Annotating as a Document Management Tool

I Presentation
Current technology (scanners, OCR) is paving the
way for the creation of vast machine-readable
document holdings (digital libraries, technical
full-text databases, CD-ROM, etc.). At the same
time, the range of potential users is also widening
considerably. It appears that under these conditions, very general language engineering tools
may be both too powerful and deficient in providing for all conceivable user situations. In addition, personalized, reader-oriented document management tools such as annotation have to be
designed for running reading products.
To reach this aim, three partners (BNF [Bibliothque Nationale de France], AIS [Advanced-Information Systems] and IRIT) have already collaborated in the past with regard to the definition of a
strategy for creating the machine-readable holding
for the BNF and specifications for the CARS
(Computer Aided Reading Station) [Chahuneau &
Al, 1992]. This work is now developed as a part
of the European MEMORIA project (Multimedia
Electronic MemORIes At hand) and is used as a
background for this article.
II Electronic annotating
Nowadays, researchers pay a particular attention
to electronic annotating [Nielsen 86], [Stiegler
94], [Virbel 93]. This is due to the convergence of
two principal factors:
(1) from a technical point of view, and particularly
in a context of hypertexts, users are allowed to
widen their documents by adding different individualized augmentations such as characterizations
of passages, attachments of various comments,
links between passages, etc.
Since all annotations could be memorized by a
system, they present the two following major advantages, compared to traditional form annotating
(with paper and pencil):
– The set of annotations itself constitutes a
material to be managed and exploited subsequently, as long as the tools necessary for
exploitation are available.
– Re-reading and later consultations of the document can actually be performed through
annotations.
(2) at the same time, the technical feasonbility of
creating very wide corpora of digitized documents
(from paper books and catalogues) and the profile
broadening of potential users of these corpora,
favour personalized management modes of these
corpora. Therefore, annotation seems to provide a
mode of personalization of the document through
its reading, since it is qualified as a highly individualized activity.
In this context, we characterize this activity as a
Dynamic Annotating contrary to the static character qualifying it in paper context. The characterization as dynamic is due to the following main
reasons:
– firstly, as in the classic context, the annotating activity is contemporary with the reading process and more generally all reading
concerns;
– secondly, various types (discursive, graphic,
sound, etc.) of annotating trail themselves
constitute a material to be managed and exploited as far as suitable tools are available;
– thirdly, the re-reading of the text could be
organized from and through annotation. So
annotation could be considered a new access
mode to the text (such as the table of contents, the alphabetic index, keywords, etc.);
– fourthly, annotating can provide users with a
sophisticated experimental tool for reading:
for example searching for new passages sharing some properties with a given one, or
creating passages automatically if they have
not been created, may be considered a relevant element of the dynamic character of
annotating;
– finally, if annotations are well managed, and
the expansion of the system is well controlled, then this evolutive system could be a
considerable aid for researchers who study
the comprehension phenomenon associated
with the cognitive activity of reading.
III Elementary typology of annotation
and the composition problem
Experiments carried out within the scope of the
BNF, have proved the importance of different
modes of annotation. It seems to be a form of
writing which is intrinsically bound to reading, as
well as a method of management of the reading
itself and therefore the “re-readings” or subsequent consultations of Digital Library.
According to these experiments, functions generated by the annotating activity seem to be numerous. They particularly concern memorization; capitalization of the reading’s results; indirect
dialogue with the author of the text under consideration; possibility of communication between
readers; management of the subsequent consultations and re-readings; scheduling of the operations to be carried out aside from purely reading.
From the observations collected from professional
or academic readers (who were in fact mainly
researchers in the fields of Humanities and Social
Sciences) using the sources of the BNF, we defined eight principal elementary events of annotation:
(1) Organize into a hierarchy: Attach degrees of
judgement according to importance, representativeness, etc.
(2) Create an architecture: Add on by explanation
of structural elements (e.g.: highlight items for
a purely discursive listing).
(3) Contextualize: Create a passage which is relevant for semantic apprehension of a term or
expression.
(4) Schedule: Plan operations to be carried out
aside from purely reading (e.g.: “to re-read”,
“to translate”, etc.).
(5) Reformulate: Modify the content of the text.
(6) Comment: Attach some text to a point, or a
passage of text.
(7) Document: Attach a document (ex: picture,
sound, film).
(8) Correlate: Establish footnotes and references
between points, passages, or zones of text or
of another text.
After selecting an object (called a passage) from a
document, the reader assigns a type of annotation
to the passage. In this context, annotated passages
can be defined either by direct selection (for example by using the mouse), automatically from
other passages which have already been created,
or by using markers, if they are present.
