Gripla - 2020, Síða 24

23
That said, when we break the problem down to address it with a stylom-
etric method, it turns out that, in fact, we are dealing with at least three
separate problems each requiring a dedicated approach to a solution. First,
there is the problem of whether A or C’s divergent text in chapters 13–18
contains greater stylistic consistency with the text of the parallel chapters.
This is a problem which stylometry is well-positioned to address, as will
be discussed. Second is the problem of whether the sections unique to
the C-redaction — the three þættir in chapters 5–12, the post-Guðmundur
section in chapters 22–31, and finally Þórarins þáttr ofsa (chapter 32) — are
stylistically related to the canonical Guðmundur chapters. This is a more
complex problem from a stylometric point of view. To date, our attempts
to test this second problem have been inconclusive.51 The third problem is
whether chapters 1–4 in both redactions should indeed be considered an
abbreviation of a now-lost text. This is also a difficult problem to approach
stylometrically, since chapters 1–4 are very short.52 In light of these mat-
ters, this article focuses on the first problem: is A-divergent more consist-
ent with the parallel chapters of both versions, as Magerøy argued? Or is
C-divergent more consistent, as Erichsen argued? As will be shown, the
stylometric evidence is sufficient to accept Erichsen’s conclusion and reject
Magerøy’s: C-divergent is more consistent with the style of the parallel
chapters in both A and C, while A-divergent is likely a retelling.
From stylistics to stylometry
Before we proceed, let us discuss stylometry in general. What is stylom-
etry and what distinguishes it from stylistics? From the point of view of
its fundamental premise, there is little separating the former from the
A STYLOMETRIC ANALYSIS OF LJóSVETNINGA SAGA
51 In particular, we applied Multidimensional Scaling to these different parts of the
C-redaction to determine their stylometric relationships. The results supported neither the
hypothesis that these sections are interpolations nor the hypothesis that they are straight-
forwardly consistent with the remaining texts. As such, further research will be required to
address this problem.
52 The word counts for these chapters alone falls well below the acceptable thresholds
discussed below. As with the previous problem, we performed some initial tests which
were inconclusive. Namely, the calculated cosine distances were highly dependent upon
parameterization (for more on these terms, see below), indicating a high likelihood that the
results could be explained as random chance.