- Options
- Login
- Desktop Style
Options
Stars in the recipe section
- Stars in the recipe section
-
Stars in the recipe section
Labyrinthinesecurity > Yesterday, 03:52 PM
In the recipe section, I annotated the stars showing a tail and the stars without a tail.
I then trained a simple statistical model to predict whether a paragraph maps to a “tail” star using only internal text features (no external information, no hand labeling tricks).
The model correctly identifies tail paragraphs about 86% of the time (balanced accuracy), which is far above chance (50%).
(I also ran permutation tests where we randomly reshuffled the data within each folio, preserving structure but destroying signal. In those tests, performance dropped to about 52%, which is basically chance. This suggests the model is not just exploiting simple artifacts like folio grouping or class imbalance.)
Is anyone aware of similar studies? I'd be interested to know if star morphometrics correlate to textual contents.
Thanks! -
RE: Stars in the recipe section
Koen G > Yesterday, 03:54 PM
I don't remember anything along these lines. But do you know anything at all about how or why the prediction is made? -
RE: Stars in the recipe section
Labyrinthinesecurity > Yesterday, 04:05 PM
(Yesterday, 03:54 PM)Koen G Wrote: You are not allowed to view links. Register or Login to view.I don't remember anything along these lines. But do you know anything at all about how or why the prediction is made?
From a syntax perspective, obviously yes because I chose the features for the model
Now from a linguistic/semantics perspective, that's another story I am exploring and that I will share when and if I have something robust.
-
RE: Stars in the recipe section
oshfdk > Yesterday, 04:21 PM
Some people suggested that stars are just a way to represent certain markers or even characters: You are not allowed to view links. Register or Login to view.
In which case there will obviously be a correlation between the star type and the contents of the paragraph.
(08-01-2016, 04:34 PM)Anton Wrote: You are not allowed to view links. Register or Login to view.There are also stars with an ink dot at the centre. They can be light-painted or unpainted; not all light-painted stars have a centre dot. These observations dismiss the hypothesis that the dot is used as a reminder to paint (or not to paint) a star. Looks like a marker for some other purpose.
Likewise, star tails are probably markers. Pelling once suggested them to represent letter "y" in a hidden fashion, but I don't consider this a plausible explanation. Usually tails look downwards, but there is at least one tail looking upwards (last star in f103r). -
RE: Stars in the recipe section
Labyrinthinesecurity > Yesterday, 04:29 PM
(Yesterday, 04:21 PM)oshfdk Wrote: You are not allowed to view links. Register or Login to view.Some people suggested that stars are just a way to represent certain markers or even characters: You are not allowed to view links. Register or Login to view.
In which case there will obviously be a correlation between the star type and the contents of the paragraph.
(08-01-2016, 04:34 PM)Anton Wrote: You are not allowed to view links. Register or Login to view.There are also stars with an ink dot at the centre. They can be light-painted or unpainted; not all light-painted stars have a centre dot. These observations dismiss the hypothesis that the dot is used as a reminder to paint (or not to paint) a star. Looks like a marker for some other purpose.
Likewise, star tails are probably markers. Pelling once suggested them to represent letter "y" in a hidden fashion, but I don't consider this a plausible explanation. Usually tails look downwards, but there is at least one tail looking upwards (last star in f103r).
Thank you, that's precious! I have tried to link the dot to actual text but so far I have failed. My only success is about tails. -
RE: Stars in the recipe section
ReneZ > 11 hours ago
The stars without tails are found on f103r, f103v, You are not allowed to view links. Register or Login to view. (one bifolio) and on f111v, where there are also a number of cases with tails.
That is four out of 23 pages, meaning (rough estimate): 17% without tails, 83% with tails.
With this, an 86% success rate may easily be achieved just by chance.
More importantly, any text property that varies according to this bifolio split will strongly correlate with the appearance of tails. That is: if such a text property exists. The recent work by Colin Layfield, presented by Lisa, suggests that it does.
A meaningful measure of success would come from how well the paragraphs on f111v match, but this is unfortunately a small sample.
