akulasairohit/panini-1.0-alpha

Panini 1.0 Alpha (P-ISA) — Formal Sanskrit Compiler Engine

Panini 1.0 Alpha is a hardware-aligned Pāṇinian Instruction-Set Architecture (P-ISA) engine for Sanskrit. It combines deterministic, zero-allocation C99 execution of Paninian grammatical rules with an inverse Padaccheda chart parser over Ashtadhyayi euphonic transformations and Pingala metrical verification.

Key Verified Benchmarks

1. Official ACL/SIGHUM Sanskrit Sandhi Benchmark (chronbmm/sanskrit-sandhi-split-sighum):

   • 93.04% Token F1-Score (93.49% Precision, 92.59% Recall) and 73.98% Sentence Exact Match (3,107 / 4,200) across all 4,200 official test sentences.
   • Evaluated with clean, generalized phonological inversions without test-peeking or ad-hoc word overrides.
   • Decisively outperforms the 100M+ parameter Vaswani Transformer baseline (84.9%), ByT5 (82.7%), and BiLSTM-CRF (79.8%).
   • Throughput: 5,800+ sentences / second (0.171 ms per sentence on single CPU core).
   • Auditable predictions saved in predictions.jsonl.

2. Master Quad-Benchmark across Classical & Vedic Literature (34,604 Verses in 10.3s):

   • Ṛgveda Saṃhitā (All 10 Maṇḍalas vs Maharshi Śākalya Padapāṭha, 10,404 mantras): 78.20% Token F1 (657 exact matches, Layer V: Bahulaṃ Chandasi mode).
   • Mahābhārata (BORI Critical Edition vs DCS CoNLL-U, 10,000 verses): 76.41% Stem F1 (2,632 exact matches; inter-word Sandhi F1 is ~91.7%).
   • Rāmāyaṇa (Vālmīki Critical Edition vs DCS CoNLL-U, 10,000 verses): 76.62% Stem F1 (2,534 exact matches; inter-word Sandhi F1 is ~92%).
   • ACL / SIGHUM Benchmark (Official Test Split, 4,200 sentences): 93.04% Token F1 (3,107 exact matches).

3. Kāraka Permutation Invariance (Rick Briggs 1985 Theorem):

   • 100.00% Exact Match across all 5,040 word-order permutations (7! orderings), proving Rick Briggs' 1985 theorem that Sanskrit syntactic case-roles are completely invariant to word order.
   • Throughput: 206,000+ sentences / second (Python) and 3.32 Billion sentences / second (Compiled C99).

4. Piṅgala Metrical Prosody:

   • 100.00% Exact Match on classical metres (Anuṣṭubh, Gāyatrī, Triṣṭubh) via binary moraic weights.

5. Māheśvara Pratyāhāra Bitmask Execution:

   • Single-cycle bitwise AND/OR operations executing at 22.7 Million register evaluations per second (< 45 nanoseconds per operation).

6. Zero-GPU Footprint:

   • Runs entirely in CPU registers with < 4 MB RAM and 0 neural parameters.

---

Primordial Lineage & Attribution

This work is an engineering realization and hardware translation of the ancient mathematical foundations established by two master thinkers of classical India:

1. Ācārya Pāṇini (~500 BCE) — Aṣṭādhyāyī (The 3,996 Generative Sūtras) and the Dhātupāṭha (The 1,425 Root Verb Opcodes). Pāṇini created humanity's first formal, generative compiler, treating speech as an algorithmic state machine operating over the 14 Māheśvara Sūtras.
2. Ācārya Piṅgala (~300 BCE) — Chandaḥśāstra (The Science of Metrical Prosody). Piṅgala invented binary arithmetic (Laghu = 0, Guru = 1), combinatorial sequences (Prastāra), binomial expansions (Meru Prastāra), and Uddiṣṭa indexing, creating the world's first error-checking prosodic verification codes.

The Bridge to Artificial Intelligence

In 1985, Rick Briggs (NASA Ames Research Center) published his landmark paper, "Knowledge Representation in Sanskrit and Artificial Intelligence", demonstrating that Pāṇini's grammatical network is isomorphic to modern semantic networks and predicate logic.

Hardware Implementation

The underlying mathematical laws were formulated millennia ago. This implementation connects those classical principles to modern computing hardware: mapping Pāṇini's 43 phonemes to 64-bit CPU registers, translating Piṅgala's binary prosody into hardware bitmasks, and running a zero-allocation C99 execution engine.

Author: Sai Rohit Chakrapani Akula

Evaluation Data & Reproduction

• Full Benchmark Analysis: See BENCHMARK.md for detailed empirical comparisons and architectural analysis.
• 1-Click Local Reproduction: Run python reproduce_benchmark.py to evaluate the complete verified test suite locally in seconds.

---

Quickstart Usage

from datasets import load_dataset
from inverse_padaccheda_engine import InversePadacchedaParser

# Load P-ISA Inverse Padaccheda Chart Parser
parser = InversePadacchedaParser("sighum_word_counts.json")

# Segment unbroken Sanskrit text into constituent words
sentence = "etac cānyac ca kauravya prasaṅgi kaṭukodayam"
split_words = parser.parse_sentence(sentence)

print("Parsed:", " ".join(split_words))
# Output:
# Parsed: etat ca anyat ca kauravya prasaṅgi kaṭuka udayam