The Moonstone and The Coquette: Narrative and Epistolary Styles

In his seminal work on Austen, John F. Burrows demonstrates
that characters can be distinguished from each another on
the basis of the frequencies of the most frequent words of
their dialogue treating the characters as if they were authors
(1987). Computational stylistics has also been used to study
the distinctive interior monologues of Joyce’s characters in
Ulysses (McKenna and Antonia 1996), the styles of Charles
Brockden Brown’s narrators (Stewart 2003), style variation
within a novel (Hoover 2003), and the interaction of character
separation and translation (Rybicki 2006). Here I address two
kinds of local style variation: the multiple narrators of Wilkie
Collins’s The Moonstone (1868) and the multiple letter writers
in Hannah Webster Foster’s sentimental American epistolary
novel The Coquette (1797).
The Moonstone has several narrators whose styles seem
intuitively distinct, though all the narrations share plot elements,
characters, and physical and cultural settings. The Coquette,
based on an infamous true story, and very popular when it
was published, is read today more for its cultural signifi cance
and its proto-feminist tendencies than for its literary merit.
Nevertheless, one would expect the coquette, the evil
seducer, the virtuous friend, and the disappointed suitor to
write distinctively. Treating the narrators and letter writers of
these two novels as different authors will test how successfully
Collins and Foster distinguish their voices and shed light on
some practical and theoretical issues of authorship and style.
Computational stylistics cannot be applied to narrators
and letter writers of these novels, however, unless they can
distinguish Collins and Foster from their contemporaries.
Experiments on a 10-million word corpus of forty-six Victorian
novels confi rm that Collins is easily distinguished from fi ve
of his contemporaries, as shown in Fig. I. For The Coquette, I
have made the task more diffi cult by comparing Foster’s letter
writers to those of fi ve other late 18th-century epistolary
novels by fi ve authors. I separated all the letters by writer and
addressee and retaining only the 22 writers with the most
letters, then combined the letters between each single writer
and addressee and cut the combined texts into 42 sections
of about 3,500 words. Both Delta and cluster analysis do a
good job on this diffi cult task, and many Delta analyses give
completely correct results. Cluster analysis is slightly less
accurate, but several analyses are correct for fi ve of the six
authors; they also show that individual letter writers strongly
tend to group together within the cluster for each novel (see
Fig. 2).
Examining the individual novels reveals a sharp contrast:
Collins’s narrators are internally consistent and easy to
distinguish, while Foster’s letter writers are much less internally
consistent and much more diffi cult to distinguish. For Collins,
cluster analysis consistently groups all narrative sections of
6 of the 7 narrators. When a modifi ed technique developed
especially for investigating intra-textual style variation is used
(Hoover 2003), the results are even better. As Fig. 3 shows,
all sections by all narrators sometimes cluster correctly (the
sections range from 4,300 to 6,900 words).
“Tuning” the analysis to produce better clustering may seem
circular in an analysis that tests whether Collins’s narrators
have consistent idiolects. But this objection can be answered by
noting the consistency of the groupings. The stable clustering
across large ranges of analyses with different numbers of
MFW that is found here is obviously more signifi cant than
frequently-changing clustering.
Note also that the sections strongly tend to occur in narrative
order in Fig. 3: every two-section cluster consists of contiguous
sections. This “echo” of narrative structure provides further
evidence that the analysis is accurately characterizing the
narrators’ styles. A fi ne-grained investigation of The Moonstone involving smaller
sections and including more narrators puts Collins’s ability to
differentiate his narrators to a sterner test, but some cluster
analyses are again completely correct. The distinctiveness of the
narrators of The Moonstone is thus confi rmed by computational
stylistics: the narrators behave very much as if they were
literally different authors. Given the length of the novel, the
generic constraints of narrative, and the inherent similarities
of plot and setting, this is a remarkable achievement.
For my analysis of The Coquette, I have added 3 more sections
of letters, so that there are 13 sections of approximately 3,000
words by 6 writers, with separate sections by a single writer
to two different addressees. Although these sections should
be considerably easier to group than those in the fi ne-grained
analysis of The Moonstone, the results are not encouraging.
