joyboseroy commited on
Commit
642d5a8
·
verified ·
1 Parent(s): 20d5e2e

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +210 -0
README.md CHANGED
@@ -1,3 +1,213 @@
1
  ---
2
  license: mit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: mit
3
+ language:
4
+ - en
5
+ pretty_name: Indus Script Corpora (indus_website and CISI)
6
+ tags:
7
+ - indus-script
8
+ - indus-valley
9
+ - undeciphered-script
10
+ - archaeology
11
+ - historical-linguistics
12
+ - computational-epigraphy
13
+ size_categories:
14
+ - 1K<n<10K
15
+ task_categories:
16
+ - text-classification
17
+ - other
18
+ configs:
19
+ - config_name: indus_website
20
+ data_files: indus_website_real_corpus.csv
21
+ - config_name: cisi_primary
22
+ data_files: cisi_real_corpus.csv
23
+ - config_name: cisi_hierarchical
24
+ data_files: cisi_real_corpus_hierarchical.csv
25
+ - config_name: cisi_allograph
26
+ data_files: cisi_real_corpus_allograph.csv
27
  ---
28
+
29
+ # Indus Script Corpora (indus_website and CISI)
30
+
31
+ ## Dataset Description
32
+
33
+ Two independently sourced Indus Valley script corpora, packaged together
34
+ as four configs of one dataset since they share a schema and are
35
+ routinely used side by side for comparison. Both are derived, cleaned
36
+ CSVs built for computational analysis, not new digitization efforts.
37
+ All credit for the underlying transcription work belongs to the two
38
+ source repositories named below; this card documents what was changed,
39
+ checked, and fixed in producing these versions, so downstream users
40
+ don't have to reverse-engineer that from code.
41
+
42
+ **This dataset does not decipher or interpret the Indus script.** It is
43
+ raw structural data (which signs appear, in what order, at which site)
44
+ for anyone doing their own statistical or computational analysis.
45
+
46
+ ### The four configs
47
+
48
+ | Config | Corpus | N (usable) | Signs | Unit |
49
+ |---|---|---|---|---|
50
+ | `indus_website` | 2,543 inscriptions, 52 sites | 2,543 | 592 | inscription |
51
+ | `cisi_primary` | 179 Mohenjo-daro unicorn seals (104 non-damaged) | 179 | 182 (142 non-damaged) | inscription |
52
+ | `cisi_hierarchical` | same 179 seals, allographs split at 3+ attestations | 179 | 234 (185 non-damaged) | inscription |
53
+ | `cisi_allograph` | same 179 seals, every allograph a distinct sign | 179 | 337 (230 non-damaged) | inscription |
54
+
55
+ Load a specific config with, e.g.:
56
+ ```python
57
+ load_dataset("<repo>", "cisi_hierarchical")
58
+ ```
59
+
60
+ **Do not directly compare raw statistics (entropy, vocabulary size,
61
+ uniqueness) between `indus_website` and any `cisi_*` config without
62
+ adjusting for scale.** Their vocabularies differ by roughly 3 to 4
63
+ times (142 to 337 signs vs. 592), and such statistics are strongly
64
+ vocabulary-size-dependent. The three `cisi_*` configs are directly
65
+ comparable to each other (same 179 underlying inscriptions), and are
66
+ specifically designed to be compared to each other; see "Why three CISI
67
+ granularities" below.
68
+
69
+ ### Companion resources
70
+
71
+ Built as part of
72
+ [joyboseroy/indus_decipher](https://github.com/joyboseroy/indus_decipher),
73
+ a falsification-oriented statistical toolkit for the Indus script. That
74
+ repository's `CORPUS_REGISTRY.md` and `CLAIM_EVIDENCE_MATRIX.md`
75
+ document, in much more depth than this card, every real corpus this
76
+ project uses (including a fifth corpus, the actual M77/IDF-80 corpus,
77
+ not distributed here due to its more involved access terms), the
78
+ specific bugs found and fixed while working with each one, and exactly
79
+ what has and has not been established using them.
