Computational Models of Narrative

Abstract: The quality and believability of a story can be significantly enhanced by the presence of compelling characters. Characters can be made more compelling by the portrayal of a distinguishable personality. This paper presents an algorithm that formalizes an approach previously described for the incorporation of character personality in narrative that is automatically generated. The approach is based on a computational model that operationalizes personality as behavior that results from the choices made by characters in the course of a story. This operationalization is based on the Big Five personality structure and results from behavioral psychology studies that link behavior to personality traits.

Abstract: We position a narrative-centred computational model for high-level knowledge representation and reasoning in the context of a range of assistive technologies concerned with visuo-spatial perception and cognition tasks. Our proposed narrative model encompasses aspects such as space, events, actions, change, and interaction from the viewpoint of commonsense reasoning and learning in large-scale cognitive systems. The broad focus of this paper is on the domain of human-activity interpretation in smart environments, ambient intelligence etc. In the backdrop of a smart meeting cinematography domain, we position the proposed narrative model, preliminary work on perceptual narrativisation, and the immediate outlook on constructing general-purpose open-source tools for perceptual narrativisation.

Abstract: We present the ongoing activities and the first results achieved in a research project concerning the understanding of narrative in the high school. We analyze the literary text I promessi sposi, extracting social networks of characters and other fundamental narrative elements. We present a reading and visualization environment that is being experienced by the students in this project.

Abstract: The large question informing our research is how narratives shape the way people think about moral issues. One aspect of this question concerns moral choice in situations that do not seem to be ambiguous. In this paper we discuss whether narratives support moral ambiguity or moral clarity. The approach we use is serial reproduction of narratives as in the telephone game or, in German, “Stille Post”. Our first findings is that serial reproduction quickly leads to a polarization of moral choice. Our second, more surprising, finding is that the entire situation of a character or reader having to make a moral choice is dropped in its entirety. Hence, we speculate that narratives seek an optimal position or perspective on moral choice. This position is one of being an observer, but not a decision-maker. Morality, within narratives, is an observer sport.

Abstract: A major concern regarding the study of narratives regards how they are indexed and retrieved. This is a question which touches on the structure of human memory in general. Indeed, if narratives capture the substance of human thought, then data that we have already collected regarding human memory is of central importance to the computational study of narrative. Fuzzy Trace Theory assumes that memory for narrative is simultaneously stores at multiple levels of abstraction and, whenever possible, decision-makers interpret a stimulus qualitatively and therefore operate on a simple–typically categorical– “gist” representation. Here, we present a computational model of Fuzzy Trace Theory applied to explain the impact of changes in a narrative upon risky-choice framing effects. Overall, our theory predicts the outcome of 19 experimental effects using only three basic assumptions: 1) preference for lowest level of gist, that is, categorical processing; 2) decision options that fall within the same categorical description are then interpreted using finer-grained (ordinal or verbatim) distinctions; and 3) once the options are mentally represented, decision preferences are generated on the basis of simple positive vs. negative valences stored in long-term memory (e.g., positive value for human lives). A fourth assumption – that negatively-valenced decision options are preferentially converted to positive decision options – is used when categories are not otherwise comparable.

Abstract: This paper describes a novel method of representing semantic networks of stories (and other text) as a two-mode graph. This method has some advantages over traditional one-mode semantic networks, but has the potential drawback (shared with n-gram text networks) that it contains paths that are not present in the text. An empirical study was devised using a false memory paradigm to determine whether these induced paths are remembered as being true of a set of stories. Results indicate that participants report false memories consistent with the induced paths. Implications for further research and two-mode semantic representations are discussed.

Abstract: Models of discourse and narration elaborated within the classical compositional framework have been characterized as bottom-up models, according to which discourse analysis proceeds incrementally, from phrase and sentence local meaning to discourse global meaning. In this paper we will argue against these models, suggesting that the assessment of discourse coherence is a top-down process, in which the construction of a situational interpretation at the global meaning level guides local meaning analysis. We will further explore the role of executive functions in coherence's establishment suggesting that, compared to other models of discourse processing which are focused on comprehension, our model may be a viable candidate for an integrated account of discourse comprehension and production.

Abstract: In this paper, we describe the narrative ontological model encompassed in the Labyrinth system. The aim of the system is to allow users to explore a digital archive by following the narrative relations among the resources contained in it. Targeted at cultural heritage applications, the Labyrinth project relies on the notion of “cultural archetype”, i.e., a core representation encompassing archetypical stories and characters, exploited as a conceptual framework for the access to archives of heterogeneous media objects.
In particular, we describe how the system leverages various types of ontological reasoning to let narrative relations emerge between artworks, and exemplify how these relations are exploited by the system to provide the user with a narrative conceptual framework she or he is familiar with in the exploration of the archive.

