The raw data for the analysis came from Joseph H. Peterson's "Esoteric Archives" CD, available at www.esotericarchives.com. Peterson has transcribed almost the complete text of Liber Logaeth (with some exceptions) into an HTML document. It should be noted up front that Peterson's transcription is known to have numerous errors. I have made corrections wherever an error has been identified, but to proof-read the entire document would be an arduous task which I have not attempted. Furthermore, due to the quality of the source manuscripts, photocopies, and Kelley's handwriting, many parts of the text are either illegible or impossible to interpret with a reasonable degree of certainty. Therefore, it has to be accepted that a certain degree of error exists within the analysis.
Liber Logaeth consists of 49 leaves. The 49th leaf contains of the following text:
loagaeth seg loxi brincr
larzed dox ner habzilb adnor
doncha larb uors hirobra
exi ur zedmp taiip chimuane
chermach lendix nor zandox
The remaining 48 leaves each have a 49 x 49 square grid on both the front and back page (96 tables in total). For the first leaf that the angels delivered, the front table was to have a word in each square. To actually write out a complete word in a tiny cell of a 49 x 49 grid was hardly feasible, so the first table of Liber Logaeth was actually recorded as a series of 49 paragraphs. The back page table of the first leaf was similarly recorded, except that the first 40 rows had a complete word in each cell, whereas the last 9 rows had a single letter in each cell.
The remainder of the tables have, at the most, a single character in each cell. (There are one or two exceptions where a double "l" or double "t" appears in a cell.) The predominant pattern of the tables is to have one letter per cell. A number of them have a checkerboard pattern in which every other cell contains a letter and the remaining cells are blank. These typically show the letters oriented at various 45-degree angles, indicating that the text runs in a diagonal pattern.
The grid formats described above make up the majority of the tables, and they are fairly straightforward to analyze. A few of the tables have odd formats that require more careful interpretation. Some of them have segments of text running in two different diagonal directions. Others have a diagonal text segment and a horizontal text segment. On a few of the tables, I am reasonably confident that some of the letters are intersected by two different segments of text.
One of the text segments has been excluded from the analysis. This is the text from table "Chi pled sagnaronph" consisting of normally-oriented letters. It begins at the top left cell and runs every other letter in checkerboard fashion. The arrangement of the letters appears to be highly irregular, and I haven't yet figured out what the pattern is. A separate analysis of this segment will have to be done at a later date.
Having worked out the patterns in the tables, I wrote a series of software routines to extract the text of Liber Logaeth from Peterson's HTML document into a plain-text file. The transliterations of the Enochian letters were standardized. "v" and "w" was changed to "u" throughout. "y" and "j" were changed to "i". "k" was changed to "c". A central question around how to analyze the data was whether the text had words carrying over from the end of one line to the beginning of the next, or if each line break clearly marked the end of one word and the beginning of the next. A cursory glance at the tables gives the impression that the text lines are cleanly separated from one another, and that there are no overlapping words. I initially began my analysis with the assumption that overlap was possible, but a certain degree of "noise" in the results convinced me that this was not the case. Therefore, in order to greatly simplify the analysis process, I restarted with the assumption of clean word breaks between the text lines.
Obviously, this type of analysis is not necessary for the section of Liber Logaeth in which each cell contains a complete word. The remainder of the tables, in which there is no more than one letter per cell, are what require interpretation. I will hereafter refer to that section of Liber Logaeth as "grid text".
I chose to scan the grid text for combinations of 4 letters or more, which occur at least twice. 2 and 3 letter combinations were ignored because their occurrence would not be statistically significant enough to warrant consideration. If we produced a string of 3 Enochian letters at random, the probability of it matching any pre-determined 3-letter string would be 1 in (21x21x21) = 9261. The total number of 3-letter chunks in the grid text is 167423. This means that, at random, any given 3-letter combination would be expected to occur an average of (167423/9261) = 18.078 times. This would create too much "noise" in the results to make a meaningful analysis. A 4-letter combination, on the other hand, would be expected to occur an average of 0.835 times in the text, so any such combination that occurs more than once is above expectation. A 4-letter combination that occurs only twice is not a particularly strong indicator, but this represents the threshold of what would be considered potentially significant. The higher the frequency of occurrence, the stronger the significance of the letter string. And the significance increases exponentially with the number of letters in the combination. A 5-letter combination would be expected to occur an average of 0.038 times, so any such combination that occurs more than once is highly unlikely to be random.
The first step in the analysis was to identify the longest recurring combination. This was found to be the 17-letter string "adnachephalzednoc". Having established that, I then did a series of scans for combinations of x number of letters from 16 down to 4. I then ran a series of comparison routines to weed out recurring text combinations which were simply sub-strings of longer combinations. In other words, if "doghouse" were to occur 10 times, then there would also be 10 instances of "house" which we want to filter out of the results, because they are simply sub-strings of results which have already been captured. However, if an instance of "house" were to be found which was not contained in any instance of "doghouse", then "house" would be considered a recurring string in it's own right, and that result would be duly noted.
