Professional Documents
Culture Documents
A Cryptanalytic Decipherment of the Indu
A Cryptanalytic Decipherment of the Indu
A Cryptanalytic Decipherment of the Indu
yajnadevam
������
ARTICLE HISTORY
Compiled November 6, 2023
Abstract
Indus inscriptions hold the key to unlocking the history of pre-Iron Age India and
all Indo-European peoples but remain undeciphered for over a century. All prior
attempts have been unsatisfactory and unfalsifiable. We decipher the Indus script
by treating it as a large cryptogram as described by Claude Shannon. We decipher
every sign sequentially using regular expressions and set-intersection. Indus script
is discovered to be proto-abugida segmental with signs for consonants and vowels.
Indus inscriptions are in grammatically correct post-Vedic Sanskrit. Variants of
74 allographs constitute most signs. Conjunct signs constitute the rest. Our deci-
pherment can read every inscription and we translate 500+ inscriptions including
the 50+ longest, 50+ shortest and 400+ medium-sized inscriptions including 100+
inscriptions with conjunct signs. We comfortably surpass Shannon’s criteria for a
credible cryptogram decipherment. Brahmi glyphs are discovered to be standard-
ized Indus signs. We find significant continuation of Indus linguistic features and
cultural elements in post-bronze age India.
KEYWORDS
Indus Valley Civilization; Indus script; Epigraphy; Brahmi; Sanskrit
1 1. Introduction
2 Indus Valley civilization was the largest bronze age civilization, spanning over one mil-
3 lion square kilometers and having an estimated population of five million(Dixit, 2019).
4 This advanced civilization featured planned cities, drainage, international trade and
5 standardized weights among many impressive accomplishments. Indus civilization has
6 left behind strings of symbols on seals, tablets and a large sign (Dholavira signboard)
7 which is termed “the Indus script”. Although the mature Indus civilization is believed
8 to have started around 2600 BCE, we see stage-by-stage evolution from the earliest sites
9 to Bhirrana and Mehrgarh. These older stages are given different names such as pre-
10 Harappan, early-Harappan and so on to distinguish them from the mature Harappan
11 stage.
2
62 Of the decipherments that claim to have decoded a large number of seals, most use a
63 combination of the erroneous methods discussed below. The erroneous methods create
64 degenerate sets used to force a reading. These degenerate sets give a shallow semblance
65 of plausibility. When these erroneous methods are applied to larger sets, they fall apart
66 like the binary-fish narrative.
67 1.3.1. Intuition
68 Most attempted decipherments of Indus script are logographic or logo-syllabic, which
69 avoids the need for syllables or phonemes to interlock into words. Logographic scripts
70 are unlikely to be satisfactorily deciphered without a Rosetta stone or some other falsi-
71 fiability criteria. A majority of signs can be claimed to be single words, most are nouns
72 with occasional generic verbs. This enables short inscriptions to be read as adjective-
73 noun or verb-adjective-noun. The method starts by assuming the value of a sign and
74 then tries to read a meaning into a short inscription. Parpola for example, considers the
75 intersecting circles sign � to represent “bangles” and three strokes sign �to represent
76 “hearth”(Parpola, 1994). The reasoning for such an assumption is unexplained. These
77 two signs may as well represent “nuts” and “three” or “wheels” and “baby teeth” or
78 any thousands of competing concepts. Occasionally, these signs occur in meaningless
79 patterns and need an alternate explanation. For example, when these two signs are
80 combined �� , the reading “hearth-bangles” would be meaningless, so a new overloaded
81 interpretation is added: “pregnancy bangles”. The derivation pathway for this seems to
82 be simply associating random strings of signs with an idea or a myth that the author is
83 familiar with. Such a reading is non-falsifiable and rests on our faith in the decipherer’s
84 intuition. Intuition can be a valid method in a falsifiable system such as mathematics
85 as demonstrated by Ramanujan but not in decipherments.
86 1.3.2. Eclectic
87 The sign values are sometimes borrowed from other scripts. Signs are assumed to
88 have the same value as any similar sign in other scripts. Mahadevan for example,
89 assumes symbols similar to Egyptian hieroglyphs and Sumerian cuneiform represent
90 the same meaning in Indus script and then tries to connect it to some Indian cultural
91 or mythological context, thus “deciphering” a symbol(Mahadevan, 2010). For example,
92 based on his analysis of M-1896 ��� and the similarity of Indus sign � with Egyptian
93 hieroglyph O.49 meaning city, Iravatham Mahadevan speculates that Mohenjodaro’s
94 ancient name could be Kukkutarma, [kukkuta = Indian fowl] or “Cocks-city” based
95 on interpreting the 1st CE Chola kingdom city Uraiyur in South India, as the city of
96 the cock. No evidence of chickens have been found in Mohenjodaro. Recent evidence
97 shows that chickens were domesticated in Thailand about 300 years after Indus Valley
98 stopped making seals(Peters et al, 2022). Mohenjodaro and Uraiyur which are separated
99 by 2500 years and 2000 km are alluded to as perhaps connected for narrative support
100 since “cocks-city” is a meaningless term not backed by any evidence.
101 It’s usually easy to find failures of this method for even short inscriptions. Using the
102 meaning of other signs from his decipherments, H-452 �� would be chicken-jar, M-795
103 �� �� would be ruler-chicken-jar-bearer. These other meaningless readings need their
104 specific narrative support. It becomes evident that eclectic decipherments very quickly
105 become untenable.
3
Table 1. Eclectic decipherment of Latin alphabet from unrelated scripts
4
138 1.3.6. Nominal
139 Phonetic and syllabic decipherments would be expected to have some grammar but
140 nearly all prior decipherment claims typically lack grammar, morphology, syntax,
141 subject-verb agreement, plurals, conjugation, declension, prepositions or any other
142 grammatical elements. One way to justify this is to claim that all Indus inscriptions
143 are names. Short inscriptions can plausibly be argued to be names since names outside
144 sentences are essentially not declined. This also means most long inscriptions often
145 won’t be readable. This is a claim of many decipherments, the prominent being S. R.
146 Rao.
174 2. Methods
5
179 of starting carving letters from the right side and so on was combined with internal ev-
180 idence such as identical inscriptions appearing as both one and two line versions. This
181 enabled the eventual determination of the direction of most inscriptions. Approximately
182 83% are determined to be right-to-left and 6.5% were left-to-right. The rest are either
183 single-sign, undetermined or damaged. A mathematical model of the gini-coefficient of
184 the Indus corpus also finds that the Indus script is written right-to-left regardless of the
185 type of script(Ashraf and Sinha, 2018).
186 Logographic scripts are characterized by low frequency and a large signary. Ma-
187 hadevan classifies the Indus script into 417 signs, ICIT classifies them into 700+ signs,
188 treating minor variants as separate signs since a sizable number (1000+) of signs is
189 suitable for a logo-syllabic analysis. This is based on the assumption that a script from
190 3500 BCE must be logo-syllabic(Wells and Fuls, 2023).
191 However, a large signary is untenable on thorough analysis. Only a small fraction
192 of signs (27.5%) are attested 10 or more times and only a tiny fraction (6%) have an
193 attestation count of 100+. A significant fraction of the rare signs seem to be stylistic
194 variants, accidentally mirrored signs or word fragments. Some are gradual abstractions
195 of complex signs such as � � � and must be seen as variants of a single sign. The
196 extreme rarity of conjunct signs like � � � indicates that these are artifacts of crowding
197 due to lack of space and must be seen as digraphs of their component signs rather than
198 as completely different unrelated signs.
199 With these considerations, the sign count drops to under a hundred and the script is
200 unlikely to be logographic. This strengthens the case for a syllabic or segmental script.
201 The pattern of repetitions of the same sign multiple times in the same inscription
202 also supports syllabic or segmental (abugida/alphabetic/abjad/phonetic) script. A non-
203 script should be expected to see repetition in the use of symbols similar to the birthday
204 collision effect. This number crosses 50% when the number of symbols in a single
205 inscription is 25 or more. No Indus inscription is this long, so we wouldn’t expect to
206 see any repeated symbols.
