xpfcorpus 0.1.0

XPF Corpus grapheme-to-phoneme transcriber

xpfcorpus

A Python package for grapheme-to-phoneme transcription based on the XPF Corpus.

Documentation: xpfcorpus.readthedocs.io

XPF Corpus Resources:

• XPF Website
• XPF Manual (PDF)
• XPF GitHub

Installation

pip install xpfcorpus

Python API

from xpfcorpus import Transcriber, available_languages

# Basic usage - languages with a default script
es = Transcriber("es")
es.transcribe("ejemplo")  # ['e', 'x', 'e', 'm', 'p', 'l', 'o']

# Languages with multiple scripts require explicit script choice
tt_latin = Transcriber("tt", "latin")
tt_cyrillic = Transcriber("tt", "cyrillic")

# BCP-47 style language codes with script/region
es_es = Transcriber("es-ES")  # Region code stripped, uses default script
yi = Transcriber("yi-Latn")   # Script extracted from code
tt = Transcriber("tt-cyrillic")  # Script name in code
zh = Transcriber("zh-Hans-CN")  # Script extracted, region stripped

# Explicit script parameter overrides code
yi = Transcriber("yi-Latn", script="hebrew")  # Uses hebrew, not latin

# Skip verification on load
es = Transcriber("es", verify=False)

# List available languages
available_languages()
# {"es": {"scripts": ["latin"], "default": "latin"},
#  "tt": {"scripts": ["latin", "cyrillic"], "default": None}, ...}

Language Code Format

The package supports BCP-47 style language codes:

• Simple codes: "es", "tt", "yi"
• With region (variants): "es-ES", "en-US" → treated as language variants
• With script (extracted): "yi-Latn", "tt-Cyrl" → extracts script
• Script names (extracted): "tt-cyrillic", "yi-latin" → extracts script
• Complex codes: "zh-Hans-CN" → extracts "hans" script, preserves "CN" region

When both a script in the code and an explicit script parameter are provided, the explicit parameter takes precedence.

Language Variants

Region codes are treated as language variants. The package will:

1. Try to load variant-specific data (e.g., es-ES.json) if available
2. Fall back to base language (e.g., es.json) with a warning if variant not found
3. Store the variant information in the variant property (only if variant file exists)

# Base language
es = Transcriber("es")
print(es.variant)  # None

# Variant request (falls back to base with warning if es-ES.json doesn't exist)
es_es = Transcriber("es-ES")
print(es_es.variant)  # None (because es-ES.json doesn't exist, fell back to es.json)

# If es-ES.json existed, then:
# es_es.variant would be "ES"

Behavior:

• If variant file exists: variant property returns the region code (e.g., "ES")
• If variant file doesn't exist: falls back to base language, variant is None

To create a variant, add a JSON file like es-ES.json to the xpfcorpus/data/languages/ directory with variant-specific rules.

Command-Line Interface

# Transcribe words
xpfcorpus transcribe es ejemplo hola mundo

# Transcribe from file (extracts first word from each line)
xpfcorpus transcribe es -f words.txt

# Transcribe from stdin
echo -e "mundo\nbueno" | xpfcorpus transcribe es
cat words.txt | xpfcorpus transcribe es -f -

# Combine command-line words and file
xpfcorpus transcribe es ejemplo hola -f more_words.txt

# List available languages
xpfcorpus list
xpfcorpus list --json

# Export language rules as YAML
xpfcorpus export es
xpfcorpus export es -o spanish.yaml

# Verify language rules
xpfcorpus verify es -v
xpfcorpus verify --all

Supported Languages

The package includes rules for 201 languages with 203 language/script combinations. Some languages have multiple scripts:

• iu (Inuktitut): latin, syllabics
• tt (Tatar): latin, cyrillic

Use xpfcorpus list or available_languages() for the full list.

Citation

If you use this package in your research, please cite the XPF Corpus:

@misc{xpf_corpus,
  title={The Cross-linguistic Phonological Frequencies (XPF) Corpus},
  author={Cohen Priva, Uriel and Gleason, Emily},
  year={2022},
  url={https://cohenpr-xpf.github.io/XPF/}
}

License

MIT