utils/ingredient_split.py

"""Split compound ingredient text into resolved individual names.

Uses the ingredient-parser-nlp CRF model which handles coordination
resolution: shared noun propagation, multi-way OR/AND lists, and
comma-separated alternatives.

    >>> from utils.ingredient_split import split_ingredient_names
    >>> split_ingredient_names("cherry or grape tomatoes")
    ['cherry tomatoes', 'grape tomatoes']
    >>> split_ingredient_names("chicken or vegetable broth")
    ['chicken broth', 'vegetable broth']
"""

from __future__ import annotations

import re

from ingredient_parser import parse_ingredient

REPEATED_PREFIX_RE = re.compile(r"\b(\S+(?:-\S+)*)\s+\1\b")


def deduplicate_prefix(name: str) -> str:
    """Fix CRF bug where a shared prefix token is doubled.

    "day-old day-old sandwich bread" -> "day-old sandwich bread"
    """
    return REPEATED_PREFIX_RE.sub(r"\1", name)


def split_ingredient_names(text: str) -> list[str]:
    """Split a compound ingredient string into resolved individual names.

    Returns a list of 1+ names. A single-element list means the input
    was not compound (or the model chose not to split it).
    """
    parsed = parse_ingredient(text)
    names = [deduplicate_prefix(n.text) for n in parsed.name]
    if not names:
        return [text]
    return names