4 A Hybrid Visual Formalism for Authoring IN
• In this section, the formalism that responds to the requirements of interactive narration is developed. Cf. also [Iurgel03d], [Iurgel04a], [Iurgel04b], [Iurgel05c], [Iurgel06a]. It constitutes the theoretical basis of the authoring framework.
“hard-coded”, easy-to-create, excellent reactions to predictable input, e.g.
for utterances of virtual characters as reactions to certain user input, while algorithmic (e.g. a generic and generative dialogue manager) modules can be used when the prediction fails.
• Intuitive. Directed graph formalisms are intuitive enough to be employed by non-computer scientists.
• Concrete. Finite states facilitate concrete authoring, because usually a state of the system can be interpreted as a particular situation in the story, and the author can then define exactly what should happen next.
In this section, it will be shown how a directed graph basis can serve as starting point for an authoring environment for IN, given the current situation of a “wicked”
state of research, where technologies, application concepts and authoring me-thods have to be developed together, and are interdependent. The directed graph serves as a basis, since it is always possible, within the hybrid formalism, to han-dle system behavior with the directed graph, and insofar as a first intuitive, con-crete approach to a certain problem can be initially and experimentally handled with directed graphs. Directed graphs are also basic in the sense that the introduc-tion of a delegaintroduc-tion chain (cf. below, Secintroduc-tion 7.3) will allow to employ directed graphs as first choices of the system, in order to produce excellent reactions of the system when a predicted input occurs, and generic, less accurate reactions when the input differs from the expected flow. Thus, directed graphs serve as a visual formalism that can always be employed, in appropriate parts of the system, and that can also be used for prototyping.
The main strategy behind the decision to address complex sequence selection questions by starting with directed graphs is to allow for a cautious migration into the sphere of “wickedness” form a starting point that is not wicked at all: Directed graphs are well understood and extensively employed for varied interactive appli-cations, including games and educational applications. An incremental and intui-tive authoring of an interacintui-tive narraintui-tive must support a stepwise enlargement of the amount of possible states, of the level of complexity of permutations, and find-ing out of appropriate structures and algorithms durfind-ing the creative process. This responds to the fact that appropriate precise story algorithms and declarative
structures, e.g. in the Proppian vain [Propp], are normally, in IN, not known befo-rehand, at the beginning of the creation process.
A lesson learnt from the Façade system is that it is paradoxically necessary to handle sophisticated natural language interaction in IN with shallow methods. But Façade does not offer any visual support for authoring of shallow natural language processing. The hybrid formalism developed in the next sections is perfectly suited for this task, but at the same time it offers a visual concept, and allows for integrat-ing external, more generic talkintegrat-ing modules (cf. also Section 6, and [Iurgel05b]).
The “shallow” method is built upon predictability, whereas a “deep” method of NLP relies on principles. Predictability means that, in a given dialogue situation, a cer-tain set of reactions from the user can be expected, and these can be handled with different grades of precision, in order to obtain appropriate reactions of the virtual characters. Predictability is a fallible principle for NLP, as many sensible inputs are possible that will not be predicted, but it can often be employed rather efficiently.
Predictability is the reason why a use of directed graphs in dialogue management for IN is possible.
The novel Hybrid Control Formalism will thus allow for experimenting with se-quencing methods, for an incremental creation process, and for a concrete author-ing. It also supports content creation in its proper sense, where the author is not required to program, and teamwork of content creators, computer scientists, and content specialists.
Some details of the formalism may require adaptations on a project basis. In the projects described in Section 6, the only project-specific addition to the formalism has been a function to exchange memory state data, cf. Section 6.2.2. Neverthe-less, also modifications to the authoring formalism on the programming and defini-tion level must be regarded as a normal part of the development of a “wicked” de-velopment process on innovative IN applications.
It is necessary to stress that a generic formalism is not a generic solution to all authoring problems of IN. When the authoring team decides to employ sequence selection methods different form directed graphs, the framework developed in the
present section and in Section 5 offers appropriate interfaces, data slots, and processes for integrating these methods. It is generic in this sense of allowing for the use of any additional sequencing paradigm, with additional specific authoring support. No assumption is made with regard of these methods or their specific au-thoring process, besides the interface and data structure definitions.
The formalism can be employed for gradually testing and developing specific algo-rithms to IN sequence selection problems, departing from the proven concepts of visual formalisms. For simplicity, every control approach that is derived from finite-state machines, i.e. that employs a finite set of finite-states and transitions between them that define the control flow, will be called here the “directed graph” approach.
In the following, any method to determine the sequence of storytelling elements that is not based on directed graphs will be called, for simplicity, an “algorithmic method”, or “algorithmic approach”. For example, the choice of a subsequent scene according to a story engine that implements an Aristotelian model is an “al-gorithmic approach”.
Given the case of a limited set of possible states, current authoring tools based on directed graphs reach their limits when
• the complexity of the connections is too high
• the guard conditions become too complicated
The measure of what counts as “too complicated” depends much on the back-ground and talent of the author. Employing the Hybrid Control Formalism based on directed graphs developed here, it is possible to approximate desired complex system behavior, and to migrate to an algorithmic, non-graph based sequencing method when necessary and when possible, i.e. when the stage of a graph that is
“too complicated” is reached.
The authoring process that is assumed enabled by the Hybrid Control Formalism initiates at an almost linear story idea with few bifurcations, which is then gradually extended by further branches, until possibly the use of some algorithmic approach appears to be more appropriate. Thus, it is possible to start with the simple use
case, and to increase complexity and involvement of computer scientists stepwise, as required (Figure 18). Examples will be given below. When things start to get wicked, i.e. when the application idea and the necessary technology are not suffi-ciently clear, it is then possible to experiment with different ideas and technologies that are added to the formalism only when and at the places where this is required, whereas other parts of the envisaged application can still employ the standard graph based approach.
Figure 18 – The approximative and experimental authoring process that is enabled by the Hybrid Con-trol Formalism. Starting with few or no interactions, specified with directed graphs, complexity can be added stepwise, until the directed graph must be partially or totally replaced by algorithmic sequence selection methods.