Using Wmatrix to investigate the narrators and characters of Julian Barnes' Talking It Over

The focus of this paper is on Wmatrix (Rayson 2008), and
how the output from this relatively new corpus tool is proving
useful in connecting language patterns that occur in Talking It
Over - a novel by Julian Barnes - with impressions of narrators
and characters. My PhD research is concerned with: 1) the
role of the narrator and author in characterisation in prose
fi ction, in particular how the narrator intervenes and guides
our perception of character – something that, in my opinion,
has not been satisfactorily dealt with in existing literature; and
2) how corpus tools can be usefully employed in this type of
investigation. In this paper I will show that Wmatrix helps to
highlight parts of the novel that are important insofar as theme
and/or narratorial style and/or characterisation. I will go on to
discuss diffi culties I encountered during my investigation, and
possible developments to Wmatrix that could be benefi cial to
future researchers.
Of course, computer-assisted approaches to the critical
analyses of novels are no longer completely new, and corpus
stylistics (as it is sometimes known) is a steadily growing area of
interest. Indeed, corpus approaches have already been used to
investigate character in prose and drama. McKenna and Antonia
(1996), for example, compare the internal monologues of three
characters (Molly, Stephen and Leopold) in Joyce’s Ulysses,
testing the signifi cance on the most frequent words in wordlists
for each character, and showing that common word usage
can form “[…] distinct idioms that characterize the interior
monologues […]” (McKenna and Antonia 1996:65). Also,
more recently, Culpeper (2002) uses the ‘KeyWord’ facility
in WordSmith Tools (Scott 1996) to demonstrate how the
analysis of characters’ keywords in Shakespeare’s Romeo and
Juliet can provide data to help establish lexical and grammatical
patterns for a number of the main characters in the play.
While my research adopts similar approaches to the two
examples above, it takes a small yet useful step beyond wordlevel
analysis, focussing instead on semantic analysis and keyconcepts.
This is possible because Wmatrix can analyse a text
at the semantic level (see below).
Wmatrix
Wmatrix (Rayson 2008) is a web-based environment that
contains a number of tools for the analyses of texts. Plain-text
versions of texts can be uploaded to the Wmatrix web-server
where they are automatically processed in three different
ways:
1) Word level – frequency lists of all the words in a text are
compiled and can be presented either in order of frequency
or alphabetically;
2) Grammatical level – using the Constituent Likelihood
Automatic Word-tagging System (CLAWS – see Garside
1987, 1996; Leech, Garside and Bryant 1994; and Garside
and Smith 1997) developed by the University Centre for
Computer Corpus Research on Language (UCRELi) at
Lancaster University, every word in a text is assigned a
tag denoting the part-of-speech (POS) or grammatical
category to which the word belongs. The words are then
presented in list form either alphabetically (by POS tag) or
by frequency (the most frequent POS tags at the top of the
list);
3) Semantic level – every word is semantically tagged using
the UCREL semantic analysis system (USASii) and then listed
in order of frequency or alphabetically by semantic tag.
USAS groups words together that are conceptually related.
It assigns tags to each word using a hierarchical framework
of categorization, which was originally based on MacArthur’s
(1981) Longman Lexicon of Contemporary English. Tags consist
of an uppercase letter, which indicates the general discourse
fi eld, of which there are 21 – see Table 1.
Table 1 The 21 top level categories of the USAS tagset The uppercase letter is followed by a digit, which indicates
the fi rst sub-division of that fi eld. The tag may also include a
decimal point followed by a further digit to indicate fi ner subdivision
of the fi eld. For example, the major semantic fi eld of
GOVERNMENT AND THE PUBLIC DOMAIN is designated by the letter
G. This major fi eld has three subdivisions: GOVERNMENT, POLITICS
AND ELECTIONS – tagged G1; CRIME, LAW AND ORDER – tagged
G2; and WARFARE, DEFENCE AND THE ARMY – tagged G3. The fi rst
subdivision (G1 - GOVERNMENT, POLITICS AND ELECTIONS) is further
divided into: GOVERNMENT ETC. – which is has the tag G1.1; and
POLITICS – which is tagged G1.2. From these initial lists, further analyses and comparisons are
possible within the Wmatrix environment. However, the focus
of this paper will be on my analysis of the USAS (semantic)
output from Wmatrix, as this output proves to be the most
interesting and the most useful interpretatively.
Data
Talking It Over, by Julian Barnes, is a story with a fairly familiar
theme – a love triangle – that is told in a fairly unusual way
– there are nine fi rst person narrators. This offers interesting
data with regard to the interaction between character and
narrator, as the story is told from a number of different
perspectives, with the different narrators often commenting
on the same events from different angles. Of the nine narrators
Oliver, Stuart and Gillian are the three main ones, not only
because they are the three people involved in the love triangle,
but also because they say substantially more than any of the
other narrators in the novel. The six other narrators are
people whose involvement in the story is more peripheral, but
come into contact with one or more of the main narrators.
Within the novel each contribution by different narrators is
signalled by the narrator’s name appearing in bold-type at the
beginning of the narration. Some chapters consist of just one
contribution from one narrator while others consist of several
“narrations”. The three main narrators make contributions of
varying lengths throughout the novel.
Approach
The exploration of narrators described in this paper adopts
a similar approach to that used by Culpeper (2002) in his
analysis of characters in Romeo and Juliet. That is, I compare
the words of one narrator with the words of all the other
narrators combined, using Wmatrix in its capacity as a tool
for text comparison. This comparison produced a number of
lists ranked by statistical signifi cance. The measure of statistical
signifi cance used by Wmatrix is log-likelihood (LL).
