This repo is a (currently WIP) digitization of the Corpus of Indus Seals and Inscriptions (CISI) by Parpola et al.
The corpus is in json format, which can be read by humans with a little training, and is consumed easily by applications. This format was chosen
due to the popularity of json in the programming community, and most programming languages have robust support for handling json data.
I'd like to thank Dr Asko Parpola for his tireless efforts into the preservation and dissemination of the Indus Valley corpus. It is largely due to his decades-long efforts that I am able to transform the priceless artefacts of the Indus Valley Civilization into this digital format. Thank you, Dr Parpola.
I'd also like to acknowledge both Dr Andreas Fuls and Bryan K Wells for their many contributions, both in theory and in pedagogy, of the Indus Valley scripts. While their work is less represented here, that is most certainly not due to any lack of significance or importance.
In each file in the corpus subdirectories, the json file represents a single artefact. The highest level is an array of "sides" of the artefact.
Each array entry is one side of the artefact, with a json object containing the object id, a short description of the artefact type, and a list of graphemes.
The graphemes are recorded from the left of the artefact side to the right with the understanding that the script is read right-to-left. Note that
for seals, I have transcribed the sealing as opposed to the seal itself, as this was the intended way the inscriptions would be used.
To get a human-readable description of a given artefact, I have provided a small utility in the artefact_finder folder. This is a Rust language command
line utility. When run from the artefact_finder folder, you can provide command line arguments like -a m102 to display the corpus entry for the
Mohenjo-Daro M102 artefact as per the Corpus of Indus Seals and Inscriptions (CISI). If you would like to use this utility but are not familiar with
the Rust prgramming language, I am happy to provide pre-built binaries upon request. The source code is provided by default here so users can compile
the utility for any operating system via the usual Rust build techniques.
My intention is to create a friendly, free, open digital dataset for Indus script studies. To this end, I will use the CISI text numbering scheme, where for instance the second Mohenjo-daro text is denoted "M-2".
The initial goal I hope to work toward is refining the graphemic and allographic understanding of the script. It is my belief that this is a fundamental pre-requisite for a more global understanding of the artefacts. By producing a corpus like the one here, I hope others will be able to use the data herein to work toward this goal. Of course, many other important insights can be gleaned from having another corpus available, such as verifying the data between the various corpuses. By releasing this corpus under the MIT and/or Apache licenses, it is my hope that others can use this data in whatever way they wish, through careful analysis and cross-referencing with other well-established information and corpora.
I will use the most inclusive allographs provided by Parpola (1982), but provide additional digital information to distinguish more finely between
graphemes that may be important. For instance, the tree sign with symmetric trifurcation at the top is allographed by Parpolo to the tree sign
with multiple repeated bifurcations along one side. I will use Parpola's numbering system, which would give sign P086 to both of these graphemes.
While Parpola does include an alternate numbering scheme, giving V126 to the first and V127 to the second, I choose to use P086 and include
extra information alongside the primary sign value.
I have also included Wells's sign list from Wells (2015), matching as closely as I could with the Parpola sign list. Where there were multiple signs from Wells that mapped to a single sign from Parpola, I ensured that there was an associated feature to distinguish between them.
The corpus itself can be found in the corpus subdirectory of this repository.
This extra graphemic information takes the shape of "grapheme features", with a "feature vector" for each allograph. For instance, with sign P086, we
have a feature vector like:
branching_factor(integer, >= 2)branch_count(integer, >= 1)branch_direction(integer, 0 = none, 1 = left, 2 = right)
So V126, the tree with a single trifurcation on top, would be represented as P086 (3, 1, 0). In contrast, a common variant of V127, the tree with
four bifurcations along the right branch, would be represented as P086 (2, 4, 2).
The feature vectors for each sign from Parpola (1982) can be found in the features subdirectory of this repository.
Note that all allographs share some common "default" features, which include "damage", "line", and "uncertainty". Damage refers to how much
of a sign might be damaged, Line refers to which textual line on the artifact in which the grapheme occurs, and Uncertainty refers to how clearly
recognizable the grapheme is on the artefact (as judged by the annotator). Default features come before all other features. So for the first example above, P086 (0, 1, 0, 3, 1, 0)
would indicate no damage, line 1, no uncertainty, trifurcating, branching only once, with no significant branching direction.
The Uncertainty feature is extremely subjective and only included to give a rough indication of my own relative certainty when assigning a primary allographic designation.
My intention is to use an integer encoding for all features in a given allograph's feature vector. While this does obscure the exact graphemic representation for a given representation, it does make the data more easily used by numerical analyses.
I have developed a small Visual Studio Code extension which allows you to hover over a number like P086, and it will give you a description of
the symbol itself, as well as a description of the feature vector for that symbol. This lives in the indus_helper_vsce directory. It is fairly
technical to build VSCode extensions, but if you'd like a copy, I can supply one.
There are a few ambiguities in Parpola's simplest allograph scheme. These are in the numerical signs, the groups of vertical strokes. In particular,
P121(V764) seems to be identical with P144(V008), P122(V765) with P145(V009), P123(V766) with P147(V010), and P124(V771) with P150(V294).
I have tried to be consistent in choosing the allographs from P144 onward for full-height vertical strokes when possible, and reserving the allographs
from P121 for partial-height strokes.
I intend to include Unicode characters to give a visual clue to the grapheme being described. Right now, I have tested the AMBILE IVC Script font, and it seems to work nicely once installed.