Introduction
The research team headed by Dr Colette Daiute at the CUNY Grad Center is conducting a followup research project to their 2018 study, “Meta-Communication Between Designers and Players of Interactive Digital Narratives.” This study addresses quantitatively analysing “meta-communication” between designer and player of an IDN product. Interactive Digital Narrative (IDN): form of digital interactive experience in which users create or influence a dramatic storyline through their actions. During the IDN research workshop, research participants designed and played Twine IDNs and communicated with think-aloud protocol. Narrative Theory , as well as the narrative tree structure branch and connection counts were used to qualitatively and quantitatively relate the type of meta-communication and the complexity of the final narrative trees.
\[branch~density=\frac{branches}{nodes}\]
One of the quantitative metrics used, “branch density”, was found to be insufficient in capturing the nature of the narrative tree structure.
My role in this research project is to create a more robust narrative tree structure complexity metric that is capable of capturing the overall complexity of a narrative tree structure. This will allow researchers to quantitatively compare the effects of meta-commentary on the treatment and control groups.
knitr::include_graphics("SaraSternTwine1.png")
Much like a biological tree, algorithmic tree or a model of a molecule, the narrative tree structure changes size and organization as it grows or shrinks. The nature of this structure is not well described by measuring its main axis of change, number of branches and number of nodes.
Think of a tall biological tree with few branches compared to a short tree with many branches, how would one determine which is more complex?
Each field of study that works with tree like structures has it’s own methods for measuring complexity, taking into account the “goal” of the tree. In computer science there is a temporal dimension that is very important. In biology volume might be more important, where time is not.
In computer science, computational complexity is used. In biology box-dimensionality is a common complexity metric. However, the most correlary tree complexity metric I have found to a potential narrative tree structure complexity is molecular complexity from chemistry.
The complexity rating of a compound is a rough estimate of how complicated the structure is. There are many different ways that chemists calculate a molecules complexity, but two publicly available methods from PubChem and DataWarrior offer their own versions, that often do not correlate with each other. Dr Dan Erlanson sums up his understanding of molecular complexity with Justice Potters ‘I know it when I see it’ defense, which is what I hope to leverage for narrative tree structure complexity.
Using expert opinion on a set of narrative tree structures, build a model that will give predictions of how an expert would score a narrative tree structures complexity.
-Which predictors will be the most correlated with expert’s assessment on complexity?
-Which meta-communication types cause the largest changes in narrative tree structure complexity?
-Is narrative tree structure complexity itself a suitable measure of a twine novices understanding of the Twine IDN?
-Is narrative tree structure complexity robust enough to be projected onto higher dimension IDNs?
-How does narrative tree structure complexity compare to the previous metrics used?
In order to create a narrative tree structure complexity metric that tracks an experts intuition on the structural complexity of a narrative tree with the least number of logical contradictions (being presented with a differential value that contradicts intuition) I propose presenting a random pairing of trees to an expert and having them choose which is more structurally complex. After a sufficiently large number of pairings, the variables innate to the tree, such as number of branches, nodes, non-leaf nodes, choice points etc… can be used to build a model that predicts the trees overall complexity.
-Create an application that allows experts to compare two randomly selected narrative tree structures and choose which they deem more complex
-record the number of wins and losses each individual tree scores and write it to a SQL database
-Calculate the “win percentage” of each tree. This will be the individual tree structures ‘complexity metric’
-Extract all predictor variables from twine files built in the research workshop
-Build a model to predict a tree’s complexity metric and asses its accuracy
By applying the model, researchers will be able to calculate an estimate of a narrative tree’s structural complexity at any point along it’s growth that will align with an expert’s complexity prediction.