Reconceiving Text Analysis

Reconceiving Text Analysis

Stephen
Ramsay

IATH, University of Virginia
sjr3@virginia.edu

John
Bradley

King's College London
john.bradley@kcl.ac.uk

Geoffrey
Rockwell

McMaster University
grockwel@mcmaster.ca

Stéfan
Sinclair

University of Alberta
Stefan.Sinclair@ualberta.ca

2002

University of Tübingen

Tübingen

ALLC/ACH 2002

editor

Harald
Fuchs

encoder

Sara
A.
Schmidt

How can we use computers to assist us in the interpretation of literary
texts?
On the one hand, this has the ring of a settled question. Half of the
humanities computing community works to make texts available to researchers
and provides facilities whereby those texts may be searched, annotated, and
linked. The other half produces tools for textual analysis that allow us to
undertake complex statistical and procedural analyses. Yet most humanists
outside of our discipline conceive of these activities as either
pre-interpretive, or else outside of the normative realm of critical
exegesis found in literary criticism, philosophy, history, and the various
other disciplines which take the text as central to the endeavor.
This panel brings together four creators of text analysis software interested
in reconceiving the activity from a theoretical standpoint. Panelists will
address a range of questions about text analysis: Can text analysis be
reconceived as fundamentally an act of text enrichment--not a taking from,
but an adding to the text being analyzed? What happens when text analysis is
thought of not as the quest for empirical data about texts, but as a
technology more in line with the readerly quest for novel patterns? How
might we re-theorize the classical modes of text analysis (searching, word
frequency analysis, stylometrics) as participating in the hermeneutics of
play?
We conceive of this session neither as a critique of existing systems nor as
as a commentary on text analysis as it is brought to bear on problems in
computational linguistics. Rather, we conceive of this session as theorizing
text analysis in literary studies and other fields with similar
hermeneutical practices with an eye toward the future of text analysis
tools. We realize that proposing such a session in Tübingen, the home of
TUSTEP, is like bringing coals to Newcastle, but we offer this panel as a
perspective from across the waters -- a different tradition of text analysis
and computing.
[Note: We have requested that this proposal be considered as a session,
despite the fact that we are proposing the delivery of four formal papers.
However, we are prepared to adjust ourselves to the time constraints as
necessary.]

Finding the Middle Ground Between "Determinism" and "Aesthetic
Indeterminacy": A Model for Text Analysis Tools
John Bradley

