Íslenskt mál og almenn málfræði - 2020, Page 296

exhibit Adv−Vfin to a lesser degree than the private letters during this period as
potential evidence to support language planning effects. In my view, there is
nothing to suggest that Adv−Vfin is excluded somehow from the newspapers
for other reasons, so the most likely reason is that writers were avoiding the fea-
ture (see also section 3.4.1.4 on potential hypercorrection).
Question 3: This is something that I wanted to address quantitatively through
modelling. There is an important paper by Hinrichs et al. (2015) that establishes
such a relationship between a number of different variables in English. They use
this to show that prescriptive dicta have an effect such that texts that avoid fea-
ture A, also tend to avoid features B and C, where all features are considered
non-standard. This is thus an additional factor in their analysis over and above
the time factor which shows a gradual shift in usage frequency — the last of
which is typically the only or at least most important factor in studies on the
effects of prescriptivism (see also my answer to HÞ’s question no. 18).
The statistical modelling of Hinrichs et al. (2015) is similar to my methodol-
ogy so I tried to test whether there was a relationship between individuals (or
individual texts) and the different variables. However, I immediately ran into a
lot of problems. My texts per individual (or individual newspaper) are usually
very short, so it is not very likely that, say, a letter writer or even a newspaper,
will turn up many examples of each of these linguistic variables. This is easier
with a corpus of longer, internally more consistent texts, like a full-length novel.
Another problem that I ran into is that there may be changes in the editors and
staff of the newspapers, so in what sense is this the same “individual” (a more
general problem)?
Moreover, quantifying the variables as a factor of comparison was far from
straightforward. The sá/hinn variable contains some “noise” as outlined in the
dissertation, which left me with the option of using instead the normalised rate
of hinn per 10,000 words, the same method as I used to quantify the univariate
variable maður. However, this way I am throwing away a lot of data points. I did
try to keep these quantified sá/hinn and maður variables to measure covariation
with Vfin−Adv/Adv−Vfin as a basis, but the models did not converge with
these effects in. This has probably everything to do with the make-up of the cor-
pora and the amount of data points per “individual”. These variables never
resulted in a converging statistical model, meaning that the result could not be
interpreted.
The student essays were my way of (going towards) addressing the criticism
voiced in Hinrichs et al. (2015) with regard to previous studies on the effects of
prescriptivism, namely that it is insufficient to find a temporal decrease, as a
(delayed) response to prescriptive dicta, to convincingly argue for an effect. In
the Reykjavík Grammar School data, we have the temporal effect in addition to
aspects such as progression of study and graduation score: freshmen might be
Heimir Freyr van der Feest Viðarsson296