417!
(417−25)!
1− = 0.52
41725
207 Birthday collision, of course, applies to uniformly random events, while segmental
208 scripts are not uniformly random due to phonotactic rules unique to every language.
209 Therefore more frequently used symbols should see more collisions in smaller-length
210 inscriptions than less frequently used symbols. We see repetitions in the shortest of
211 inscriptions to medium to long inscriptions. About 17% of inscriptions have a repeated
212 sign if we include immediately repeated and bracketed signs. The jar and fish signs are
213 prolific repeaters and the spoked wheel sign occurs 4 times the Dholavira inscription.
214 This is additional support for a non-logographic script. This number is likely to be
215 much higher once allographs are discovered. By comparison in a logosyllabic script,
216 repetitions are extremely rare in short inscriptions. For example, of the 488 subway
217 station names in Beijing, only 3 (0.6%) have a repeated Mandarin sign and none with
218 3 or more occurrences(wikipedia, 2021).
6
219 2.2. Cryptanalytic Decipherments
220 Some scripts such as the Copiale cipher have been deciphered using cryptanaly-
221 sis(Knight, Megyesi, and Schaefer, 2011). Ephron was able to re-decipher the Ventris-
222 Chadwick decipherment of Linear-B using cryptanalysis(Ephron, 1961). He only found
223 one sign that had a different value, which was later acknowledged as correct by the orig-
224 inal decipherers, showing the superiority of cryptanalytic decipherments. Cryptanalysis
225 using frequency analysis is possible if we have an idea of the set of allographs, without
226 which complex tools may need to be specially written to analyze them as in the case
227 of the Copiale cipher. Attempts to use frequency analysis on the Indus script did not
228 yield any results other than the possibility of some relation to the Brahmi script(Kak,
229 1988). The dictionary method of solving cryptograms is superior to frequency analysis
230 since it’s immune to frequency changes due to the passage of time and does not require
231 apriori knowledge of allographs(Moreno, 2005).
7
263 sponding consistency conditions in the cryptogram. The key gives a certain amount of
264 freedom to the cryptogram, but as more and more letters are intercepted, the consistency
265 conditions use up the freedom allowed by the key. Eventually there is only one message
266 and key which satisfy all the conditions and we have a unique solution(Shannon, 1945).
267 The redundancy in a language is simple to explain: There are about 3742 five-letter
268 words in English of 265 = 11881376 possible combinations. Most of the symbol space
269 therefore is empty. The reason for this is phonotactics and the way words are made:
270 Certain combinations of sounds and letters are impossible in a given language. The
271 first character of a word and every letter following its previous letter is a matter of
272 probabilities. For example, if the first 2 letters of a five-letter word are TH then there
273 are only 30 possible words out of a possible 17576 symbol combinations. If the first
274 3 letters of a five-letter word are THY, then the next two are definitely ME from the
275 word THYME of a possible 262 = 676 possibilities. At a level above words, morphol-
276 ogy, syntax and idioms dictate how the sequence of phonemes in a language can flow.
277 While the phonotactics and syntax of every language are unique, each has a comparable
278 amount of redundancy. Fixed states with fixed probabilities of transition can be used
279 to construct a Markov process, which serves as a graph model for the language.
280 Shannon derives a quantitative metric to determine if a claimed decipherment is
281 credible or suspect. After a certain number of letters are deciphered, equivocation
282 becomes zero and the cryptogram has a unique solution. This point is called the unicity
283 distance. Shannon’s formula for unicity distance is |K|/D, where |K| is the size of the
284 key space and D is the redundancy of the language which measures excess information
285 content:
286 For the English language, the key space is the number of possible ways to rearrange
287 the alphabet to create the cryptogram alphabet. The number of keys is 26×25×...2×1 =
288 26! The information contained in the key space is log10 (26!). The redundancy of English
289 can be experimentally determined to be 0.7, that is to say, 70% of English letters are
290 redundant or can be guessed based on the remaining 30% of letters.
291 From |K| = log10 26! and D = 0.7, we calculate the unicity distance for a cryptogram
292 in the English language: log10 (26!)/0.7 = 30 letters. Shannon’s criteria is any cryp-
293 togram whose length is well beyond the unicity distance (say, 50 for English) has a
294 unique solution and is correctly deciphered:
295 In general, we may say that if a proposed system and key solves a system for a length of
296 material considerably greater than the unicity distance the solution is trustworthy. If the
297 material is of the same order or shorter than the unicity distance the solution is highly
298 suspicious(Shannon, 1945).
8
310 variances introduced by encoding, we can transliterate to Latin before compression.
311 This roughly gives 0.7 for most natural languages. Any minor differences may be seen
312 as insignificant since we are looking at way past the unicity distance for determining
313 correctness.
314 This gives a uniform formula for unicity distance to check the correctness of deci-
315 pherment claims in any language:
316 |K|/D = log10 (S 417 )/0.7, where S is the number of phonemes in the language.
317 For claims of Indus script deciphered as Sanskrit with 35 phonemes:
318 Unicity Distance of Sanskrit for Indus script:
319 |K|/D = log10 (35417 )/0.7 ≈ 920.
320 We can use a multiplier of 1.5 to determine “well-past unicity” giving the decipher-
321 ment size:
322 920×1.5 = 1380.
323 Note that while this allows for the same phoneme to be represented by multiple
324 graphemes, graphemes representing multiple phonemes add to the equivocation and
325 may be adjusted for by multiplying with the equivocation added. Reusing ~30 signs of
326 unaspirated phonemes for aspirated adds about 10%, as does reusing ~20 dental signs
327 for retroflex. Pooling all sibilants into ~20 signs increases the equivocation by 150% for
328 those signs. The net effect would be adding (30+20)×0.1 + 20×1.5 = 35 signs for a
329 total of 452 signs.
330 Unicity with folded aspirates, retroflex and pooled sibilants:
331 log10 (35452 )/0.7 ≈ 997.
332 Similar adjustments for unicity must be made when more deviations are added. For
333 example, if inscriptions are read as a five vowel abjad, then its unicity distance is
334 log10 (35(417∗5) )/0.7 ≈ 4599.
9
354 All information in a decipherment result needs to pre-exist in the plain text or the
355 decipherment key. This means that it’s mathematically impossible to force a language
356 of our choice onto a cryptogram when the corpus is larger than the key. The closest we
357 can do is the frequency-fit described in Section 1.3.4, which is essentially a 1st order
358 approximation of a Markov process. Higher-order approximations using an incorrect
359 language cannot be constructed as they would conflict with the frequencies of the source
360 language of the corpus.
361 This can be illustrated by trying to force-fit modern English as the language of the
362 Indus Valley and demonstrating how quickly it fails. Let us pick the seal M-1837 ���.
363 The only English word that matches this pattern is EEL with � resolved as E and
364 � as L. When we try to resolve the next symbol � using K-153 ��, we don’t get any
365 matches since there are no two-letter English words that have L as the second letter.
366 We could abandon this thread of investigation and try with other signs, but eventually,
367 they will all reach a dead-end. It becomes clear that English cannot be force-fit as the
368 language of Indus inscriptions. This will be true for force-fitting any language onto any
369 cryptogram, including languages that are close to the source language. This is due to
370 the mathematical property of Markov processes: the graphs representing the Markov
371 process of different languages are utterly unique because the set of states and their
372 transition probabilities are unique.
10
Table 2. Names of Meluhhan persons and goods as recorded in Sumerian cuneiform
401 Vibhra, Atibhra etc. but unlikely in other candidates like Dravidian.
402 In addition, 90 identified words in Sumerian may be borrowed from Sanskrit(Vyas,
403 2020). This may be seen as sufficiently strong evidence to attempt decoding the Indus
404 language as Sanskrit. We use Monier-Williams abbreviations throughout the paper to
405 denote the source of attestation(Monier-Williams, 1899a). For example, Soma स�म[RV]
406 means “Soma as attested in Rigveda” and Samara समर[MW] means “Samara, attested
407 as a proper name in Monier-Williams dictionary.”
