SasakScript: A Linguistically-Aware Bidirectional Transliteration Library and Benchmark for Sasak

---

1. Introduction

SasakScript is a Python library and standardized evaluation benchmark for bidirectional transliteration between the Latin alphabet and the traditional Sasak script (Aksara Sasak, encoded in the Balinese Unicode block U+1B00–U+1B7F).

To the best of our knowledge, existing computational studies on Sasak script have predominantly focused on rule-based Latin-to-Sasak transliteration and image-based character recognition, while a reproducible, linguistically-aware, bidirectional transliteration library accompanied by a dedicated benchmark remains insufficiently explored.

SasakScript integrates phonological mutation recovery, morphological boundary parsing, maximal onset syllabification, dialectal demonstrative variation, and an extensive multi-tier evaluation framework.

---

2. Motivation

Sasak (basa Sasak) is an Austronesian language spoken by approximately 3 million people on the island of Lombok, Indonesia. While traditionally inscribed onto palm-leaf manuscripts (lontar), the script faces severe endangerment:

• Low-Resource Scarcity: Computational linguistic tools and standardized digital corpora for Aksara Sasak have remained minimal.
• Orthographic Nuance: Aksara Sasak is an abugida where consonants possess an inherent vowel [a], vowels are written as diacritics (pangangge), consonant codas take specialized signs (cecek, surang, bisah), and clusters require conjunct consonants (gantungan) formed via the virama (adeg-adeg U+1B44).
• Phonological Complexity: Active verb derivation invokes homorganic nasal substitution, epenthesis before liquids/glides, and morpheme boundary sandhi that naive transliterators fail to capture.

---

3. Installation

From PyPI

pip install sasakscript

From Source (Editable Mode)

git clone https://github.com/kodetr/sasakscript.git
cd sasakscript
pip install -e .

With REST API Support

pip install -e ".[api]"

---

4. Quick Start

Python API

import sasakscript as sk

# 1. Latin to Aksara Sasak
res_sasak = sk.latin_to_sasak("tiang lalo jok pasar")
print(res_sasak.output)
# ᬢᬶᬬᬂ ᬮᬮᭀ ᬚᭀᬓ᭄ ᬧᬲᬃ

# 2. Aksara Sasak to Latin
res_latin = sk.sasak_to_latin("ᬩᬮᬾ")
print(res_latin.output)
# bale

# 3. Bidirectional Round-Trip Validation
rt = sk.round_trip("kanak bajang nulis surat")
print(f"Exact match: {rt.exact_match} | CER: {rt.cer:.4f}")

# 4. Morphological Analysis
morph = sk.analyze_morphology("pemangan")
print(morph)
# {'word': 'pemangan', 'root': 'mangan', 'prefix': ['pe-'], ...}

# 5. Pipeline Explainability Trace
explained = sk.transliterate("bale", explain=True)
print(explained.explanation["stages"])

# 6. Bilingual Sasak -> Indonesian Dictionary & Glossing
print(sk.translate_word("bale"))    # "rumah, tempat tinggal"
print(sk.translate_word("mangan"))  # "makan"

Command-Line Interface (CLI)

# Transliterate text
sasakscript translate "tiang lalo jok pasar"
# ᬢᬶᬬᬂ ᬮᬮᭀ ᬚᭀᬓ᭄ ᬧᬲᬃ

# Lookup Sasak word meaning in Indonesian
sasakscript lookup bale
# Kata:       bale
# Arti (ID):  rumah, tempat tinggal

# Lookup with JSON output
sasakscript lookup mangan --json

# Transliterate file
sasakscript translate input.txt --direction latin-to-sasak --output output.txt

# Morphological analysis
sasakscript analyze pemangan --json

# Run benchmark suite
sasakscript benchmark

# Inspect Unicode characters
sasakscript validate-unicode "ᬩᬮᬾ"

FastAPI REST Service

Launch the local API server:

uvicorn sasakscript.api.app:app --host 0.0.0.0 --port 8000 --reload

Interactive Swagger API documentation will be available at http://localhost:8000/docs.

