Using Communicative Acts to Plan the ...

Using Communicative Acts to Plan the Cinematographic Structure of Animations Kevin Kennedy and Robert E. Mercer Cognitive Engineering Laboratory, Department of Computer Science The University of Western Ontario, London, Ontario, Canada [email protected], [email protected]

Abstract. A planning system that aids animators in presenting their cinematic intentions using a range of techniques available to cinematographers is described. The system employs a knowledge base of cinematic techniques such as lighting, color choice, framing, and pacing to enhance the expressive power of an animation. The system demonstrates the ability to apply cinematography knowledge to a pre-defined animation in a way that alters the viewer’s perceptions of that animation. The tool is able to apply cinematography techniques to communicate information, emotion, and intentions to the viewer. The application of cinematography knowledge to animations is performed by a communicative act planner that draws on techniques from Rhetorical Structure Theory (RST). The RST-guided planning paradigm generates coherent communicative plans that apply cinematography knowledge in a principled way. An example shows how the system-generated cinematography structure can enhance the communicative intent of an animation.

1

Introduction

In our research we are building a system that will aid animators in presenting their cinematic intentions using the large range of techniques which are available to cinematographers. An experienced animator can use techniques far more expressive than the simple presentation of spatially arranged objects. Our system employs a knowledge base of cinematographic techniques such as lighting, color choice, framing, and pacing to enhance the expressive power of an animation. By harnessing cinematography information with a knowledge representation system, the computer can plan an animation as a presentation task and create cinematography effects on behalf of the animator. The system described in this paper contains a knowledge base of cinematography effects and tools, which is employed by a planner to automate the presentation of animations. The planner reasons directly about the communicative intent that the animator desires to express through the animation. Utilizing this system, an animator can take an existing animation and ask the computer to create a communicative plan that uses cinematography techniques to reinforce the content of the animation, or create new meanings which

This research was funded by NSERC Research Grant 0036853.

R. Cohen and B. Spencer (Eds.): AI 2002, LNAI 2338, pp. 132–146, 2002. c Springer-Verlag Berlin Heidelberg 2002

Using Communicative Acts to Plan the Cinematographic Structure

133

would otherwise not be expressed. The cinematography planning system acts to increase the visual vocabulary of an animator by acting as an expert assistant in the domain of cinematography technique. The computer program that has been implemented as part of this research does not create animations, but works as a semi-automated tool to assist in generating an animation. The techniques presented here are intended to enhance the communication skill of a inexperienced animator when working in this medium or to assist the communication skill of an experienced animator by presenting multiple cinematographic presentation choices. 1.1

Animation

Animation is the process of generating a motion picture one frame at a time using methods such as drawing, photographing, photocopying, or clay modelling. The illusion of smooth motion is created by making slight alterations to each successive frame in a series and playing them quickly in their proper sequence. For much of its history, animation has been a painstaking task heavily burdened with manual labour. Computers have been applied to animation to help automate some of the more mechanical aspects of animation production with some success [15]. A successful area for applying computers to animation is the art of what is generally called computer animation. In a computer animation, the computer takes over the entire act of synthesizing the images, though humans generally control the image content. In this sense computer animation refers to the use of computer-generated images in forming an animation. Computer animation’s success is the result of mature computer graphics algorithms for synthesizing (or rendering) images from geometric models. Today it is possible to render very realistic images from a geometric model of the objects, backgrounds, lighting, and atmospheric effects present in a scene. However, rather than being interested in these aspects of computer animation, we are interested in communicating information, emotions and themes that the human director would like to convey with their animation. 1.2

Cinematography as a Means of Communication

Cinematography is the art of photographing a moving film image. The art encompasses placing the camera, choosing the film and colour filters, and controlling (and compensating for) the lighting. The cinematographer, usually referred to as the director of photography, controls the equipment that captures light onto the film media and manipulates the visual look of the film. In the realm of computer animation, the mechanism of filming is significantly simplified. A virtual camera can be operated without care for shutter speed, focus, or physical size and weight (of course these things could be simulated if so desired). Inside the computer there are no cloudy or sunny days, and the cameraman never makes mistakes. This lack of physical constraints reduces the role of the virtual cinematographer to its very essence: controlling the visual message of the film.

