metadata
license: mit
language:
- sa
task_categories:
- token-classification
- other
tags:
- sanskrit
- samasa
- sandhi
- compounding
- vigraha
- vyakarana
Sanskrit Samas (Compound) Dataset — V1
Grammar-grounded training data for Sanskrit samas (compounds) covering all 7 samasa types, with laukik vigraha (natural paraphrase) and alaukik vigraha (Pāṇinian sUP analysis) on every row.
Companion to sanskrit-sandhi-boundaries-v2
(sentence-level external sandhi). Merge both for a full sandhi+samas boundary
training set.
Dataset
- Rows: 258,408
- Checksum:
e7c0a804c109 - Builder:
benchmarks/build_samas_data.py(deterministic, seed 42) - Format: JSONL (one compound per line)
Schema (per row)
| Field | Type | Description |
|---|---|---|
surface |
string | Natural merged SLP1 surface (internal sandhi applied at the seam) |
devanagari |
string | Devanagari rendering |
pieces |
list[str] | Surface partition at the compound seam (splitter contract; "".join == surface) |
bnd |
list[int] | Boundary positions (training format; last-char indices) |
gold_words |
list[str] | Laukik vigraha — natural paraphrase with explicit case forms |
gold_alaukik |
string | Alaukik vigraha — Pāṇinian analysis: stems + sUP tags (e.g. rAma + Nas + puruSa + su~) |
samasa_type |
string | One of the 7 types |
stem1, stem2 |
string | The constituent stems |
Samasa types covered
| Type | Meaning | Example (laukik) | n |
|---|---|---|---|
| tatpuruṣa | determinative (X of Y) | rājasya puruṣaḥ | 62,893 |
| karmadhāraya | appositional (X that is Y) | nīlam utpalam | 62,893 |
| dvandva | copulative (X and Y) | rāmaḥ ca kṛṣṇaḥ ca | 59,924 |
| bahuvrīhi | exocentric "having X" | cakraṃ pāṇau yasya saḥ | 60,172 |
| avyayībhāva | adverbial | śaktim anatikramya | 7,097 |
| amreḍita | reduplicative | dine dine | 5,091 |
| dvigu | numerical | trayo lokāḥ | 338 |
Construction (Pāṇinian)
- First member(s) lose their case endings → bare stem (prātipadika).
- Last member keeps its own case ending.
- Internal sandhi applies at the seam (same rules as external sandhi).
- The compound takes ONE case ending on the last member.
Alaukik vigraha is generated from vidyut prakriya derivation history —
the exact sUP tags (su, am, wA, Ne, Nasi, Nas, Ni × Eka/Dvi/Bahu) are
recovered from the Pāṇinian derivation steps.
Integrity
- 0 rows fail
"".join(pieces) == surface. - Every seam is rejoin-verified by the sandhi engine.
- Builder deterministic (seed 42): reproducible byte-for-byte.