1
00:00:15,140 --> 00:00:18,493
Hello and welcome to this
episode of Voynich Talk.

2
00:00:18,913 --> 00:00:21,400
If you are watching this
video, there's a good chance

3
00:00:21,425 --> 00:00:24,133
you've heard about the
Voynich manuscript before.

4
00:00:24,298 --> 00:00:27,000
You may also have heard
that it's been solved.

5
00:00:27,839 --> 00:00:29,072
And then solved again.

6
00:00:29,452 --> 00:00:30,458
And again.

7
00:00:30,759 --> 00:00:31,759
And again.

8
00:00:32,330 --> 00:00:35,336
It's been solved many times
by many different people

9
00:00:35,357 --> 00:00:36,703
in many different ways.

10
00:00:37,303 --> 00:00:39,503
You may even have
solved it yourself.

11
00:00:39,923 --> 00:00:41,263
Today we will learn

12
00:00:41,303 --> 00:00:43,770
why all of these
solutions are wrong.

13
00:00:46,009 --> 00:00:48,703
Because despite all these
theories being different,

14
00:00:49,237 --> 00:00:52,536
they also fall short
in the exact same way.

15
00:00:52,949 --> 00:00:54,276
That being said, there are a few

16
00:00:54,301 --> 00:00:56,183
theories this video
does not cover.

17
00:00:56,650 --> 00:00:58,330
For example, Timm and Schinner

18
00:00:58,337 --> 00:01:01,319
believe that the text of the
manuscript is meaningless,

19
00:01:01,343 --> 00:01:04,367
and was generated through
some predetermined process.

20
00:01:04,723 --> 00:01:08,633
Similarly, Gordon Rugg argues
that glyph tables and a grill

21
00:01:08,896 --> 00:01:11,723
can be used to quickly
obtain meaningless text

22
00:01:12,010 --> 00:01:14,340
compatible with the
format of Voynichese.

23
00:01:15,057 --> 00:01:17,723
Approaches like these
are not part of this video.

24
00:01:17,930 --> 00:01:21,583
No, we are interested in the
most common kind of solutions.

25
00:01:22,177 --> 00:01:24,634
The ones you see in the media
every couple of years,

26
00:01:24,757 --> 00:01:27,855
or enthusiastically proclaimed
on forums, blogs, or

27
00:01:27,880 --> 00:01:29,676
even in peer-reviewed journals,

28
00:01:30,123 --> 00:01:32,196
that sometimes forget
to peer-review.

29
00:01:32,809 --> 00:01:35,983
Let's have a closer look at
this average Voynich solution.

30
00:01:36,443 --> 00:01:38,116
How does it actually function?

31
00:01:38,956 --> 00:01:40,303
Most Voynich theories,

32
00:01:40,523 --> 00:01:41,782
way more than you think,

33
00:01:42,176 --> 00:01:44,629
are driven by two
main mechanisms.

34
00:01:44,869 --> 00:01:47,282
Substitution and
creative writing.

35
00:01:48,103 --> 00:01:50,036
Both are equally problematic.

36
00:01:50,650 --> 00:01:54,331
I will explain exactly why
substitution cannot work

37
00:01:54,356 --> 00:01:56,450
as a solution
to the Voynich manuscript.

38
00:01:57,103 --> 00:01:58,996
In a later video we will turn

39
00:01:59,023 --> 00:02:01,576
to the wonderful world
of creative writing

40
00:02:01,957 --> 00:02:04,083
and broader issues
of methodology.

41
00:02:04,603 --> 00:02:07,366
Substitution, in its
most simple form,

42
00:02:07,383 --> 00:02:09,700
is called: simple substitution.

43
00:02:10,070 --> 00:02:12,933
This is when each letter
in your original text,

44
00:02:13,177 --> 00:02:14,990
which we call the plain text,

45
00:02:15,157 --> 00:02:17,877
gets replaced or substituted

46
00:02:18,290 --> 00:02:21,234
by its corresponding
letter in the ciphertext.

