RFC-0002: Sanskrit Processing Engine (vyasa-sanskrit)

Status

Accepted / In Implementation (Phases 1 & 2 Complete)

Summary

This RFC specifies the architecture of vyasa-sanskrit, a suite of modular, zero-dependency Rust crates inside the vyutils workspace (vyasa-phonetics, vyasa-lipi, vyasa-patha). The suite supports:

Pre-Pāṇinian & Pāṇinian Phonetics: Śaunaka’s Ṛgveda-Prātiśākhya (Śaiśirīya sequence), Pāṇini’s 14 Māheśvara Sūtras with dynamic Pratyāhāra computation, and Śikṣā articulatory classifications.
Universal Vedic Multi-Script Transliteration: Lossless round-trip conversion across Indic abugidas (Devanagari, Telugu, Kannada, Grantha, Malayalam, Bengali), Roman transliteration schemes (Strict ISO 15919, Standard IAST), and ASCII formats (SLP1, Harvard-Kyoto, WX) with suprasegmental pitch accent (svara) preservation.
Recitation Mode Generation: Mathematical generation of Prakṛti (Saṁhitā, Pada, Krama) and Aṣṭa-Vikṛti (Jaṭā, Ghana, etc.) pāṭhas with strict Pragṛhya detection and Parigraha (iti) clauses.

Motivation & Gap Analysis

Existing computational Sanskrit tools (e.g. indic_transliteration, sanscript.js, Heritage du Sanskrit) are almost exclusively designed around unaccented classical Sanskrit. They fail when applied to Vedic literature due to four critical gaps:

Gap 1: Vedic Accent Corruption

In Vedic recitation, every vowel possesses a tone: Udātta (high), Anudātta (low, \u0952), or Svarita (falling, \u0951). Existing tools treat combining accent marks as non-spacing noise or strip them entirely. vyasa-sanskrit treats pitch accents as first-class phonetic properties of the syllable (Akṣara).

Gap 2: Conflation of ISO 15919 with IAST

Standard IAST does not distinguish short and long e and o, nor does it accurately represent vocalic liquids or the Vedic retroflex lateral flap (ळ). Strict ISO 15919 uses under-rings (r̥, r̥̄, l̥, l̥̄) and explicit macrons (ē, ō), whereas IAST uses under-dots (ṛ, ṝ, ḷ, ḹ) and bare vowels (e, o). vyasa-lipi enforces this distinction strictly.

Gap 3: Absence of Prātiśākhya Sequences

Classical toolkits assume Pāṇini’s Śiva Sūtras (a-i-u-ṇ…). The Ṛgveda-Prātiśākhya formulates rules around the Śaiśirīya order (a, ṛ, i, u, e, o, ai, au) and defines Nāmin vowels (all vowels except a and ā) which trigger retroflexion (Nati). vyasa-phonetics implements these pre-Pāṇinian models directly.

Gap 4: Naive Recitation Generation

Existing pāṭha scripts perform naive string concatenation without applying forward/reverse sandhi, without recomputing Vedic accents, and without inserting canonical “iti” clauses (Parigraha / Sthita-Upasthita) for Pragṛhya vowels.

Architecture of Workspace Crates

crates/
├── vyasa-phonetics/     # Sound representations, Prātiśākhya & Pāṇinian models
├── vyasa-lipi/          # Universal multi-script transliteration engine
└── vyasa-patha/         # Recitation generator (Prakṛti & Aṣṭa-Vikṛti pāṭhas)

Invariants

Zero Runtime Dependencies: All crates compile with no_std + alloc support to guarantee immediate compilation and WebAssembly export to vyasa-viewer (vyasav).
Lossless Round-Trip Guarantee: Transliterating text A → B → A produces byte-identical strings for all supported scripts.
Akṣara Intermediate Representation (IR): Transliteration is performed via a structured phonemic AST, never ad-hoc regex cascades.

Roadmap

Phase 1 (Complete): vyasa-phonetics (Śaiśirīya sequence, Pratyāhāras, Śikṣā) + vyasa-lipi (Devanagari, Telugu, Kannada, ISO 15919, IAST).
Phase 2 (Complete): Expanded Indic scripts (Grantha, Malayalam, Bengali) + ASCII schemes (SLP1, Harvard-Kyoto, WX).
Phase 3: vyasa-patha for Prakṛti pāṭhas (Saṁhitā, Pada, Krama) with Pragṛhya detection and Parigraha (iti) clauses.
Phase 4: Aṣṭa-Vikṛti generators (Jaṭā, Ghana, etc.) with forward/reverse Vedic sandhi.
Phase 5: WASM build and Svelte UI components in vyasa-viewer.