Abstract: Interested in formally modelling similarity between narratives, we investigate judgements of similarity between narratives in a small corpus of film reviews and book–film comparisons. A main finding is that judgements tend to concern multiple levels of story representation at once. As these texts are pragmatically related to reception contexts, we find many references to reception quality and optimality.

We conclude that reviews cannot inform current formal models of narrative and vice versa, but that other models may be necessary and that overall, narrative similarity must be modelled in task-related pragmatic terms.

Abstract: The understanding of story variation, whether motivated by cultural currents or other factors, is important for applications of formal models of narrative such as story generation or story retrieval. We present the first stage of an experiment to elicit natural narrative variation data suitable for evaluation with respect to story similarity, to qualitative and quantitative analysis of story variation, and also for data processing. We also present few prelimary results from the first stage of the experiment, using Little Red Riding Hood and Romeo and Juliet as base texts.

Abstract: The semi-formal analysis of Russian folk tales carried out by Vladimir Propp has often been used as theoretical background for the automated generation of stories. Its rigour and its exhaustive description of the constituent elements of Russian folk tales, and the enumeration of the patterns they follow, have acted as inspiration for several story generation systems, both sequential and interactive. Yet most of these efforts have attempted to generalize Propp's account to types of stories beyond the corpus that it arose from. In the process, a number of the valuable intuitions present in the original work are lost. The present paper revisits Propp's morphology to build a system that generates instances of Russian folk tales. Propp's view of the folk tale as a rigid sequence of character functions is phrased as a grammar that can be employed as a plot driver. Unification is used to incrementally build a conceptual representation of discourse by adding to an ongoing draft story actions that instantiate the character functions. Story actions are defined by pre and post conditions on the state of the plot to account for the causal relations crucial to narrative. The potential of the resulting system for providing a generic story generation system is discussed and possible lines of future work are discussed.

Abstract: Narratives are often used to form, convey, and reinforce memberships in social groups. Our system, called Chimeria, implements a model of social group membership. Here, we report upon the Chimeria Social Narrative Interface (Chimeria-SN), a component of the Chimeria system, that conveys this model to users through narrative. This component is grounded in a sociolinguistics model of conversational narrative, with some adaptations and extensions in order for it to be applied to an interactive social networking domain. One eventual goal of this work is to be able to extrapolate social group membership by analyzing narratives in social networks; this paper deals with the inverse of that problem, namely, synthesizing narratives from a model of social group membership dynamics.

Abstract: The ability to identify similarities between narratives has been argued to be central in human interactions. Previous work that sought to formalize this task has hypothesized that narrative similarity can be equated to the existence of a common summary between the narratives involved. The present work offers tangible psychological evidence in support of this hypothesis. Human participants in our empirical study were presented with triples of stories, and were asked to rate: (i) the degree of similarity between story A and story B; (ii) the appropriateness of story C as a summary of story A; (iii) the appropriateness of story C as a summary of story B. The story triples were selected systematically to span the space of their possible interrelations. The gathered empirical evidence overwhelmingly supports the position that the higher the latter two ratings are, the higher the first rating also is. Thus, while this work does not purport to formally define either of the two tasks involved, it does argue that one can be meaningfully reduced to the other.

Abstract: We present a web-based environment – an Ethics Workbench – which allows a reader’s ethical judgments to be solicited while reading a narrative. Preliminary results show generally consistent scores across subjects and test conditions, and suggest that it is possible to measure how individual readers respond to texts in terms of ethical judgments, how the linearity inherent in narrative plays a role in affecting ethical judgments, and how readers appear to synthesize judgments over the course of a text. Applications of the model include the empirical analysis of the ethical aspects of reading, the more detailed study of ethical issues, the potential for eliciting ethical discussions, and a means of dynamically planning texts to achieve maximum effect with respect to reader judgments.

Abstract: Developing a coherent computational model for narratives across multiple cultures raises the question of the components and structure of a framework within which African narratives can be conceptualised and formalised. It is well known that narratives are influenced by cultural, linguistic, and cognitive factors. We identify and define entities, elements, and relations necessary for the adequate description of Yorùbá narratives. We also discuss these theoretical issues in the context of designing a formal framework that is amenable to computational modelling.

