Mapping Differences in 1st Century Greek Style

Mapping Differences in 1st Century Greek Style

David
L.
Mealand

University of Edinburgh
D.Mealand@ed.ac.uk

2002

University of Tübingen

Tübingen

ALLC/ACH 2002

editor

Harald
Fuchs

encoder

Sara
A.
Schmidt

A series of studies has built up a set of 178 samples of Greek texts in a format
which now provides a dataset comprising 178 rows and 35 columns. The previous
studies by this author and relevant works by others are listed in Mealand
(1999). There are 30 effective linguistic features being tested for each sample,
and to these we need to add classification categories such as author, genre,
subgroup, text and sample. The 178 samples and 30 linguistic variables alone
provide 5340 observations. We are now in a position to make report on detailed
stylistic comparisons of a wide range of texts from 300 BCE to 200 CE but
focused on texts from 100 BCE to 100 CE. Some manual intervention and a set of
programmes eventually turns complexities of the Greek texts from the TLG into
thousand row columns of single words, and simpler routines then allow a pentium
to make much shorter work of the word counts.
The texts include a range of Greek historical writers: Polybius, Dionysius of
Halicarnassus, Plutarch; also a set of Greek technical writers, some samples of
letters and other documents preserved on papyri, and a range of biblical and
other Jewish and Christian writings of the period. The latter include samples
from the Septuagint (4 Kingdoms, Deuteronomy, Isaiah, 2 & 4 Maccabees
and the like), most of the New Testament books, Philo, Josephus, Clement of
Rome, Ignatius of Antioch and Polycarp.
The variables include a set of stylistic features which are mostly function
words: a set of sentence connectives equivalent to but, for, and, therefore, on
the one hand; a set of conjunctions which introduce clauses such as if, while,
in order that, where, whenever, and the like; a large set of prepositions such
as away from, into, out of, in, on account of, according to, after, with, about,
on, towards; and a set of (genitive) participle endings. These variables were
chosen as a set of function words suitable for inspecting the style of samples
of 1000 words. (It would be possible to extend the range of the linguistic
features by choosing much larger samples of those texts which have the necessary
bulk. Some, but not all, of the works are extensive.) Variables were selected in
the light of recent studies showing the effectiveness of high frequency words
and or features linked to syntax.
The statistical method used was Correspondence Analysis as this plots both
samples and variables and allows one to see not only relations between texts,
but also those linguistic features which associate or separate the texts. CA was
run on a Sun machine shared by scores of other users. Correspondence Analysis
also differs from some other multivariate analysis in not requiring the data to
be normally distributed. This issue is a problem when samples from large numbers
of different authors are involved. Many stylometric studies limit themselves to
a handful of authors or even just two or three. This study is attempting to make
comparisons across a wide range of ancient literature.
Using all the data we obtain a complete plot (OHP Cell 1) showing the location of
178 samples and 30 variables. This was the plot to which much careful attention
was given. But it is not an easy plot to interpret and for display purposes a
supplementary strategy provided an overview plot of cumulated samples - mostly
of 5000 words from 5 samples from each author combined together. This preserves
the broad shape and outline of the more complex plot. Each author still appears
in the same general segment of the plot as in the original. We can, however,
more easily see how the authors relate to each other and to the linguistic
variables. We can also interestingly see how the original plot reveals the wider
spread of 1000 word samples compared with the cumulated 5000 word samples. This
gives added insight into variation within the work or works of each author
The main interpretation of the plots is that the more fluent Greek writers (i.e.
Polybius, Dionysius, Plutarch and their associates) lie to one side (right) of
the plot while the Septuagint translations of biblical texts lie to the other
side. (I use the term fluent to capture the more complex
observation that the writers who appear on the right of the plot are those
whose work would be regarded as more acceptable in style by an educated
Graeco-Roman public. Jerome is on record as being embarrassed by some
features of biblical Greek.) As the horizontal axis represents
Dimension 1 this captures the largest single component of the inertia. The
