Para

Para is a small, boring, and transparent toolkit for working with Burmese text. It detects whether text is encoded in Zawgyi or Unicode and converts Zawgyi to Unicode using a rule-based approach. Para never invents a new encoding and keeps its APIs explicit.

Goals

• Be Unicode-first and never invent a new encoding.
• Offer stable, explicit APIs without side effects or magic imports.
• Provide deterministic Zawgyi vs Unicode detection.
• Convert Zawgyi to Unicode with maintainable, rule-based logic (Parabaik-style), not machine learning.
• Stay batch-friendly for spreadsheets, CSVs, and plain text.
• Avoid heavy native dependencies.
• Be honest about limitations and edge cases.

Installation

pip install paraencoder

For Office document support (.docx, .xlsx, .odt):

pip install paraencoder[office]

Supported File Formats

Plain Text (built-in, no extra dependencies)

• Text files: .txt, .text, .log, .md, .rst, .asc
• Web/markup: .html, .htm, .xhtml, .xml, .json, .yaml, .yml, .csv, .tsv
• Documentation: .tex, .latex, .adoc, .org, .wiki, .mediawiki
• Config: .ini, .cfg, .conf, .properties, .env, .toml, .lock
• Source code: .py, .js, .ts, .java, .c, .cpp, .h, .cs, .php, .rb, .go, .rs, .sh, .bat, .ps1, .sql
• Subtitles: .srt, .vtt, .sub
• Other: .po, .pot, .texi, .man, .nfo, .readme, .eml, .mbox

Office Documents (requires paraencoder[office])

• Microsoft Word: .docx, .docm
• Microsoft Excel: .xlsx, .xlsm
• OpenDocument: .odt

Usage

from para.detect import is_zawgyi, detect_encoding
from para.convert import zg_to_unicode
from para.normalize import normalize_unicode

text = "\u1031\u1010\u1004\u103a"  # Zawgyi-encoded string
if is_zawgyi(text):
    cleaned = zg_to_unicode(text)
    cleaned = normalize_unicode(cleaned)

CLI

Detect encoding:

echo "\u1031\u1010\u1004\u103a" | para detect

Convert Zawgyi to Unicode:

echo "\u1031\u1010\u1004\u103a" | para convert > output.txt

Process a file in place (write to stdout by default):

para convert --input input.txt --output output.txt

Convert Office documents (requires paraencoder[office]):

para convert --input "Document.docx" --output "Document_Unicode.docx"
para convert --input "Spreadsheet.xlsx" --output "Spreadsheet_Unicode.xlsx"

Windows / PowerShell note

PowerShell's default encoding corrupts Myanmar text in pipes. Before piping Burmese text, set UTF-8 encoding:

$OutputEncoding = [System.Text.Encoding]::UTF8
[Console]::InputEncoding = [System.Text.Encoding]::UTF8
[Console]::OutputEncoding = [System.Text.Encoding]::UTF8
echo "ျမန္မာ" | para convert

Or use file-based input/output to avoid pipe issues:

para convert --input input.txt --output output.txt

API surface

• para.detect.is_zawgyi(text: str) -> bool

  • Input: text string.
  • Output: True only when the detector score prefers Zawgyi; otherwise False.
  • Guarantee: Never raises on empty/ASCII-only input; returns False for those.

• para.detect.detect_encoding(text: str) -> Literal["zawgyi", "unicode", "unknown"]

  • Input: text string.
  • Output: One of the three labels. Ties or insufficient evidence → "unknown" (no auto-conversion).
  • Guarantee: Deterministic, no network/ML, explicit tie handling.

• para.convert.zg_to_unicode(text: str, *, normalize: bool = True, force: bool = False) -> str

  • Input: text string.
  • Output: Converted Unicode string when detection prefers Zawgyi (or when force=True). Otherwise passes through (optionally normalized).
  • Guarantee: Ordered, test-backed regex rules; no Unicode→Zawgyi path; force=False avoids silent conversion on ambiguous text.

• para.normalize.normalize_unicode(text: str) -> str

  • Input: text string.
  • Output: NFC-normalized string with simple Myanmar ordering tweaks.
  • Guarantee: Idempotent on already-normalized Unicode Burmese.

• para.io.read_text(path: str, *, encoding: str = "utf-8") -> str

• para.io.write_text(path: str, data: str, *, encoding: str = "utf-8") -> None

• para.io.convert_file(...) -> str

  • Batch helpers for files; never guess encodings beyond the provided encoding argument.

Detection approach

Detection is deterministic and rule-based. Para scores the input with Zawgyi-specific patterns (e.g., U+1031 prefix order, U+105A, stacked medials) and Unicode-only patterns (e.g., valid ordering of medials, U+103A usage). The side with the higher score wins; ties produce "unknown". No machine learning, no network calls.

Conversion approach

Conversion uses an ordered list of regex replacements derived from Parabaik-style mappings. The rules are explicit, unit-tested, and live in para.rules. The converter does not attempt Unicode-to-Zawgyi; it only supports Zawgyi-to-Unicode because Unicode is the target canonical encoding.

Limitations

• Ambiguous short strings (e.g., ASCII-only) return "unknown" and pass through unchanged.
• Extremely malformed Zawgyi text may require manual cleanup.
• The converter focuses on common Zawgyi usage; rare legacy ligatures may need additional rules.

Non-goals

• Creating or endorsing any new Burmese encoding.
• Unicode-to-Zawgyi conversion.
• ML-based detection or probabilistic auto-conversion.
• Silent mutation of text when detection confidence is low; ties stay "unknown".

Contributing

Issues and pull requests are welcome. Keep changes readable and testable.

Packaging

• Build a wheel/sdist locally: python -m pip install build then python -m build.
• Publish to PyPI (once ready): python -m pip install twine then twine upload dist/*.
• The package metadata in pyproject.toml is PyPI-ready (MIT license, explicit packages, CLI entrypoint).

License

MIT