Modulating Style (and Expectations): An Experiment with Narrative Voice in Faulkner's The Sound and the Fury

Faulkner’s The Sound and the Fury provides an interesting
test bed for stylistic analysis of narrative
and authorial voice. Readers of the text understand that
the work contains at least four different “voices,” and
if we include the Appendix that Faulkner wrote after
the book’s initial publication, then we have five distinct
voices, namely those of Benjy, Quentin, Jason, the
Omniscient narrator of the fourth section, and then the
somewhat mysterious voice of the Appendix that some
have called Faulkner’s own voice. As part of Dr. Matthew
Jockers’s humanities computing seminar at Stanford
this fall, the authors of this proposal conceived of
an experiment to investigate narrative voice and utilize
authorship attribution techniques to explore the extent
to which Faulkner is able to “create” distinct narrative
styles within The Sound and the Fury.
Starting with a traditional reading of Faulkner’s novel,
our group employed the traditional tools of literary analysis
as a sort of “control.” We explored the differing
styles content, and structure of the five separate narrative
voices in The Sound and the Fury. We paid close attention
to the stylistic changes, and we hypothesized about
what qualities separate the different sections. We put all of these observations into an aggregated document; a traditional
“reading” of style in Faulkner’s text.
We then began a phase of speculation based upon our understanding
of the sort of stylistic data that can be generated
by a computer algorithm. We speculated about how
or if these qualities identified in our traditional reading
might be detectable via computer-based text analysis.
We then drew up a series of hypotheses and predictions
about what sort of differences might be made evident by
a quantitative, computer analysis of the text.
Despite the significant shifts in style detected by readers,
our first prediction was that in terms of the most common
words, Faulkner’s narrative would be fairly consistent.
Our prediction was informed by current studies in authorship
attribution, which suggest that, with frequently
occurring words at least, differences between authors are
greater than difference between works by the same author.
With The Sound and the Fury, we had a particularly
compelling test case since Faulkner worked very hard
to create unique narrative perspectives, even changing
between first and third person.
Our second prediction involved the very noticeable
changes a reader detects when moving from section
to section. These changes constitute a sort of “narrative
dissonance” that repeatedly “jars” the reader as he
moves between sections. We speculated that these effects
would be difficult for a computer to detect.
We formulated these and other predictions/hypotheses
into a secondary document, and then began the more
practical work of developing the tools necessary to test
our hypotheses. This work involved leveraging existing
tools such as Patrick Joula’s JGAAP and also creating
new tools. Most challenging has been our effort to create
a “dissonance” detector, a tool which at the time of
this writing is showing great promise but still under active
development.
The paper we propose here represents a sort of experiment
in which we are both the subjects and the investigators.
While we are overtly exploring narrative voice
in Faulkner we are simultaneously investigating the
role that a computer-based methodology might play in
a conventional study of literature. On this point we are
inspired by Steve Ramsay’s observation that texts are
“seldom . . .transformed algorithmically as a means of
gaining entry to the deliberately and self-consciously
subjective act of critical interpretation.” In our final
analysis, we transform our text into a new sort of literary
artifact, a statistical matrix that allows us to work in a
reverse order—starting with objective, quantified facts
and moving to subjective interpretations of those data.
Ultimately, we compare our objective and subjective
methods and their respective conclusions to determine to
not only explore the formal aspects of an author’s narrative
style, but to imagine what weight quantified data can
bring to the traditional literary enterprise.