11
429 Patterns in the Indus corpus may be represented as regular expressions enabling
430 us to search in dictionaries and determine their values. Solving cryptograms using
431 regular expressions is a known art and we will only touch upon the techniques used for
432 deciphering the Indus script(Goibhniu, 2007).
444 The sign σ has the value of one of the few letters that are doubled at the end of a four-
445 letter-word such as bass, pull, cuff, etc. represented by πασσ . The regular expression
446 / ̂(.)(.)(.)\3$/ can be used to check for candidates that match this pattern. The
447 periods in parentheses are capture groups that are used to extract letters in positions
448 1-3. The final \3 indicates that the last letter is repeated. Note that we are not guessing
449 any value here, but simply extracting all possible words that match this pattern from
450 a dictionary. The above expression returns 95 words:
451 The possible values for each sign are captured by the regex group (period in paren-
452 theses):
453 π in {A, B, C, D, F, G, H, I, J, K, L, M, N, P, R, S, T, W, Y}
454 α in {A, E, H, I, L, N, O, R, S, U, W}
455 σ in {E, F, L, O, R, S, T, Z}
456 For the next word σαπσ , we can substitute the list of possibilities instead of the
457 period to capture the words that have the same letters in the new positions. The regex
458 / ̂(E|F|L|O|R|S|T|Z)(A|E|H|I|L|N|O|R|S|U|W)([A-DF-NPR-TWY])\1$/
459 gives 41 words:
12
466 α in {A, O, U}
467 σ to {S, T}.
468 Repeating with the last word ασπσ we resolve all three:
469 π in {P}
470 α in {A}
471 σ to {S}.
13
511 FOREMAN could be the right pick.
512 The abundance of choices on how to translate an inscription is generally a good sign
513 that the decipherment is nearly correct. If many signs are right but a few of the signs
514 are incorrect, they would typically force a break or the translation would not be in the
515 same themes as the rest of the translation. For example, if the letter E was incorrectly
516 deciphered as T, we would have MAYBTFORTMAN, forcing a word break near the
517 bad letter and we may attempt to read as MAY BT FORTMAN. Sometimes the bad
518 letter may be innocuous in one inscription, for example, if O was deciphered as I :
519 MAYBE FIREMAN. While this instance is grammatically correct, many others would
520 have breaks and the text would be nonsensical or incongruous with the general themes
521 of the rest of the inscriptions. These errant signs can be detected using the intersection
522 of the signs in all broken inscriptions and the decipherment process is repeated for such
523 signs with greater thoroughness.
524 If a decipherment is utterly incorrect, it won’t even be possible to make any words,
525 let alone morphologically consistent words and grammatically correct phrases that can
526 be read meaningfully. For example: mapagakajh cannot be made into a Sanskrit or
527 English phrase no matter how we dice it. This is because p or ph never follow m in
528 Sanskrit phonotactics, nor do words end in jh. Even plausible sounding short strings
529 like hena, hoja etc are non-words and it is unlikely that we get many matches because
530 the symbol space of a language is mostly filled with non-words.
531 A small subset of short inscriptions may be read as an abjad and with wide latitude
532 for word forms, ignoring morphology or syntax with a partially correct decipherment.
533 The grammatically correct reading of the longest inscriptions thus becomes a litmus
534 test for decipherment accuracy.
14
559 a capture group.
15
Table 3. Six extensively stylized signs
602 When individual variants have sufficient attestations, we may prove each variant
603 individually for example, all 3 variants �� � may be independently proven to have
604 the value /ka/ क. Where sufficient attestations don’t exist we can narrow down the
605 possibilities to a handful and attempt to read all the inscriptions meaningfully. For
606 example, by applying the regex method to identify � we do not get a unique value,
607 but narrow it down to /ra/ र or /va/ व. Based on sign similarity and readability in
608 inscriptions, we can confirm the sign is a variant of � /ra/ र.
609 This method is not only helpful in identifying variants, but it is also self-correcting
610 and an incorrect identification will be easy to identify:
611 (1) two signs which are different are assumed to be the same
612 (2) two signs that are the same are assumed to be different
613 In the first case, regular expression search yields two disjoint sets of inscriptions with
614 two different phonetic values. For example, if we presume the signs � and � are variants
615 of the same sign, An attempt to narrow the matches from the first set using a member
616 of the second set will yield a null set. This immediately alerts us to the fact that the
617 signs are indeed distinct.
618 In the second case, both sets will yield the same phonetic value. In fact, this is how
619 we discover allographs as shown in Table 4.
620 While an optimal reachability graph can be constructed by analyzing the inscriptions,
621 we can use the heuristic of solving signs in descending order of frequency with a few
622 detours as needed.
16
623 2.14. The first sign
624 The first sign we decipher is the jar sign � representing an. It also represents a variety
625 of nasal sounds including anusvara, which is a post-vocalic nasal sound. From the
626 dictionary itself, a regex for H-764B ��� we get a single match ananan, representing
627 ananaṃ, which is an accusative of anana. Technically, there are other possible matches
628 like jajaja, lalala which are legitimate matches. The seal Dmd-1 � single jar sign
629 matches ana and eliminates other alternatives like ja and la. Finally, many inscriptions
630 with terminal jar signs representing the accusative ending aṃ resolve the value of the
631 sign.
632 After we decipher the first sign, we simply look for other inscriptions that enable
633 us to decipher more signs by substituting the value of the first sign. The vertical bar
634 sign � is decipherable with just the jar sign and we keep repeating the steps till all
635 signs are resolved. We continue the process roughly going in the descending order of
636 frequency of signs, to optimize the probability of finding sufficient attestations needed
637 for deciphering the sign. The complete derivation may be followed step by step in
638 Section 8. A computer program that partially reproduces the derivation is available in
639 the supplementary section.
17
667 (1) Let two inscriptions sa = gi xa gk xc ... and sb = gj xa gl xf ...
668 (2) Let set of symbol matches Gxa = /gi (.+)gk . + / ∩ /gj (.+)gl . + /
669 (3) Repeat intersection with new inscription above until Gxa is of unit length
670 (4) x a = G xa
671 (5) G = G ∪ xa
672 (6) Repeat until G = X
673 3. Results
674 Indus script is made of 74 allographs representing the most common signs shown in
675 table 6. The complete derivation can be traced in section 8. A compact one-pager is
676 given in table 5. Long inscriptions and inscriptions with conjunct signs are listed in
677 sections 5 and 6 respectively. Some signs have variants that mark reading direction or
678 word boundaries and they are listed in table 11. Consonant clusters using syncope are
679 shown in table 10. Table 7 shows Indus signs standardized into Brahmi.
Table 5. Indus script phoneme glyphs
Phoneme Glyphs
अ a �� � �� � � � � � �� � � � �
आ ā �� � ��� � � � ����
इ i ��
ई ī ��
उ u �����
ऋ ṛ �
ए e �
ओ o ��
अन� अ� ṃ ��������
अस� अ� ḥ ��������������
कख k kh �� � � � � � � � � �
गघ g gh �
ङ ṅ �
चछ c ch ��
जझ j jh �� � ��
ञ ṅ �
तथटठ t th ṭ ṭh ��������������
�
दधडढ d dh ḍ ḍh � � �� � � � � � � � � � � � � � �
नण nṇ � � � � � � � � � � �� � �
पफ p ph �� � �
बभ b bh ������
म m ����↑�������������������
य y ����������
र r ��������������������
ल l ��
व v � �� � �
शषसह śṣsh � � � � � � � � � � � � � � � � �� � � � � � �
18
680 3.1. Sign values and their reconstructed names
अग aga mountain[Kirat] � 3
आ ā repeated अ a � � � � � �� �
� � 7
ई ī doubled इ �� 9
ए e एक eka one[RV] � 12
an aṃ
g gh
जझ झर jhara waterfall[Prab] �� � 23
26
19
Phoneme Reconstructed Sign Name Sign Glyphs
न n न�ल nāla mat of reeds[BhP] � 36
नल nala reed[Bhp] � 39
यव yava barley[RV] �� 54
�व vi two[RV] � 63
20
Table 8. Allograph derivation for अ /a/ and क /k/
ट ṭ ठ ṭh ड ḍ ढ ḍh ण ṇ
𑀝 𑀞 � 𑀟 � 𑀠 � 𑀡 𑀭𑁉
𑀝
त t थ th द d ध dh न n
𑀢 � 𑀣 � 𑀤 � 𑀥 � 𑀦 𑀦
प p फ ph ब b भ bh म m
𑀧 � 𑀨 � 𑀩 � 𑀪 � 𑀫 �
य y र r Brahmi ल l व v
𑀬 � 𑀭 � vs 𑀮 � 𑀯 �
श ś ष ṣ Indus स s ह h
𑀰 � 𑀱 � 𑀲 � 𑀳 �
21
Table 10. Consonant clusters using schwa and double syncope
681 4. Discussion
682 Indus script is a segmental script that may be described as proto-abugida. The major
683 difference from Brahmi is that retroflexes use the same signs as dentals and aspirated
684 and unaspirated stops use the same signs. Signs have a default vowel of अ a unless
685 overridden by an immediately following vowel sign. No diacritics are used.