80
+
81
+ ## Dataset Structure
82
+
83
+ ### Data Instances
84
+
85
+ `indus_website`:
86
+ ```
87
+ inscription_id,sign_sequence,site,object_type,line_count,damaged,reading_direction,motif,mean_uncertainty,cisi_number
88
+ S2531,G140 G8 G342,Mohenjo-daro,seal,1,False,R-L,Bull1:W,0.0,M-1734
89
+ ```
90
+
91
+ `cisi_primary` / `cisi_hierarchical` / `cisi_allograph` (same schema,
92
+ sign IDs differ in granularity):
93
+ ```
94
+ inscription_id,sign_sequence,site,object_type,line_count,damaged,reading_direction,motif,mean_uncertainty
95
+ M-1A,P121 P202 P385 P073 P108,Mohenjo-daro,seal,1,False,L-R,unicorn_I,0.0
96
+ ```
97
+
98
+ ### Data Fields
99
+
100
+ All four configs share one schema:
101
+
102
+ | Field | Type | Description |
103
+ |---|---|---|
104
+ | `inscription_id` | string | Real ID from the config's own source (database key for `indus_website`; real CISI ID, e.g. `"M-1A"`, for the three `cisi_*` configs) |
105
+ | `sign_sequence` | string | Space-separated sign IDs, left-to-right as stored (see `reading_direction`) |
106
+ | `site` | string | Excavation site. 52 distinct sites in `indus_website`; always `"Mohenjo-daro"` in the `cisi_*` configs |
107
+ | `object_type` | string | Seal, tablet, etc. where recorded; `"unknown"` otherwise |
108
+ | `line_count` | int | Number of physical lines on the object |
109
+ | `damaged` | bool | `indus_website`: a per-inscription flag from the source. `cisi_*`: true if ANY grapheme in the inscription has a nonzero damage code |
110
+ | `reading_direction` | string | `"R-L"` means use `sign_sequence` as stored; `"L-R"` means reverse it before use. A loader-convention flag, not a claim about the script's actual reading direction. `indus_website` has a genuine mix of both values (per-inscription, source-provided); all `cisi_*` rows are `"L-R"` |
111
+ | `motif` | string | `indus_website`: iconographic field-symbol code (e.g. `"Bull1:W"`), `"unknown"` for 921 of 2,543 rows. `cisi_*`: always a unicorn variant (e.g. `"unicorn_I"`) |
112
+ | `mean_uncertainty` | float | `indus_website`: always `0.0` (no per-grapheme uncertainty field in that source). `cisi_*`: mean, across an inscription's graphemes, of the original annotator's 0-100 subjective uncertainty score, distinct from `damaged` |
113
+ | `cisi_number` | string | `indus_website` only. Real Mahadevan/CISI catalog number (e.g. `"M-1734"`, `"H-322"`) for 2,375 of 2,543 rows (93%); `"unknown"` otherwise. Not applicable to the `cisi_*` configs, whose `inscription_id` already is the real CISI number |
114
+
115
+ ### Data Splits
116
+
117
+ No splits in any config.
118
+
119
+ ## Dataset Creation
120
+
121
+ ### Source Data
122
+
123
+ `indus_website`: MySQL dump (`population-script.sql`) from
124
+ [yajnadevam/indus-website](https://github.com/yajnadevam/indus-website).
125
+ Scale matches the "ICIT/Yajnadevam digitization" cited in independent
126
+ literature on Indus sign-system statistics.
127
+
128
+ `cisi_*`: JSON files from
129
+ [mayig/indus-valley-script-corpus](https://github.com/mayig/indus-valley-script-corpus)
130
+ (MIT licensed), an in-progress, transparently documented
131
+ hand-transcription of Parpola's CISI corpus, credited to Parpola,
132
+ Wells, and Fuls's sign lists.
133
+
134
+ ### Corrections made during conversion
135
+
136
+ - **Reading direction (`indus_website`).** The original conversion had
137
+ sign order backwards. Caught by a diagnostic checking for a published
138
+ fingerprint (final sign position should show lower entropy than
139
+ initial position); confirmed correct after reversal.
