Paper Two: Plot as Network: Quantifying the Evolution of Dramatic Style

In the last decade or so, quantitative evidence has
become part of literary study in a variety of ways
– from the material realities of book history [library
holdings, or translation flows], to the linguistic macropatterns that have renewed the study of attribution,
genre differentiation, and stylistics. Plot, however, has
proved much more difficult to quantify. This paper
looks for a possible solution in network theory, whose
concepts re-define plot as a system where characters
are the vertices, and their interactions the edges of
a narrative network. Focusing, for now, on dramatic
literature, such "network narratology" brings to light
the striking discontinuities between the structure of
ancient and Renaissance tragedy, and suggests a
reconceptualization of literary characters in terms of
their "connectedness" and their position within the
network.
This paper grew out of Franco Moretti’s network
analysis of three Shakespeare plays: Hamlet,
Macbeth, and King Lear. Preliminary work for the
project included php-extraction of speaker-receiver
data from MIT’s xml-encoded Shakespeare corpus.
The program R was then used to generate network
analyses. A team of Stanford graduate students handcorrected receiver attributions to a level of 100%
accuracy. The paper’s methodological component
discusses the difficulties of automated recipient
attribution in dramatic texts, and touches on issues
of “quantifying” performance literature, including plays
lacking an authoritative text. The idea of (performance)
time turned into space is grounded in narrative
theory: specifically, Alex Woloch’s concept of a
character-system made of many character-spaces
(The One vs. the Many). In addition to modeling “plot
as network” based in dialogue, I also utilize stage
directions to map networks as shared “space.” Here,
space includes more ambiguous character presence
such as disguised identity and plays-within-plays, as
well as the auditory presence of eavesdropping. Thus,
overlapping networks of character position -- both
physical and verbal -- model multiple dimensions of
plot through form and performance.
The paper draws on extant and original corpora of over
70 plays, from classical through Renaissance drama.
They include Shakespeare’s complete dramatic works,
as well as select plays by Marlowe, Jonson,
and Webster. The classical plays are drawn from
Project Perseus’ database. Evaluating our script’s
performance across literary periods has revealed that
classical dramatic dialogue follows a linear progression
more frequently than dialogue in Renaissance drama.
In other words, the script always assumes that Speaker
A will address Speaker B who addresses Speaker C,
etc, and therefore inaccurately assigns the receiver
when Speaker B responds back to Speaker A. Here,
methodology illuminates dramatic structure. The linear
model of “progressive” dialogue correctly attributes a
higher percentage of utterances (and words) in Greek
than Shakespearean tragedy. Dialogue in ancient
plays more strictly observes a recursive model that
occurs less frequently within Shakespeare’s more
populated scenes. The script correctly identifies more
receiver tags in plays with smaller casts (i.e., classical
drama) and scenes with two interlocutors, such as
Iago and Roderigo in Othello I.i. When there are
more than two interlocutors, the script misattributes
a response to a previous speaker as an address to
the following speaker. However, accuracy is usually
measured at the unit of the utterance, or speech-tag.
When measuring accuracy by words, Act I of Othello
presents a different case. Word-based accuracy is
consistently lower than utterance-based accuracy in
the Greek plays and in Hamlet, but in Othello, the
reverse is true. Thus, inaccurately attributed speeches
—those given in response to a preceding speech--
are longer in Hamlet and the Greek plays than in
the opening act of Othello. It follows that in Othello,
characters tend to say more when they address the
next character who speaks. If there's more chain-like
dialogic progress in Act I of Othello than in all of
Hamlet, this corroborates scene-based thematics that
Digital Humanities 2011
50
characterize Othello: its beginning in medias res and
the transmission of news through increasingly public
scenes (from “a street” (I.i) to “another street” (I.ii) to
“a council-chamber” (I.iii)). In the second section of this
paper I explore such network patterns diachronically:
at levels of scene, act and play, and in comparative
context.
This formal figuration of dialogue’s progress
supplements the paper’s central findings: weighted
network visuals, at levels of the play entire, act
and scene. At the most macroscopic level, we
observe “synchronic” patterns and discontinuities: for
example, larger scenes, with more characters and
therefore usually a more “public” nature occur near the
beginning and the ending of Hamlet, Lear, and Othello.
Networks create a hierarchy of centrality among
characters. This model inevitably calls into question the
binaries with which we usually think about characters:
protagonist vs. minor characters, or "round" vs. "flat":
nothing here supports these dichotomies, and, in fact,
the hierarchical re-conceptualization of characters is
another promising research area opened by network
theory.
The third section focuses not on character as
determined by location in narrative structure, but by
semantic analysis of dialogue, using word frequency
and semantic field analysis. For example, relativized
word frequency lists reveal that “Cassio” is among
the top ten most frequent words that Brabantio, the
Duke, Lodovico and Montano speak, before any other
character’s name, including “Othello,” “Desdemona,”
or “Iago.” This semantic analysis detects the linguistic
virulence of Iago’s revenge scheme, “promoting”
Cassio’s name in the constructed adultery plot.
Semantic analysis and weighted word networks also
reveal that antagonists’ confrontations are consistently
verbose in Greek tragedy, while Shakespearean
antagonists may never exchange a word. In this paper,
I consider dramatic plot as a function of network
connections--coded as relations between speaker
and addressee—through both historical and generic
developments.
References
Albert-László Barabási, (2003). Linked. NY: Penguin.
Mark Granovetter,. 'The Strength of Weak Ties'.
American Journal of Sociology. May 1973.
R. Alberich, J. Miro-Julia, F. Rosselló. 'Marvel Universe
looks almost like a real network'. http://arXiv:cond
-mat/0202174v1.
M. E. J. Newman (2006). 'Finding community structure
in networks using the eigenvectors of matrices'.
Physical Review E 74. 036104.
M. E. J. Newman (2003). 'The Structure and Function
of Complex Networks'. SIAM Review. .
Mark Newman, Albert-László Barabási, Duncan J.
Watts (eds.) (2006). The Structure and Dynamics of
Networks. Princeton UP.
Mark Steyvers, Joshua Tenenbaum (2005). 'The
Large-Scale Structure of Semantic Networks:
Statistical Analyses and a Model of Semantic Growth'.
Cognitive Science. .
James Stiller, Daniel Nettle, Robin I.M. Dunbar (2003).
'The small world of Shakespeare's plays'. Human
Nature. vol. 14, no. 4.