vertical axis seems to separate narrative texts lower down the plot from speech,
reported speech, treatises, and ordinary letters in the centre of the plot.
Paul's letters are even higher up the plot though neither to the extreme left
nor right. Paul's style is therefore between that of the Septuagint and that of
Dionysius on the one dimension, and much higher than the others on the vertical
dimension. This may reflect the very argumentative style of Paul's dictation in
his letters.
The plots from Correspondence Analysis show the location of the variables in
relation to the location of the samples and from this we can draw inferences
which we can confirm by inspecting boxplots of means and other statistics. This
reveals the relative weight of usage of our main criteria in the different
writers. So for example we can see that Paul makes heavier use of alla, gar, ei,
ou, mh, hina (but, for, if, not, not, in order that) and to some extent dia and
eis (into), while making moderate use of de, oun, and men (but, therefore, on
the one hand). He has low usage of kai, meta and peri (and, after/with, about).
Much work on Paul has focused on internal differences of style. This study shows more clearly than before just how Paul
differs widely from most, but not all, of the other authors selected.
The Septuagint passages from Biblical texts have higher use of kai, apo and epi
(and, from, on)and to some extent hews (while/until), moderate use of ou (not),
and low usage of alla, gar, de, men, oun, ei, mh, kata and peri (but, for, but,
on the one hand, therefore, if, not, according to/against, about). But the Septuagint texts themselves differ and I conclude
that the differences are not just due to the style of samples from 4
Kingdoms, nor just to differences between translated texts and those
composed in Greek. The samples from Genesis and Proverbs lie between these
other Septuagintal groups. The more fluent Greek writers have higher
use of de, men, kata, peri (but, on the one hand, according to/against, about)
and the genitive participle endings, moderate use of alla, gar, ou, kai, and oun
(but, for, not, and, therefore), and low usage of hina and mh (in order that,
not). Here it is worth noting that despite his diffidence
Josephus is close on stylometric grounds to Dionysius and Plutarch. It is
also worth noting that the few selected samples from non-literary papyri are
not as far distant from Dionysius as some might expect. Finally we should
note that the Greek technical writers do not form a coherent group, as
Alexander and others tend to assume, but lie in several different directions
around the group including Dionysius and Plutarch. All these provisional
conclusions suggest lines for further investigation.
Therefore (oun) is more heavily used in some NT texts esp. some Johannine texts.
The conjunction hopws (in order that) is mainly used by the more fluent Greeks.
The preposition ek (out of) is more heavily used in Revelation, moderately in
the Septuagint. The word en (in/by) is used more heavily in Colossians and
Ephesians and in some of the Johannine texts, and moderately high usage is also
found in the Septuagint and Paul. Philo and Josephus often side with the more
fluent Greeks on most criteria, but do have a raised usage of pros (towards).
The results reported here make use of a larger selection of samples than any of
the previous work I have published on the stylometry of 1st Century Greek texts.
We can see not only that the writers fall into different groupings, but also
which literary criteria tend to distinguish their styles one from another.
Some of the authors used could not provide further samples as their works are
relatively short. Others could provide massive amounts of further text for
analysis. That leaves plenty of scope for further researchers to explore the
more voluminous authors further, and perhaps, by taking much larger samples from
them, to scrutinize stylistic features of their work which appear slightly less
frequently than the highest frequency words and features which are the main
criteria used here.

Bibliography:

David
L.
Mealand

Style, Genre and Authorship in Acts, the Septuagint,
and Hellenistic Historians

Literary & Linguistic Computing

14
4
479-505 esp. 501-502
1999

a list of 29 relevant works will be found in

Plots:
Plot of large text samples
This plot shows the relative location of the different large cumulated samples of
text from the main authors. The samples from Genesis and Proverbs appear near
Acts on the more detailed plots: 2 and 4 Maccabees nearer Dionysius. Samples
from the papyri appear near Hebrews. Various Greek technical writers appear on
the more detailed plots above and below Polybius, well above Philo, to the right
of Acts and below Acts.
Plot of literary variables