"The Currier languages revisited" revisited

[kckluge]

your using a different transcription alphabet

I use the GC transliteration for my analysis work. 101-C characters I convert to ee . I only used paragraph text, including 'Pb' text. Otherwise no labels, radial or circular text.

relative frequencies of prefixes & suffixes, but didn't describe how they were defined

The prefixes I used were the most frequent prefixes to words on the pages labelled as language B. A word such as  okeedy would contribute 5 prefixes to the list.   o, ok, oke ,okee ,okeed . There is no harm in adding the longer strings since they can be expected to appear low in the frequency list. Similarly, suffices to contribute to the list would be  y, dy, edy, eedy, keedy .

Other measures that might be useful to try: frequency of long words ( 6 or 7 GC-101 characters ), frequency of words containing  e or some other character, ratios of  t to k , ratios of  iin to in .

In my opinion the PCA plots do have some limitations. I just wanted to highlight that they could lead to wrong conclusions. Correlation maps such as the ones I gave on  You are not allowed to view links. Register or Login to view.  are more useful to me to visualise how closely pages relate to each other.

You're absolutely right that heat maps are a better way to visualize the structure of pairwise distances in a higher dimensional space. The split in the starred paragraph section pops out nicely in the plots there, with one set of pages looking fairly Quire-13-like.

You may want to do more preprocessing of the v101 transcription -- Glen's intent in having (for example) '9' and '(' variants for EVA 'y', or '7' and '8' for EVA 'd' (or, by my count, 6 variants of EVA 'r') wasn't to claim those were actually different glyphs, he was just giving you the option.