Bibliotheca Arnamagnæana - 01.06.1999, Page 234

214
Part Two
non, he calculates deviations from a supposed norm, which is based on
the purely theoretical assumption that occurrences of the phenomenon
in question are equally distributed within the corpus. Secondly, the ob-
served deviations are divided by theoretical standard deviation,1 * * * * * 7 thus
obtaining a “reduced deviation” (écart réduit), which in its tum may
serve as a base for classification (Muller 1968: 93-94; cf. 1981:
135-36). The effect of this operation is an increase in the deviation ob-
served in long texts, at the expense of the deviation in shorter texts,
which are “anonymized” by being removed in the direction of the ex-
pected norm. In this way the risk of getting extreme deviations merely
as a result of the uncertainty brought about by the shortness of a text is
reduced. According to Muller, “this method can be used in comparing
the frequency of some lexical units in one text with the frequency of the
same units in another corpus, taken as a point of reference”,8 which is
exactly our case, where the frequency of the particle in Eddie texts is
compared to the frequencies in skaldic poetry taken as a reference cor-
pus.
Table 6 sets out the names of “major” skalds (or the name of the poem
in cases where the identity of the poet is unknown or uncertain) ranked
according to their age. Furthermore, the table lists the observed (obs.)
and expected (exp.) number of particles, the factual deviation between
these values (dev.), the deviation reduced by Muller’s method (red.
dev.), and, finally, rank according to frequency of particle.9
1 Theoretical standard deviation, er, is calculated according to the following formula: cr =
Vnpq, where p is the probability of the occurrence of a certain event, q is the probability
that it does not occur (p + q = 1), and n is the number of trials.
In the case of the occurrence of the particle in skaldic poems, p is the probability that a
chosen skaldic verse contains one occurrence of the particle (the possibility of more than
one particle in any one verse having been discarded).
It should be noted that according to Muller the method does not work perfeetly when p
is very small, as it is here (Muller 1968: 66). I am unable to assess the importance of this
objection, however.
8 “Cette méthode peut s’appliquer å la comparaison entre la fréquence des vocables dans
un texte et la fréquence des mémes unités lexicales dans un corpus pris comme référence”
(Muller 1968: 94; cf. 1981: 136).
9 When skalds are assigned to identical periods of time, the table gives the arithmetical
mean of the ranks in question. Thus, the contemporaneous skalds Holmgpngu-Bersi and
VIga-Glumr are both assigned the rank 6.5, which is the mean of ranks 6 and 7; Glumr
Geirason and Gisli Sursson receive the rank 8.5, which is the mean of ranks 8 and 9, etc.