In Danielle Miller's review of the book The Legacy
of Northrop Frye, she notes Imre Salusinsky's
observation that the textual theorist is the "true liberal who
positions himself in the middle ground between 'determinism' and
'aesthetic indeterminacy'". Over the past several years I have
proposed a model for text analysis tools that balances the ability
of the computer to carry out a set of formal tasks on a text against
the need of the human user to introduce rather more
non-deterministic material into an analysis. Although it is meant to
reflect a view on some aspects of how a text is analyzed that is not
specifically "computer based", it is also based on certain
developments in the computing world over the past several years, and
on developments in software in a sister field to the humanities: the
social sciences. The model is based on XML, and more specifically
TEI -- surely a solid foundation upon which text analysis tools
should be built. It, however, goes against several of the current
developments in XML -- turning, instead, to a view of XML that is, I
think, truer to the thinking behind TEI than these current
developments.
The World Wide Web, with its model of servers and clients, often
dominates thinking about the role of computers in certain groups
within the computing humanities community. In the web, the server
has a resource that can be made available to a community of users --
the clients. The nature of browser-based interaction means that the
user can use the web to select displays of results, and can also
take advantage of search engines that a
server might make available -- posing a query through a form which
allows the server machine to select material from the resource to be
presented. In the Humanities, then, the WWW encourages the view of
scholarly materials as a resource that
the WWW makes available to a collection of scholars -- no wonder,
perhaps, that conferences like the UK's Digital
Resources in the Humanities have begun to appear.
The nature of the server/client interaction in the WWW is
transactional. A user sends a request (either in the form of a
request for a page of material, or in the form of a query), and the
server sends back a response. Not surprisingly, given that XML came
from the W3C -- setters of standards for the WWW -- there have been
a flurry of activities that are based on XML that support this
transactional model of interaction. Standards like W3C's SOAP (which
is XML based) work best in a what in computing is called a
"peer-to-peer" context: where a computer system belonging to one
organisation (say a purchasing system) needs to send an order to a
peer machine in another (say, supplier) organisation.
The transactional model is of course appropriate for certain kinds of
humanities resources. In many kinds of linguistics-based work, for
example, it is often sensible to view a corpus as a resource to be
queried. However, this model does not suit other aspects of the
traditional model of humanities scholarship nearly so well. Indeed,
the TEI, originally developed before the WWW was available, takes
quite a different, and much more intimate, view of the relationship
between the user and his/her text. In the TEI's "analytic mechanism"
and related schemes such as "feature structures" one sees an attempt
to express in SGML (and nowadays XML) connections between text and
analysis that are tightly connected to the text itself in that they
rely on the insertion of markup by the scholar directly onto the
text base. This kind of activity is not so much like the
transactional model, which would have the scholar interacting with a
text as a remote resource, but instead should be thought of as being
closer to a form of ownership -- the scholar gradually makes some
aspect of the text his/her own by attaching material ("annotations")
that represent her/his personal interests onto a text.
I believe there is some evidence to suggest that this "enrichment
model" is closer to the interaction between a scholar and his/her
text and provides a better model for computer support of humanities
scholarship than the transactional one does. During the presentation
of this paper we will examine Willard McCarty's "Analytical
Onomasticon to the Metamorphoses of Ovid" as an example of textual
enrichment. The Onomasticon represents a blending of traditional
annotation with computer processing which helps to reveal and assist
in the imposition of a unified, yet rich, vision of personification
on the text. McCarty's recent analysis of the commentary also
proposes a model for the digital commentary in which an enrichment
approach is implicit. Furthermore, there are useful models to
examine in the social sciences, where there has been the blossoming
of tools to provide computer assistance to the kind of textual
analysis that is often needed for their texts (e.g. interviews).
Particularly interesting in this regard are the packages Nud*ist, NVivo
and Atlas.ti, all tools that suggest
some characteristics of enrichment that would suit humanities
scholarship as well.
Tools to support textual enrichment have been available for some
time, and some of them are remarkably powerful. TuStep, for example,
has its origins in the 1970s -- predating even SGML -- and provides
an integrated set of tools to support a broad range of scholarly
activities. Much more recently, the "EyeContact" model, proposed by
Geoffrey Rockwell and me, operates in a broadly similar fashion --
emphasising a set of tools that can process text and text-related
materials, and that can be combined in many ways. I have come to
believe, however, that the enrichment model is better served when
the emphasis provided by the environment to the user is on viewing
of the text and scholarly annotations rather than on the task of
assembling of the tools to do work. Thus, it seems to me that
"XML/SGML editors" such as XMetal or emacs/psgml provide a good
starting point for envisaging a model of what is needed. They are at
least aware of XML/SGML constructs and can assist with ensuring
XML/SGML conformity. However, on their own they cannot help the user
much if s/he tries to introduce one of TEI's more sophisticated
analytic mechanism models onto a text -- the TEI guidelines
themselves suggest that there is a need for software beyond an
XML-aware text editor to facilitate the introduction of the
analytical models that they propose. It is in the combining of XML
structures (already understood by XML editors) with software objects
more closely related to the scholarly tools being modeled that
development work would need to be done.
The development of software that supports an enrichment model of text
is supported by some developments that are XML based. Many of these
standards are still in their early days and are as of yet, from an
enrichment perspective, hobbled by the transactional focus of the
members of the committees that develop them, since they are tailored
first to meet the needs of those who process relatively short,
relatively simple XML documents that characterise transactional
processing. Standards such as XLink -- arising as it does out of
HyTime, provides one starting point, and perhaps because of its
origin, is less specific to transactional-based work. XSLT and XSL's
formatting objects provide another starting point, and the DOM
provides a third. A serious development effort, based on these
standards, could be undertaken, but I believe one would find that in
the end further enhancement of them to more closely match the
complex needs of modeling within the humanities would be needed.