686 Many signs have stylistic variations that have no phonetic distinctions. While signs
687 are clearly distinct in most cases, occasional ambiguities do exist, which are discussed
688 below. Signs may be classed into several groups based on characteristics specific to
689 them.
22
Table 12. Provisional readings of mixed IVC/Brahmi inscriptions from subsection 1.1
23
Table 13. Provisional readings for non-Sanskrit seals from Mesopotamia and Dilmun(Laursen, 2010)
Authors Finding
24
734 of simple line strokes.
735 For example, the sign M-734 reads � � ��� वरण-जज varaṇa-jaja, “chosen[Uṇādi] war-
736 rior[Śiś].” The clear space between � � is unambiguously read as two � ज ja signs, rather
737 than a single �� sibilant sign, which would need to be read as वण�स�, “color[MBh].” Seal
738 M-1822 ��� uses incline differences to distinguish � � स� sā to avoid being read as � ज
739 ja.
740 When strokes are arranged in rows, if the upper row has an equal or higher number
741 of strokes than the lower row, then the sign is read as representing the total number of
742 strokes. Inscriptions use spacing to ensure clear separation between distinct signs.
743 For example, the sign M-1904 reads � जज jaja, “warrior.” The three-stroke sign
744 representing the first syllable is marked as distinct from the other by horizontal and
745 vertical spacing. If there was no clear spacing, the seal would look like � and be read
746 as a sibilant.
747 When the upper row has fewer strokes than the lower row, spacing or other indications
748 of separateness become important for an accurate reading. For example, M-948 reads
749 ����� अववज्य avavajya “off[RV] journey[√वज� ]”, rather than अवपय avapaya “desiring[RV]
750 water[RV]” due to slight offset of two rows of strokes. While signs with five or more
751 strokes can appear in one or two rows, signs with two and four always appear in a
752 single row. If they appear in two rows, they are distinctly different signs with distinctly
753 different sound values. Note the differences between �व va, � ज ja, � च ca versus� न
754 na, �त ta, � श śa.
25
777 than a fairly meaningless अण्वस aṇva-sa “interstice bestowing”.
26
Table 15. Directional markers on jar signs
817 Indus signary evolved to address this challenge by deliberately adding asymmetry to
818 some signs. For example the elephant head sign � and the wasp sign � are asymmetric
819 and the direction is obvious at a glance and the inscribed direction flows from the
820 pointed side to the blunt side.
821 The most common sign, the � jar sign repeated sufficiently within inscriptions to
822 enable a directional marker of its own, by adding one to four small strokes inside the
823 jar as shown in the first section of Table 15. The strokes appear in non-ascending order.
824 The first jar always has the highest number of strokes and the next sign can either
825 decrease or maintain the number of strokes. Typically, the final jar sign will end up
826 with no strokes. This pattern holds for over two hundred inscriptions with just a handful
827 of exceptions as in the last section of Table 15. The improbability of these strokes to
828 have any kind of phonetic meaning is easily verified by testing the inscriptions against a
829 dictionary. These strokes are a directional marker similar to a fuel gauge, tracking the
830 remaining text portions from full to empty. The directional mark is placed on a nearby
831 sign if the starting sign is numeric or otherwise unsuitable for a directional indicator.
832 A directional marker may be used even when it’s the only jar sign in the inscription to
833 avoid accidental reading in the wrong direction as we see in the second section since the
834 � jar sign is so often a terminal sign. When the leftmost sign has a directional marker
835 and the rightmost sign doesn’t, the seal may be preferentially read left to right as we
836 see in the last section of Table 15. This is most useful for multi-line boustrophedon
837 inscriptions but occasionally also occurs in single-line seals.
838 Signs other than jar signs can also have one to four strokes to indicate direction or
839 word boundary. Usually, these occur as the initial sign in the inscription and act as a
840 directional indicator and are listed in Table 11.
27
841 4.13. Stylistic variants
842 The different glyphs of the same allograph without directional function in Table 6 are
843 likely to be stylistic variations, based on the fact that different forms of a single allograph
844 rarely occur together in the same inscription. Typically when the inscription needs to
845 use the same allograph twice, the scribe seems to choose the same variant or one very
846 close to it. There are a few obvious exceptions, such as fish sign, where variants do
847 occur together.
28
881 4.17. Religious continuity
882 We see several linguistic and cultural features that continued post-Indus phase. The
883 deities in Indus inscriptions are the same as Vedic. The Sun, Soma, Rudra, Indra, Agni,
884 Ushas, Ashlesha and Bharani constellations, the horse ashva, Ardhanareeshwara (seals
885 depicting one-breasted human) and pipal tree continue to be revered to this day. Select
Rigvedic references related to long inscriptions are noted in Section 5.
