Bringing Stories Alive: Generating Interactive Fiction Worlds

Abstract

Interactive fictions—also called text-based games—are games in which a player interacts with a virtual world purely through textual natural language. In this work, we focus on procedurally generating interactive fiction worlds. Generating these worlds requires (a) referencing everyday and thematic commonsense priors in addition to (b) being semantically consistent, (c) interesting, (d) coherent throughout, all while (e) producing fluent natural language descriptions of places, people, and things. Using existing story plots from books as inspiration, we present a method that first extracts a partial knowledge graph encoding basic information regarding world structure such as locations and objects. This knowledge graph is then automatically completed utilizing thematic knowledge and used to guide a neural language generation model that fleshes out the rest of the world. We evaluate generated worlds with human-participant studies, comparing our technique against rule-based and human-made baselines

Envisioning Narrative Intelligence: A Creative Visual Storytelling Anthology

“…Each item in the anthology contains all text written by each author for the four facets that accompanied a given image sequence. 2 Table 1 exemplifies one item in the anthology: a complete set of author responses for each facet for a particular image-set. The remainder of this section outlines the facets, crowdsourcing interface, image origins, and anthology statistics.…”

“…An example of such interaction between imagery and stories is apparent in the Tell Tale Card game, 1 where players analyze pictures, form connections among what they see, and then articulate language to fellow players to tell stories based on how they interpreted the images. While the stories vary player to player and game to game, there are three consistent factors that guide their formation: (1) the environment and presentation in the imagery; (2) the narrative goal of the storyteller; and (3) the audience [50]. The environment and presentation refer to the content and quality of the imagery, which can spark a range of interpretations depending on the order and speed at which images are presented during the story writing, as well as the mood that the storyteller detects when viewing the images.…”

“…In this context, creativity is defined as "the ability to generate novel and valuable ideas" [7]. Prominent applications of creative computational models and narrative intelligence [40] in the form of automatic story generation include entertainment [7,54,83], gaming [2,28,75], education and training [8,28,29], and reconnaissance [88]. Alhussain et al have surveyed the textual story generation space, spanning resources, corpora, evaluation methods, and factors of story "interestingness," which feed into applications [1].…”

Story Shaping: Teaching Agents Human-Like Behavior with Stories

“…A knowledge graph is a set of ⟨subject, relation, object⟩ tuples. Knowledge-graph based reinforcement learning agents have been shown to be state-of-the-art in text-based games (Ammanabrolu and Riedl 2018; Ammanabrolu et al 2020a;Ammanabrolu and Hausknecht 2020;Ammanabrolu et al 2020b;Xu et al 2020;Peng, Riedl, and Ammanabrolu 2022). These agents infer objects and relations from text observations and use this knowledge graph as a long-term memory of the world state as a means of handling partial observability.…”

Creative Wand: A System to Study Effects of Communications in Co-creative Settings

“…While most text generation with transformer-based language models starts with a user-provided first sentence as a prompt, a number of research efforts in story generation attempt to modify the AI system to allow for a greater degree of control over what is generated. One means of controlling story generation is to condition generation on high-level plot outlines (Fan, Lewis, and Dauphin 2018;Peng et al 2018;Rashkin et al 2020) or story in-filling (Donahue, Lee, and Liang 2020;Ammanabrolu et al 2020). Tambwekar et al (2019 and Alabdulkarim et al (2021) retrain the language with a specific, provided end-goal.…”