कोश
Gasuns Sanskrit Dictionary · the Sanskrit data-hub

Sanskrit NLP Data & Tools Directory

A single curated entry point for Sanskrit computational linguistics: our openly-licensed derived datasets (downloadable), plus the external stacks and APIs the project builds on — what each does, how to call it, and its license.

19public datasets
7restricted (on request)
8external tools & stacks
↓ Our datasets ↓ External tools Repository Features Index

Our datasets

Derived from the Cologne Digital Sanskrit Dictionaries and the DCS corpus. Public assets are CC BY-SA 4.0, downloadable from the kosha data releases. Rendered from datasets.json — the single machine-readable source.

MW verbal-root inventory

public

SLP1 root + explicit verb_type (750 genuineroot + 1363 root)

2,113 rows · 75.3 KB · TSV

WhitneyRootscsl-atlasRussianTranslation verb worklist
Download ↓

Headword→root derivation table (Pāṇinian, 10 dicts)

public

SLP1 headword → root

9,377 rows · 1.5 MB · TSV

WhitneyRootscsl-atlas
Download ↓

DCS lemma ↔ CDSL headword crosswalk

public

DCS lemma id ↔ CDSL headword (SLP1), 81.4% linked

15,902 rows · 618.5 KB · TSV

simple-searchcsl-atlasVisualDCS
Download ↓

Cross-dictionary union headword index

public

SLP1 headword + per-dict provenance flags

323,425 rows · 11.8 MB · TSV

kosha crosswalkSanskritSpellCheck (union=N tags)PWG→EN pilot (94,753 subset)
Download ↓

MW ↔ Heritage (INRIA) entry-level crosswalk

public

MW entry id → Heritage anchor (25,140 covered, 97.6% anchor-resolved)

185,803 rows · 3.1 MB · TSV

kosha ingest (queued)csl-atlas witness column (queued)
Download ↓

DCS lemma frequency sidecar (whole-corpus + per-period)

public

SLP1 lemma; whole-corpus counts + per-period vectors + core-vocab coverage

83,277 rows · 4.7 MB · TSV

kosha evidence layerNagari teaching order
Download ↓

Zaliznyak-style compact grammar-token index over all PWG

public

SLP1 headword → paradigm token (335 tokens)

98,639 rows · 5.8 MB · TSV

pwg_ru nominal layerkosha P4 grammar tokensgramdict mapping
Download ↓

Böhtlingk Indische Sprüche subhāṣitas

public

per-saying records, public domain

7,537 rows · 6.9 MB · JSONL

PWG <ls> citation links
In source repo ↗

DCS akshara/varṇa/ligature frequency tables

public

akshara 7,347 · varṇa 48 · conjunct 999; corpus-wide + per DCS time-slot

8,394 rows · 4.0 KB · CSV

Nagari script-teaching order
In source repo ↗

Consonant series (varga) by DCS period + Cramér's V

public

5 vargas × 5 DCS time-slots; column shares + counts + Δ(I→V)

5 rows · 4.0 KB · CSV

GasunsDhatu 2026 §2.6 (Table 5 replacement, L7 fix)
In source repo ↗

DCS syntagmatic table — all corpus lemmas (Приложение 7)

public

lemma; total occurrences; co-occurrence count; ranked collocate list per lemma

82,800 rows · 25.7 MB · CSV

In source repo ↗

DCS parallel-passage alignments, full-size export (PARA/Polnorazmernye)

public

one file per source-passage id (`<id>_<n>--<m>.csv`), 245 files; each row = a candidate parallel passage with a GOOD/PARTLY/… match verdict

506,787 rows · 56.2 MB · CSV

In source repo ↗

DCS verb roots grouped by class (with attested forms)

public

root,count — one file per verb class (1.csv…10.csv), roots that have attested corpus forms

463 rows · 4.4 KB · CSV

In source repo ↗

DCS verb-form frequency by class, with prefixed forms

public

root;count — one file per verb class (1.csv…10.csv), includes prefixed verb forms

8,454 rows · 110.0 KB · CSV

In source repo ↗

DCS syntagmatic tables of frequent lexical cores, by historical period (Приложение 6)

public

lemma; frequency; collocates — one file per historical period (1.csv…7.csv)

19,076 rows · 22.9 MB · CSV

In source repo ↗

DCS parallel-passage alignments, 'stop-word' variant export (PARA/Stopovye)

public

one file per source-passage id, same naming scheme as dcs-parallel-passages-full; includes a split 7z archive member

1.1 GB · CSV

In source repo ↗

DCS verb-form frequency in corpus (preliminary)

public

verb form; frequency count, corpus-wide

106 rows · 8.0 KB · CSV

In source repo ↗

DCS compound (samāsa) dictionary derivation

public

CompDic.csv (compound headword list, 37,333), cmps.csv (per-compound member breakdown, 401,478), names.csv (168,880 attested compounds w/ frequency vector, already noted in DATA_LAYERS_CENSUS.md), parts.csv (2,667), verbx.csv (3,400, verbal compounds); cmp400000.csv is present but empty (0 bytes)

613,758 rows · 104.8 MB · CSV

In source repo ↗

Sanskrit-stem co-occurrence table (full)

public

stem-pair co-occurrence rows, ID range 1–222342

353,352 rows · 34.6 MB · CSV

In source repo ↗

Restricted & in preparation

Rights-encumbered or unbackuped local-only assets — listed for discovery; available on request as rights clear.