Table 16. Post-IVC attestation of script elements(Balasubramaniam, 2005; Bhatt, 1998; Hultzsch, 1925)
886
29
901 5. Long inscriptions with their scriptural references
रव rava roarer[√रु + अच� ]; अमम� amam ams. the powerful[RV]; मन mana honor[√म्न� + ल�ट� 2s.];
सक्ष sakṣa capable[TS]; नर� naraṃ ams. man[TS]; जठल jaṭhala vms. Ocean[Sāy] =सम�द्र=Shiva[MBh];
ध�र dhāra vms. Sustainer[MBh 13.14.13]; रह raha yield, release[√रह� + ल�ट� 2s.];
द�मन� dāman giver[RV]; मन mana vns. great[=मह�न� mahān Vā.]; क�स� kaṃsaṃ ans. cup/jar[AV];
न� naḥ for us[RV dp. of अस्मद� asmad]; रन्धन randhana vms. destroying[Bhp];
द�मन� dāman giver[RV]; मन mana vns. great[=मह�न� mahān Vā.]; क�स� kaṃsaṃ ans. cup/jar[AV];
आस āsa taken[√अस� + �लट� 1s./3s.]; नदन� nadan roaring[√नद� + शत��];
कम� kam indc. good/well[TS]; म�ख� mākhaṃ ans. oblation[Hariv]; क�स� kaṃsaṃ ans. cup[AV];
द�त dāta vms. purified[Pāṇ]; नदन� nadan roaring[√नद� + शत��];
30
Seal-Id Inscription Translation
धक्क dhakka vns. destroyer[√धक्क� + अच� ]; वह vaha carry[√वह� + ल�ट� 2s.]; म�� सज māṃsaja fat[Suśr];
आश āśa food[ŚBr]; वरम� varam ams. boon[RV];
तत� tat ans. that[RV]; ददनम� dadanam ans. present[√दद� + ल्य�ट�]; रव rava vns. roarer[√रु + अच� ];
अमम� amamam ams. powerful[RV]; अञ्जस� añjas Soma[RV]; सरम� saram ams. liquid[VS];
शतद� śatadā giver of a hundred[SV]; रव rava vns. roarer; आम āma serve[√अम� + ल�ट� 1s.];
मह� mahaṃ ams. great[RV]; म�ख� mākhaṃ ans. oblation[Hariv];
31
Seal-Id Inscription Translation
दद� dadā grant[RV 4.36.9]; तव tava your[RV]; वशम� vaśam ams. power[AV];
अञ्ज añja shining[√अञ्ज� + अच� ]; दह्र dahra vns. fire[Uṇvṛ];
तन tana roarer[√तन� + अच� RV]; भव bhava Rudra[BhP,AV]; वह vaha carry[√वह� + ल�ट� 2s.];
अमम� amam mighty[RV]; वरम� varam ams. choicest[MBh];
हट haṭa vns. shining one[√हट� + अच� ]; ममद mamada enjoyed[√मद� + ल�ट� 1s.];
म�ख mākha oblation[Hariv]; वरम� varam ams. choicest[MBh];
32
Seal-Id Inscription Translation
शत�त� śatāt hundred, abnp. शतम� śatam[RV]; अवमम� avamam ams. youngest[RV];
स sa indc. with[RV], मञ्जन mañjana vms. shine[√मञ्ज� + ल्य�ट�];
इद्ध iddha shining; अचल acala vns. immovable[RV,BG]; व�ह vāha vms. bearer[BhP];
न�म nāma from न�मन� nāman water[Naigh]; नश� naśaṃ ams. loss[MW]; तर tara overcome[√त� + ल�ट� 2s.];
आरव ārava vns. Roarer[Pāṇ]; श�स� śāsaṃ ams. ruler[RV]; अञ्ज añja honored[√अञ्ज� + अच� ];
सभ� sabhaṃ ans. council[RV];
33
Seal-Id Inscription Translation
34
Seal-Id Inscription Translation
धक्क dhakka vns. destroyer[√धक्क� + अच� ]; व va indc. like; अञ्ज añja shining[√अञ्ज + अच� ];
वरम� varam ams. gift[RV]; आच āca I seek[√अच� + ल�ट� 1s.];
हत hata Slayer[RV]; मदन madana vms. Kāmā[MBh]; अञ्जन� añjan shining[√अञ्ज� + शत��];
आय[म� ] āya[m] ams. arrival[RV]
अन� anā indc. verily[RV]; आनष� ānarṣa set flow[√ऋष� + �लट� 3s.]; स�त� sātā gifts[RV];
द� daḥ nms. giver[MBh];
दस्म dasma vms. extraordinary[RV]; वह vaha bring[√वह� + ल�ट� 2s.]; अञ्जस� añjas Soma[RV];
भ�� bhāṃ ans. shine[RV];
35
Seal-Id Inscription Translation
यम yama Lord of death[RV]; हन� han killer[RV]; अ� श� aṃśa ams. portion[RV 2.19.5];
शम śama calm[MBh]; -मन� -man vms. containing[R];
अथ atha indc. now/hereby[RV]; अचल acala immovable[Bhag]; त्व tva vms. one[RV];
म�न māna honored[MBh]
36
Seal-Id Inscription Translation
अन� anā indc. verily[RV]; आनष� ānarṣa set flow[√ऋष� + �लट� 3s.];
स�त�न� sātān amp. gifts[RV];
39 C-24 ������� � �
�� O Roaring destroyer, salutations
रसतस नमन� स� ध�र O possessor of all things
rasa-tasa naman saṃ-dhāra
RV 4.8.1 Your envoy who possesses all
रस rasa roarer[√रस� + अच� ŚBr]; तस tasa vms. destroyer[√तस� + अच� Dhātup];
नमन� naman saluting[√नम� + शत��]; स� saṃ indc. all[RV]; ध�र dhāra vns. possessing[BG];
37
Seal-Id Inscription Translation
अहम� aham ns. I[RV]; सह� sahaṃ ams. capable[Kālid]; म�क्षम� mokṣam ams. salvation[MBh];
नम� namīṃ ams. Nami, son of Sapya[RV 6.20.6]; जठल jaṭhala vms. Ocean;
श�य[म� ] śāya[m] ans. sleeping[RV];
38
Seal-Id Inscription Translation
शम śama calm[MBh]; शर śara arrow[RV,MBh]; अङ्ग aṅga body[Uṇ]; तम्र tamra vms. red/dark[RV];
[Sri Rudram: 1.4 य�त इष�� �शवतम�, 1.11 यस्त�म्र�]
39
Seal-Id Inscription Translation
40
902 6. Sign Variants and Conjuncts
903 Variants are stylistic and abstracted evolution of symbols. Conjunct signs combine two
904 or more signs and are read as adjacent signs: firstly in the direction of writing, secondly
905 top to bottom and lastly the base character followed by the ligatured character.
906 (1) Ligatured symbols that are simply touching but otherwise written in normal size
907 and position are read as normal. For example, � is simply read as ���.
908 (2) Symbols arranged vertically are read from top to bottom. � is read as �� �र as
909 opposed to � which is read as �� ईर
910 (3) The base character, i.e., the large character is read first and the embedded char-
911 acter is read subsequent. � becomes �� रस� .
अ H-829→ � �� अवर
� �
a avara
4
youngest[RV]
41
Sign Variant Inscription Sanskrit
अद� Ad-6 �/�\��� तद्धनद�म
/� \ ���
adā tat-dhana-dāma
12
giver of that wealth[RV,RV,RV]
आप H-413 � आप
� ��
āpa āpa
18
vedic deity(Vasus)[MBh]
अम M-969 �� आमय
� ��
ama serve[√अम� + ल�ट� 2s.] āmaya
23
42
Sign Variant Inscription Sanskrit
अर K-45→ ���� अण�पय
� ��
ara arṇapaya
27
power[RV] of flood[RV]
अश M-219 �� अश्म[न� ]
� ��
aśa aśma[n]
40
sky[RV]
आश M-2088 �� आश
� �
āśa āśa
41
food[ŚBr]
43
Sign Variant Inscription Sanskrit
आश M-1887 ���� म�म� आश�
� ��
āśa mām āśaṃ
42
obtaining me[RV]
ईर M-1170 � ईर
� ��
īra īra
45
rising [ईर� :RV]
ज M-1848 �� जष
� �
ja jaṣa
55
O destroyer[√जष� Dhātup]
जर M-898 �� जरम�
� ��
jara jaram
56
praise[√ज� RV]
44
Sign Variant Inscription Sanskrit
जस� Ns-4 �� यजस�
� ��
jas yajas
57
sacrifice[RV]
/ त K-11 /�
���� आतटद�
� �
ta ā-taṭa-daṃ
59
one who crosses[MBh]
ध� M-605 � ध�
� ��
dhī dhī
70
devotion[RV]
45
Sign Variant Inscription Sanskrit
न H-295 �� सन
� �
na sana
72
ancient[RV]
न M-1277 �� नश
� �
na naśa
73
destroyer[√नश� ]
प H-1024 ���� मव पण
� �
pa mava paṇa
79
bind[√मव� ] the wager[MBh]
पद M-777 ������
� असम�मनपद
��
� ��
pada asama-āmana-pada
80
unequalled[RV] friendly[TS] sign[MBh]
बत Frm-1329 �� भटन�
� ��
bata employing[√भट� +शत��] bhaṭan
82
46
Sign Variant Inscription Sanskrit
म M-523 ��� समर
� �
ma samara
87
confluence[RV]
↑ म H-2575 ↑� वम
�
ma vama
88
emitter[√वम� :RV]
म M-1367 �� मर
� �
ma mara
89
death[AV]
म� M-649 �� म�र
� ��
mī mīra
94
ocean[Uṇ]
य M-831 �� यत
� �
ya yata
97
controlled[RV]
47
Sign Variant Inscription Sanskrit
र K-18 �� ��� तद� वरल
� � ra tad varala
102
that[AV] swan[Mṛga]
48
Sign Variant Inscription Sanskrit
रर H-405 ���� �� रर वहन�मन�
� ��
rara rara vahanāman
117
give, O sign[RV] of bearer[RV]
रज K-122 ��� रर जट
� ��
raja rara jaṭa
119
granted[√र�] O Shiva[MBh]
�र M-918 �� �� रव र�र
� ��
ri rava rīra
122
O Roarer[√रु], O Shiva[Cat]
लल M-751A �� ललक
� ��
lala lalaka
124
little delight[√लल� ]
श M-207 �� शद
� �
śa śada
126
Destroyer[√शद� MBh]
49
912 7. Very Short Inscriptions
50
913 8. Derivation
� = अन� अ� · an am (1)
� = अ · a from 1 (2)
� = द · d from 1, 2 (3)
51
917 8.4. � · अ · a from आय� āyu man
� = अ · a from 1, 3 (4)
� = इ · i from 3, 4 (5)
� = द · d from 1, 2, 5 (6)
� = त · t from 1, 6 (7)
52
921 8.8. � · श · s from श�क्र śukra seed
� = श · s from 1, 5 (8)
� = न · n from 1, 2, 8 (9)
�� = ई · i� from 1, 5 (10)
� = ज · j from 1, 3, 4 (11)
53
925 8.12. � · र · r from रथ ratha chariot
� = र · r from 1, 2, 5, 11 (12)
� = र · r from 5, 11 (13)
� = च · c from 1, 5, 13 (14)
54
928 8.15. � · अ · a from अजशृङ्ग� ajaśṛṅgī goat’s horn
�= व · v from 1, 12 (17)
55
932 8.19. � · र · r from रथ��रन� rathārin chariot wheel
� = ब · b from 5, 9, 17 (20)
56
936 8.23. � · न · n from न�ल�क� nālīka arrow
938 8.25. � · अम · ama from अङ्क + मत्स्य ama conjunct curve + fish
57
940 8.27. � · न · n from न�व� nāvā boat
� = स · s from 1, 5 (29)
� = क · k from 1, 9, 12 (30)
58
944 8.31. � · अ · a from अङ्क aṅka curve
� = अ · a from 1, 8, 25 (32)
� = अ · a from 1, 2, 16 (34)
59
948 8.35. � · म · m from मन्थ mantha firesticks
60
952 8.39. � · श · s from श�खर śākhāra squirrel
61
955 8.42. � · र · r from रथद�रु rathadāru Dalbergia tree
62
958 8.45. � · द · d from ध�नक�� dhānakāḥ variant of coins �
63
961 8.48. � · द · d from ध�नक�� dhānakāḥ coins
64
964 8.51. � · त · t from त�ड� ल tāḍula fighter
65
967 8.54. � · श · s from श�खर śākhāra variant of squirrel �
66
970 8.57. � · म · m from मय maya horse
� = अ · a from 1, 17, 19, 21, 24, 27, 30, 38, 39, 42, 50 (59)
67
973 8.60. � · ल · l from लत� latā creeper
68
976 8.63. � · अ · a from अग aga mountain
� = भ · b from 1, 2, 4, 8, 11, 12, 13, 16, 27, 29, 31, 35, 52 (64)
� = ष · s from 1, 6, 12, 13, 16, 17, 19, 21, 23, 31, 41 (65)
69
979 8.66. � · श · s from �शख� śikhā variant of peacock crest �
70
983 8.70. � · क · k from क�तम� kṛtam axe
71
986 8.73. � · उ · u from उ��म udyāma variant of �
72
989 8.76. � · भ · b from भक्षपत्त्र� bhakṣapattrī betel leaf
73
993 8.80. � · क · k from क�ष kṛaṣa ploughshare
� = क · k from 2, 5, 8, 21 (80)
74
996 8.83. � · म · m from मय maya variant of horse �
� = द · d from 1, 6, 14, 15, 20, 30, 31, 36, 41, 45, 52, 55, 72, 77, 79 (85)
75
999 8.86. � · द · d from धन्वन� dhanvan variant of bow �
� = म · m from 1, 67 (89)
76
1003 8.90. � · त · t from त्र tra three
� = उ · u from 1, 4, 5, 11, 13, 16, 23, 42, 50, 55, 61, 65, 79 (91)
77
1006 8.93. � · ए · e from एक eka one
78
1010 8.97. � · य · y from य�ष्ट yaṣṭi pearl necklace
� = त · t from 1, 2, 4, 9, 12, 16, 17, 24, 31, 55, 63, 77, 98 (99)
79
1013 8.100. � · झ · j from झर jhara variant of rain; waterfall �
80
1016 8.103. � · म · m from मन्द�र mandāra a flower
� = अस� · as from 1, 13, 14, 16, 17, 24, 30, 38, 41, 49, 60, 64 (105)
81
1019 8.106. � · त · t from तण्ड� ल taṇḍula rice plant
82
1023 8.110. � · श · s from �शख� śikhā peacock crest
� = म · m from 1, 2, 13, 14, 16, 17, 24, 34, 39, 40, 82, 106 (111)
83
1026 8.113. � · र · r from रथद�रु rathadāru variant of Dalbergia �
1028
84
1030 8.117. � · ओ · o from ओपश opaśa tuft of hair
85
1032 9. Derivation Sequence
86
Phoneme Reconstructed Name
44 � ह h स=ह
45 � द d ध�नक�� dhānakāḥ variant of coins �
46 � र r रथव� ratharvī multi/split snake
47 � य y यव yava barley
48 � द d ध�नक�� dhānakāḥ coins
49 � त t त�ड्य tāḍya drum
50 � म m मत� ग mataṅga elephant head
51 � त t त�ड� ल tāḍula fighter
52 � स s सप्तन� saptan seven
53 � र r रथ��रन� rathārin variant of wheel �
54 � श s श�खर śākhāra variant of squirrel �
55 � प p पञ्चन� pañcan hand
56 � द d दन्त danta teeth
57 � म m मय maya horse
58 � म m मय maya variant of horse �
59 � अ a आज�न ājani variant of stick �
60 � ल l लत� latā creeper
61 � त t त�ल tāla small cymbal
62 � अन� an अ� श� aṃśu variant of drinking vessel �
63 � अ a अग aga mountain
64 � भ b भक्षत्र bhakṣatra oven
65 � ष s षण� ṣaṇ six
66 � श s �शख� śikhā variant of peacock crest �
67 � उ u उ��म udyāma coil of rope
68 � इ i इष�क iṣīkā variant of stalk of grass �
69 � व v वरट� varaṭī wasp
70 � क k क�तम� kṛtam axe
71 � क k क�ष kṛaṣa ploughshare
72 � प p पञ्चन� pañcan five
73 � उ u उ��म udyāma variant of �
74 � द d ध�नक�� dhānakāḥ variant of coins �
75 � छ c छत्त्र chattra mushroom
76 � भ b भक्षपत्त्र� bhakṣapattrī betel leaf
77 � अ a अङ्क aṅka variant of curve �
78 � त t त�� tardū wooden ladles
79 � म m मन्द�र mandāra variant of flower �
80 � क k क�ष kṛaṣa ploughshare
81 � ध d ध�न dhāna receptacle
82 � झ j झञ्झ�न� jhañjān rain and wind
83 � म m मय maya variant of horse �
84 � त t तद� tarda Indian blackbird
85 � द d धन्वन� dhanvan variant of bow �
86 � द d धन्वन� dhanvan variant of bow �
87 � स s �शख� śikhā variant of �
88 � म m मन्द�र mandāra variant of flower �
89 � म m म�न्दर mandira dwelling
87
Phoneme Reconstructed Name
115
88
1033 10. Conclusion
1034 The ability to read well beyond the unicity distance alone should be sufficient proof
1035 of correct decipherment of the Indus script. This decipherment has many additional
1036 compelling attributes. This is the only cryptanalytic decipherment and the only one
1037 that uses well-established mathematical models and methods instead of guessing sign
1038 values based on their appearance. This decipherment is the only full decipherment
1039 and the only one where every sign and every stroke has been resolved, the only one
1040 that is programmatically reproducible, the only one where the decipherment can be
1041 followed sign-by-sign by the reader, the only decipherment that reads Semitic and mixed
1042 inscriptions in addition to native IVC inscriptions, the only one that reads over 500
1043 inscriptions including all 50 longest inscriptions grammatically correctly in an attested
1044 language, the only one that validates research spanning almost a century from Hunter to
1045 Heggarty. In addition, we have uncovered a remarkable number of additional evidence
1046 such as reconstructed names of the signs, reasons for their allographs, and the clear
1047 correspondence of derived sound values to known Brahmi values. We also show how
1048 the constraints and habits of the Indus script carry on to Brahmi inscriptions of the
1049 early historic era. Such a strong result is a first in any ancient script decipherment and
1050 should be taken as plenary proof of decipherment of the Indus script.
