labels as words

[MarcoP]

I have made a new campaign of histogram stats, trying to find a subset of Voynichese that behaved similarly to Labelese. I only found partial matches, still there are data that I found interesting. I don't a simple solution to discuss, but several minor points that seem hard to explain clearly, but I'll try.


The main features of Labelese when compared with average Voynichese are:

• the frequency of prefix o- is doubled
  
• the frequency of prefix q- is about one third of the normal
  

Time ago we discussed You are not allowed to view links. Register or Login to view.. Examining the graphs for o- and q-, it is clear that no word ending provides a context that doubles o- (even if -r, -s and -t provide a reduction of q- occurrences close to that observed in labels).
   


I then turned to subsets identified not by the character ending of the previous word, but by the position of the word inside the text. The results I consider interesting are illustrated in the following complex histogram. I will discuss each subset individually, comparing prefiexes with Labelese (the Orange bars).
I have used Zandbergen's  IVTT ZL transcription instead of Takahashi's, so there are a few very minor differences with the graphs I previously posted.

   

• [Green] All text. These are the basic numbers we have already discussed. I have here only considered paragraph text, but the exclusion of labels doesn't have a considerable impact. The general statistics mainly differs from Labelese because it has a much higher qo- frequency and much lower ot- and ok- frequencies.
  
• [Dark Blue] First word of the first line of a paragraph (i.e. first word of a paragraph) This is the only subset that gets close to the low qo- frequency in Voynichese. Labels have a 2.5% of qo- prefixes, the first words of paragraphs 3.9%, the whole text 10.3%. But o- occurs quite rarely at the beginning of paragraphs, while there are several occurrences of otherwise rare prefixes like po-, pc-, to-. This is a major difference from Labelese. The other major difference is that ch- and sh- are almost completely absent, while in Labelese they are well represented (even if much rarer than in normal text).
  
• [Pink] Last word of the first line of a paragraph. This subset seems uninteresting at first sight, but it turns out it has the best overall correlation with Labelese (with a very small margin). I guess its strength is the close match for op-. This prefix is frequent in the first line of a paragraph, but as the first word of the paragraph p- seems to be preferred. This subset also is the one which gets closer to Labelese ch- stats.
  
• [Light Blue] First word of the last line of a paragraph. These stats forced me to include several rare suffixes in the graph, because it has spikes in very uncommon prefixes (dc- yc-). As Emma pointed out above, there are cases in which Voynichese favors prefixes which are typical of line starts (e.g. yt- yk- so-) but other line-start prefixes are uncommon in Labelese.
  
• [Brown] Last word of the last line of a paragraph (i.e. last word of a paragraph). This is the subset of Voynichese that gets closer to the high ok- rate of Labelese. Yet its more striking feature is the huge spike for ch-. About 25% of the paragraphs end with a ch- word: look at paragraph endings You are not allowed to view links. Register or Login to view.. Labels have a slight dislike for ch- (less than 10%) so this is a huge difference.
  

The suffixes histogram is much more uniform. The clearest spikes correspond to the well known line ending -m. The last word of the first line of a paragraph [Pink] has a large -am spike, but it compares well with Labelese because -ly and -ry are common suffixes both at line end and in Labelese. -in also is a good match between last-paragraph-words and Labelese. These good matches compensate the difference due to the -am spike when computing correlation.
   


Conclusions

• The most extensive Labelese peculiarity is the high frequency of the o- pefix (in particular, ot- and ok-). This feature isn't paralleled by any language subset I examined (neither subsets following a specific ending, see radar chart, nor text position subsets, see histograms).
  
• The prefix anomaly qo- is approximated by the subset lacking a left context (first words of paragraphs). A similar distribution also occurs after some specific endings (-r, -s, -t).
  
• The suffix anomalies of Labelese -ly and -ry are approximated by subsets lacking a right context (line endings and paragraph endings). The same subsets also correlate with the da- prefix stats.
  
• As observed by Emma, some Labelese common prefixes are also frequent as line starts (yt-,yk-). But there are frequent line-start prefixes that aren't common in labels (e.g. dc- yc-).
  
• The relatively high frequency of op- in labels is only matched by first-paragraph-line statistics, but not by the stats for the very first words of paragraphs (since they dislike o- and tend to start directly with p- and other gallows).