47
00:02:21,737 --> 00:02:25,001
For example, you may have
heard of a Caesar cipher,

48
00:02:25,023 --> 00:02:27,576
where each letter is
shifted in the alphabet.

49
00:02:27,957 --> 00:02:30,023
But you could also
use a scrambled key.

50
00:02:30,500 --> 00:02:32,900
The effect remains
exactly the same.

51
00:02:33,057 --> 00:02:34,725
Simple substitution ciphers

52
00:02:34,750 --> 00:02:36,863
are very, very easy to crack.

53
00:02:37,450 --> 00:02:39,800
And I will demonstrate
this in the next video.

54
00:02:40,023 --> 00:02:42,210
But we can go much
broader than that

55
00:02:42,270 --> 00:02:44,492
and expand our
definition of substitution

56
00:02:44,516 --> 00:02:45,963
beyond simple ciphers.

57
00:02:46,130 --> 00:02:48,098
Converting your text
to an alphabet

58
00:02:48,123 --> 00:02:50,049
the reader is not familiar with,

59
00:02:50,177 --> 00:02:51,950
in this case for example Greek,

60
00:02:52,530 --> 00:02:54,990
that is also substitution
because each letter

61
00:02:55,010 --> 00:02:57,723
can be substituted by
a different letter form.

62
00:02:58,023 --> 00:03:00,166
But you could also try
something more complex

63
00:03:00,167 --> 00:03:03,563
like removing vowels
and only converting consonants.

64
00:03:03,810 --> 00:03:06,233
Your code may become
harder to crack,

65
00:03:06,370 --> 00:03:08,270
but it's still substitution,

66
00:03:08,750 --> 00:03:11,210
and still completely unsuitable

67
00:03:11,230 --> 00:03:13,453
as a solution to the
Voynich manuscript.

68
00:03:13,477 --> 00:03:16,930
Just to give you an idea
of how pervasive this issue is,

69
00:03:16,983 --> 00:03:18,305
let's have a look at this list

70
00:03:18,330 --> 00:03:20,857
of Voynich solutions
compiled by Tavi.

71
00:03:21,090 --> 00:03:24,367
These are all people who
— each and every one —

72
00:03:24,383 --> 00:03:26,663
are convinced they
cracked the code.

73
00:03:26,863 --> 00:03:29,597
Each in their own
mutually exclusive way.

74
00:03:29,857 --> 00:03:32,437
I started going through them
in order to get an idea

75
00:03:32,462 --> 00:03:34,989
of the variety of solutions
we're dealing with,

76
00:03:35,403 --> 00:03:37,737
but I gave up soon
for two reasons.

77
00:03:38,510 --> 00:03:42,109
One is that many theorists
don't communicate clearly

78
00:03:42,123 --> 00:03:43,223
about their methods.

79
00:03:43,248 --> 00:03:45,996
They act like they come down
to us, mere mortals,

80
00:03:46,023 --> 00:03:49,703
with a mysterious black box
full of secret knowledge.

81
00:03:50,196 --> 00:03:51,562
Sometimes, to be honest,

82
00:03:51,616 --> 00:03:53,876
I get the impression
that they hardly know

83
00:03:53,900 --> 00:03:55,263
what they're doing themselves.

84
00:03:55,403 --> 00:03:56,683
The second reason is

85
00:03:57,167 --> 00:03:59,320
that when the method
was finally revealed,

86
00:03:59,400 --> 00:04:03,133
it always turned out
to be substitution.

87
00:04:03,463 --> 00:04:06,000
Apart from the
aforementioned exceptions,

88
00:04:06,203 --> 00:04:09,200
all Voynich solutions
are based on substitution.

89
00:04:09,603 --> 00:04:11,666
But the problem is
that the Voynich text

90
00:04:11,770 --> 00:04:14,763
cannot be the result
of a substitution cipher.