1052 A programmatic decipherment of the first 40 signs is openly available in the GitHub
1053 repository at https://github.com/yajnadevam/ScriptDerivation. This paper uses the
1054 Indus script font from the National Fund for Mohenjodaro under open license(Kumb-
1055 har and Buriro, 2017). Brahmi and Devanagari fonts are from Google Fonts under Open
1056 Font license(Google, 2021, 2022). Adinata font for Tamil Brahmi is under Open Font
1057 license (Rajan, Sharma, and Sankar, 2021). The Indus corpus reference used is Interac-
1058 tive Corpus of Indus Text(Wells and Fuls, 2023). Corpus of Indus seals and inscriptions
1059 volumes are also the primary reference(Parpola et. al., 1991). Rigveda translations in
1060 section 5 are from Griffith(Griffith, 1896). The dictionary used for decipherment deriva-
1061 tion is the downloadable Monier-Williams dictionary(Monier-Williams, 1899b). Attes-
1062 tation data of individual words are from Monier-Williams dictionary(Monier-Williams,
1063 1899b), the Purana Index(Dikshitar, 1955) and Wisdomlib(Hiemstra, 2023).
References
89
ing, India. https://archive.org/details/in.ernet.dli.2015.207529/page/
n195/mode/2up?view=theater.
ASI. 2004. Indian Archaeology 1998-1999 - A review. Archaeological Survey of In-
dia. https://web.archive.org/web/20120508064754/http://asi.nic.in/nmma_
reviews/Indian%20Archaeology%201998-99%20A%20Review.pdf.
Association Assyrophile de France. 2006. “Sureth Dictionary.” https://www.
assyrianlanguages.org/akkadian/list.php.
Bacon, Roger. 1401-1599. “Voynich manuscript.” https://collections.library.
yale.edu/catalog/2002046?child_oid=1006204.
Balasubramaniam, R. 2005. Story of the Delhi Iron pillar. Foundation Books pvt Ltd.
https://books.google.com/books?id=FruHOlKlbJAC.
Banerjee, N. R., and Soundara K. V. Rajan. 1960. “Sanur 1950, 1952: A megalithic
site in district Chingleput.” Ancient India 15: 2–28. https://asi.nic.in/Ancient_
India/Ancient_India_Volume_15/article_1.pdf.
BBC. 1999. “’Earliest writing’ found.” Last accessed 18 March 2022, https:
//web.archive.org/web/20220318234228/http://news.bbc.co.uk/2/hi/sci/
tech/334517.stm.
Bhatt, S. K. 1998. Nishka-the Rig Vedic Money. Academy of Indian Numismatics and
Sigillography.
Bonta, Steven. 2023. “A Partial Decipherment of the Indus Valley Script: Proposed
Phonetic and Logographic Values for Selected Indus Signs and Readings of Indus
Texts.” academia.edu https://web.archive.org/web/20230913171235/https:
//www.academia.edu/105163460/A_Partial_Decipherment_of_the_Indus_
Valley_Script_Proposed_Phonetic_and_Logographic_Values_for_Selected_
Indus_Signs_and_Readings_of_Indus_Texts.
CDLI. 2023. “The CDLI Collection.” https://cdli.ucla.edu/.
CoinIndia. 2010. “The Coin Galleries: Mauryan Empire.” http://coinindia.com/
galleries-maurya.html.
Devi, Yashoda. 1933. The history of Andhra country (1000 AD - 1500 AD). Gyan
Publishing House.
Dikshitar, Ramachandra. 1955. “The Purana Index.” https://www.
sanskrit-lexicon.uni-koeln.de/scans/PUIScan/2020/web/index.php.
Dixit, K. N. 2019. “Origin of Early Harappan Cultures in the Sarasvati
Valley: Recent Archaeological Evidence and Radiometric Dates.” Jour-
nal of Indian Ocean Archaeology 88–141. https://web.archive.org/
web/20170118032736/http://server2.docfoc.com/uploads/Z2015/11/21/
vESLakMBYz/45a03572f94e7a873d7c350293cca188.pdf.
Ephron, H. D. 1961. “MYCENAEAN GREEK: A LESSON IN CRYPTANALYSIS.”
Minos: Revista de Filología Egea 63–100. https://revistas.usal.es/dos/index.
php/0544-3733/article/view/2088/2140.
Gelb, I. J. 1957. Glossary of old Akkadian. University of Chicago Press. https://isac.
uchicago.edu/sites/default/files/uploads/shared/docs/mad3.pdf.
George, Andrew. 2003. “CUNEIFORM INSCRIPTION OF DEFEAT OF
OMAN and INDUS VALLEY.” https://www.schoyencollection.com/
history-collection-introduction/sumerian-history-collection/
cuneiform-indus-valley-ms-2814.
Goibhniu. 2007. “Perl uses for Cryptograms - Part 1: One-liners and
Word Patterns.” https://web.archive.org/web/20170223171834/https://www.
perlmonks.org/?node_id=636818.
Google. 2021. “Noto Serif Devanagari font.” This work is licensed under the
90
Open Font Licence. To view a copy of this license, visit https://scripts.sil.
org/cms/scripts/page.php?site_id=nrsi&id=OFL, https://fonts.google.com/
noto/specimen/Noto+Serif+Devanagari/about.
Google. 2022. “Noto Sans Brahmi font.” This work is licensed under the Open
Font Licence. To view a copy of this license, visit https://scripts.sil.
org/cms/scripts/page.php?site_id=nrsi&id=OFL, https://fonts.google.com/
noto/specimen/Noto+Sans+Brahmi/about.
Griffith, Ralph T. H. 1896. The Hymns of the Rigveda. Munshiram Manoharlal Publ.
Vedaweb version https://vedaweb.uni-koeln.de/rigveda/view/index/0, http://www.
sanskritweb.net/rigveda/griffith.pdf.
Haryana Directorate of Archaeology and Museums. 2023. “Excavation Sites A and M.”
https://web.archive.org/web/20230330043817/http://archaeologyharyana.
nic.in/sites/default/files/Excavation%20sites%20A%20and%20M.pdf.
Heggarty, Paul, Cormac Anderson, Matthew Scarborough, Benedict King, Remco
Bouckaert, Lechosław Jocz, Martin Joachim Kümmel, et al. 2023. “Language trees
with sampled ancestors support a hybrid model for the origin of Indo-European
languages.” Science 381 (6656): eabg0818. https://doi.org/10.1126/science.abg0818,
https://www.science.org/doi/abs/10.1126/science.abg0818.
Hiemstra, Gabe. 2023. “Wisdom Library.” https://wisdomlib.org.
Hultzsch, E. 1925. Inscriptions of Asoka. Clarendon Press. https://archive.org/
details/InscriptionsOfAsoka.NewEditionByE.Hultzsch/page/n110/mode/1up?
view=theater.
Hunter, G. R. 1934. The script of Harappana and Mohenjodaro and its connection
with other scripts. Kegan Paul, Trench, Trubner Co Ltd. https://ufdc.ufl.edu/
AA00013642/00001.
Kak, Subhash. 1988. “A Frequency analysis of the Indus script.” Cryptologia 129–142.
https://www.ece.lsu.edu/kak/IndusFreqAnalysis.pdf.
