This writing is somewhat a response to this paper published by some researchers on the Google Brain team. The paper in review is “Shaping the Narrative Arc: An Information-Theoretic Approach to Collaborative Dialogue,” which can be read here.
https://arxiv.org/pdf/1901.11528.pdf
Defining The Dialogue Object
Let’s create a thing called the Dialogue Object. Let’s define and look at the Dialogue Object as a game with players. In the simplest case, we have a two-player game. (It’s hard to have a one-person dialogue. Even the crazy who talks to himself arguably has two voices at play. “Talks to himself.” I-thou relationship. That’s two entities.) In this game, we try to create an artificial agent with computers that can be a substitute to one of these players. There’s good motive to create an agent. Among many possible use cases, Therapy is a valuable dialogue.
Our task when constructing the object class (when constructing the Dialogue) is to construct the rules of the game, define the players, define the moves the players can make, and define what a “winning” game is.
In a Dialogue, an object of art, we steer away from a classical win-lose architecture. Classical examples of “winning” are tallying maximum points (like in basketball) or defeating the other player (making a sequence of moves to take out the king or eliminate all other pieces). Often the metrics of “winning” are the determinants of an ending. To win is to end.
But, we can also look at “Winning” as having completed a duration, and defining what the duration of the object is. The clock ran out. There are no more pages in the book. There are no more minutes in the film. Kermodes’ “Sense of an Ending” articulates the importance of beginnings and ends, and the feeling of the passage of time in an object like a “Book” or a “Film”. He has contributed to a whole field of study called Literary Theory, which, I believe, has fought to legitimize what we can summarize by instantiating the Dialogues Object class.
When “winning” is defined by duration and not by points, a player’s strategy adapts itself, and, at the local level--the human level--the player’s very idea of “competition” changes. The player dynamics change. The popularized aesthetic appreciation for “tension” becomes a part of what is beautiful when players compete in a dialogue with duration, rather than points.
To define our Dialogue Object, and its shape-taking iterative form through the contributions of two players, we will borrow from Charles Wooldridge’s Bang Game.
In this paper, we recreate the Dialogue Object’s architecture from Google’s paper. They did not specifically articulate it. It lurked in the subconscious of the paper, assumed with no credit given. There is not just one architecture and there needs to be room for distinction. We will show what their architecture is, then formalize their architecture to begin identifying the features of the Dialogue Object.
For our purposes, we build from their standard because...they’re Google. Their influence is significant and their methods are tried and it sets precedence for others - we are an Other. Once the architecture of their game-play is defined, the features can then reveal themselves and, I predict, there are many other features in the Dialogue Object that have gone unexamined.
After we’ve constructed it, I want to do a thorough scan over the Dialogue Object to then develop an exhaustive list of classifying tools to capture the aesthetic traits of the Dialogue Object.
By the end, the purpose of the analysis is to allow discussion around the Dialogue Object. By recognizing it, we recognize that they are not all the same in architecture or in game-play. While the Google paper operates within a single game-play model primarily influenced by improvisational theater, we want to allow creativity in developing new Architectural Models for the Dialogue Object. This would allow the “Architecture” to be designed by Dialogue artists as a plug-and-play feature, and then inserted into the Dialogue Object with the guided intention of the players. There could be “Therapy” dialogues, “Educational” dialogues, “Interviewing” dialogues, “Brainstorming” dialogues, “Improvisational” dialogues. This approach propels us down a focused path to refine the conception of the Dialogue Object and Language Aesthetics by allowing for the identification of new aesthetic features in the Dialogue Object.
This approach is in line with popular Systems Thinking, Category Theory, Design Thinking, Algorithmic Thinking, Buberian philosophy and, last but not least, Object-Oriented Ontology.
As a speaker, we can only cast the Narrative Arc paper in the light of thinking about the Dialogue Object as it is now understood. We cannot unlearn what we already know. We have the responsibility of behaving with the knowledge. The following is an assessment:
The Dialogue Object
The Dialogue Object is simple and doesn’t require a lot of discussion. I can simply offer an architecture, and the complexity of the data inserted, or the equations used, are determined by the programmer or designer.
Object Class- Dialogues
Time
Duration - ie; t = x pages, t = x minutes
Structure - Linear (most stories), Stacked (Inception), Broken (Pulp Fiction), Perspective-based (Hateful Eight)
Players
Number of players - static: p = 1, p = 4, dynamic: p = 3a - 2b
Possibility space
Options - sets of available choices
Choices - choice-making logic, a determinant
Game play - Sequential, simultaneous, etc.
Knowledge space - Perfect information, imperfect information
The Dialogue Object Settings in the Narrative Arc Paper
The Narrative Arc paper sets its features of the Dialogue’s Object Class to: Players = 2, Options = Movie Dataset (sentence-level tokenizations), Choices = Concealing or Revealing utterances, Duration = 20 lines of utterances, Structure = Linear, Game Play = Sequential, Knowledge = Perfect Information.
To judge if it is “winning” or not, it is judged by a third-party spectator. (The participants, the audience.)
The Role Google Serves in the Global, Humanity “Dialogue Object”
Google’s position as a company is to elevate the Players’ Knowledge setting in a Dialogue to “Perfect Information”.
