Indus Script: Search for Grammar1

1
Indus Script Search for Grammar1

• Nisha Yadav
• Tata Institute of Fundamental Research
• Collaborators
• Mayank Vahia, Iravatham Mahadevan, Hrishikesh
  Joglekar

1 Lecture given at a two day seminar on The
Indus Script Problems and Prospects, Chennai
2
Contents

• Indus Script - An Overview
• Various Approaches
• Our Approach
• Dataset
• Preliminary Analysis
• Analysis - 1 Check against random order
• Analysis - 2 Positional analysis of Frequent
  Sign combinations
• Text Beginners and Text Enders
• Segmentation of Indus Texts
• Summary

Note In the lecture, unless specified otherwise,
all text examples are from Mahadevan 1977 and
all images are from Parpolas UNESCO volumes of
Indus seals.
3
1) Indus Script An Overview
4
Indus Valley Civilization
From Mahadevan, 1977
5
Indus Script Pointers to understand

• Indus script is one of the few scripts that defy
  decipherment.
• Inscriptions found only on small objects like
  seals.
• The inscriptions are very brief average length
  4-5 signs.
• There are only 417 signs in the script as per
  Mahadevans Concordance (1977).
• The script is pictographic with signs showing
  human, fish etc.
• Signs are modified by joining or by strokes and
  many signs appear as combination of other simple
  signs.
• The direction of the script is variable (mostly
  right to left 83 of times).
• In general the seals are of 1 to 2 square inches
  in size.
• There are no bi-lingual texts to aid
  decipherment.

6
Direction indicators of the script

• Cramping or overflow of signs at the left end
• Orientation of asymmetric signs
• Sequence of frequent combinations of signs
• Split sequences

A split sequence indicating direction
7
Scale of a typical seal
For the most part, seals are between 1 inch or 2
inches square.
From Professor John C. Huntingtons ppt
8
SEAL
SEAL IMPRESSION
SEAL
SEAL IMPRESSION
From Professor John C. Huntingtons ppt
9
Specimens of Indus Texts on different objects
Text No. Text
From Mahadevan, 1977
10
INDUS INSCRIPTIONS
Courtesy www.harappa.com
11
Indus Script Signs (1 to 110)
From Mahadevan, 1977
12
2) Various Approaches
13
Indus Script

• Scientists from a variety of disciplines have
  attempted
• to read the Indus script with no clear answer.
• Various attempts so far include
• I. Mahadevans analytical work Creation of
  first Published Concordance (1977)
• Gift Siromoneys statistical work
• A. Parpolas comparison with Dravidian
• Russian groups comparison with Dravidian
• Subbarayappas interpretation as pure numerals
• S. R. Raos interpretation as Vedic literature
• Others (Ref. Possehl,1996)

14
3) Our Approach
15

• We make no assumption about its content or
  meaning.
• Our first emphasis is to attempt to WRITE IN THE
  SCRIPT RATHER THAN READ.
• We search for rules of writing without assigning
  meanings or interpretations.
• We ignore variation due to archaeological context
  of sites, stratigraphy and type of objects.

16
4) Dataset
17
Dataset

• Unambiguous data subset (EBUDS) was created for
  analysis of the grammar of Indus writing, from
  the original electronic dataset of Mahadevan
  (1977) partially modified as M80.
• EBUDS Extended Basic Unique Dataset, excludes
• All ambiguous lines
• All texts from sides having multiple lines
• All duplicates (keeping their single occurrence)
  
• Thus, EBUDS consists of 1548 lines of texts, with
  7000 sign occurrences.

18
5) Preliminary Analysis
19
Frequency distribution of Indus Signs

• Only 67 (16 of total no. of signs) signs
  account for over 80 of the writing.

20
Conclusions from Preliminary Analysis

• The frequency distribution of the signs in EBUDS
  is consistent with M77.
• The manner of choosing the data set has not
  changed the pattern of occurrence of various
  signs and the results are consistent with the
  analysis of M77.
• Only 67 signs (16 of total no. of signs) account
  for over 80 of the writing.

21
6) Analysis 1Check against Random Order
22
Methodology

• We take 1548 unique texts (7000 signs) present in
  EBUDS.
• We randomise their appearance keeping the
  frequency of each sign as in EBUDS.
• We split this long random string (of 7000 signs)
  into texts of 1 to 14 signs as in EBUDS.
• We create 10 such random databases.
• We then compare the frequency of their sign
  pairs, triplets etc. with Genuine Indus database
  (EBUDS) to check if Indus texts have any
  significant sequencing.