91
00:04:15,397 --> 00:04:17,136
Perhaps the biggest issue is,

92
00:04:17,230 --> 00:04:20,436
as experienced Voynich
researchers often point out,

93
00:04:20,523 --> 00:04:23,423
the text's low conditional
character entropy.

94
00:04:23,557 --> 00:04:26,096
This simply means that
the characters tend to appear

95
00:04:26,121 --> 00:04:28,391
in the same order within a word.

96
00:04:28,557 --> 00:04:30,117
That when you get one character

97
00:04:30,150 --> 00:04:32,070
it's easy to predict
the next one.

98
00:04:32,650 --> 00:04:33,770
Let's try something.

99
00:04:34,530 --> 00:04:36,469
I'm thinking of a
word in English,

100
00:04:36,657 --> 00:04:39,257
and I'm telling you
it starts with a P.

101
00:04:40,297 --> 00:04:42,344
Can you tell me
which letter is next?

102
00:04:43,603 --> 00:04:45,676
It could be A as in park,

103
00:04:45,737 --> 00:04:46,916
E as in pet,

104
00:04:46,957 --> 00:04:49,304
I as in picture, L as in play,

105
00:04:49,783 --> 00:04:52,076
O as in poster, R as in print,

106
00:04:52,097 --> 00:04:53,457
U as in publish.

107
00:04:53,910 --> 00:04:56,236
If we allow words that
originally came from Greek,

108
00:04:56,237 --> 00:04:58,550
we can even include H, S and T

109
00:04:58,563 --> 00:05:01,230
as in photo, psychology
and pterodactyl.

110
00:05:01,550 --> 00:05:05,390
There are just too many options,
several of them quite likely.

111
00:05:05,470 --> 00:05:06,750
And it doesn't end there.

112
00:05:07,163 --> 00:05:09,696
If P is the second
letter in a word,

113
00:05:09,857 --> 00:05:13,167
then the letter following it
is completely unpredictable.

114
00:05:13,450 --> 00:05:16,350
We could move P to the
third place, fourth place,

115
00:05:16,497 --> 00:05:19,789
any place, and it would
still be able to be followed

116
00:05:19,967 --> 00:05:22,703
and preceded by a
whole range of letters

117
00:05:22,723 --> 00:05:24,857
producing perfectly valid words.

118
00:05:25,490 --> 00:05:28,343
Voynichese does
not work like this.

119
00:05:29,217 --> 00:05:30,843
If I give you this character,

120
00:05:31,043 --> 00:05:33,043
you know almost for certain

121
00:05:33,297 --> 00:05:34,844
that this is the next one.

122
00:05:34,970 --> 00:05:37,757
Then, you can guess
with surprising accuracy

123
00:05:37,777 --> 00:05:40,721
that one of these two
tall characters is next.

124
00:05:40,990 --> 00:05:42,450
Having arrived at this point

125
00:05:42,503 --> 00:05:46,570
over two thirds of cases are
followed by either one of these.

126
00:05:46,783 --> 00:05:48,143
It goes on like that.

127
00:05:48,330 --> 00:05:51,616
Additionally, many characters
appear almost exclusively

128
00:05:51,623 --> 00:05:54,923
in certain positions like at
the beginning or end of words.

129
00:05:55,137 --> 00:05:56,870
Once you pick a character,

130
00:05:56,963 --> 00:05:59,396
there are not many
options for the next,

131
00:05:59,503 --> 00:06:01,403
and the next, and the next.

132
00:06:01,937 --> 00:06:04,337
It's all very rigid
and predictable.

133
00:06:05,377 --> 00:06:08,190
The implications of this
are enormous.

134
00:06:08,310 --> 00:06:12,690
As René Zandbergen points out
in a paper published in 2019,

135
00:06:13,443 --> 00:06:16,387
he focuses on the writings
of Gerard Cheshire,

136
00:06:16,737 --> 00:06:18,483
but the same problems exist

137
00:06:18,517 --> 00:06:20,876
for every possible
substitution solution

138
00:06:20,897 --> 00:06:23,003
in any known human language.