References

Melina
Alexa

Cornelia
Zuell

A Review of Software for Text Analysis

Manneheim
ZUMA
2000

ISBN 3-924220-16-6

John
Bradley

Tools to augment scholarly activity: an
architecture to support text analysis

Dino
Buzzetti

Giuliano
Pancaldi

Harold
Short

Augmenting Comprehension: Digital Tools (and
Resources) for the History of Science and Philosophy; Papers
from the conference "Informatica umanistica: filosofia e
risorse digitali" (Bologna, September 2000)

OHC: Oxford and King's College London
forthcoming, expected 2002

Willard
McCarty
et al
An Analytical Onomasticon to the Metamorphoses
of Ovid
(Online Sampler 29/8/99)

Willard
McCarty

The DIY commentary; or, what the reference and
the link told each other

from Paper for ACH/ALLC 2001 New York
University 14 June 2001

2001

Geoffrey
Rockwell

John
Bradley

Eye-ConTact: Towards a New Design for Research
Text Tools

Computing in the Humanities Working Papers

A
4

February 1998

This online refereed journal is located at: URL:

Toward an Algorithmic Criticism
Stephen Ramsay

There is a left-hand tradition in literary analysis which most
literary critics think of as incompatible with mainstream scholarly
activity. That tradition manifests itself in Hebrew gematria and in
the bibliomancy of the ancient Chinese; in the anti-art poetics of
the Dadaists and the algorithmic writing of the Oulipo; in Ferdinand
de Saussure's secret quest for anagrams in Saturnian poetry and in
Emily Dickinson's injunction to read her poems backwards, so that "a
certain Something overtakes the mind." These traditions are often
deliberately anarchic, mystical, and even irrational in their
approach to the text. Instead of the hermeneutics of illumination,
in which the goal is some clear statement of meaning or an unfolding
of the truth, this tradition asserts the hermeneutics of play.
That this tradition should seem so far afield from the normative
practices of literary exegesis is itself revelatory of certain
distinctive features of the illuminative mode. Literary criticism
asserts a rhetoric of explanation intended to reveal both the
internal and extrinsic logic--the meaning--of a textual artifact;
the more ancient ludic traditions deemphasize this aspect of
hermeneutics, often relinquishing this rhetoric entirely, content
simply to let alternative formations exist without the scaffolding
of explicit interpretation.
But these types of interpretive activities, in which one proceeds
from text to playful reordering and "refactoring" (to borrow a term
from software engineering) to interpretation, only serve to make
manifest a progression that is always at work in literary critical
method, even when the dominant rhetorics of interpretation attempt
to conceal the playfulness beneath. In order to create and
communicate meaning, the critic must remap, reenvision, and re-form
(even "deform" it, as Jerome McGann and Lisa Samuels have suggested)
into some alternate arrangement. In essence, one must create the
text anew in order to illuminate the original.
There is also a right-hand tradition in literary analysis which most
literary critics think of as preinterpretive or else unallied with
the real work of generating critical interpretation. That tradition
manifests itself in programs which display search results; in Busa's
Index Thomisticus and in concordancing software; in statistical
analysis of word frequency distributions and in the algorithms of
authorship attribution. Instead of the hermeneutics of play, in
which the goal is simply to facilitate engagement and enable
insight, these traditions assert the hermeneutics of the algorithm.
These types of interpretive activities have always possessed the
sheen of scientism--a feature which practitioners of text analysis
have sometimes emphasized and sometimes deemphasized as the tools
have moved across the disciplines and in and out of academic
fashions. Yet it may be argued that its closest family resemblance
lies not with conventional literary criticism, but with its ludic
cousin. The algorithmic analysis of text may be thought of as
lending a positivistic slant to critical activity, but it may just
as easily be thought of as yet another critical practice at the
interstices of work and interpretation--a critical act of
deformation, neither mystical nor anarchic, and yet invested with
the same power to facilitate engagement and enable insight.
Both the ludic and scientistic traditions of interpretive activity
exist on the margins of literary critical practice. Mainstream
literary critical culture tends to consign the former to the realm
of literary artifact while viewing algorithmic criticism as merely
part of the pre-interpretive organizations deemed necessary for
certain limited types of analysis. In this way, the Muse and the
mathematician each abut the boundaries of a central position
(mainstream literary critical practice) which is too rational for
the former and too mysterious for the latter. As long as the work
generated by these tools is perceived as pre-interpretive--or worse,
positivistic--humanities computing in literary studies will continue
to operate outside of mainstream discussions in the discipline.
I would like to suggest that we reenvision text analysis from the
theoretical standpoint of the ludic tradition--envisioning
computer-assisted text analysis in literary studies as
preinterpretive in the strong sense of exploration and play.
Reforming and refactoring to enable insight, notice aspects, and
reveal codes.
I have developed a set of software components intended to demonstrate
these principles called the D-Machines. The D-Machines consist of a
set of program modules that allow one to perform discrete
deformations of text: e.g. print backwards and forwards, switch
gender terms, colorize word frequencies, show only nouns or only
verbs, and so on. I will demonstrate this system and show how it may
be used to enact the principles I have set forth concerning the
reconception of text analysis as an activity which participates in
the creation of those alternative textualities which undergird all
literary critical acts.