Cluster analyses based on the 300-700 MFW produce very
similar but not very accurate results, in all of which letters by
Boyer and Eliza appear in both main clusters (see Fig. 4). Lucy’s
letters to Eliza also cluster with the last section of Eliza’s letters
to Lucy in all of these analyses a misidentifi cation as strong as
any correct one. In a real authorship attribution problem, such
results would not support the conclusion that all of Boyer’s
or Eliza’s letters were written by a single author. Perhaps one
could argue that Boyer’s letters to Selby should be different
from his letters to Eliza, and perhaps it is appropriate that
Eliza’s last section includes only despairing letters written
after Boyer has rejected her. Yet such special pleading is almost
always possible after the fact. Further, any suggestion that
Boyer’s letters to Eliza should be distinct from those to Selby
is impossible to reconcile with the fact that they cluster so
consistently with Eliza’s early letters to Lucy. And if Eliza’s early
and late letters should be distinct, it is diffi cult understand the
clustering of the early letters with those of Boyer to Selby and
the consistent clustering of the late letters with Lucy’s letters
to Eliza. It is diffi cult to avoid the conclusion that Foster has
simply failed to create distinct and consistent characters in
The Coquette.
In contrast, Fanny Burney, whose Evelina is included in the
novels compared with The Coquette, above, creates very
distinct voices for the letter writers. Although only four of
the writers have parts large enough for reasonable analysis,
Evelina writes to both her adoptive father Mr. Villars and her
friend Miss Mirvan, and Villars writes both to Evelina and
to Lady Howard. One might expect signifi cant differences
between letters to these very different addressees. Evelina’s
style might also be expected to change over the course of this
bildungsroman. However, analyses of all 34 sections of letters
from Evelina (approximately 2,500 words long), show that
Burney’s characters are much more distinct and consistent
than Foster’s, as a representative analysis shows (see Fig. 5).
This dendogram also strongly refl ects the narrative structure
of the novel. Burney, like Collins, is very successful in creating
distinctive voices for her characters.
Criticism of Foster’s novel has paid little attention to the
different voices of the characters, but what commentary there
is does not suggest that the patterns shown above should
have been predictable. Smith-Rosenberg, for example, suggests
a contrast between Eliza and “the feminized Greek chorus of
Richman, Freeman, and Eliza’s widowed mother, who, at the
end, can only mouth hollow platitudes” (2003: 35). Although
these women and Julia are often considered a monolithic
group urging conventional morality, the distinctness of Julia’s sections, especially from Lucy’s (see Fig. 5) might suggest
a reexamination of this notion, and might reveal how the
styles of Julia and the other women are related to more
signifi cant differences of opinion, character, or principle.
Various suggestions about changes in Eliza over the course of
the novel might also benefi t from a closer investigation of the
language of the letters. Because Foster’s novel is of interest
chiefl y on cultural, historical, and political grounds rather than
literary ones, however, such an investigation is more likely to
advance the theory and practice of computational stylistics
than the criticism of The Coquette. It is clear, at any rate, that
computational stylistics is adequate to the task of distinguishing
narrators and letter writers, so long as the author is adequate
to the same task.
References
Burrows, J. (1987) Computation into Criticism. Oxford:
Clarendon Press.
Hoover, D. (2003) Multivariate Analysis and the Study of Style
Variation’, LLC 18: 341-60.
Smith-Rosenberg, C. (2003) Domesticating Virtue: Coquettes
and Revolutionaries in Young America’, In M. Elliott and C.
Stokes (eds),American Literary Studies: A Methodological Reader,
New York: NewYork Univ. Press.
Rybicki, J. (2006). “Burrowing into Translation: Character
Idiolects in Henryk Sienkiewicz’s Trilogy and its Two English
Translations,” LLC 21:91-103.
Stewart, L.”Charles Brockden Brown: Quantitative Analysis
and Literary Interpretation,” LLC 2003 18: 129-38.