CodeIsAbstract/sanskrit-samas-v1

1---2license: mit3language:4  - sa5task_categories:6  - token-classification7  - other8tags:9  - sanskrit10  - samasa11  - sandhi12  - compounding13  - vigraha14  - vyakarana15---16 17# Sanskrit Samas (Compound) Dataset — V118 19Grammar-grounded training data for **Sanskrit samas (compounds)** covering all207 samasa types, with **laukik vigraha** (natural paraphrase) and **alaukik21vigraha** (Pāṇinian sUP analysis) on every row.22 23Companion to [`sanskrit-sandhi-boundaries-v2`](https://huggingface.co/datasets/CodeIsAbstract/sanskrit-sandhi-boundaries-v2)24(sentence-level external sandhi). Merge both for a full sandhi+samas boundary25training set.26 27## Dataset28 29- **Rows**: 258,40830- **Checksum**: `e7c0a804c109`31- **Builder**: `benchmarks/build_samas_data.py` (deterministic, seed 42)32- **Format**: JSONL (one compound per line)33 34## Schema (per row)35 36| Field | Type | Description |37|-------|------|-------------|38| `surface` | string | Natural merged SLP1 surface (internal sandhi applied at the seam) |39| `devanagari` | string | Devanagari rendering |40| `pieces` | list[str] | Surface partition at the compound seam (splitter contract; `"".join == surface`) |41| `bnd` | list[int] | Boundary positions (training format; last-char indices) |42| `gold_words` | list[str] | **Laukik vigraha** — natural paraphrase with explicit case forms |43| `gold_alaukik` | string | **Alaukik vigraha** — Pāṇinian analysis: stems + sUP tags (e.g. `rAma + Nas + puruSa + su~`) |44| `samasa_type` | string | One of the 7 types |45| `stem1`, `stem2` | string | The constituent stems |46 47## Samasa types covered48 49| Type | Meaning | Example (laukik) | n |50|------|---------|------------------|---|51| tatpuruṣa | determinative (X of Y) | rājasya puruṣaḥ | 62,893 |52| karmadhāraya | appositional (X that is Y) | nīlam utpalam | 62,893 |53| dvandva | copulative (X and Y) | rāmaḥ ca kṛṣṇaḥ ca | 59,924 |54| bahuvrīhi | exocentric "having X" | cakraṃ pāṇau yasya saḥ | 60,172 |55| avyayībhāva | adverbial | śaktim anatikramya | 7,097 |56| amreḍita | reduplicative | dine dine | 5,091 |57| dvigu | numerical | trayo lokāḥ | 338 |58 59## Construction (Pāṇinian)60 611. First member(s) lose their case endings → bare stem (prātipadika).622. Last member keeps its own case ending.633. Internal sandhi applies at the seam (same rules as external sandhi).644. The compound takes ONE case ending on the last member.65 66Alaukik vigraha is generated from **vidyut prakriya derivation history** —67the exact sUP tags (su~, am, wA, Ne, Nasi~, Nas, Ni × Eka/Dvi/Bahu) are68recovered from the Pāṇinian derivation steps.69 70## Integrity71 72- 0 rows fail `"".join(pieces) == surface`.73- Every seam is rejoin-verified by the sandhi engine.74- Builder deterministic (seed 42): reproducible byte-for-byte.75