References

Jerome
J.
McGann

Lisa
Samuels

Deformance and Interpretation

New Literary History

30
1
25-56
1999

Warren
F.
Motte

OuLiPo: A Primer of Potential Literature

Normal, IL
Dalkey Archive Press
1998

Raymond
Queneau

Cent Mille Milliard de Poems

Paris
Gallimard
1997

Jean
Starobinsky

Words upon Words: The Anagrams of Ferdinand de
Saussure

New Haven
Yale UP
1979

Richard
Rutt

The Book of Changes (Zhouyi)
Durham East-Asia Series

1
Richmond, UK
Curzon
1996

What is text analysis, really?
Geoffrey Rockwell

In a mock confrontation between Allen Renear and Jerome McGann at the
ACH/ALLC in 1999 at the University of Virginia, two views as to what
a text really is were put forward. Renear put forward, for the sake
of the confrontation, the OHCO (ordered hierarchy of content
objects) perspective while McGann practiced a view of text as
performance.[1] Susan Hockey, Chair, with Allen
Renear and Jerome J. McGann, Panel: "What is text? A debate on
the philosophical and epistemological nature of text in the
light of humanities computing research", ACH-ALLC, June 9-13
1999, at the University of Virginia. In the context of a
humanities computing conference this confrontation was designed to
highlight the relationship between theories of text and ways of
representing texts digitally. Renear's Platonic view of the text as
a real abstract OHCO fits nicely with the dominant practice for the
digital representation of texts, namely the guidelines of the TEI.
McGann instead gave us an example of a reading that was both a
performance itself and pointed to the (hypertextual) links within
and around the text. McGann's challenge to Renear was to show how a
reading of a text both was the text and could not be captured by an
OCHO. The confrontation succinctly opened again the question of the
relationship between how we represent texts, how we use them, and
our theories of textuality.
What does this have to do with literary text analysis and computing?
What was not made clear in the confrontation was the role of the
tools we use for accessing and manipulating digital texts; tools
which I will call text analysis tools. If we are to take McGann's
public performance of a reading as an analogue for what we wish to
achieve digitally, we have to think not just about how we represent
the text but also about analysis and the tools that are used to
perform the analysis on a computer. The logic of the tools, despite
(or because of) their tendency to become transparent in use, can
enhance or constrain different types of reading which in turn makes
them a better or worse fit for practices of literary criticism.
Another way of saying this is that we have a model of
computer-assisted literary text analysis that is guided by a view of
what a text is and how we should use it that does not match the
practice of many contemporary literary critics. (It should be noted
that this is not true in the field of computational linguistics and
may not be true in literary criticism in the future). Consequently,
as others have pointed out, text analysis tools and the practices of
literary computer analysis have not had the anticipated impact on
the research community. This is often blamed on the absence of
easy-to-use tools, especially tools that take advantage of OCHO, but
I will argue in this paper that there are two other issues that have
to be taken into account. First, the tools we have (and even those
we anticipate) have emerged out of a particular critical tradition
that I will call an "editorial" tradition going back to tools for
editors of concordances. To understand the current state of text
analysis tools we need to review their history in terms of the
practices they complement and the theories of textual practice they
augment. Second, I will argue that the moment when humanities
computing could have an impact on literary criticism through the
provision of critical tools (and relevant theories of what computer
based text analysis) is passing as server-based text access tools
that provide access to licensed digital archives seem to satisfy
