-level terrain definition information required.   (such as resources, or settlements), the non-player charac-
                                                                ters in the region (including details of their species, or in-
Copyright © 2021 for this paper by its authors. Use permitted   teractions available to them), and quest information for the
under Creative Commons License Attribution 4.0 International    player. The first iteration will employ a partially planning-
(CC BY 4.0).
based approach to quest generation.
   Each concrete (content) generator (see Fig. 2) defines a
category of objects and thereby also what properties and
relations may be specified for an object of that category.
For example, a content generator for vegetation may de-
fine a plant as a type of content, with properties like ed-
ible and poisonous defined as unary predicates (for in-
stance, edible(plant)), and defining locations as binary
predicates (for example via, at(plant, area)). Similarly,
the agent generator controls available interactions in the
world, for instance, avoids(agent, agent), dazed(agent),
or initiates(agent, agent, F ight|Dialogue). These con-
tent generators are registered with the narrative system, en-
abling the abstract generators to see them as object cate-
gories.

                                                                  Figure 4: An expansion on Fig. 3 with two connected areas.
                                                                  Please note that the image on the right is somewhat mislead-
                                                                  ing as it does not reflect that the two areas are connected via
                                                                  specific bridge parts.


                                                                                          References
                                                                  Mateas, M., and Stern, A. 2005. Structuring content in the
                                                                  façade interactive drama architecture. In AIIDE, 93–98.
                                                                  Propp, V. 2010. Morphology of the Folktale, volume 9.
                                                                  University of Texas Press.




Figure 2: An example of interface code (in C#) between a
concrete generator and the abstract generators. A fact can be
added by calling the matching predicate method.




Figure 3: An example of how the generator from Fig. 2
might be used to construct a definition of the world, a sim-
plified sketch of the result and how it looks once instantiated
into the game world.

   Currently we are in the process of iterating on game me-
chanics and improving the content generators in the game.
We are actively collaborating with an academic group to de-
sign a cohesive system that will be underpinned by Answer
Set Programming, Linear Logic, and/or AI planning tech-
niques. We are interested in sharing our current approach
with the EXAG community and receiving feedback on our
initial prototypes.