2D and 3D Visualization of Stance in Popular Fiction

In analyzing literary style through statistical methods, we
often show our results by plotting graphs of various types, e.g.
scatterplots, histograms, and dendrograms. While these help
us to understand our results, they always necessarily “fl atten”
the data into a 2- dimensional format. For the purposes of
visualization of complex data, we might be better off trying to
look at it three dimensionally if a suitable vector representation
of the data is found fi rst. This paper investigates stance in three
groups of popular fi ction, namely romance fi ction, detective
fi ction written by female authors and detective fi ction written
by male authors, and presents the results using both 2
dimensional and three dimensional visualization.
Romance fi ction and detective novels are both characterized
by the fact that they all have the same basic story. In romance
fi ction it is the story of how the hero and heroine meet,
how their relationship develops and how the happy ending
is achieved. In detective fi ction it is the story of the murder,
the unraveling of the case, the misleading clues and the fi nal
solution. The same story is being retold over and over again,
just as in the oral storytelling tradition (Radway 1984:198). The
reader is not left in suspense of the fi nal outcome and each
story is different from the others in the details of the events
and the way the story is told. Through this the reader becomes
involved in the story and partakes in the emotions, attitudes
and thoughts of the protagonist. These feelings, emotions and
moods are marked by syntactic and semantic features often
referred to as markers of stance.
Stance refers to the expression of attitude and consists of
two different types of expressions of attitude: evidentiality
and affect (Biber and Finegan 1989). Evidentiality means that
the reader becomes privy to the speaker’s attitudes towards
whatever knowledge the speaker has, the reliability of that
knowledge and how the speaker came about that knowledge.
Affect refers to the personal attitudes of the speaker, i.e. his/
her emotions, feelings, moods, etc. Biber and Finegan (1989)
investigated 12 different categories of features deemed to
mark stance: certainty/doubt adverbs, certainty/doubt verbs,
certainty/doubt adjectives, affective expressions, hedges,
emphatics and necessity/possibility/predictive modals. They
showed that different text types are likely to express stance
in different ways. Opas and Tweedie (1999) studied stance
in romance fi ction and showed that three types of romance
fi ction can be separated by their expression of stance. This
paper continues these studies, paying special attention to the
visualization of the results.
Materials and methods
Our total corpus is 760 000 words. It consists of three parts:
romance fi ction, female-authored detective fi ction and maleauthored
detective fi ction, all published in the 1990s. The
romance fi ction comprises a total of 240 000 words from the
Harlequin Presents series, the Regency Romance series and
Danielle Steel’s works, which are often classifi ed as women’s
fi ction or ’cross-overs’. The female-authored detective fi ction
part of our corpus includes Patricia Cornwell, Sue Grafton,
P.D. James, Donna Leon, Ellis Peters, and Ruth Rendell. These
texts make up 295 000 words. The rest of the corpus (229 000
words) is made up of male-authored detective fi ction texts,
including Colin Dexter, Michael Dibdin, Quintin Jardine, Ian
Rankin and Peter Tremayne.
Principal components analysis was used to reduce the 12
markers of stance to three dimensions which describe 52.4%
of the variation in the data.
Results
In a previous study Opas and Tweedie (2000) concluded
that the detective stories seem to mark evidentiality, i.e. the
characters express their certainty and doubt. The romance
stories, on the other hand, seem to mainly mark affect, i.e.
the characters express their emotions and moods. In Biber
and Finegan’s terms (1989), the detective fi ction texts show
‘interactional evidentiality’ and the romance fi ction texts show
‘emphatic expression of affect’.
The results in Figures 1 and 2 below show the female-authored
detective stories in shades of red and yellow, the male-authored
detective stories in shades of blue and the romance stories in
shades of black. While in Figure 1 the female-authored texts
are perhaps slightly more clearly separatable from the others,
it still seems that all these texts are far more overlapping and
less clearly distinguishable as three groups than the texts in
previous studies were. However, broadly speaking, the romance
texts are in the lower half of the graph, the female-authored
detective texts in the upper half and the male-authored
detective texts in the middle. What is surprising though is that
no features seem to “pulling” texts downwards, towards the
lower half of the graph; and that the feature that “pull” the
texts upwards include both markers of certainty/doubt and
affect. Figure 1. PC1 and PC2 for detective and romance fi ction
Figure 2. PC2 and PC3 for detective and romance fi ction
Figure 2 seems to show similar results. Here, perhaps, the
female-authored detective texts are even slightly more easily
separated from the others, but the general impression of
the male-authored texts and the romance texts overlapping
remains. Yet again it seems that there are hardly any features
accounting for the texts on the left-hand side of the graph and
that the feature “pulling” the texts to the right include both
features marking evidentiality and those marking affect.
These results are quite surprising. Either male-authored
detective stories mark evidentiality and affect in the same
manner as romance fi ction, and here is seems that they don’t
show many features that would mark stance, or there is
something more complex at work here. To help us understand
these phenomena better, we would suggest visualizing the
workings of the markers of stance in a 3 dimensional model.
To this end, we have built a tool that takes the principal
components analysis data, reduces dimensionality down to
three components with the most energy and presents the data
with these components. The software tool is implemented
in the MATLAB® environment (The MathWorks, Inc.,
Massachusetts, USA) utilizing its 3D graphical functions. The
tool is an interactive one, allowing the researcher to turn the
3D model and look for the angles that best show the clustering
structure and the differences between the texts. We will
demonstrate how tools such as this one signifi cantly improve
the researcher’s ability to visualize the research results and to
interpret them.
References
Biber, D. and E. Finegan (1989) Styles of Stance in English:
Lexical and Grammatical Marking of Evidentiality and Affect.
Text 9.1: 93-124.
Opas, L.L. and F.J. Tweedie (1999) The Magic Carpet Ride:
Reader Involvement in Romantic Fiction. Literary and Linguistic
Computing 14.1: 89-101.
Opas, L.L. and F.J. Tweedie (2000) Come into My World: Styles
of stance in detective and romantic fi ction. Poster. ALLCACH2000.
Glasgow, UK.
Radway, J. A. (1984) Reading the Romance. Chapel-Hill: University
of North Carolina Press.