---

5. Linguistic Pipeline

SasakScript processes text through eight deterministic, explainable stages:

INPUT
  ↓
[Stage 1: NORMALIZATION]       (Unicode NFC, apostrophe → q, diacritic standardization)
  ↓
[Stage 2: TOKENIZATION]        (Word boundaries, punctuation, numbers, whitespace)
  ↓
[Stage 3: MORPHOLOGY]          (Root recovery, prefix/suffix/infix/circumfix decomposition)
  ↓
[Stage 4: PHONOLOGY]           (Active nasal mutations, liquid epenthesis, hiatus resolution)
  ↓
[Stage 5: SYLLABIFICATION]     (Maximal Onset Principle: V, CV, CVC, CCV, CCVC, VC)
  ↓
[Stage 6: ORTHOGRAPHIC RULES]  (Prioritized YAML rules: coda signs, initial vowels, gantungan)
  ↓
[Stage 7: UNICODE SYNTHESIS]   (Canonical Balinese codepoint assembly, virama U+1B44)
  ↓
OUTPUT

---

6. Unicode Architecture

Sasak script is encoded in the Balinese Unicode Block (U+1B00–U+1B7F).

Disputed Codepoints Policy (U+1B45–U+1B4B)

In accordance with Unicode Technical Note #51 (Balinese and Sasak Orthography in Unicode, Perdana 2024), seven dedicated codepoints proposed in Unicode 5.0 for Sasak are marked UNVERIFIED by default because historical Sasak lontar manuscripts predominantly employ the standard consonant bases combined with the Rerekan diacritic (U+1B34). SasakScript strictly prevents the injection of invented rules without primary literature verification.

---

7. Dialect Module

SasakScript models the 5 canonical Sasak dialect clusters identified by Austronesian dialectology (Teeuw 1958; Mahsun 2006, 2007):

1. meno-mene: Central Sasak (Praya, Janapria, Kopang) — VERIFIED Standard Reference.
2. ngeno-ngene: East Sasak (Selong, Masbagik, Keruak).
3. ngeto-ngete: Northeast Sasak (Sembalun, Suela, Sambelia).
4. mriaq-mriku: South Sasak (Pujut, Kedaro, Sekotong).
5. kuto-kute: North Sasak (Bayan, Gangga, Tanjung).

---

8. Dataset: SasakScript-Bench

SasakScript-Bench is an annotated benchmark suite partitioned into six core linguistic categories:

• lexical: Base roots, kinships, and high-frequency terms.
• morphological: Productive affixation, nasalization, and reduplication.
• sentence: Complete declarative and interrogative sentences.
• dialect: Dialectal demonstrative pronouns.
• cultural: Classical lontar manuscripts and seloka (pantun).
• hard_case: Complex onset clusters, Rerekan loans, hiatus, and glottal codas.

Quality and Zero-Leakage Guarantee

• No Synthetic Gold: Synthetic data is strictly barred from the gold test set.
• Lexical-Disjoint Split: Guarantees zero root morpheme overlap between train and test sets to evaluate true generalization rather than vocabulary memorization.

---

9. Benchmark & Baselines

SasakScript provides implementations for 5 standard baselines (Section 16):

• Baseline 1: Naive 1-to-1 character mapping table.
• Baseline 2: Standard rule-based syllabic engine.
• Baseline 3: Morphology-aware rule engine.
• Baseline 4: Phonology-aware rule engine.
• Proposed: Full SasakScript unified pipeline.

Reproducibility

Execute the complete benchmark suite with one command:

python3 experiments/run_all.py

This automatically produces:

• results/baseline.csv
• results/ablation.csv
• results/dialect.csv
• results/roundtrip.csv
• results/error_analysis.csv

---

10. Limitations

1. Dialect Phonetic Generalization: While demonstratives are empirically verified, generative dialectal vocalic harmony across arbitrary vocabulary is marked UNVERIFIED pending field survey recordings.
2. Manuscript Paleographic Variance: The current engine targets the standard orthography established by the Department of Education & Culture (Depdikbud NTB) and primary lontar traditions; highly regional scribal idiosyncratic ligatures may require custom rule additions.

---

11. Citation

If you use SasakScript or SasakScript-Bench in your research, please cite:

@software{sasakscript2026,
  author    = {kodetr and SasakScript Working Group},
  title     = {SasakScript: A Linguistically-Aware Bidirectional Transliteration Library and Benchmark for Sasak Language and Script},
  year      = {2026},
  url       = {https://github.com/kodetr/sasakscript},
  note      = {Website: \url{https://kodetr.com}},
  version   = {0.8.0}
}

---

12. Authors & Maintainers

• Author & Maintainer: kodetr (https://kodetr.com)
• Research & Linguistic Contributors: SasakScript Working Group

---

13. License

This project is licensed under the MIT License. The dataset annotations are released under CC-BY-4.0.