134

Kevin Kennedy and Robert E. Mercer

Viewer Attention The most basic method of communication with cinematography is directing the viewer’s attention. When telling a story visually, it is important that it be easy for the viewer to follow the action. There are many ways that the film-maker can do this, including directed lighting, colour contrasts, camera focus, and framing. By strongly lighting important objects, for example, the film-maker can restrict the viewer’s attention. Information The use of cinematography to provide information to the viewer is also pervasive. Many effects are used by film-makers to tell the viewer something without any direct statements. In a novel, such information would be provided by the narrator, but in film it can be presented in a more subtle, subliminal way. The size and weight of objects is implied by the framing of objects on the screen. A tall character is often shot from a low position to enhance her height or menace. The time of day is indicated by the lighting quality of the scene. Passing time is communicated through slow dissolves and fade-outs. Themes and Emotion The emotional predisposition of a viewer can be altered by applying the proper visual effect. Though the psychological effects of film techniques are hard to quantify, they are still relentlessly sought by film-makers. There is probably a certain level of learned response involved, as audiences are repeatedly exposed to the same stimuli in similar situations [20]. The broad approaches to setting mood are well understood. To set up a happy scene, one should use bright lights and vivid colours to create a “feeling-up” mood. A sad, dramatic, or scary scene should use low-key, “feeling-down” effects. De-saturation of light and colour can be used to draw viewers into the scene, or high saturation can make them feel like outside observers. These techniques are applied constantly in the medium to give films a specific mood.

2

Achieving a Communicative Act

The way in which cinematography can be used to communicate is best shown with an example. Figures 1 and 2 demonstrate how the manipulation of an animation’s cinematography can enhance the story and provide the viewer with more information. Figure 1 shows two still frames from a simple animation. “Superball” and “Evilguy” are squared off in a tense confrontation. Superball sizes up the situation and concludes that discretion is the better part of valour and thus makes a hasty exit to stage right. Figure 1 shows how this animation can be shown with a dispassionate outside observer’s point-of-view. However, cinematography gives us better tools to present this animation. By changing the cinematography, the animator can add information, emotions, and drama to this simple scene. Figure 2 shows an alternative presentation of the animation that expresses several communicative acts through cinematography. Figure 2(a) sets up the scene by showing the confrontation in screen center with a more dramatic lighting effect than Figure 1(a). Figure 2(b) gives us a


(a)

(b)

Fig. 1. Original Flat Animation

(a)

(b)

(c)

(d)

Fig. 2. Communicative Act Animation

135

136


close-up of Evilguy who is looking suitably menacing. Superball’s reaction, and his uncomfortable feelings are shown in Figure 2(c). Finally, Superball beats his retreat which is enhanced by showing him zooming off into the distance (Figure 2(d)). Figures 2(a)–(d) are sample frames from all four shots of the new cinematographically-enhanced animation. These frames were generated automatically by our cinematography planning tool. The only user inputs were a geometric description and a set of communicative acts to be achieved by the animation. To create the enhanced animation the planner was told to communicate the following information to the viewer: – – – –

Superball is diminutive and frightened Evilguy is domineering the scene is dramatic Superball makes a hasty exit to stage right

The planning system created the output animation shown in Figure 2 based on this limited set of inputs. In Figure 2(a), drama is created by stark lighting choice, and a slightly pulledback camera. Figure 2(b) shows Evilguy’s domineering nature by placing the camera close to him, looking upwards. The effect is further enhanced by positioning him to touch the top edge of the frame. Superball’s reaction of fear is shown through a harsh spotlight from above which casts a dark shadow beneath him. His short stature is shown with a high camera looking down. Finally, the exit to stage right is enhanced by aligning the action with the z-axis (into the frame) as Superball recedes into the distance.

3

System Description

The implemented system described here is meant to be used as an interactive tool to refine the cinematography plan for an animation. A pre-defined description of character action is interactively re-filmed to achieve the animator’s requested communicative acts. The animator can change the communication goals or request alternate cinematography plans to perfect the animation. The current system, which is intended as a proof of concept, has a traditional AI architecture consisting of a knowledge base, a planner, and an acting agent, the latter being a graphical renderer in this case. A real animation production environment which uses a mixed initiative strategy [11] would be an obvious next step. The current system has been shown to work on several examples of short action sequences, and is able to recreate scenes that are similar to pre-existing animations. The knowledge base stores knowledge about space, time, solid objects, lights, colours, cameras, scenes, shots, and cinematography effects. This knowledge is implemented using the language LOOM [6]. The planner creates a plan that implements the desired communicative acts. The renderer transforms the cinematography plan created by the planner into a sequence of animation frames which constitute the output of the program. The renderer makes use of the POVray ray-tracing system to create the graphical images.