Abstract: Although narratives often contain detailed descriptions of space and setting and readers frequently report vividly imagining these story worlds, evidence for the construction of spatial representations during narrative processing is currently mixed. In the present study, we investigated 7-year-old children’s ability to construct spatial representations of narrative spaces and compared this to the ability to construct representations from non-narrative descriptions. We hypothesized that performance would be better in the narrative condition, where children have the opportunity to construct a multi-dimensional situation model built around the character’s motivations and actions. Children listened to either a narrative that included a character traveling between 5 locations in her neighbourhood or a description of the same 5-location neighbourhood. Those in the narrative condition significantly outperformed those in the description condition in constructing the layout of the neighbourhood locations. Moreover, regression analyses revealed that whereas performance on the narrative version was predicted by narrative comprehension ability, performance on the description version was predicted by working memory ability. These results suggest the possibility that building spatial representations from narratives and non-narratives may engage different cognitive processes.

Abstract: Considering the existence of “narrative DNA”, we used sequence mining methods used in bioinformatics to learn more about the nature of tale types as a corpus. The complete Aarne-Thompson-Uther catalogue (2249 tale types in 7050 variants) was listed as individual motif strings based on the Aarne-Thompson motif index and scanned for similar subsequences. Next, using machine learning algorithms, we built and evaluated a classifier which predicts the tale type of a new motif sequence. Our findings indicate that the classification model was able to predict magic tales, novelles and jokes to the best extent. The probabilistic scaffolding of tale types is suitable for this type of analysis.

Abstract: This paper models narrative as a complex adaptive system in which the temporal sequence of events constituting a story emerges out of cascading local interactions between nodes in a social network. The approach is not intended as a general theory of narrative, but rather as a particular generative mechanism relevant to several academic communities: (1) literary critics and narrative theorists interested in new models for narrative analysis, (2) artificial intelligence researchers and video game designers interested in new mechanisms for narrative generation, and (3) complex systems theorists interested in novel applications of agent-based modeling and network theory. The paper is divided into two parts. The first part offers examples of research by literary critics on the relationship between social networks of fictional characters and the structure of long-form narratives, particularly novels. The second part provides an example of schematic story generation based on a simulation of the structural balance network model. I will argue that if literary critics can better understand sophisticated narratives by extracting networks from them, then narrative intelligence researchers can benefit by inverting the process, that is, by generating narratives from networks.

Abstract: Personal narratives typically involve a narrator who participates in a sequence of events in the past. The narrator is then present at two narrative levels: (1) the extradiegetic level, where the act of narration takes place, with the narrator addressing an audience directly; and (2) the diegetic level, where the events in the story take place, with the narrator as a participant, usually the protagonist. Although story understanding is commonly associated with semantics of the diegetic level (i.e. understanding the events that take place within the story), personal narratives may also contain important information at the extradiegetic level that frames the narrated events and is crucial for capturing the narrator's intent. We present a data-driven modeling approach that learns to identify subjective passages that express mental and emotional states of the narrator, placing them at either the diegetic or extradiegetic level. We describe an experiment where we used narratives from personal weblog posts to measure the effectiveness of our approach across various topics in this narrative genre.

Abstract: The challenge of narrative automatic generation is to produce not only coherent, but interesting stories. This study considers the problem within the Simplicity Theory framework. According to this theory, interesting situations must be unexpectedly simple, either because they should have required complex circumstances to be produced, or because they are abnormally simple, as in coincidences. Here we consider the special case of narratives in which characters perform actions with emotional consequences. We show, using the simplicity framework, how notions such as intentions, believability, responsibility and moral judgments are linked to narrative interest.

Abstract: This paper seeks to investigate some of the defining elements of narrative. The underlying assumption of my discussion is that the terms “narrative” and “story” do not refer to clearly defined, self-enclosed genres. Rather, they are part of a spectrum which embraces all forms of texts. Similarly, narratives and stories are not independent discourses but rather are an integral part of virtually all forms of discourse, be it day-to-day conversation or more specialized discourses. In order to analyze the relationship between narratives and other modes of discourse, we introduce the concept of narrativity. Narrativity refers to a collection of textual attributes. All texts exist along a continuum of greater or lesser narrativity, depending on the number and prominence of the narrative attributes they contain. When we refer to a text as a story, we mean that it contains a critical mass of narrativity. Most theorists of narrative have defined narrativity purely in terms of “dynamism”—that is, the extent to which a text portrays transition and change. To this I have added the quality of “specificity.” Specificity refers to the extent to which a text focuses on a particular time or place, a unique event, or individual people and objects. Many if not most texts contain a certain degree of narrativity.
We established, however, that in order to be considered a story the text must present a sequence of at least two interrelated events that occurred once and only once in the past. In other words, a story must have a certain degree of dynamism in that it portrays the transition from at least one event to another.
It must also have specificity at least to the degree that the text narrates events that happened at a fixed time in the past. This theoretical framework allows us to chart the relationship between different types of texts within a single discourse. It also gives us a vocabulary for discussing different parts of more complex narratives which often contain elements of varying narrativity.
The paper then goes on to discuss the concept of narrative structure, arguing that narrative structure is not an inherent attribute of narrative texts but a framework that the reader imposes on the text in order to make it intelligible in terms of other narratives. The structure which the reader abstracts from a given narrative will be heavily dependent on the context of the narrative with in a wider discourse.