140
+ - **CISI/Mahadevan numbers (`indus_website`).** The source SQL has a
141
+ `CISI` column that an earlier version of the conversion script never
142
+ extracted. Fixed: 93% of rows now carry this real crosswalk number,
143
+ spot-checked against physical seal photographs for consistent sign
144
+ counts.
145
+ - **Reading direction (`cisi_*`).** Same kind of fix, independently
146
+ confirmed for this separate source.
147
+
148
+ ### Why three CISI granularities
149
+
150
+ The `cisi_hierarchical` config exists to distinguish two competing
151
+ explanations for a result found using `cisi_allograph`: does allograph
152
+ choice genuinely matter, or is naive full fragmentation just too sparse
153
+ on 104 inscriptions to trust? `cisi_hierarchical`'s 3-occurrence
154
+ threshold was built specifically to test this. It did not resolve to
155
+ "it's just sparsity," which is why all three configs are provided
156
+ rather than one being treated as the "real" version.
157
+
158
+ ## Considerations for Using the Data
159
+
160
+ - **The three `cisi_*` configs give different answers to the same
161
+ structural question, and that is a documented finding, not an
162
+ inconsistency to average away.** Using a basic language-vs-non-language
163
+ statistical classifier (see companion repo), `cisi_primary` classifies
164
+ this corpus one way; `cisi_hierarchical` and `cisi_allograph` both
165
+ classify it the other way, and this is not simply because the latter
166
+ two are too sparse (`cisi_hierarchical` was built to be conservative
167
+ and still agrees with the fully fragmented version). Tracing why found
168
+ that the single most common sign at primary granularity is actually
169
+ several independently well-attested, visually distinct forms bundled
170
+ under one ID. Report all three when it matters, don't pick the one
171
+ that supports your preferred conclusion.
172
+ - **`cisi_*` is small (104 non-damaged inscriptions) and entirely one
173
+ motif at one site.** Statistics requiring higher-order sequential
174
+ structure did not produce a reliable signal on it in the companion
175
+ repository's own testing, most likely due to sample size. It is also
176
+ not a representative cross-section of the Indus corpus as a whole.
177
+ - **`motif` and `cisi_number` are both partial in `indus_website`.**
178
+ Treat `"unknown"` as missing data, not a meaningful third category.
179
+ - This dataset alone cannot support a claim that the Indus script is or
180
+ is not a language, an administrative code, or any other specific kind
181
+ of system. See the companion repository's synthetic-continuum and
182
+ external-language-calibration experiments for why that inference is
183
+ harder than it looks.
184
+ - **`indus_website`'s source repository maintainer is also associated
185
+ with an independently disputed decipherment claim for an unrelated
186
+ script.** The raw glyph sequences and site metadata here are ordinary
187
+ transcription data, checked directly (reading-direction fingerprint,
188
+ CISI-number cross-reference) rather than taken on trust; no
189
+ sign-to-meaning mapping from that other work is used anywhere here.
190
+
191
+ ## Additional Information
192
+
193
+ ### Licensing
194
+
195
+ `indus_website`: source repository license terms apply to the
196
+ underlying transcription data; this derived CSV is provided under MIT
197
+ for the conversion and structuring work. `cisi_*`: source repository is
198
+ MIT licensed; this derived CSV is provided under the same terms.
199
+
200
+ ### Citation
201
+
202
+ If you use this dataset, please cite the relevant original source:
203
+
204
+ ```
205
+ indus_website config: yajnadevam/indus-website (GitHub), underlying ICIT/Yajnadevam digitization
206
+ cisi_* configs: mayig/indus-valley-script-corpus (GitHub), digitization of Parpola's CISI corpus
207
+ ```
208
+
209
+ and, if relevant, the companion analysis toolkit:
210
+
211
+ ```
212
+ joyboseroy/indus_decipher (GitHub)
213
+ ```