A section from chapter 2 of my dissertation:
Metalinear Cinematic Narrative: Theory, Process, and Tool.

4.4.3 How to Interpret Feedback
As stated earlier, the writing process is a cyclic one. Rethinking and rewriting are the most time consuming tasks of the writing process. Agent Stories supports this process by offering the writer feedback about their story ­ feedback which the writer can then use to make necessary changes in the other environments. The feedback is in the form of short textual explanations for why the story agent made the choice it did for each clip.

The feedback text comes in two categories: explaining what the agent could not find, and explaining what the agent did find. The explanation text for each clip box can include text from both categories. Listed first is what the agent could not find, then what the agent could find. The text explaining what the agent could not find is provided in a statements such as:

- Could not find any Main POV Supportive Conflicts
- Could not find any Alternative Oppositional Character Intros

For each attempt and failure at finding a certain type of clip, the Agent Stories story engine provides feedback about it so a writer could go back and edit their story material, adding, for instance, more conflicts or character introductions.

The reporting of successes in the WFE comes with a little more detail and finesse. Since the goal is not just to blurt out negative comments, but also to provide the writer with detailed positive statements, the reporting of success happens as a concatenation of sentence parts depending on what was found. There are four parts to each success sentence: the opening, the number detail, the link attribute from the Representational Environment, and the primitive type from the Structural Environment.

There are two different success sentence openings: Found and Randomly chose. Found is used in three cases:

* When the agent was looking for one clip (i.e. a single supportive conflict) and found exactly one clip.
* When the agent was looking for a clip from a specific character (i.e. a speaker intro by Bob)
* When the agent was looking for multiple clips which satisfy a certain criteria and found them.

The opening phrase Randomly Chose is used only when the agent was looking for a single clip of a certain type and found many of them. The intention is to let the writer know that the agent could have made a number of different choices at that point. If the agent randomly chose a clip which has a strong bearing on what happens later in the story (like a conflict which is linked to specific resolutions), then this random choice signifies that this agent would very likely tell the story quite differently next time because the different conflict could also mandate a different resolution and thus a different story ­ a useful thing for the writer to know.

A success sentence can have one of three different number detail pieces. They are:

- this
- this one of X available
- this one of X requested

The number detail piece this is used simply in the case when a single clip was requested and a single clip was found; i.e. Found this...

The "X" in the second and third pieces refer to the number of found clips of the particular type that the agent was looking for. one of X available is used when the agent was looking for a single clip, but found more than one, whether the request was for a specific character or not. For example: Randomly chose this one of 3 available... or Found one of 2 available...

The number detail piece this one of x requested is used when many clips of a particular clip type are found according to the agent's behavior, and when the agent has found as many as it could. There is one clip box on the screen for each clip the agent finds. The agent will try to satisfy the number of requested clips and display as many as it finds up to the requested number. For example, in the success sentences under each of those clip boxes, the final line would read, Found this one of X requested... The writer would know how successful the agent was in satisfying this behavior by comparing the number requested to the number of clip boxes on the screen. Though there are no clip number restrictions, it appears that the number of clips requested would typically be small, on the order of 2 to 5.

The link attribute part of the success sentence refers to the type of link made between clips in the Representational Environment. It also refers specifically to the behaviors in the Agent Scripting Language as described in section 4.6. Since the story agents have behaviors for choosing a main POV for each construction and weaving their construction around that POV, their success sentences also reflect that main POV choice. Some examples of this sentence part are: Main POV Single Supportive, Alternative Multiple Oppositional, Main POV Single Oppositional, etc. Examples of the success sentence for the three parts up to this point would be:

- Found this Main POV Single Supportive
- Randomly chose this one of 3 Alternative Single Oppositional

The final piece of the success sentence is the primitive type that the agent was looking for, as stipulated by the active story framework in the Structural Environment. The same seven narrative primitive types that are in the Structural Environment are used here: Speaker Introduction, Character Introduction, Conflict, Negotiation, Resolution, Diversion, and Ending.

When all four of the success sentence parts are concatenated, they resemble the following examples:

1. Found this one of 2 requested Alternative Multiple Supportive Conflicts.
2. Randomly chose this one of 5 available Main POV Single Supportive Character Intros.

If a writer using the WFE received example 1 as feedback, it would mean that the agent was trying to find two conflicts supportive for the main POV character's conflict, but not actually belonging to the main POV character. This sentence would ideally be at the bottom of two clip boxes. If this sentence is on the bottom of just one clip box, then it means that the story agent was unsuccessful at finding both clips. The story agent's inability to find another clip is not listed as a failure however, because it is not a complete failure ­ it did find one of the clips.

If a writer using the WFE received example 2 as feedback, it would mean that the agent was trying to find only one character intro for the main POV character and instead found five. The agent then made a random choice between the five to produce the one chosen clip. This type of feedback in important because it is crucial for the writer to know what parts of the story he has a lot of material for and what parts remain lacking. If the agent is making a random choice between several clips in one part of the story, and able to find no clips for another part of the story, then the writer will be able to see where she should focus her efforts.

There is one remaining possible feedback which the Agent Stories system is able to give the writer ­ if no clip could be found for a particular purpose. In such a case, an empty clip box is placed on the screen in the proper sequence location where a successfully chosen clip would be. Underneath the clip box are listed all the appropriate failure sentences, indicating the sorts of clips the story agent tried to find but was unable to. The one piece of text inside the clip box, where the clip text would normally be, is the simple statement:

Could not find a clip for this XXXX.

The feedback simply tells the writer that a clip was not chosen and for lists the reasons why. The writer is then free to go back to the Representational Environment or the Structural Environment to make changes so this event will not happen again. The writer also has the other story agents available. Where one of the agents could not find a clip for a particular place in the story framework, one or more of the other agents could very well find a clip for that same place, as their behaviors have them making different choices. By polling the various agents or even by scripting their own agent and having that agent construct the story, the writer has the ability to see many different sides of their story and rework the different parts as they see fit.

The familiar cyclic writer's pattern of change­read­rethink-change-read­rethink is maintained in Agent Stories because the software agents output food for rewriting thought. The resulting process is one where the story evolves during the writing process. Not only does the text become better, as it should over any writing process, but the representation and the body of work becomes stronger as well. The details of the writing process and case study examples will be discussed in chapter 5, Authoring and Evaluation.