Sa→Ru word-aligned corpus lexicon

restricted

SLP1 surface key (~190k keys), per verse-pair alignment

1,091,528 rows · 276.6 MB · JSONL

SanskritRussian glossarypwg_ru TM
available on request / in preparation

Ranked Sa→Ru glossary (surface/lemma/root layers)

restricted

SLP1; surface 190,838 · lemma 40,370 · root 2,021; 87% token coverage

233,229 rows · 147.0 MB · JSONL

live search sitepwg_ru gates
available on request / in preparation

kosha unified lookup database

restricted

9 tables: 444,773 entries · 323,425 lemmas · 692,403 senses · 1,378,401 forms · 6,916,522 inflections

444,773 rows · 1.5 GB · SQLITE

kosha APIP2 static cache
available on request / in preparation

DCS 2026 full corpus database

restricted

5,688,416 tokens · 754,726 sentences · 180,176 lemmas · 270 texts

5,688,416 rows · 878.2 MB · SQLITE

VisualDCS dashboardskosha frequencyakshara stats
available on request / in preparation

SamudraManthanam Sa↔Ru verse-parallel corpus

restricted

574,939 corpus lines · 148 sources, verse-aligned

574,939 rows · 590.7 MB · SQLITE

pwg_ru gate dictsmw_ru validationcorpus-lexicon upstream
available on request / in preparation

Sanskrit Heritage (INRIA) local mirror

restricted

Heritage entry anchors; VH↔SLP1 bridge validated

MIXED (DICO HYPERTEXT, TSV, OCAML BANKS)

mw-heritage-crosswalkHeadwordLists coverage studies
available on request / in preparation

Heritage (INRIA) form-level crosswalk extras (gloss, disagreements, forms-only)

restricted

heritage_dico_gloss (mw_key1 → Heritage anchor + FR gloss), heritage_forms_oracle_disagreements (form → Heritage/kosha lemma disagreement class), heritage_only_forms (form → lemma, Heritage-only coverage)

1,037,239 rows · 34.5 MB · TSV

available on request / in preparation

External tools & stacks

Call these, don't clone them. Rendered from external_tools.json.

vidyut

external

Ambuda (ambuda-org)

Rust Sanskrit toolkit: the kosha FST lexicon, a prakriyā (derivation) generator, the cheda segmenter, and a chandas meter identifier. The paradigm/lemma engine behind our RU translation kits and the Zaliznyak grammar index.

Our relation: We call vidyut for declension/conjugation paradigms and PPP validation rather than hand-rolling paradigm generation; it produces the paradigm tokens in zaliznyak-grammar-index.

Call / docs ↗ MIT

Ambuda

external

Ambuda (ambuda-org)

Open-source Sanskrit reading platform and library (proofread texts, integrated dictionary + parser lookups). The reading front-end sibling of vidyut, built by the same community.

Our relation: Reference open-source reader; shares the vidyut engine we already consume. Not a data dependency of kosha.

Call / docs ↗ AGPL-3.0 (platform); texts vary

Sanskrit Heritage (INRIA)

external

Gérard Huet, INRIA

The Heritage dictionary + morphology generator + segmenter: DICO hypertext dictionary, MW-aligned entry pages, frequency TSVs, and OCaml morphology banks. A morphology oracle for form → lemma resolution.

Our relation: We align MW ↔ Heritage entries (see the mw-heritage-crosswalk dataset). Pull from the GitHub mirror — sanskrit.inria.fr is Anubis bot-walled. Mirror data itself is LGPLLR-pending in our restricted tier; the crosswalk is ours.

Call / docs ↗ LGPLLR

Samsaadhanii / SCL

external

Amba Kulkarni, University of Hyderabad

The Sanskrit Computational Linguistics stack: morphological analyzer, sandhi splitter/segmenter, compound processor, Pāṇinian dependency parser, Amarakośa semantic net, and Dhātupāṭha resources. Already consumes the Cologne dictionaries.

Our relation: We call the live JSON endpoints and cross-validate against them; we do not clone the GPL source. It reuses our dictionary data downstream.

Call / docs ↗ GPL

DharmaMitra

external

Sebastian Nehrdich et al., UC Berkeley

AI-driven Sanskrit stack: machine translation, deep research with references, parallel-passage exploration, segmentation, and OCR. A GPU morphology supplier and translation service.

Our relation: We reuse their MT error taxonomy (wired into the pwg_ru/mw_ru QA judges) and are a prospective Kosh-API consumer/supplier. GPU-morphology supplier to csl-atlas.

Call / docs ↗ Free web service; models vary — see site

Digital Corpus of Sanskrit (DCS)

external

Oliver Hellwig

A sandhi-split, morphologically and lexically analysed diachronic corpus of Sanskrit (~5.7M tokens, 270 texts) with per-token lemma, POS and morphology in CoNLL-U. The reference corpus for lemma frequency and attestation.

Our relation: Source of the dcs-cdsl-xref crosswalk and the kosha-lemma-frequency sidecar. We ingest the canonical CoNLL-U once — never re-parse.

Call / docs ↗ CC BY-SA 4.0

VedaWeb

external

University of Cologne / UZH (C-SALT)

Accented Rig-Veda with per-word morphology (Casaretto word-split, Lubotsky padapāṭha), aligned translations, and a query API. The validation set for Vedic accent.

Our relation: The accented-Vedic validation set that unblocks the Vedic-accent axis of the Zaliznyak index — bulk-export once, cross-validate.

Call / docs ↗ CC BY 4.0

Cologne Digital Sanskrit Dictionaries (CDSL)

external

University of Cologne (Cologne / sanskrit-lexicon)

The canonical digitised Sanskrit dictionaries (MW, PWG, AP90, and 40+ more) as XML/text with scan-anchored entries and a lookup API. The lexical bedrock the whole project is built on.

Our relation: kosha collapses every CDSL dictionary's entry for a headword onto one scan-anchored page; union-headwords, mw-roots and mw-etymology all derive from CDSL source.

Call / docs ↗ CC BY-SA 4.0