The Order of the Canterbury Tales: Praxis of Computer Analysis

This paper is designed as a complement of Matthew Spencer's proposal:
"Reconstructing the Stemma of a Textual Tradition from the Order of
Sections in Manuscripts." It will discuss the results and implications
of our use of computer programs to produce stemmata based on the tale
order of the Canterbury Tales.

It is well known that Chaucer left the Canterbury Tales unfinished
when he died. But the degree to which they were unfinished becomes
more relevant: Chaucer never assigned a final order to his
Tales. Traditionally, scholars have seen the tales as divided into
fragments or sections1, which are more or less consistent from
manuscript to manuscript. When we take into account these fragments,
we can see that probably what Chaucer did was to work on a particular
group of tales and place them in a certain position with respect to
the others. However, he worked more in some fragments than in
others. In practical terms this means that a group of tales such as
Fragment I (General Prologue, Knight's, Miller's, Reeve's and Cook's
tales) is much more consistent in the relationships between the tales
and its position in the manuscripts in general, than others. It seems
clear that Chaucer had not made a definitive decision about tales such
as the Man of Law's, the Merchant's or the Franklyn's, while he had
worked more on the relationship between the Wife of Bath's, the
Friar's and the Summoner's tales --although their position as a group
varies among manuscripts. For this reason some of the fragments are
difficult to locate and scribes and their supervisors, editors, and
scholars have had a hard time trying to make sense of the order of the
Canterbury Tales.

Most modern scholars have attempted to "fix" the order of the tales in
order to come nearer to Chaucer's intentions, and they have speculated
as to which if the extant orders is truly Chaucerian. When we started
this research our aim was not to discover Chaucer's intentions or
which manuscript reflected them in a more precise way. We wanted to
see if there was any kind of relationship between the order of the
tales and the text extant in each manuscript, or in other words if the
tradition of the tale order and that of the text go hand in hand or
not. As part of the STEMMA -- Studies of Textual Evolution of
Manuscripts by Mathematical Analysis-- Project's research we have been
using phylogenetic programs to reconstruct textual stemmata, and we
thought that it would be a good idea to try and use them with a
different aspect of the Canterbury Tales tradition. Phylogenetic
analysis2 has been successfully used by Peter Robinson to explain the
textual tradition of the Canterbury Tales,3 but we did not know if the
same kind of analysis would yield results when applied to a different
aspect of the tradition. However, the advantages, if we succeeded
seemed remarkable. In the first place, we would overcome the long-time
prejudice that has blinded the critics. They have assumed that
Ellesmere, one of the best extant manuscripts, has the best tale order
and also that it represents Chaucer's intentions, therefore any other
orders are less valuable if not irrelevant. In the second place, we
thought that the programs might cast some light on manuscripts that
have been unclassified before and also that they could make evident
relationships that have not been seen before.

The data that we used came from a series of tale-order tables, that
were based on the one produced by Manly and Rickert4 for their
analysis of the Canterbury Tales. The original table was modified to
conform to the Canterbury Tales Project notation; therefore most of
the links became separate units that had to be taken into
consideration5. In this sense, our data shows a profound difference
with that analyzed by Manly and Rickert because the number of discrete
units is about double that in their original table. Besides, when we
take into account all the links, we realize that they are liable to be
moved just as the tales are.

The tables were converted into computer readable data that was fed
into SplitsTree. The results yielded by this program were not as good
as we expected and we decided to attempt the same analysis with
PAUP. The results that we obtained show a relationship between the
textual tradition and that of the order of the tales. What this means
is that they suggest that even if there are rearrangements on tale
order these are not completely independent of the textual
tradition. The stemmata produced by the cladistic programs also shows
that the manuscripts that Manly and Rickert judged unclassified appear
to be related in different ways. The use of computerized tools has
allowed us to overcome our prejudices and guided us in new directions
that we will continue to explore.

1 . These fragments are a tale or group of tales that appear as a unit
in the manuscripts. What usually changes is the position of the
fragments.

2 . Dr. Matthew Spencer explains the technical aspects of the programs
in his proposed paper.

3 . Dr. Robinson confirmed the applicability of cladistic programs to
textual traditions by using it with Norse sagas, specifically in his
analysis of the textual tradition of Svipdagsmál.

4 . John M. Manly and Edith Rickert, eds. The Text of the Canterbury
Tales: Studied on the Basis of All Known Manuscripts. 8 vols. Chicago:
Univ. of Chicago Press, 1940.

5 . The CTP notation was devised by Norman Blake (Cf. Norman Blake and
Peter Robinson, eds. The Canterbury Tales Project Occasional Papers
Volume II. OCH: London, 1997). He made decisions about which tales
function independently of the tales and which ones worked as units.