137

Fig. 3. Module Interactions

Figure 3 shows the interactions between the various modules. The knowledge base acts as a source of information about cinematography, and as a store of knowledge while the animation plan is being built up. It is also used to store the planning rules that the planner assembles to create a solution. Hence the interaction between the planner and the knowledge base is bi-directional. The renderer, on the other hand, only retrieves information from the knowledge base, it does not add knowledge. There is also a direct connection from the planner to the renderer because the planner provides the renderer with the complete scene description that is to be rendered. Finally, the renderer outputs data to the POVray ray-tracer, which is an external program. The virtual director of photography acts as a tool to be used by a human operator or director. The human and computer work together as a team, just like the cooperation between a film director and her director of photography on a movie set. The director must do a great deal of the work involved with specifying the action, character blocking, and narrative. The human must also express her narrative goals in terms of the communicative acts that the system understands. The computer will, however, assemble these elements together, position the camera and lights, and generate the sequence of images that create the animation. Since this is meant as a semi-automated approach, the director has the capability to overrule the computer and tell it to keep searching for a better cinematography “solution”. 3.1

Cinematography Knowledge Base

The knowledge representation language used for this paper (LOOM) is a description logic that allows a programmer to use aspects of several knowledge representation schemes. LOOM is primarily a frame-based language; however it encodes knowledge in a hierarchy of concepts and roles (also known as objects and relations) that is very similar to conceptual graphs. Concept definitions can contain simplified predicate logic statements to provide a limited element of

138


logic-based representation. LOOM also provides the ability to encode rules and actions just like a production system. The knowledge base is our attempt to capture the “common sense” of cinematography. The following areas are represented in the knowledge base: – – – – – –

cameras, camera positions, field of view lights, fill lights, spot-lights, lightsets colors, color energies, color saturation scenes, foreground/background, stage positions spatial objects/relationships, 3D vectors, occlusion moods and themes; color/light effects to present them

The source we have chosen for our common-sense knowledge is a Film Studies textbook by Herbert Zettl [20]. This book is written as an introduction to cinematography and contains many rules which lend themselves to our knowledge representation technique. Figure 4 shows an example of some of the knowledge presented by Zettl in several chapters on cinematography lighting. In this figure we have broken down the techniques described into their major classifications, arranging them from left to right according to the visual “energy” they convey. The terms written below each lighting method are the thematic or emotional effects that Zettl associates with these techniques. It is these effects that the animator can select when constructing a scene with our program. It should be noted that, though the lighting techniques are modelled in detail by the knowledge base, the thematic and emotional effects are not. For example, there is no attempt to model the meaning of “wonderment” in the knowledge base. These are simply text tags which are meaningful only to a human operator. In addition to lighting techniques, the knowledge base represents camera effects like framing, zooms, and wide-angle or narrow-angle lenses. Colour selections for objects and backgrounds, as well as their thematic meanings, are also contained in the knowledge base. These three major techniques (lighting, colour, and framing) can be used to present a wide variety of effects to the viewer. The actions that make up the animation can be any combination of move, jump, stretch, squash, turn, and tilt. This allows expression of simple cartoonlike animations. Cinematography Techniques To accomplish the task laid out above, the computer cinematographer must have a knowledge of cinematography and must be able to apply it. The types of cinematography knowledge the system contains is described below. Lighting Lighting is used to set mood, direct viewer attention, and provide information. The computer cinematographer can apply lighting to characters and backgrounds independently. The quality of lighting can be adjusted to alter the amount and sharpness of shadows. The brightness and direction of lighting is changed to achieve communicative acts as required.


