Implementation:Explodinggradients Ragas QuotedSpans Module

Field	Value
source	Explodinggradients_Ragas (GitHub)
domains	Metrics, Citation
last_updated	2026-02-10 00:00 GMT

Overview

The quoted_spans module provides a lightweight citation alignment metric that measures what fraction of quoted spans in model-generated answers can be found verbatim in the provided source passages.

Description

This module extracts quoted text spans from answers using a regular expression that supports both straight and curly quote characters. Each extracted span is checked against concatenated source passages using substring matching after optional whitespace normalization and case folding. The primary function quoted_spans_alignment iterates over parallel lists of answers and sources, producing a score in the range [0, 1] together with raw matched and total counts. The helper _extract_quoted_spans filters spans by a configurable minimum word length to avoid spurious matches.

Usage

Call quoted_spans_alignment with matched lists of answers and their corresponding source passages. The function returns a dictionary containing the alignment score, the number of matched spans, and the total number of spans considered.

Code Reference

Item	Detail
Source Location	src/ragas/metrics/quoted_spans.py L27-124
Signature (quoted_spans_alignment)	def quoted_spans_alignment(answers: Sequence[str], sources: Sequence[Sequence[str]], *, casefold: bool = True, min_len: int = 3) -> Dict[str, float]
Signature (_extract_quoted_spans)	def _extract_quoted_spans(answer: str, *, min_len: int = 3) -> Sequence[str]
Import	from ragas.metrics.quoted_spans import quoted_spans_alignment

I/O Contract

Inputs

Parameter	Type	Description
answers	Sequence[str]	List of model-generated answers (length N)
sources	Sequence[Sequence[str]]	List of source passage lists (length N), one per answer
casefold	bool	Whether to normalize text by lower-casing before matching (default True)
min_len	int	Minimum number of words required for a quoted span to be considered (default 3)

Outputs

Key	Type	Description
citation_alignment_quoted_spans	float	Fraction of quoted spans found in sources (0.0-1.0)
matched	float	Number of spans that matched a source passage
total	float	Total number of quoted spans considered

Usage Examples

from ragas.metrics.quoted_spans import quoted_spans_alignment

answers = [
    'The report states "revenue increased by 15 percent" and "costs remained stable".',
]
sources = [
    ["The annual report notes that revenue increased by 15 percent over the prior year."]
]

result = quoted_spans_alignment(answers, sources)
print(result["citation_alignment_quoted_spans"])  # 0.5 (1 of 2 spans matched)
print(result["matched"])   # 1.0
print(result["total"])     # 2.0

from ragas.metrics.quoted_spans import _extract_quoted_spans

spans = _extract_quoted_spans('She said "hello world example" and "ok"', min_len=3)
print(spans)  # ['hello world example']

Related Pages

• MetricResult_Class - Result wrapper used by metrics
• MetricValidators_Module - Validation framework for metric outputs
• Validation_Module - Dataset-level validation utilities