most of our colleagues while we keep on imagining personal tools. In
other words, we will soon be Googled out of theoretical relevance -
the text tools developed outside the scholarly community (for
digital library access) may prove a closer fit to the practices of
our colleagues than our elaborate analytical tools.
While I doubt we can resist the commercial forces that lead to the
bundling of limited tools and texts, we can understand this process
in terms of its relevance to the practices of our colleagues and
imagine an alternative that is relevant to contemporary literary
criticism. This paper will therefore conclude with (yet another)
proposal for a model for text analysis tools, a portal model. The
portal model provides us a way of taking advantage of the trend away
from personal tools towards community tools while also engaging a
different critical practice of playful criticism. The theory of
analysis illustrated is in a hermeneutical tradition which
incorporates play in method and which is best expressed in the work
of Gadamer. A portal for text analysis can finesse the problems of
ease-of-use while also providing a virtual play-pen for contemporary
critics to try computer-assisted techniques beyond those provided by
the commercial publishers of e-texts. The portal ironically could be
the backdoor through which our colleagues could be introduced to the
playful work of humanities computing.
That said, we should be honest and admit that much of our discourse
around tools is for our own sake. It is our humanities computing
play with tools and texts. Does it matter if anyone else ever uses
these tools as long as they help us understand the practice of
reading digital representations? The portal prototype to be
demonstrated, while it may have practical applications, is for
humanities computing an attempt to illustrate a particular
relationship between a theory of texts and analysis on the one hand
and an interface for text analysis that implements that theory on
the other hand.
In conclusion, this paper will do the following:
1. Present a short history of text analysis tools as they
evolved from batch concording tools to server-based digital
library access tools. This history will focus on the
relationship between the form of the tools with the
practices they enabled.
2. Present an alternative definition of analysis building
on Gadamer's hermeneutics of play.
3. Demonstrate a text-analysis portal prototype that is
designed to enable playful practice.

Bibliography

J.
Bradley

G.
Rockwell

Watching Scepticism: Computer Assisted
Visualization and Hume's Dialogues

Research in Humanities Computing

5
Oxford
Clarendon Press
1996
32-47

Hans-Georg
Gadamer

Truth and Method

Trans.

W.
Glen-Doepel

2nd ed.

New York
Crossroad
1985

Johan
Huizinga

Homo Ludens: A Study of the Play-Element in
Culture

Boston
Beacon Press
1950

I.
Lancashire

J.
Bradley

W.
McCarty

M.
Stairs

T.
R.
Wooldridge

Using TACT with Electronic Texts

New York
The Modern Language Association of America
1996

R.
G.
Potter

Literary Criticism and Literary Computing: The
Difficulties of a Synthesis

Computers and the Humanities

22
2
91-97
1988

G.
Rockwell

J.
Bradley

Eye-ConTact:Towards a New Design for Research
Text Tools

Computing in the Humanities Working Papers

1998

URL: . Also at URL: .

G.
Rockwell

J.
Bradley

Empreintes dans le sable: Visualisation
scientifique et analyse de texte

A.
Vuillemin

M.
LeNoble

Litterature, informatique, lecture

Paris
Pulim
1999
130-160

G.
Rockwell

The Visual Concordance: The Design of
Eye-ConTact

Text Technology

10
1
73-86
2001

Computer-Assisted Text Exploration
Stéfan Sinclair

Mon plaisir peut très bien prendre la forme d'une dérive.
La dérive advient chaque fois que je ne respecte pas le
tout, et qu'à force de paraître emporté ici et là au gré des
illusions, séductions et intimidations de langage, tel un
bouchon sur la vague, je reste immobile, pivotant sur la
jouissance intraitable qui me lie au texte.
-- Roland Barthes, Le Plaisir du
texte