Still, there are some interesting questions related to the distribution of stars with/without tails. -
RE: Stars in the recipe section
Jorge_Stolfi > 10 hours ago
(Yesterday, 03:52 PM)Labyrinthinesecurity Wrote: You are not allowed to view links. Register or Login to view.In the recipe section, I annotated the stars showing a tail and the stars without a tail.
I then trained a simple statistical model to predict whether a paragraph maps to a “tail” star using only internal text features (no external information, no hand labeling tricks).
The model correctly identifies tail paragraphs about 86% of the time (balanced accuracy), which is far above chance (50%).
(I also ran permutation tests where we randomly reshuffled the data within each folio, preserving structure but destroying signal. In those tests, performance dropped to about 52%, which is basically chance. This suggests the model is not just exploiting simple artifacts like folio grouping or class imbalance.)
Is anyone aware of similar studies? I'd be interested to know if star morphometrics correlate to textual contents.
Thanks!
It does not directly relate to your question, but you may find You are not allowed to view links. Register or Login to view. useful. (You will have to download the file and display it with some monospace font, like Courier.)
Column 4 says whether the star has a core (dot or small circle) or not. But that datum is not available for stars which are painted red. Column 5 says the way the paint was applied, and column 6 is its color. The next two columns are the count of rays and the count of tails, usually 1 or 0 ('-').
Some random thoughts:
- The red paint is opaque and probably late, like all strongly colored paint on the VMS. The mostly transparent light yellow paint (apparently used on the stars of Cosmo and Zodiac, nymph hair, etc.) may be original.
- The red and yellow colors mostly alternate, but there are some hiccups, and some unpainted stars. My hunch is that the stars were originally alternating yellow and unpainted, with some hiccups. Much later the Painter set out to paint the unpainted stars with red, but when he got to the hiccups he either broke the alternation or left a star unpainted.
- Surely each parag was meant to have a star besides its head line. But the Scribe may have written the text first and then he drew all the stars on the page. That would explain why the stars are not aligned with the lines.
- Some stars are more than one line away from their corresponding head lines.
- There are three big blocks of text on f108r, f108v, and f111r, where apparently the Scribe ran a dozen parags together, without any line breaks between them. The number of stars in the margin of those blocks hopefully tells how many parags are in them.
- There are a few other cases where the boundary between successive parags is not clear, because the last line of the first spans the whole width of the page. In some of those cases, the tail of the star points east instead of south, and seems to be indicating the line which should be the head line of the second parag. But it is not clear whether those tails are original, or guesses by a later owner.
- However, most tails point south instead of east, and in many cases the tail is continuous with the outline of the star.
- The ray count tends to repeat in stretches, changing only every half a dozen stars or so. Therefore, that number cannot be conveying much information. Less than one bit per parag.
- The red paint is opaque and probably late, like all strongly colored paint on the VMS. The mostly transparent light yellow paint (apparently used on the stars of Cosmo and Zodiac, nymph hair, etc.) may be original.
-
RE: Stars in the recipe section
Labyrinthinesecurity > 3 hours ago
(11 hours ago)ReneZ Wrote: You are not allowed to view links. Register or Login to view.The stars without tails are found on f103r, f103v, You are not allowed to view links. Register or Login to view. (one bifolio) and on f111v, where there are also a number of cases with tails.
That is four out of 23 pages, meaning (rough estimate): 17% without tails, 83% with tails.
With this, an 86% success rate may easily be achieved just by chance.
More importantly, any text property that varies according to this bifolio split will strongly correlate with the appearance of tails. That is: if such a text property exists. The recent work by Colin Layfield, presented by Lisa, suggests that it does.
A meaningful measure of success would come from how well the paragraphs on f111v match, but this is unfortunately a small sample.
Still, there are some interesting questions related to the distribution of stars with/without tails.
Fair point, that's why I mentioned balanced accuracy, it accounts for class imbalance. -
RE: Stars in the recipe section
Labyrinthinesecurity > 1 hour ago
René here is the confusion matrix, the predictor works much better than chance:
Predicted
TAIL noTAIL
Actual TAIL 224 37
Actual noTAIL 11 69
what's more the in folio permutations also show that the test is robust despite class imbalance.