20iv49
ab-az-ab 1 1 6
ab-eg 1 1 7
ad-ab-eg cut cut
------------------------------------
The original table of mine had 2 entries, one the total of units
from complete groups, the other the total of fractional values. These
were then added together in the published form.
The fractional values are made necessary, apart from (A) alternative
readings and doubtful signs (which might be all treated as xx , as you
suggest, without affecting the frequency figures), by (B) the fact that all
my frequencies have been plotted, in the first instance, as occurrences in
groups of definite length. I have therefore in each case had from inspection
to try and fix fractions which allow for the most probable length.
I gather that in -- ab-ar /- your rules would require the group
to be treated as ///-ab-ar-/// . This means that, in any frequency count, it
will be counted as a 4-letter word (of which the probability is only about
¼) and ab will not get any credit as an initial (for which the probability
is about 1/2). The frequency table should, by addition, be our source for the
relative proportions of groups of different length; and taking our 2 inter-
pretations of the index above we get:-
Groups. Syllables. Average length.
Yours: 6 19 3.3
Mine: 7 20.5 2.9
That is to say, your groups come out longer than, in my opinion, the pro-
bability calls for.
In fact, for the Pylos and Knossos material no great divergence would
probably occur, and it is the "A" spelling which is responsible for most of the
complication. Unless you can see the end of the line before, at HT, you can
never be 100% certain that the first sign in a line really is an initial.
At the moment, then, our manipulations of the 'secondary index' are:-
Mine: Fractionalise length and alternative signs on the basis of estimated
probability; count up complete and certain unit totals in one column,
fractions in another.
Yours: Take the longest alternative in each set; treat all doubtful signs
as xx ; all values at unity.
My only criticism of your method is that as soon as frequencies by
position are plotted one must treat xx (illegible or alternative) and
/// as two different things. The compromise I would suggest is to retain
xx for the first category, and to include in it all signs not absolutely
certain; and to use, for the 2nd category, uncertainties of length and
position, the following notation in the 'secondary index':-
ab -12
01- ab-ar -01 etc.
indicating that in the first case alternatives must be plotted with either
1 or 2 additional letters; in the second that either 0 or 1 additional let-
ters must be assumed before and after (ie. 4 combinations). If in the first
case two additional letters are more probable by far, write:
ab -(1)2
and count ab-xx at 1/3 and ab-xx-xx at 2/3. It may sound rather
tedious to plot fractional values, but in practice it is quite straight-
forward if you adopt a notation to enter into the relevant boxes:-
· = 1 unit
_ ½ unit
_ 1/3 unit
+ ¼ unit
_ 1/5 unit , or something of the sort.
In adding up overall frequencies (ie. not by position), the values for
each sign of course join up to unity again.
The 'compromise' form of 'secondary index' would then be as fol-
lows, in which provision is made for fractional scoring, and in which the
by occurrence scoring is identical with that in the table you gives: