Implementation:Protectai Llm guard Output Language

Knowledge Sources	Protectai_Llm_guard
Domains	Language_Detection, NLP
Last Updated	2026-02-14 12:00 GMT

Overview

Language is an output scanner that verifies LLM responses are written in one of the specified allowed languages by delegating to the input-side InputLanguage scanner.

Description

The Language output scanner is a thin wrapper around the corresponding input scanner InputLanguage. It classifies the language of the LLM output and checks whether it belongs to the list of valid_languages. The scanner uses a language detection model to determine the language of the text and compares the result against the allowed languages. The threshold parameter sets the minimum confidence required for language classification. The match_type parameter controls whether the entire output is evaluated as a whole (FULL) or split into individual sentences (SENTENCE) for per-sentence language detection.

Usage

Use this scanner when your application must respond in specific languages only. This is important for localized applications, compliance with language regulations, and ensuring consistent user experience. For example, a French-language customer service bot should not produce responses in other languages.

Code Reference

Source Location

• Repository: Protectai_Llm_guard
• File: llm_guard/output_scanners/language.py
• Lines: 1-44

Signature

class Language(Scanner):
    def __init__(
        self,
        valid_languages: list[str],
        *,
        model: Model | None = None,
        threshold: float = 0.7,
        match_type: MatchType | str = MatchType.FULL,
        use_onnx: bool = False,
    ) -> None: ...

    def scan(self, prompt: str, output: str) -> tuple[str, bool, float]: ...

Import

from llm_guard.output_scanners import Language

I/O Contract

Inputs

Name	Type	Required	Description
prompt	str	Yes	The input prompt
output	str	Yes	The LLM output to check for language compliance

Constructor Parameters

Name	Type	Required	Default	Description
valid_languages	list[str]	Yes	N/A	List of allowed language codes (e.g., ["en", "fr", "de"])
model	None	No	None	Custom language detection model
threshold	float	No	0.7	Minimum confidence for language classification
match_type	str	No	MatchType.FULL	Matching strategy: FULL (entire text) or SENTENCE (per-sentence)
use_onnx	bool	No	False	Whether to use ONNX runtime for inference

Outputs

Name	Type	Description
sanitized_output	str	The output (potentially modified)
is_valid	bool	Whether the output is in one of the valid languages
risk_score	float	Risk score (-1.0 to 1.0)

Usage Examples

Basic Usage

from llm_guard.output_scanners import Language

scanner = Language(
    valid_languages=["en", "fr"],
    threshold=0.7,
)

prompt = "Translate this to French"
output = "Bonjour, comment allez-vous?"

sanitized_output, is_valid, risk_score = scanner.scan(prompt, output)

if is_valid:
    print("Output is in an allowed language")
else:
    print(f"Output language not allowed (risk: {risk_score})")

Related Pages

• Principle:Protectai_Llm_guard_Language_Detection