Fig. 4. Semantic Deconstruction of Zettl’s Lighting Models

139

140


Colour The computer has limited control of colour for scene objects. When object models are created they are also created with colour sets which can be applied to them before rendering. The colour sets fall into several overlapping classes of colour energy and saturation. The system can select a specific colour set which satisfies constraints imposed by the director’s communicative goals. The system described here does not contain a general model for the aesthetics of colour, but relies on the the programmer to classify colours in terms of energy, temperature, and saturation. Camera Placement The computer director of photography takes ownership of the virtual camera and its point of view. Given a scene containing background objects and characters, the system will orient the camera in a way that achieves the desired effects. The system presented here can only function with objects that are “well behaved”. The analogy used is that of a stage or small set. The computer can deal with objects that move through different positions on this small set, and arrive at proper camera placement solutions. An animation that involves large sweeping movements, interaction of convoluted objects, or highly constrained environments cannot be expected to work correctly. Framing Closely related to camera placement is the framing of objects within the two dimensional field. When prompted by the director’s communication goals, the computer will attempt to frame objects in certain zones of the screen to achieve corresponding visual effects. Shot Structure In what is a step outside of the duties of a director of photography, the system takes on some of the duties of a film editor. Given overall goals of pacing and rhythm, the computer will make decisions about where to place cuts in the film time-line. To assemble an overall viewer impression of the scene environment, the computer will assemble short sequences of shots that portray important objects and relationships within a scene. The director can choose either an inductive or deductive approach to shot sequencing. The animator must also supply information about which objects and characters are important and what meaningful relationships exist between these objects. Qualitative Representation and Reasoning Qualitative physics is the basis for the knowledge representation of most spatial qualities and physical measurements used in this paper [19]. For example, the size of an object is specified as being tiny, small, medium-sized, large, or very-large. Reasoning about where to place cameras is based upon these simplified measurements. Qualitative measurements are sufficient to capture the knowledge required for this domain. Though the use of standard quantitative representations would allow finer control of physical quantities and measurements, such greater accuracy would not benefit the reasoning process. The camera positioning really only needs to distinguish between close-up, medium, and long shots. Whether a camera is 4.0 or 4.1 units from a target is of little significance. Similarly, the


141

actual physical quantities like the number of lumens emitted by a light are not important; they only need to be bright, medium-strength, or dim. Much of the information presented in Zettl is qualitative in nature and lends itself to representation in this format [20]. For example, colour energies are discussed in terms of low-energy, neutral, and high-colour-energy. This type of knowledge is easily represented and reasoned about, using qualitative methods. 3.2

Rhetorical Structure Theory

In a strict sense, the system presented in this paper is concerned with the translation from communicative acts into animation instructions to be rendered. The communicative acts are the information, emotions, and themes that the human director would like to convey with their animation. The computer takes these communicative acts and transforms them into visual effects that can be rendered into digital images. The set of communicative acts understood by the computer cinematographer acts as a sort of vocabulary that the human animator/director can use to add to the basic action they have specified to take place in their animation. The communicative acts understood by the computer includes things like: – Show that character A is important – Increase viewer involvement – Promote viewer discomfort The task of the computer is to sort out competing goals, apply standard “default techniques”, and create final rendered images. Like a true director of photography, the computer acts as an assistant to the director. The tasks of directing the actors and creating the narrative are left to the human in charge. The computer takes control of the camera, colour, and lighting and presents final animations that comply with the director’s vision. In our research, Rhetorical Structure Theory is used to guide our representation and planning of communicative acts. Rhetorical Structure Theory (RST) is concerned with describing the structure of text [16]. The main focus of RST is that of textual coherence. A coherent text is one in which every segment of a text has a reason for being there. A coherent text consists of well-phrased sentences devoid of non-sequiturs and gaps. RST can describe the structure of any coherent text. When describing the structure of a text, RST uses a framework of nucleus-satellite relations. A nucleus is the central idea or fact that is being presented by a portion of text. The satellite is a secondary phrase or sentence that in some way supports the rhetorical purpose of the nucleus. These nucleus-satellite relations are called rhetorical relations. There are approximately 50 rhetorical relations recognized for textual analysis. The following are some examples of rhetorical relations (RR), nuclei (N), and satellites (S): – RR: Background; N: text whose understanding is being facilitated; S: text for facilitating understanding.