139
00:06:23,637 --> 00:06:26,823
Cheshire assigns
this character to D.

140
00:06:27,143 --> 00:06:28,143
Big mistake.

141
00:06:28,267 --> 00:06:31,166
Now D can only be used
in the beginning of words,

142
00:06:31,267 --> 00:06:34,240
and it can only be followed
by this character,

143
00:06:34,807 --> 00:06:37,180
which he conveniently
reads as O.

144
00:06:38,313 --> 00:06:40,628
So in Latin, these common words

145
00:06:40,653 --> 00:06:43,290
can no longer be
formed in Voynichese,

146
00:06:43,315 --> 00:06:46,215
because the D-character
must be word initial,

147
00:06:46,589 --> 00:06:49,455
and it must be followed
by the O-character.

148
00:06:50,095 --> 00:06:51,095
Keep in mind

149
00:06:51,123 --> 00:06:53,116
that these are some of
the most frequent words

150
00:06:53,141 --> 00:06:54,854
found in any Latin text.

151
00:06:55,168 --> 00:06:57,248
The system cannot produce them.

152
00:06:57,442 --> 00:07:00,034
I was about to grant
that Cheshire's system

153
00:07:00,202 --> 00:07:02,328
could form the word domus,

154
00:07:02,828 --> 00:07:05,128
but it turns out that this
is not the case either,

155
00:07:05,588 --> 00:07:07,636
which just shows how unforgiving

156
00:07:07,703 --> 00:07:09,390
the structure of Voynich text is.

157
00:07:10,290 --> 00:07:11,552
Using the table we see

158
00:07:11,577 --> 00:07:13,896
that domus should be
written like this.

159
00:07:14,430 --> 00:07:16,310
The problem is that
when you have a series

160
00:07:16,335 --> 00:07:19,195
of these I-shaped
minims in Voynichese,

161
00:07:19,802 --> 00:07:21,668
they are preceded by this shape.

162
00:07:22,155 --> 00:07:24,670
So, while Cheshire's Proto-Romans

163
00:07:24,737 --> 00:07:26,836
can't live in a proper domus,

164
00:07:27,337 --> 00:07:30,343
they can at least build a domais.

165
00:07:31,203 --> 00:07:32,796
I used Latin examples here,

166
00:07:32,850 --> 00:07:34,233
but the exact same problems

167
00:07:34,258 --> 00:07:36,543
would arise for any
human language.

168
00:07:37,010 --> 00:07:39,336
If your language has a D sound,

169
00:07:39,367 --> 00:07:42,200
you want to be able to
use it in various positions,

170
00:07:42,549 --> 00:07:44,337
and at least you want to be able

171
00:07:44,362 --> 00:07:46,383
to pair it up with
more than just O.

172
00:07:47,083 --> 00:07:49,533
The problem is not the
chosen language or dialect,

173
00:07:49,563 --> 00:07:51,483
the problem is Voynichese,

174
00:07:51,830 --> 00:07:53,718
and our insistence
to keep treating it

175
00:07:53,743 --> 00:07:55,423
as a substitution cipher.

176
00:07:55,630 --> 00:07:57,303
The fact that Voynichese letters

177
00:07:57,323 --> 00:07:59,657
have a preferred
order within words

178
00:07:59,903 --> 00:08:04,070
is unlike anything we see
in any known writing system.

179
00:08:04,583 --> 00:08:07,600
It has more in common with
something like Roman numerals,

180
00:08:08,197 --> 00:08:10,644
where MMXXIV

181
00:08:10,961 --> 00:08:14,587
is a valid order,
but VXMIXM isn't.

182
00:08:15,534 --> 00:08:17,156
A similar effect could be reached

183
00:08:17,223 --> 00:08:20,166
by alphabetically sorting
the letters within each word,

184
00:08:20,297 --> 00:08:23,170
like this, but then
more repetitive.

