Function word analysis and questions of interpretation in early modern tragedy

The use of computational methods of stylistic analysis to
consider issues of categorization and authorship is now a
widely accepted practice. The use of computational techniques
to analyze style has been less well accepted by traditional
humanists. The assumption that most humanists make about
stylistics in general, and about computational stylistics in
particular, is that it is “concerned with the formal and linguistic
properties of the text as an isolated item in the work” (Clark
2005). There are, however, two other points of emphasis that
are brought to bear on a text through a cognitive approach.
These are: “that which refers to the points of contact between
a text, other texts and their readers/listeners”, and “that which
positions the text and the consideration of its formal and
psychological elements within a socio-cultural and historical
context” (Clark 2005). Urszula Clark (2005) argues that these
apparently independent strands of analysis or interpretive
practice are an “integrated, indissolvable package” (Clark
2005).
Computational stylistics of the kind undertaken in this
study attempts to link statistical fi ndings with this integrated
indissolvable package; it highlights general trends and features
that can be used for comparative purposes and also provides
us with evidence of the peculiarities and creative adaptations
of an individual user. In this case, the individual user is Elizabeth
Cary, the author of the earliest extant original play in English by
a woman, The Tragedy of Mariam, The Fair Queen of Jewry (1613).
As well as Mariam, the set of texts in the sample includes
the other 11 closet tragedies associated with the “Sidney
Circle”, and 48 tragedies written for the public stage. All
plays in the study were written between 1580 and 1640.1 The
only other female authored text in the group, Mary Sidney’s
Antonius, is a translation, as is Thomas Kyd’s Cornelia2.Alexander
Witherspoon (1924), describes the Sidnean closet tragedies
as “strikingly alike, and strikingly unlike any other dramas in
English” (179). He attributes this to the extent to which the
closet writers draw on the work of French playwright Robert
Garnier as a model for their own writing. In contrast to other
plays of the period closet tragedies have not attracted much
in the way of favourable critical attention. They are, as Jonas
Barish (1993) suggests, “odd creatures” (19), and Mariam is
described as one of oddest.
Mariam, as it turns out, is the closet play that is most like a play
written for the public stage, in terms of the use of function
words. But this isn’t immediately obvious. Some textual
preparation was carried out prior to the analysis. Homographs
were not tagged, but contracted forms throughout the texts
were expanded so that their constituents appeared as separate
words. The plays were divided into 2,000 word segments and
tagged texts were then run through a frequency count using
Intelligent Archive (IA). A total of 563 two-thousand word
segments were analysed, 104 of which were from closet plays,
and 459 from plays written for the public stage. A discriminant
analysis on the basis of the frequency scores of function words
demonstrates that there are signifi cant differences between
the two groups of plays. Table 1 shows the classifi cation results
for a discriminant analysis using the full set of function words.
In this test, 561 of the 563 segments were classifi ed correctly.
One segment from each group was misclassifi ed. Thus 99.6%
of cross-validated grouped cases were correctly classifi ed on
the basis of function words alone. The test also showed that
only 38 of the 241 function word variables were needed to
successfully discriminate between the groups.
Table 1
Classifi cation results for discriminant analysis using
the full set of function words in 60 tragedies (1580-
1640) closet/non-closet value correctly assigned a Cross validation is done only for those cases in the
analysis. In cross validation, each case is classifi ed by the
functions derived from all cases other than that case.
b 99.6% of original grouped cases correctly classifi ed.
c 99.6% of cross-validated grouped cases correctly classifi ed. Figure 1. Discriminant scores for correctly identifi ed
public and closet play segments from 60 tragedies
(1580-1640) in 2000 word segments on the basis
of 38 most discriminating function words Figure 2. Principal component analysis for 60 tragedies (1580-
1640) in 4000 word segments for 54 most discriminating
function words selected from 100 most frequently occurring
function words - word plot for fi rst two eigenvectors
Figure 3. Principal component analysis for 60
tragedies (1580-1640) in 4000 word segments for
54 most discriminating function words selected from
100 most frequently occurring function words
A principal component analysis (PCA) gives additional
information about the differences and similarities between
the two sets. An Independent-samples T-test was used to
identify the variables most responsible for the differences. This
process picked out 54 variables that were signifi cant at the
level of 0.0001 and these 54 variables were used for the PCA.
PCA looks to defi ne factors that can be described as most
responsible for differences between groups. For this text, the
texts were broken into 4000 word segments to ensure that
frequencies remained high enough for reliable analysis. This
produced 268 segments in total (50 closet segments and 218
public play segments). Figure 2 plots the 54 most discriminating
words-types for the fi rst two eigenvectors (based on factor
loadings for each variable) and shows which words behave
most like or unlike each other in the sample.
In Figure 3 the eigenvalues from the component matrix have
been multiplied through the standardised frequencies for each
of the 4,000 word segments to show which segments behave
most like or unlike each other on the fi rst two principal
components. The scores that produce Figure 3 are “the sum of
the variable counts for each text segment, after each count is
multiplied by the appropriate coeffi cient” (Burrows and Craig
1994 68). High counts on word variables at the western end of
Figure 2 and low counts on word variables at the eastern end
bring the text segments at the western end of Figure 3 to their
positions on the graph. The reverse is true for text segments
on the eastern side of Figure 3. We can see that the far western
section of Figure 3 is populated predominantly with public play
segments, and that the eastern side of the y-axis is populated
exclusively with segments from stage plays. It is clear that in
the case of these most discriminating variables, the segments
from Mariam are the closet segments most intermingled with
the segments written for the public stage.