142


– RR: Elaboration; N: basic information; S: additional information. – RR: Preparation; N: text to be presented; S: text which prepares the reader to expect and interpret the text to be presented. – RR: Contrast; N: one alternate; S: the other alternate. The above description of RST shows its application as a tool for analyzing text; however, it is also useful for generating text. For example, the ILEX system [17] creates hypertext descriptions of objects in a virtual museum tour. This paper, however, is not concerned with the generation of texts; it is concerned with the generation of visual effects to communicate with the animation viewers. RST is used as a way to connect the wishes of the animator with the actions of the renderer. It provides a methodology for transforming intent into action. The communicative acts are not comprised of sentences, but are assembled from the structure and presentation of the scene. As an example of how RST concepts can be used to plan a scene, consider the RST relation elaborate. Elaboration occurs when one or more pieces of text present additional detail about another portion of text. Our animation planner would use this approach when trying to present a non-obvious concept. For example, when asked to show that a character is unhappy, the planner could elaborate by using both dim lighting and a muted color scheme around the character. Other systems have also used RST in the visual domain. André and Rist [1] used RST theory to aid in the creation of diagrams. These diagrams were intended to provide instruction to people in the operation of mechanical devices. In this case RST was used because it gave a structure for reasoning about communicative acts. Each image in an instruction manual has a very specific communication intent; for example, “turn this knob” or “pay heed to this indicator”. RST provides a straightforward structure for codifying these concepts. The transformation to the visual domain requires adapting many rhetorical relations and creating new ones where necessary. Transforming RST to the area of animation requires three analogies to be created between text and animations: (1) Author ⇒ Animator. (2) Text ⇒ Images. (3) Phrases ⇒ Scene Presentation and Structure. These analogies allow us to apply RST principles to the concepts of visual communication. The scene presentation and structure being referred to here are specifically the techniques outlined previously as being a part of cinematography. RST provides a structure for reasoning about high-level communicative acts, and producing coherent visual “texts” for performing these acts. 3.3

RST/Planning Integration

The planner itself is a forward-chaining planner. Planners were initially envisioned as action planners for agents interacting with the real world. The planning performed for this paper is somewhat like the planning that is required to produce a paragraph of text. Although a paragraph of text does unfold over time, the arrangement of sentences is guided more by the requirements of logic and rhetoric than by the strict requirements of cause and effect.


143

A planner works by assembling discrete plan steps together in a logical framework that achieves its goals. The plan steps are actually RST-style rhetorical relations. The planner works by combining these relations into an RST tree that achieves the communicative act “goals” of the animator who is presenting the animation to the viewer. The RST plan is thus a rhetorical structure that implements a plan of action to communicate something to the viewer. The main task of the planner is to create a plan that does not contain contradictions. The communicative acts requested by the animator are matched to high-level RST plan steps that have the desired effect when implemented. Higherlevel RST plan steps will contain lists of lower-level plan steps that are required to attain the goal of the higher-level plan step. These lower-level plan steps are usually expressed as AND/OR/SOME combinations, meaning all/one/morethan-one lower-level plan steps must be successful for the higher level plan step to be considered accomplished. At some point, the RST-style plan steps give way to direct cinematography actions that must be attempted. At this point, a contradiction is easily encountered because of conflicts with earlier branches of the RST plan tree. For example, an earlier plan step might have called for a calming mid-left screen placement for a character, but a new plan step calls for placement near a screen edge. The planner must prevent these conflicts from taking place and must circumvent them by backtracking. Although the planner does act at two distinct levels, that is, the RST plan step level and the cinematography action level, the planner cannot be considered a hierarchical planner. This is because the lowest level of the tree, a cinematography action, does not actually require any planning. Cinematography actions are called for by RST plan steps and they must be carried out for the RST plan step to be fulfilled. Hence all actual planning is restricted to the RST planning and does not occur for the lower levels of the plan tree. Future work in this area may necessitate a move to an explicitly hierarchical planner that can plan about communicative acts at the levels of narrative and acting as well as cinematography.