185
00:08:23,403 --> 00:08:24,523
People will often object

186
00:08:24,570 --> 00:08:27,333
that maybe the statistical
problems can be solved

187
00:08:27,390 --> 00:08:29,070
by special conditions,

188
00:08:29,103 --> 00:08:32,490
like using some obscure
language or even a dialect.

189
00:08:32,647 --> 00:08:36,060
But switching from standard French
to any Romance dialect

190
00:08:36,213 --> 00:08:37,786
doesn't move the needle at all.

191
00:08:38,087 --> 00:08:41,567
Both languages are equally
distant from Voynichese.

192
00:08:41,713 --> 00:08:45,633
It simply does not behave
like human language.

193
00:08:46,170 --> 00:08:48,976
Similarly, assuming
an abjadic system,

194
00:08:49,001 --> 00:08:52,270
so a system without vowels,
does not help.

195
00:08:52,910 --> 00:08:56,610
Human language needs a
certain amount of flexibility,

196
00:08:56,643 --> 00:08:59,443
which Voynichese
is not able to offer.

197
00:08:59,457 --> 00:09:01,310
What about abbreviations?

198
00:09:01,743 --> 00:09:03,176
Well, that's even worse.

199
00:09:03,637 --> 00:09:06,410
Typically, abbreviation
will leave out

200
00:09:06,423 --> 00:09:08,870
the repetitive and
predictable parts,

201
00:09:08,983 --> 00:09:12,311
exactly because they are
repetitive and predictable.

202
00:09:13,057 --> 00:09:15,390
That's why they
can be abbreviated.

203
00:09:15,677 --> 00:09:17,543
So if you abbreviate a text,

204
00:09:17,710 --> 00:09:20,443
whatever remains
will be rich and varied,

205
00:09:20,530 --> 00:09:21,950
dense in information.

206
00:09:22,510 --> 00:09:25,136
Voynichese is the
exact opposite of that.

207
00:09:25,257 --> 00:09:28,923
It looks like it's made up of
parts that should be abbreviated.

208
00:09:30,043 --> 00:09:32,623
Here's a concrete example
of how bad it is.

209
00:09:32,710 --> 00:09:33,867
Let's take this character.

210
00:09:33,910 --> 00:09:39,367
In Herbal A it occurs
a total of 1262 times.

211
00:09:39,543 --> 00:09:43,476
Literally 99% of those occur
at the end of the word.

212
00:09:44,343 --> 00:09:47,803
And 95% are preceded
by this character.

213
00:09:48,677 --> 00:09:53,034
Of those endings 90% are in
turn preceded by this character,

214
00:09:53,690 --> 00:09:55,423
resulting in this or this.

215
00:09:55,957 --> 00:09:58,570
This means that assigning
a unique phonetic value

216
00:09:58,595 --> 00:10:00,210
to this character is madness,

217
00:10:00,235 --> 00:10:03,110
since whatever letter you
choose to correspond to it

218
00:10:03,297 --> 00:10:05,708
can now only occur
at the end of words.

219
00:10:06,735 --> 00:10:09,152
But, you might say,
what if there is a language

220
00:10:09,177 --> 00:10:11,936
with certain sounds that only
occur at the end of words?

221
00:10:12,863 --> 00:10:13,771
Well, sure,

222
00:10:13,797 --> 00:10:16,676
but then those sounds
would also always be preceded

223
00:10:16,701 --> 00:10:17,816
by fixed sounds,

224
00:10:18,123 --> 00:10:21,996
and those in turn preceded
by yet another fixed sound.

225
00:10:22,957 --> 00:10:24,783
Similarly, as soon as you assign

226
00:10:24,808 --> 00:10:26,966
a plaintext letter
to this character,

227
00:10:27,357 --> 00:10:29,470
your chosen letter
can now only occur

228
00:10:29,497 --> 00:10:30,650
at the start of words,

229
00:10:31,023 --> 00:10:32,592
 and it can only be followed

230
00:10:32,617 --> 00:10:34,737
by whatever letter
you picked for O.