Looking at Figure 2 there is evidence of an “emphasis on
direct personal exchange” in western section of the graph.
In the opposite section of the graph there is evidence of a
more disquisitory style of language that is “less personal”, with
“markers of plurality, past time, and a more connected syntax”
(Burrows and Craig 1994 70). It may be that the results refl ect
the kinds of observations that critics have long made about
early modern closet tragedy and tragedy written for the public
stage, suggesting that “word-counts serve as crude but explicit
markers of the subtle stylistic patterns to which we respond
when we read well” (Burrows and Craig 1994 70). It may also
be the case, however, that we can link these statistical results
with more interpretive work.
Returning to Mariam, the function words which most distinguish
the Mariam segments from the rest of the segments of both
closet and public plays, are the auxiliary verbs did (8.4) and
had (6.9). In the case of did over 500 of the segments have
scores of between -1 and 2. Six of the eight of the Mariam
segments are extremely high (the two middle segments of
Mariam have fairly average z-scores for did). The lowest score
occurs in segment 4, when Mariam is conspicuously absent
from the action. A very similar pattern emerges for had. In
conventional grammars do and other auxiliaries including be
and have are viewed as meaningless morphemes that serve
a grammatical purpose. Langacker argues that serving a
specifi able grammatical function is not inherently incompatible
with being a meaningful element (1987 30).
Cognitive linguistics suggests that function word schemas
interact with each other to produce what Talmy calls a
“dotting” of semantic space (1983 226), and that they “play
a basic conceptual structuring role” (Talmy 88 51). In this
framework, auxiliaries are viewed as profi ling a process
– they determine which entity is profi led by a clause and
impose a particular construal. Langacker argues, for example,
that do always conveys some notion of activity or some kind
of volitionality or control on the part of the subject. Have
designates a subjectively construed relation of anteriority and
current relevance to a temporal reference point (Langacker
1991 239). Talmy argues that have can be understood in terms
of force dynamics patterns; it “expresses indirect causation
either without an intermediate volitional entity…or… with such an entity” (1988 65). Talmy goes further to suggest that
the “concepts” of force dynamics are “extended by languages
to their semantic treatment of psychological elements and
interactions” (1988 69).
Bringing the tools of cognitive linguistics to bear on the results
of computational analysis of texts can provide a framework
that validates the counting of morphemes like did and had.
The same framework may also shed light on questions of
interpretation. This approach appears to provide a disciplined
way of identifying and analyzing the linguistic features that are
foregrounded in a text, while supporting their interpretation
as part of an integrated, indissolvable package.
Notes
1 Thomas Kyd wrote for both the public and the private stage. Kyd’s
Cornelia and The Spanish Tragedy are included in the study.
2 John Burrows (2002) explores some of the issues around
translation and whether it can be “assumed that poets stamp their
stylistic signatures as fi rmly on translation as their original work”
(679). Burrows found that Drydan was able to “conceal his hand”, but
in other cases it appeared that a “stylisitic signature”, even in the case
of a translation, remained detectable (696).
References
Barish, J. (1993). Language for the Study: Language for
the Stage. In A. L. Magnusson & C. E. McGee (Eds.), The
ElizabethanTheatre XII (pp. 19-43). Toronto: P.D. Meany.
Burrows, J. (2002). The Englishing of Jevenal: Computational
stylistics and translated Texts. Style, 36, 677-750.
Burrows, J. F., & Craig, D. H. (1994). Lyrical Drama and the
“Turbid Mountebanks”: Styles of Dialogue in Romantic and
Renaissance Tragedy. Computers and the Humanities, 28, 63-86.
Cooper, M. M. (1998). Implicature and The Taming of the Shrew.
In J. Culpeper, M. H. Short & P. Verdonk (Eds.), Exploring the
Language of Drama: From Text to Context (pp. 54-66). London:
Routledge.
Clark, U. (2005). Social cognition and the future of stylistics, or
“What is cognitive stylistics and why are people saying such good
things about it?!” Paper presented at the PALA 25: Stylistics
and Social Cognition.
Connors, L. (2006). An Unregulated Woman: A computational
stylistic analysis of Elizabeth Cary’s The Tragedy of Mariam,
The Faire Queene of Jewry. Literary and Linguistic Computing,
21(Supplementary Issue), 55-66.
Sweetser, E. (1990). From Etymology to Pragmatics: Metaphorical
and cultural aspects of semantic change. Cambridge: Cambridge
University Press.
Langacker, R. W. (1999). Losing control: grammaticization,
subjectifi cation, and transparency. In A. Blank & P. Koch (Eds.),
Historical Semantics and Cognition (Vol. 13, pp. 147-175). Berlin:
Mouton de Gruyter.
Talmy, L. (1983). How language structures space. In H. Pick
& L. Acredolo (Eds.), Spatial orientation: Theory, research, and
application. New York: Plenum Press.
Talmy, L. (1988). Force Dynamics in Language and Cognition.
Cognitive Science, 12. 49-100.
Witherspoon, A. M. (1924: rpt. 1968). The Infl uence of Robert
Garnier on Elizabethan Drama.