BFM Old French Text Corpus: Current State and Prospective Developments

The BFM (Base de Français Médiéval) Old French Corpus
was founded in 1989 by Prof. Ch. Marchello-Nizia, and
its compilation continues. Céline Guillot has been the project
leader since 2006. At present, the main corpus, BFM1, includes
74 complete Old and Middle French texts (approx. 3 000 000
words).
The texts included in the BFM cover a considerable geographic
area and an extensive time span, with texts from the 9th century
(including the first known French text, the Serments de
Strasbourg) to the end of the 15th century. Both verse and prose
texts are represented, as well as different domains and genres.
The initial aim of the project was to provide scholars with
reliable text data on the oldest period of the history of the
French language. It was meant to complement the Middle
French Text database (14th and 15th centuries) and
FRANTEXT database (16th – 21st centuries), both developed
in Nancy by the ATILF laboratory. However a number of
Middle French texts have been added to the database for
different projects.
By now, about 50 theses and a significant number of research
books and articles have been prepared with the use of the BFM.
All texts in the main corpus are digitized critical editions. The
choice to use editions, and not manuscript transcriptions, was
made in order to create an extensive corpus in a relatively short
period of time: transcribing a manuscript requires much more
time and funds than digitizing a modern edition. However, a
subcorpus of manuscript transcriptions is being developed. One
of its aims is to provide a precise evaluation of the reliability
of critical editions as a source of data for different kinds of
linguistic research. In general we argue that using editions or
manuscripts are complementary and not mutually exclusive
approaches to creating old text corpora (cf. Heiden & Lavrentiev
2004).
All BFM texts are XML encoded with the tags recommended
by the TEI. However some of them contain only a very limited
markup (headers with some metadata and page breaks or lines in verse for reference purposes). A richer TEI encoding is
applied to a little more than a half of the BFM texts. Its
principles are described in (Heiden & Guillot 2002) and a
complete description is available at the BFM website.
The BFM texts are not directly accessible to users. They can
be searched by means of precise queries (e.g., discrete lexical
items, word and phrase concordances, co-occurrences, statistical
analyses, etc.) via Weblex search and analysis engine (using
CQP query language).
At present, several directions have been chosen for the
development of the BFM.
The first direction is concerned with the elaboration of precise
text typology, which is necessary to evaluate the
representativeness of a text database, a crucial question for all
corpus studies. The texts in the BFM are characterized by a
number of variables, such as the date, the region, the author,
the domain, the genre, etc. The definition of almost each
variable is in fact connected with a number of methodological
and technical issues. The date can be for instance that of the
original text composition or that of the manuscript. Author’s
age can also be a factor to take into consideration. Only
approximate dating is available for many texts and manuscripts.
Domain and genre are the main variables that contribute to
characterizing the reprensentativeness of a corpus (cf. Lee
2001). It is however extremely difficult to set a unified genre
taxonomy valid over a number of centuries. Many texts belong
simultaneously to several domains (e.g. religious and literary,
historical and literary).
To deal with this complexity, most of the variables have been
encoded by means of multiple fields in the metadata database.
Depending on the nature of a query, these can be used in
different ways. If the aim of the query is to get a list of texts
corresponding to certain criteria, “informal” fields with keyword
value type can be used. If the aim is to “cut” the corpus for
some kind of contrastive analysis, a unique value is selected
for each text on the basis of a formal procedure. If the aim is
to place a text in a multidimensional typological environment,
it is necessary to model the relations between the different
values of a variable and to quantify these values.
The work on text typology in the BFM is in progress, and we
will present its current state by the time of the conference.
Another direction in the BFM development is its linguistic
annotation. A few texts have been morphologically tagged in
a semi-manual mode (with SATO software developed by
François Daoust at the University of Québec in Montréal). A
complete automatic morphological annotation optimized on
the basis of text typology is currently envisaged. Annotation
of particular linguistic features (e.g. semantic features of
demonstrative adjectives) is conducted in the framework of
related linguistic research projects.
The development of the BFM is closely related to that of
Weblex. A completely new platform making possible
personalized corpus creation and online annotation is under
construction.
An important effort is being made to work out consensual text
encoding and description procedures with the other projects of
diachronic French corpora. The TEI Recommendations are in
fact too general to ensure real corpora interoperability. A
Consortium for Medieval French Corpora (CCFM) was created
in 2004 with the purpose of consolidating efforts of different
projects. The BFM team participates in this consortium along
with Laboratoire de français ancien (University of Ottawa),
DMF team (ATILF laboratory, Nancy, France), University of
Stuttgart (Germany), University of Zürich (Switzerland),
Anglo-Norman Dictionary project (UK) and others.
Web
BFM: <http://bfm.ens-lsh.fr>
Weblex: <http://weblex.ens-lsh.fr/wlx/>
Bibliography
Heiden, S., and C. Guillot. "Capitalisation des savoirs par le
web: une application de la TEI pour l’encodage et l’exploitation
des textes de la Base de Français Médiéval." Ancien et moyen
français sur le Web, enjeux méthodologiques et analyse du
discours. Ed. Pierre Kunstmann, France Martineau and Danielle
Forget. Ottawa: Les éditions David, 2002. 77-92.
Heiden, S., and A. Lavrentiev. "Ressources électroniques pour
l’étude des textes médiévaux: approches et outils." Revue
française de linguistique appliquée IX.1 (2004): 99-118.
Lee, David Y. W. "Genres, Registers, Text Types, Domains
and Styles: Clarifying the Concepts and Navigating a Path
Through the BNC Jungle." Language Learning & Technology
5.3 (2001): 37-72.