The results of the analysis are recorded in two accompanying text files - the "short list" and the "long list". The short list simply gives all recurring combinations along with a figure indicating the total number of times that the string occurs, plus a figure indicating the number of times that the string occurs where it is not contained in an instance of a larger recurring string. For instance, the result "oldah - 6 - 3" indicates that the string "oldah" appears a total of 6 times in the grid text. In 3 instances, it appears within a larger recurring string (which in this case happens to be "holdah"), and there are 3 instances where it occurs independently.
The long list contains the same information as the short list, but for each recurring string it also includes a list of all larger recurring strings in which it is found (plus that string's total number of occurences), and a list of all smaller recurring strings which it contains.
One of the most striking observations from these results is that we have quite a number of strings with a total frequency count in the dozens or hundreds (up to 1644), but very rarely do we find a string that occurs more than 2 or 3 times outside of a larger string. This low independence of individual strings would seem to suggest that either there are a whole lot of recurring phrases in the text, or that the language is heavily structured around the use of compound words.
Additional analysis:
I ran a series of scans to see if any known Enochian words turned up in Liber Logaeth. I used an already existing word list which is found at http://www.geocities.com/Athens/Thebes/5777/enochian.txt. I added to this list the names of the archangels Michael, Raphael, Gabriel, and Uriel, because some time back I had spotted by chance the word "raphael" in the grid text. "Michael" and "uriel" turned up as well after doing the analysis.
First, I compared the list of known words to the text of leaf 1 of Liber Logaeth, in which we have one word per cell. The one complicating factor in this analysis is that leaf 1 has a number of words which appear to be multiple syllables hyphenated together, such as "gedos-lubah", or "gu-na-pa-ge-pha". It is difficult to interpret whether these represent a single word or a series of words strung together. Because of this ambiguity, my approach was to run two comparisons. The first comparison treated the hyphenated strings as a single word, and then searched for any word from the known word list which either matched or was a substring of any word of the leaf1 text. The other comparison broke up the hyphenated strings as if they were separate words, and then searched for any word from leaf1 which was a substring of any word in the known list. There are a number of exact matches. Most of the results are probably not significant, but the list is not very large so I have included it in full in two accompanying files.
I also ran a search for words in leaf 1 that occurred in the grid text. The file "Words_in_leaf1_found_in_grid_text.txt" treats each segment of the hyphenated strings as an individual word. However, many of these hyphenated strings consist of numerous, short segments, which give the impression of being syllables rather than words. So I decided to test this possibility by scanning every sequential combination of these hyphenated segments through the grid text. (For instance, where the text of leaf 1 shows "la-zu-red", I scanned for "lazu", zured", and "lazured".) The results are given in the file "Syllable_combinations_in_leaf1_found_in_grid_text.txt". This came up with a suprisingly large number of hits. Those strings which were identical to strings in the previously mentioned file were filtered out, but there were actually very few of these.
Next, I compared the list of known words against the grid text. I was intrigued by the fact that "graph" and "drux" - the names for the Enochian letters "E" and "N" - occurred with a very high frequency. This would seem to suggest that the Enochian alphabet is similar to the Hebrew alphabet in that the names of the letters are words in their own right. In spite of my general policy against scanning for strings of less than 4 letters, I decided to run the names of all the Enochian letters through the grid text. Keep in mind that the expected number of random occurrences for any given letter combination is as follows:
5 letters - 0.038
4 letters - 0.835
3 letters - 18.078
2 letters - 390.786
The results of the scan on the letter names was such:
pa - 685
ueh - 10
ged - 450
gal - 346
or - 1710
un - 637
graph - 76
tal - 359
gon - 78
na - 2312
ur - 804
mals - 1
ger - 50
drux - 51
pal - 167
med - 5
don - 423
ceph - 1
uan - 334
fam - 15
gisg - 0
"Ueh", "mals", "med", "ceph", "fam", and "gisg" all came in at less than or approximately equal to the expected average. The rest of the letter names showed up at around twice the expected average or higher. There is one note of caution in interpreting these figures, however. The expected averages are based on a random distribution of letters. Yet we know that the content of the grid text is clearly not random. As with most spoken languages, phonetic combinations which are unwieldy to the tongue are generally avoided. This means that certain letter combinations are inherently more likely to appear than others. For instance, "bedbug" is much more likely than "fpsrgc". We have to keep this in mind when evaluating the significance of these frequency figures. The 2-letter combinations are especially suspect in this regard. Nevertheless, "ged", "gal", "tal", and "don" all have a remarkably strong showing. These all follow a consonant-vowel-consonant pattern, which is a fairly typical phonetic structure for a spoken syllable. If we were to produce random syllables following this pattern, the likelihood for any given syllable of this type would be (16x5x16) = 1280. Out of a total of 167423 3-letter chunks in the grid text, we would expect this to occur 130.799 times. Compare this to actual occurrences of 450, 346, 359, and 423. "Uan" is similar enough in phonetic structure that it can also be considered quite strong at 334 occurrences. The frequency of "drux" - and "graph" especially - are well beyond the threshold of significance.
Finally, a manual search of the words in leaf 49 against the grid text came up with the following occurences:
loxi - 21
larzed - 1
adnor - 17
doncha - 17
larb - 3
uors - 28
zandox - 1