Abstract: In this paper we present SNACS, a novel method for creating Social Narratives that can be Adapted using information from Crowdsourcing. Previous methods for automatic narrative generation require that the primarily author explicitly detail nearly all parts of the story, including details about the narrative. This is also the case for narratives within computer games, educational tools and Embodied Conversational Agents (ECA). While such narratives are well written, they clearly require significant time and cost overheads. SNACS is a hybrid narrative generation method that merges partially formed preexisting narratives with new input from crowdsourcing techniques. We compared the automatically generated narratives with those that were fully created by people, and with those that were generated semi-automatically by a stateof-the-art narrative planner. We empirically found that SNACS was effective as people found narratives generated by SNACS to be as realistic and consistent as those manually created by the people or the narrative
planner. Yet, the automatically generated narratives were created with much lower time overheads and were significantly more varied, making them more suitable for many applications.

Abstract: One of the approaches to generate narrative consists in modeling narrative in terms of a deep structure, as introduced by narrative theories in the middle of the 20th century.
This papers revisits this computational approach, and raises the central issue of dramatic tension: Would it be possible to build a computational model of dramatic tension, where tension could be managed according to the well known ascending/descending dramatic curve?
The paper describes a new computational model of narrative, based on a set of structural narrative elements (goals, tasks, obstacles, side-effects), a hierarchical and modular approach, a paradox-based model of dramatic tension and a solution for managing endings.
The papers illustrates this theoretical model with a full example.

Abstract: Multi-authoring is currently a common practice in the field of contemporary storytelling but producing consistent stories that share a common narrative space when multiple authors are involved is not a trivial task. Inconsistencies, which are not always well-received by readers are sometimes expensive to fix. In this work we attempt to improve the consistency of stories and narrative spaces by introducing a set of rules based on a formal model. Such a model takes into account the reader’s concept of consistency in storytelling, and acts as a framework for building tools to construct stories grounded in a common narrative space with a reinforced sense of consistency. We define a model (the Setting) and deploy it through a tool (CrossTale); both based on previous research, and discuss some user evaluation, with an in-depth analysis of the results and their implications.

Abstract: In the emergent narrative approach to Interactive Storytelling, narratives arise from the interactions between player- or computer-controlled characters in a simulated story world. This approach offers much freedom to the players, but this freedom may come at the cost of narrative structure. In this paper we study stories created by children using a storytelling system based on the emergent narrative approach. We investigate how coherent these stories actually are and which types of character actions contribute the most to story coherence, defined in terms of the causal connectedness of story elements. We find that although the children do produce goal-oriented story lines, overall the stories are only partially coherent. This can be explained by the improvised nature of the children's storytelling with our system, where the interactive experience of the players is more important than the production of a coherent narrative. We also observe that the communication between the children, external to the system, plays an important role in establishing coherence of the created stories.

Abstract: Audiovisual collections of narratives about war-traumas are rich in descriptions of personal and emotional experiences which can be expressed through verbal and nonverbal means. We complement a commonly used verbal analysis with a nonverbal one to study emotional developments in narratives. Using automatic text, vocal, and facial expression analysis we found that verbal emotional expressions do not correspond much to nonverbal ones. This observation may have important implications for the way narratives traditionally are being studied. We aim to understand how different modes of narrative expression relate to each other, and to enrich digital audiovisual interview collections with emotion-oriented tags.

Abstract: In legal cases, stories or scenarios can serve as the context for a crime when reasoning with evidence. In order to develop a scientifically founded technique for evidential reasoning, a method is required for the representation and evaluation of various scenarios in a case. In this paper the probabilistic technique of Bayesian networks is proposed as a method for modeling narrative, and it is shown how this can be used to capture a number of narrative properties.

Bayesian networks quantify how the variables in a case interact. Recent research on Bayesian networks applied to legal cases includes the development of a list of legal idioms: recurring substructures in legal Bayesian networks. Scenarios are coherent presentations of a collection of states and events, and qualitative in nature. A method combining the quantitative, probabilistic approach with the narrative approach would strengthen the tools to represent and evaluate scenarios.

In a previous paper, the development of a design method for modeling multiple scenarios in a Bayesian network was initiated. The design method includes two narrative idioms: the scenario idiom and the merged scenarios idiom. In this current paper, the method is extended with a subscenario idiom and it is shown how the method can be used to represent characteristic features of narrative.