4

Related Work

Much recent work has investigated the problem of automatically placing a camera to view a 3D environment. Gleicher and Witkin built a system that allows the user to control the camera by placing constraints on the image that the camera should produce [12]. Philips et al. created a system that integrates camera movement with the movement of virtual human characters so that the camera views the tasks undertaken by the virtual humans. The CINEMA system developed by Drucker et al. provides a language for users to specify camera movements procedurally [10]. Bares et al. have written several papers on the problems of positioning a camera correctly in complex 3D worlds [2,4]. They use a real-time constraint solver to position a camera in a complex interactive 3D world in a way that views characters and fulfills a cinematic goal. Another approach they have taken is to model the user’s preferences to create a user-optimal camera place-

144


ment to view a virtual 3D world [3]. Bares et al. also use a constraint solver to find a solution to various user-imposed camera viewing requirements [5]. Halper and Olivier use a genetic algorithm to find a good camera placement in highly constrained situations [13]. Christiansen et al., using established camera placement idioms from film textbooks, created a language called the Declarative Camera Control Language (DCCL) for encoding cinematography idioms [8], for example, a three shot sequence depicting one actor approaching another. They also created a virtual cinematographer that used a finite state machine approach to operationalize cinematography idioms used in filming conversations [14]. Tomlinson et al. [18] integrate a cinematography agent into a virtual animated agent environment. This cinematography agent uses lighting changes and camera angles to reflect emotions of the virtual agents. Butz [7] created a system to create short animations to explain the function of technical devices. Communicative goals are achieved using effects like pointing the camera and shining spotlights to highlight certain components and actions. Whereas these other works deal with automation of technical aspects of cinematography, the system that we present combines these techniques to bolster the communicative nature of cinematography. Some of these systems could be integrated with the system outlined in this paper in a way that would enhance the generality of the system to handle more difficult camera placement requirements. The camera placement idiom research is applicable to this paper in that it provides a general way to represent cinematography idioms, a task that is not attempted in this paper. This paper makes use of idioms that achieve certain communicative goals and applies them directly, without concern for a global representation of such idioms.

5

Conclusion and Future Work

This paper has described the area of computer animation, and how it bridges both traditional animation and real-life film. The approach taken in this paper is to take the cinematography techniques of traditional film, and apply them using automated methods in the computer graphics medium. Animation is fundamentally a form of communication between animator and viewer. By codifying the communicative acts that can be achieved with cinematography, we have created a tool that helps animators in enhancing the communicative power of their animations using cinematography techniques. In performing this task we have identified a large body of knowledge in cinematography textbooks that can be captured using a qualitative approach to knowledge representation. The applied techniques of cinematography are governed by discrete and concise rules that can be captured with modern AI techniques. The main task of the knowledge engineering lies in capturing the meaning of the visual vocabulary of cinematography. Our research achieves this in several important areas of cinematography.


145

Operationalization of this cinematography knowledge requires the ability to plan the application of the knowledge to the problem domain. By considering the act of creating an animation as a type of visual communication, the application of cinematography knowledge to real animations is performed by a communicative act planner that draws on techniques from Rhetorical Structure Theory. This planner allows an animator to enact specific communicative acts through applied cinematography. By using an RST-inspired planning paradigm, the animation assistant can generate coherent communicative plans that apply cinematography knowledge in a rational fashion. Future areas of research extending from this paper could involve: – Robust reasoning about temporal organization. The current system handles the arrangement of cuts between shots using a script-like system. It would be more interesting to reason more fully about the mechanisms of shot sequence structures and the influence of animation action on choosing cut timing. This aspect of the research would also require solving problems dealing with the narrative structure. Others have developed methods for dealing with space and time in image sequences (such as the logic developed in [9]). However, these methods are interested in media in which real time and media time are essentially equivalent. In cinematography time can be used to create effects or to map parallel events to sequential camera shots. These issues need to be captured in a temporal reasoner. – More robust camera placement. Other cinematography research has dealt with the issue of placing cameras in tightly constrained situations, and actively evaluating placements based on the output images. It would be interesting to coordinate this research with such a system to create a more general cinematography tool. Such an extension would also allow a wider range of actions or settings to be expressed. – Narrative and virtual characters. It is possible to imagine a system that integrates animation presentation issues with character actions, intentions and beliefs. The organization of these characters in front of the camera, all in the name of satisfying communicative acts, poses an interesting problem.