231
00:10:35,897 --> 00:10:37,592
The most Voynichese-like thing

232
00:10:37,617 --> 00:10:39,734
many of us are
probably familiar with

233
00:10:39,803 --> 00:10:43,676
is the spelling convention
that Q should be followed by U,

234
00:10:44,023 --> 00:10:45,705
as is the custom in
several languages

235
00:10:45,730 --> 00:10:47,283
that use the Latin script.

236
00:10:47,990 --> 00:10:49,263
For example, in English,

237
00:10:49,288 --> 00:10:53,067
we could easily assign
the QW sound to Q alone,

238
00:10:53,383 --> 00:10:54,810
without causing any trouble.

239
00:10:55,197 --> 00:10:57,530
The U is redundant
and predictable,

240
00:10:57,763 --> 00:10:59,343
and it could easily be omitted.

241
00:10:59,557 --> 00:11:02,863
It is merely a relic of
earlier spelling conventions,

242
00:11:02,903 --> 00:11:06,603
but it does remind us a bit of
the behavior of Voynich text.

243
00:11:06,943 --> 00:11:10,343
And still, Voynichese
is much, much worse.

244
00:11:10,963 --> 00:11:14,476
For starters, QU is just one part

245
00:11:14,530 --> 00:11:16,333
of any language that uses it.

246
00:11:16,690 --> 00:11:17,810
For Voynichese,

247
00:11:17,870 --> 00:11:19,416
predictable groupings like this

248
00:11:19,437 --> 00:11:21,416
form the majority of the text.

249
00:11:22,077 --> 00:11:24,810
But even if we focus
only on the use of Q

250
00:11:25,067 --> 00:11:27,709
and compare this to
this Voynichese pair,

251
00:11:28,190 --> 00:11:29,923
the situation in Voynichese

252
00:11:29,963 --> 00:11:31,800
turns out to be much more rigid.

253
00:11:32,263 --> 00:11:34,925
I pulled up some random
text in various languages

254
00:11:34,950 --> 00:11:36,510
I had sitting around
on my hard disk

255
00:11:36,537 --> 00:11:37,949
from a previous project.

256
00:11:38,497 --> 00:11:42,310
Here's 10,000 words of Jacob
van Maerlant's Middle Dutch.

257
00:11:42,910 --> 00:11:47,700
He uses Q 16 times,
and 16 times it's followed by U.

258
00:11:48,577 --> 00:11:50,163
It's the most frequently found

259
00:11:50,183 --> 00:11:53,530
in forms of the adjective quaet,
which means something like bad.

260
00:11:54,283 --> 00:11:56,723
But elsewhere, it
occurs in the third,

261
00:11:56,810 --> 00:11:59,210
fifth or sixth position
in the word.

262
00:12:00,083 --> 00:12:01,610
Turning to Pliny's Latin,

263
00:12:01,630 --> 00:12:04,030
we see that Q
is expected to perform

264
00:12:04,037 --> 00:12:05,703
at any position in the word.

265
00:12:06,570 --> 00:12:08,830
No matter which
language you choose,

266
00:12:09,143 --> 00:12:13,683
whether it be Latin, Turkish,
Chinese, Swahili, Esperanto,

267
00:12:13,723 --> 00:12:16,636
or the unique
Proto-Judeo-Roman dialect

268
00:12:16,661 --> 00:12:19,916
spoken by two Ethiopian
nuns on a desert island,

269
00:12:20,190 --> 00:12:23,030
the situation always
remains the same.

270
00:12:23,630 --> 00:12:26,574
As soon as you start
mapping glyphs to sounds,

271
00:12:26,903 --> 00:12:29,391
you can no longer
express the language

272
00:12:29,416 --> 00:12:31,689
using the system of Voynichese.

273
00:12:32,390 --> 00:12:34,483
You are locking
yourself into a system

274
00:12:34,510 --> 00:12:37,126
that's much too
rigid and predictable,

275
00:12:37,267 --> 00:12:39,400
unsuitable for a human language.