Existing text-analysis tools can be very useful if one knows what
questions to ask (and how to ask them). In general, they presuppose
a researcher who has read a text, who has formulated some questions
about it, who then sets the text aside while using analysis tools to
attempt to answer the questions (with a text in electronic form that
is rarely viewable in its entirety). Thus, data completely displaces
the text, at least temporarily, as the object/objective of study in
text-analysis.
Etymologically, analysis denotes breaking something up or loosening
it. Computer-assisted text-analysis tools have fully exploited the
flexible, digital nature of the electronic medium to allow texts to
be segmented in innumerable ways. It has proven far trickier to
reconstitute the divided parts into meaningful units, in large part
because this step depends on an interpretive intention that is
beyond the capabilities of current tools. Connotatively, analysis
includes this interpretive or synthetic phase that completes the
circle of segmentation and unification, but text-analysis tools have
historically stranded the literary critic at the half-way, base arc,
part of the process (though many imaginative and resourceful
colleagues have made their own way back up).
The computer need not manifest signs of intelligence to play a role
in completing the analytic process, it need only help the critic do
so. One fairly simple strategy for this is to create paths from the
data (segmented text) back to the integral text. That functionality
was precisely the motivation behind the creation of the first
version of HyperPo, an online text-analysis and exploration tool
(see <>). HyperPo can generate many of the usual types of data in
text-analysis, such as frequency, collocation and distribution
lists, but it can also create links from those data back to the
text, displaying both simultaneously (see Sinclair 1997 and 1998 for
more details on these functions). As such, HyperPo can break down a
text into constitutive parts and do comparative analyses, but it can
also reconstitute a text from any of those parts.
Because there is a high degree of speculation and experimentation
involved, I have found it more useful to view such decomposing and
recomposing methods less as analysis and more as exploration. I
navigate through the text and data the way one might explore the
streets of an unknown city or the trails in an expansive parkland;
various things along the way may prompt me to change directions, and
though I often don't know where I am going, I know that I am somehow
accumulating a broader representation of the terrain.
The notion of text-analysis as exploration has recently led me to
develop some more adventurous functions for HyperPo (functions that
may not be available at the above web address until August 2002).
These new functions are less concerned with the immediate analysis
of a text and more concerned with multiplying the means for its
traversal, its discovery, and enjoyment (perhaps even Roland
Barthes' jouissance). Playing with a text may not contribute
directly to its analysis, but I believe it can contribute to its
appreciation and perhaps its understanding (in ways that might be
next to impossible to measure). As Malcolm McCullough states in a
manner that is fully compatible to the development of a literary
interpretation: "play often lacks any immediately obvious aim, other
than the pursuit of stimulation, but functions almost instinctively
to serve the process of development" (223).
Like HyperPo itself (HYPERtexte POtentiel), the new functions are
inspired by the work of the Oulipo (OUvroir de LIttérature
POtentielle; see <>
for more information on this group). Interestingly, the Oulipo
divides its activities (in characteristically ironic terms) between
"l'analoupisme" (the analytic) and "le synthoulipisme" (the
synthetic) and states that "the synthetic branch is more ambitious;
it is the primary vocation of the Oulipo. It is a matter of opening
to our predecessors new and unexplored ways" (17, my translation).
Such are my ambitions too.

References

Roland
Barthes

Le Plaisir du texte

Paris
Seuil
1973

Malcolm
McCullough

Abstracting Craft: The Practiced Digital Hand

Cambridge, MA
MIT Press
1996

Oulipo

La Littérature potentielle

Paris
Gallimard
1973

Stéfan
Sinclair

HyperPo: The Next Generation

ACH-ALLC '99 Conference Proceedings

Virginia
University of Virginia
1999

Stéfan
Sinclair

L'HyperPo: Exploration des structures lexicales
à l'aide des formes hypertextuelles

Greg
Lessard

Michael
Levison

ACH-ALLC '97 Conference Abstracts

Kingston, ON
Queen's University Press
1997