References 1. E. André and T. Rist. The Design of Illustrated Documents as a Planning Task. In M. T. Maybury, editor, Intelligent Multimedia Interfaces, pages 94–116. American Association for Artificial Intelligence, 1993. 142 2. W. H. Bares, J. P. Gregoire, and J. C. Lester. Realtime Constraint-Based Cinematography for Complex Interactive 3D Worlds. In Proceedings of the Tenth National Conference on Innovative Applications of Artificial Intelligence, pages 1101–1106. American Association of Artificial Intelligence, July 1998. 143 3. W. H. Bares and J. C. Lester. Cinematographic User Models for Automated Realtime Camera Control in Dynamic 3D Environments. In Proceedings of the Sixth International Conference on User Modelling, pages 215–226, June 1997. 144

146


4. W. H. Bares and J. C. Lester. Intelligent Multi-Shot Visualization Interfaces for Dynamic 3D Worlds. In Proceedings of the 1999 International Conference on Intelligent User Interface, pages 119–126, 1999. 143 5. W. H. Bares, S. Thainimit, and S. McDermott. A Model for Constraint-Based Camera Planning. In Smart Graphics: Papers from the 2000 AAAI Symposium, pages 84–91. AAAI Press, 2000. 144 6. D. Brill. LOOM reference manual version 2.0. Technical report, University of Southern California, Los Angeles California, 1993. 136 7. A. Butz. Anymation with CATHI. In AAAI’97 Proceedings of the 14th National Conference on Artificial Intelligence, pages 957–962. AAAI Press, 1997. 144 8. D. B. Christianson, S. E. Anderson, L. He, D. H. Salesin, D. S. Weld, and M. F. Cohen. Declarative Camera Control for Automatic Cinematography. In Proceedings of the Thirteenth National Conference on Artificial Intelligence and the Eighth Innovative Applications of Artificial Intelligence Conference, volume 1, pages 148– 155. AAAI Press/The MIT Press, 1996. 144 9. A. Del Bimbo and E. Vicario. Specification by-Example of Virtual Agents Behavior. IEEE Transactions on Visualization and Computer Graphics, 1(4):350–360, December 1995. 145 10. S. M. Drucker, T. A. Galyean, and D. Zeltzer. CINEMA: A System for Procedural Camera Movements. In Proceeedings 1992 Symposium on Interactive 3D Graphics. ACM, March–April 1992. 143 11. G. Ferguson, J. Allen, and B. Miller. TRAINS-95: Towards a Mixed-Initiative Planning Assistant. In Proceedings of the Third Conference on Artificial Intelligence Planning Systems, pages 70–77. American Association of Artificial Intelligence, May 1996. 136 12. M. Gleicher and A. Witkin. Through-the-Lens Camera Control. Computer Graphics, 26(2):331–340, July 1992. 143 13. N. Halper and P. Olivier. CAMPLAN: A Camera Planning Agent. In Smart Graphics: Papers from the 2000 AAAI Symposium, pages 92–100. AAAI Press, 2000. 144 14. L. He, M. F. Cohen, and D. H. Salesin. The Virtual Cinematographer: A Paradigm for Automatic Real-Time Camera Control and Directing. In Computer Graphics Proceedings, Annual Conference Series, pages 217–224. ACM, August 1996. 144 15. P. C. Litwinowicz. Inkwell: A 2 1/2-D Animation System. Computer Graphics, 25(4):113–121, July 1991. 133 16. W. C. Mann and S. A. Thompson. Rhetorical Structure Theory: Toward a Functional Theory of Text Organization. Text, 8(13):243–281, 1988. 141 17. C. Mellish, M. O’Donnell, J. Oberlander, and A. Knott. An Architecture for Opportunistic Text Generation. In Proceedings of the Ninth International Workshop on Natural Language Generation, 1998. 142 18. B. Tomlinson, B. Blumberg, and D. Nain. Expressive Autonomous Cinematography for Interactive Virtual Environments. In Proceedings of the Fourth International Conference on Autonomous Agents, pages 317–324. ACM Press, June 2000. 144 19. D. Weld and J. d. Kleer. Qualitative Reasoning about Physical Systems, ed. Morgan Kaufmann, 1990. 140 20. H. Zettl. Sight Sound Motion: Applied Media Aesthetics. Wadsworth Publishing Company, 1990. 134, 138, 141