Knight, Kevin, Beáta Megyesi, and Christiane Schaefer. 2011. “The Copiale Cipher.”
Association for Computational Linguistics 2–9. https://web.archive.org/web/
20111112013300/http://www.aclweb.org/anthology-new/W/W11/W11-12.pdf#
page=12.
Kumbhar, Shabir, and Amar Fayaz Buriro. 2017. “Indus script font.” License ”Indus
Script font is available to be downloaded for further studies, computational exercises
and statistical analysis, free of charge; the only encumbrance is that user acknowl-
edge our website.”, https://web.archive.org/web/20210308035522/http://www.
mohenjodaroonline.net/index.php/indus-script/corpus-by-asko-parpola.
Lahafian, Jamal. 2013. “Rock Art in Kurdistan Iran.” https://www.researchgate.
net/publication/276033365_Rock_Art_in_Kurdistan_Iran.
Lal, B. B. 1960. “From the Megalithic to the Harappa: tracing back the grafitti on the
pottery.” Ancient India 16: 4–24. https://asi.nic.in/Ancient_India/Ancient_
India_Volume_16/article_1.pdf.
Lashari, K., Sindh (Pakistan). Department of Culture, Tourism, and Sindh Archives.
2020. Studies on Indus Cript: Conference on Indus Script Mohenjodaro 2020. National
Fund for Mohenjodaro, Culture, Tournism, Antiquities and Archives Department,
Government of Sindh. https://books.google.com/books?id=gjWCzwEACAAJ.
Laursen, Steffen Terp. 2010. “The westward transmission of Indus Valley seal-
ing technology: origin and development of the ‘Gulf Type’ seal and other ad-
ministrative technologies in Early Dilmun, c.2100–2000 BC.” Arabian archaeol-
ogy and epigraphy 96–134. https://onlinelibrary.wiley.com/doi/10.1111/j.
1600-0471.2010.00329.x.
91
Lien, L. T. 2013. “Brahmi Inscription from Archaeological Sites in South-
ern Vietnam during the 1st Millennium CE.” Advancing Southeast Asian Ar-
chaeology 402. https://publications.spafajournal.org/index.php/spafapub/
catalog/view/37/53/158-1.
Mahadevan, Iravatham. 1977. The Indus Script. Archaeological Sur-
vey of India. https://archive.org/34/items/TheIndusScript.
TextConcordanceAndTablesIravathanMahadevan/The%20Indus%20Script.
%20Text%2C%20Concordance%20and%20Tables%20-Iravathan%20Mahadevan.pdf.
Mahadevan, Iravatham. 2010. “Akam and Puram : ‘Address’ Signs of the Indus Script.”
https://rmrl.in/wp-content/uploads/42-Akam-and-Puram.pdf.
Monier-Williams. 1899a. “Monier-Williams dictionary Sanskrit abbreviations.” https:
//en.wiktionary.org/wiki/Appendix:Sanskrit_abbreviations.
Monier-Williams. 1899b. “Monier-Williams Sanskrit-English Dictionary, 1899.” https:
//www.sanskrit-lexicon.uni-koeln.de/scans/MWScan/2020/web/index.php.
Moreno, Marsha Lynn. 2005. “Frequency Analysis in Light of Language In-
novation.” https://mathweb.ucsd.edu/~crypto/Projects/MarshaMoreno/
TimeComparisonFrequency.pdf.
Munjal, Sanjay, and Arvin Munjal. 2005. “(Brahmi inscription on Anthropomorphic
figure).” Prāgdhārā .
Parpola, Asko. 1994. Deciphering the Indus script. Cambridge University Press.
Parpola, Simo, Asko Parpola, and Robert H. Brunswig Jr. 1977. “The Meluḫḫa Vil-
lage: Evidence of Acculturation of Harappan Traders in Late Third Millennium
Mesopotamia?” Journal of the Economic and Social History of the Orient 20 (2):
129–165. http://www.jstor.org/stable/3631775.
Parpola et. al. 1991. Corpus of Indus seals and inscriptions, Vol 1,2,3.1,3.2,3.3. Suo-
malainen Tiedeakatemia.
Peters et al. 2022. “The biocultural origins and dispersal of domestic chickens.” PNAS
119 (24). https://www.pnas.org/doi/pdf/10.1073/pnas.2121978119.
Raghupathy, P. 1987. Early settlements in Jaffna: An archaeological survey. P. Raghu-
pathy.
Rajan, Vinod, Shriramana Sharma, and Udhaya Sankar. 2021. “Adinatha Tamil-
Brahmi font.” This work is licensed under the Open Font Licence. To view a copy
of this license, visit https://scripts.sil.org/cms/scripts/page.php?site_id=
nrsi&id=OFL, http://www.virtualvinodh.com/projects/adinatha.
Ramakrishna, K. A., N. K Swain, M. Rajesh, and N. Veeraraghavan. 2018. “Ex-
cavations at Keeladi, Sivaganga District, Tamil Nadu (2014 - 2015 and 2015 -
16).” Heritage: Journal of Multidisciplinary Studies in Archaeology 6 30–72. http:
//www.heritageuniversityofkerala.com/JournalPDF/Volume6/2.pdf.
Rao, Rajesh P. N., Nisha Yadav, Mayank N. Vahia, Hrishikesh Joglekar,
R. Adhikari, and Iravatham Mahadevan. 2009. “A Markov model of the
Indus script.” Proceedings of the National Academy of Sciences 106 (33):
13685–13690. https://doi.org/10.1073/pnas.0906237106, https://www.pnas.org/
doi/abs/10.1073/pnas.0906237106.
Rao, S. R. 1980. “Indus script and language.” Annals of the Bhandarkar Oriental Re-
search Institute 61: 157–188.
Reddy, Bhaskar, and V. Sakunthalamma. 2023. “Indian Numismatics:
Punchmarked Coins of the Janapadas.” https://epgp.inflibnet.ac.
in/epgpdata/uploads/epgp_content/S000829IC/P001738/M028921/ET/
1523266853P09-M04-PunchMarkedCoinsoftheJanapadas-ET.pdf.
Shannon, C. 1945. A Mathematical Theory of Cryptography. Alcatel-Lucent. https:
92
//www.iacr.org/museum/shannon/shannon45.pdf.
Singh, Unpinder. 2006. Delhi: Ancient History. Berghahn Books. https://en.
wikipedia.org/wiki/Iron_pillar_of_Delhi.
Sivanantham, Dr. R., and M. Seran. 2019. Keeladi: An Urban Settlement of Sangam
Age on the banks of river Vaigai. Department of Archaeology, Government of Tamil
Nadu. https://web.archive.org/web/20231015164905/https://www.vinavu.
com/wp-content/uploads/2019/09/Keeladi-Book-English-18-09-2019.pdf.
Solomon, Richard. 1998. Indian Epigraphy. Oxford University Press.
Srivatsava, O. 2021. (New Dimensions of History and Archeology). B. R. Publishing
Corporation.
Subrahmanian, T. S. 2010. “Tamil Brahmi potsherds found at urn burial site.”
https://www.thehindu.com/features/friday-review/history-and-culture/
Tamil-Brahmi-potsherds-found-at-urn-burial-site/article16483663.ece.
Sullivan, Sue. 2011. Indus script dictionary. Sullivan, Sue.
UPenn. 2006. “The Pennsylvania Sumerian Dictionary.” http://psd.museum.upenn.
edu/nepsd-frame.html.
Vyas, Shail. 2020. “Indus Musicians in Mesopotamia.” https://osf.io/preprints/
socarxiv/kce5x/download&usg=AOvVaw228Fmhab2qMXq19nY5AenT&opi=89978449.
Wells, Bryan, and Andreas Fuls. 2023. “Interactive Concordance of Indus Texts (ICIT):
An Online Database of Indus Inscriptions and Iconography.” https://www.indus.
epigraphica.de.
wikipedia. 2021. “List of Beijing Subway stations.” https://en.wikipedia.org/wiki/
List_of_Beijing_Subway_stations.
93
Appendix A. Evolution of Brahmi signs
94