The process of comparison involved, to some extent,
disassembling the novel, in order to extract, into separate text
fi les, the narrated words of each of the three main narrators.
The disassembly process had to take into account the fact
that within the different narrations there were the words of
other people and characters. That is to say, the narrations
contained speech, writing and thought presentation (SW&TP).
This needed to be accounted for in the analysis in order to
be as precise as possible about descriptions and impressions
of narrators based on the Wmatrix output. The best way to
achieve this was to produce a tagged version of the novel,
from which relevant portions could be extracted more exactly
using computer-tools. These extracted portions could then be
analysed using Wmatrix.
My paper will discuss the tagging process, the Wmatrix output
and the subsequent analysis and show how key semantic
concepts highlighted for each of the three main characters
draw attention to important themes within the narrations, and
to styles of narration which can then be linked to character
traits. Included in my discussion will be issues concerning
Wmatrix’s lexicon and the type of automated semantic analysis
Wmatrix carries out.
Conclusions
Interpretative conclusions
The list of semantic groups produced by Wmatrix for each
of the three main narrators show a number of differences
in narrator characteristics and narratorial styles. Stuart’s list
contains semantic groups that relate to his job. An investigation
of highly signifi cant categories in the list showed that Stuart’s
attitude relating to particular themes or concepts change
during the novel. For example, Stuart’s relationship with money
alters, which is refl ected in a change of attitude toward love.
Stuart also becomes less worried about disappointing people
and more concerned about not being disappointed himself.
An investigation of items in Gillian’s list of semantic categories
identifi ed a more conversational style to her narration
when compared to the rest of the narrators, as well as a
determination to give an account of the story that was
accurate and complete.
The top item in Oliver’s list of categories contains the words
that Wmatrix could not match. While on one hand this could
be seen as a short-coming of Wmatrix, in the case of this study,
the result gave a clear indication of the number of unusual,
foreign and hyphenated words Oliver uses, and showed that
he has very broad vocabulary as well as knowledge of a wide
variety of topics. The list of unmatched words also highlighted
Oliver’s creativity with language, and that creativity is very
much part of his style. He uses unusual, technical and poetic
words in place of more commonplace synonyms and this
could be seen as evidence towards Oliver being showy and
fl amboyant.
Wmatrix development
The results from this study, in particular those relating to
Oliver’s narration and the failure of Wmatrix to successfully
categorise many of the words used in it, raise issues regarding
the tool’s lexicon and the semantic categories it uses.
These issues will be discussed as potential avenues for the
development of this useful corpus tool. For instance, the
Wmatrix-lexicon is currently progressively updated and
expanded, meaning that, in practice, there is no one static point
from which comparisons of texts can be made. While there
is a case for a lexicon that is as comprehensive as possible,
the present way of managing the lexicon can cause diffi culties
for research projects that continue over an extended period
of time. However, creating a ‘standard’ fi xed lexicon is not
without diffi culties and raises questions about what counts as
‘standard’. Even though Wmatrix allows users to defi ne their
own lexicon, a possible way forward might be to have multiple lexicons, such as a scientifi c lexicon or a learner-English
lexicon, which researchers could select depending on their
research needs.
Notes
i For further information about UCREL see http://www.comp.lancs.
ac.uk/ucrel/
ii For further information about USAS, see http://www.comp.lancs.
ac.uk/ucrel/usas/
References
Culpeper, J. (2002) “Computers, language and
characterisation: An Analysis of six characters in Romeo and
Juliet.” In: U. Melander-Marttala, C. Ostman and Merja Kytö
(eds.), Conversation in Life and in Literature: Papers from the
ASLA Symposium, Association Suedoise de Linguistique Appliquee
(ASLA), 15. Universitetstryckeriet: Uppsala, pp.11-30.
Garside, R. (1987). The CLAWS Word-tagging System. In: R.
Garside, G. Leech and G. Sampson (eds), The Computational
Analysis of English: A Corpus-based Approach. London: Longman.
Garside, R. (1996). The robust tagging of unrestricted text:
the BNC experience. In J. Thomas and M. Short (eds) Using
corpora for language research: Studies in the Honour of Geoffrey
Leech. Longman, London, pp 167-180.
Garside, R., and Smith, N. (1997) A hybrid grammatical tagger:
CLAWS4, in Garside, R., Leech, G., and McEnery, A. (eds.)
Corpus Annotation: Linguistic Information from Computer Text
Corpora. Longman, London, pp. 102-121.
Leech, G., Garside, R., and Bryant, M. (1994). CLAWS4: The
tagging of the British National Corpus. In Proceedings of the
15th International Conference on Computational Linguistics
(COLING 94). Kyoto, Japan, pp622-628.
McKenna, W. and Antonia, A. (1996) “ ‘A Few Simple Words’ of
Interior Monologue in Ulysses: Reconfi guring the Evidence”.
In Literary and Linguistic Computing 11(2) pp55-66
Rayson, P. (2008) Wmatrix: a web-based corpus processing
environment, Computing Department, Lancaster University.
http://www.comp.lancs.ac.uk/ucrel/wmatrix/
Rayson, P. (2003). Matrix: A statistical method and software
tool for linguistic analysis through corpus comparison. Ph.D.
thesis, Lancaster University.
Scott, M. (1996) Wordsmith. Oxford:Oxford University Press