23
Comparison of EBUDS with Random Datasets
24
Result of Analysis 1
25
Conclusions from Analysis 1

• String lengths of 2, 3 and 4 signs appear with
  frequency far higher than expected by random
  chance.
• The signs are ordered in a specific manner.
• It is justifiable to state that Indus texts
  followed certain rules and thereby meant
  something significant and meaningful.

26

• 7) Analysis 2
• Positional analysis of Frequent Sign Combinations

27
Positional Analysis of Frequent Two-sign
Combinations
28
Positional Analysis of Frequent Three-sign
Combinations
29
Positional Analysis of Frequent Four-sign
Combinations
30
Conclusions from Positional analysis

• The most frequent two-sign, three-sign and
  four-sign combinations appear at fixed positions.
• The exact location varies from combination to
  combination.
• However, frequently occurring two-sign,
  three-sign and four-sign combinations may be
  incomplete except of course when they occur as
  solo texts.
• It can be seen that two-sign, three-sign and
  four-sign combinations which are complete have
  typically one of the text-enders (mostly 342
  or 211 ) at the end. This is confirmed by
  the solo occurrences of such texts.

31
8) Text Beginners and Text Enders
32
Indus Text Beginners and Enders
33

• Consider an Indus Text with Signs
• G F E D C B A
• (In order of their statistical significance)

Frequent Text Beginners
Frequent Text Enders
34
Specimens of Indus Texts illustrating syntactical
patterns
From Mahadevan (1986)
35
Conclusions for Indus Script

• There are well defined text-enders though
  text-beginners are not that well-defined.
• Sign distribution within the strings seems to be
  ordered as per some specific rules. The
  distribution is far more significant than would
  arise by chance.
• This indicates existence of patterns and rules
  that need to be dug out.

36
9) Segmentation of Indus Texts
37
Segmentation Approach

• There can be various methods which can be used
  for segmenting an
• Indus text namely
• Comparing texts
• Using frequent combinations of signs
• Using Pair Frequencies
• Using Single Signs (Enders, Beginners, Auxiliary
  Enders)
• These methods are overlapping and hence it is
  decided to select an
• approach which takes into consideration the
  effect of each of these.
• A cumulative method based on statistically
  significant units, is thus
• formulated.

38
Segmentation using Comparison of texts
Longer text can be shown to consist of 2 or more
shorter texts occurring as complete texts
elsewhere indicating the boundaries
From Mahadevan (1986)
39
Segmentation using Frequent Sign Combinations
A few highly frequent signs form stable
combinations with other signs. Then these
sign-combinations can be treated as separate
segments.
From Mahadevan (1986)
40
Segmentation using Pair frequencies
Comparison of frequencies of successive adjacent
pairs reveals boundaries at the weakest
junctions.
From Mahadevan (1986)
41
Segmentation Process
Percent of texts split (for texts of 5 or more
signs)
INDUS TEXT
Look for pair, triplet and quad texts successively
55 split
Look for frequent 4, 3 and 2 sign combinations
successively
77 split
Look for Enders, Beginners and Auxiliary Enders
successively
88 split
TEXT SEGMENTS
42
Segment Length vs. Segment Frequency in EBUDS
before and after segmentation
43
EBUDS before and after segmentation
EBUDS after Segmentation
EBUDS before Segmentation
44
Few Examples of Segmentation
45
Conclusions from segmentation

• It is possible to segment 88 of Indus texts of
  length 5 and above into segments of length 4 and
  below by using statistically significant signs
  and their combinations in addition to all the
  texts of length 2, 3 and 4.
• Many frequent sign combinations make their
  appearance as independent texts.
• The Indus texts after segmentation can be viewed
  as permutations of the identifiable units
  (segments) of 2, 3 or 4 signs.
• The identifiable units may or may not be
  standalone (or complete) pieces of information.

46
10) Summary
47
Summary

• The writing is highly ordered.
• Typical length of information containing units is
  2, 3 or maximum 4 signs.
• However, they are not always complete enough to
  exist as standalone pieces of text.
• This suggests a more complex grammar in the
  writing where information units need proper
  beginners or enders.
• The present study shows that Indus writing seems
  to have specific ordering as would be expected if
  sophisticated information is coded. This is
  consistent with the general level of
  sophistication associated with the Indus culture.

48
End