Journal on Policy & Complex Systems Volume 3, Issue 2 | Page 96

invention from his contemporaries , only the North American Review excerpts were released while he was alive . Following his death , a broader selection of excerpts appeared , but a complete version of the text was not attempted for a century . The text was known to scholars and circulated in different forms , but the unexpurgated autobiography remained largely hidden from public view .
In this paper , I treat Twain ’ s Autobiography as a communications technology and test the impact this technology has on the topics of discussion taken up by his contemporaries . Simulating the introduction of Twain ’ s Autobiography draws attention to the effect of his decision to withhold the text from his contemporaries and opens a window into Twain ’ s sense of his place in the world . The model also outlines an approach that may be used to explore the movement of information through an audience and to identify assumptions upon which this movement may depend .
Description of The Model

The model world I am working with

is established in NetLogo by asking
1,089 patches to sprout a turtle . Each turtle attempts to communicate with a neighbor once each tick and the results of these interactions are tracked using the variables below :
Policy and Complex Systems
1
For Chronicling America , see http :// chroniclingamerica . loc . gov /.
turtles-own [ topic-of-conversation ;; the topic of conversation speaking ;; how often a turtle starts a conversation listening ;; how often a turtle listens conversationfailed ;; how often a turtle fails to start a conversation prior-topic ;; prior topic of conversation repeatconvo ;; indicates if turtle is making a second attempt at a conversation with a topic ]
Aggregating the turtle-specific information collected above provides an overview of the conversation dynamics unfolding in the environment .
Once the simulation starts , every turtle picks a topic of conversation from a list of available topics . Choices are made in proportion to a probability distribution obtained by topic modeling a selection of historic newspapers found in the Chronicling America Project at the Library of Congress . I used the Machine Learning for LanguagE Toolkit ( MALLET ) via the Software Environment for the Advancement of Scholarly Research ( SEASR ) to identify 30 topics and their distributions over the month of January 1906 . 1 The SEASR flow I designed finds a minimum of one and a maximum of five active topics in each day of coverage , with an average of 2.28 and a standard deviation of 1.39 active topics per day . The distribution of topics in the corpus is skewed to the right : days with few topics are more common and days with multiple active topics are less common in the data . Figure 1 summarizes the distribution of topics produced using topic modeling .
This approach provides a representation of topics present in the news for each day of the month . For example , January 4 and January 6 are each described by five active topics ; while coverage for
92