Concerning the question of composition, we define a sophisticated model of complex annotating.
In fact, we can say that the power of an annotating
system can be measured by the intrinsic quality of
the kind of annotations defined, and by their capacity to enter in composition in order to represent
complex annotating acts ([Mazhoud et al 95]).
We define a complex annotative act as a combination of elementary acts with respect to an annotation composition mode. We distinguish the following modes:
(1) “Accumulation”, which is tied to the spatial
layout (inclusion, overlapping, etc.) of considered passages.
(2) “Reapplication”, where a reader can create
annotations on top of annotating acts already
performed by the reader (e.g. to comment the
act of criticizing a given passage rather than
the textual content tied to the critique).
(3) “Macro-annotating” is the means of composing more than one elementary event in the
same annotating act.
(4) “Chain”: as annotating events can be classified
into two classes, characterizations and attachments, this mode concerns only the second
class. Indeed, events that produce new texts
can be the object of a continuity of annotation
performing.
As some combinations of elementary acts are not
relevant, a CARS must provide users with a composition control unit which ensures that the reader’s compositions are well formulated. To be sure,
this control acts only on a “syntactical” level.
IV Annotating and reading activities
We present in this communication the Annotating
activity and its relationships with other reading
activities, in particular Mark-up, Prospecting and
Forming into corpora. During the CAR, it appears
that these activities are strongly correlated. An
example of a reading session will be given to show
this interdependence.
(1) Mark-up i.e.: the association to the text of
various markers denoting units of its logicolinguistic structure (sentences, paragraphs,
acts, scenes, etc.). Here author-structure is in
question, with regard to this sort of readerstructure where a part of annotation is composed.
(2) Prospecting, which may mean the set of possibilities offered to carry out detailed investigations of the text, in lexical, syntactic,
semantic and stylistic terms, etc.
(3) Forming into Corpora, i.e. the classification of
(segments or units of) texts, that is the composition of textual entities coming from various
texts into new sets (corpora), arguments of
textual operations.
(4) Annotating: the reader defines relevant passages which will become arguments of textual
operations.
It was seen above that annotating activity is defined as an event which consists of the selection of
a passage in a document, and of assigning a type
of annotation to this passage. Therefore, how can
a passage be selected? Different methods, which
are classics in reference theory, also run for referring to passages:
200
– direct reference:
– showing (e.g. with mouse),
– naming (proper name),
– determining a set of properties;
– indirect reference:
– determining a passage by properties shared with another passage.
As Mark-up and Forming into corpora functionalities may be represented and tailored in terms of
SGML DTD’s (e.g. TEI), it would be very interesting to also represent annotating activities and
composition in the same frame. This hypothesis
has not be evaluated for now.
V Towards an “experimental” reading
A reader-oriented document management system
with the annotating functions described above
must have the following main features:
– a form of memorization of different kinds of
reading activities such as consultation, sorting, etc., with respect to reading sessions;
– at the same time, a form of “opportunism”,
in fact every instantaneous idea or intuition
must be immediately recorded, and its effects
are systematically related to the set of the
concerned area;
– a form of exhaustiveness, i.e. the possibility
of formulating and testing in near real time
various hypotheses that otherwise remain
unchecked and, in consequence, the possibility of recording the results of these tests.
These features lead readers to a new way of reading. It is no longer a matter of simply following
lines in screen, but rather of discovering new
notions in the text by employing new concepts of
approximations and controlled reiterations. This is
what we call “experimental” reading.
References
Chahuneau F., Lécluse Ch., Stiegler B., Virbel J.
Prototyping the Ultimate Tool for Scholarly
Qualitative Research on Texts. 8eme Confrence Annuelle du New Oxford English Dictionary, Waterloo, 18-20 Octobre 1992.
Mazhoud O., Pascual E., Virbel J. Reprsentation
et gestion d’annotations. Hypertextes et Hypermédias, Hermes, 1995, 127-138.
Nielsen J. Online Documentation and Reader Annotation. International Conference on Work
with Display Units, Stockholm, 12-15 May
1986.
Virbel J. Reading and Managing Texts on the
“Bibliothèque Nationale de FranceÈ Station.
In Delany P., Landau G. eds. The Digital
Word: Text based computing in the Humanities, MIT Press, 1993, 31-52.
Stiegler B. Machines à écrire et matières à penser.
Genesis, 1994.
Sperberg-McQueen G.M, Lou Burnard eds. Guidelines for the Encoding and Interchange of
Machine-Readable Texts, ACH-ACL-ALLC,
1990.