Implementation:Evidentlyai Evidently Legacy Words Descriptor

Knowledge Sources	Evidentlyai_Evidently
Domains	ML Monitoring, NLP, Text Analysis
Last Updated	2026-02-14 12:00 GMT

Overview

Provides descriptor classes for checking word inclusion, exclusion, matching, and non-matching between text columns using configurable word lists and lemmatization.

Description

This module defines four descriptor classes that wrap the corresponding feature classes from evidently.legacy.features.words_feature:

• ExcludesWords - Checks whether a text column excludes specific words from a provided word list. Supports "all" (default) or "any" matching mode and optional lemmatization.

• IncludesWords - Checks whether a text column includes specific words from a provided word list. Defaults to "any" matching mode with lemmatization enabled.

• WordMatch - Compares words in two columns and evaluates whether there is a match. Takes a with_column parameter to specify the second column for comparison.

• WordNoMatch - The inverse of WordMatch; evaluates whether words in two columns do not match.

All four classes extend FeatureDescriptor and implement the feature() method, which accepts a column_name parameter and returns the corresponding GeneratedFeature from the words_feature module. Each descriptor supports configurable mode ("any" or "all"), lemmatize toggle, and an optional display_name inherited from the base class.

Usage

Use these descriptors when building Evidently reports or test suites to check text quality constraints such as whether generated text avoids forbidden words (ExcludesWords), contains required words (IncludesWords), or matches/diverges from a reference column (WordMatch/WordNoMatch). They are particularly useful in LLM monitoring to validate output against expected vocabulary.

Code Reference

Source Location

• Repository: Evidentlyai_Evidently
• File: src/evidently/legacy/descriptors/words_descriptor.py

Signature

class ExcludesWords(FeatureDescriptor):
    words_list: List[str]
    mode: str = "all"
    lemmatize: bool = True
    def feature(self, column_name: str) -> GeneratedFeature: ...

class IncludesWords(FeatureDescriptor):
    words_list: List[str]
    mode: str = "any"
    lemmatize: bool = True
    def feature(self, column_name: str) -> GeneratedFeature: ...

class WordMatch(FeatureDescriptor):
    with_column: str
    mode: str = "any"
    lemmatize: bool = True
    def feature(self, column_name: str) -> GeneratedFeature: ...

class WordNoMatch(FeatureDescriptor):
    with_column: str
    mode: str = "any"
    lemmatize: bool = True
    def feature(self, column_name: str) -> GeneratedFeature: ...

Import

from evidently.legacy.descriptors.words_descriptor import ExcludesWords
from evidently.legacy.descriptors.words_descriptor import IncludesWords
from evidently.legacy.descriptors.words_descriptor import WordMatch
from evidently.legacy.descriptors.words_descriptor import WordNoMatch

I/O Contract

Inputs

ExcludesWords / IncludesWords:

Name	Type	Required	Description
words_list	List[str]	Yes	The list of words to check for inclusion or exclusion.
mode	str	No	Matching mode: "all" (every word must match) or "any" (at least one word must match). Default varies by class.
lemmatize	bool	No	Whether to apply lemmatization before matching. Defaults to True.
display_name	Optional[str]	No	Custom display name for the generated feature (inherited from BaseDescriptor).
column_name	str	Yes (at call time)	The text column to evaluate, passed to the feature() method.

WordMatch / WordNoMatch:

Name	Type	Required	Description
with_column	str	Yes	Name of the second column to compare words against.
mode	str	No	Matching mode: "any" or "all". Defaults to "any".
lemmatize	bool	No	Whether to apply lemmatization before matching. Defaults to True.
display_name	Optional[str]	No	Custom display name for the generated feature.
column_name	str	Yes (at call time)	The primary text column, passed to the feature() method.

Outputs

Name	Type	Description
feature() return	GeneratedFeature	A words_feature instance (ExcludesWords, IncludesWords, WordMatch, or WordNoMatch) ready for feature generation.

Usage Examples

from evidently.legacy.descriptors.words_descriptor import (
    ExcludesWords, IncludesWords, WordMatch, WordNoMatch
)

# Check that responses exclude forbidden words
excludes = ExcludesWords(
    words_list=["error", "fail", "crash"],
    mode="all",
    lemmatize=True,
    display_name="No Error Words"
)
feature = excludes.feature("response")

# Check that responses include at least one required keyword
includes = IncludesWords(
    words_list=["thank", "welcome", "please"],
    mode="any",
)
feature = includes.feature("response")

# Match words between prediction and reference columns
match = WordMatch(with_column="reference_text", mode="any")
feature = match.feature("generated_text")

# Check for non-matching words
no_match = WordNoMatch(with_column="reference_text", mode="any")
feature = no_match.feature("generated_text")

Related Pages

• Environment:Evidentlyai_Evidently_Python_Core_Environment
• Evidentlyai_Evidently_Legacy_Generated_Features - Base classes (FeatureDescriptor, GeneratedFeature) that these descriptors extend.