Implementation:Protectai Llm guard Output LanguageSame

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

Overview

LanguageSame is an output scanner that verifies the LLM response is written in the same language as the input prompt.

Description

The LanguageSame output scanner is not a thin wrapper; it has its own standalone implementation. It detects the language of both the prompt and the output independently, then checks whether there is an overlap between the detected languages. The scanner uses the same language detection model as the Language scanner (default: papluca/xlm-roberta-base-language-detection). It runs the classification pipeline on both the prompt and the output in a single batch call ([prompt, output]), then intersects the sets of detected languages for each. The threshold parameter controls the minimum confidence required for a language to be included in the detected set. If the intersection of detected languages between prompt and output is non-empty, the output is considered valid; otherwise, it is flagged as a language mismatch.

Usage

Use this scanner when you want to ensure the LLM responds in the same language the user used for their query. This is important for multilingual applications where the response language should match the input language automatically, without requiring explicit language configuration. It prevents scenarios where a user writes in Spanish but receives a response in English.

Code Reference

Source Location

• Repository: Protectai_Llm_guard
• File: llm_guard/output_scanners/language_same.py
• Lines: 1-87

Signature

class LanguageSame(Scanner):
    def __init__(
        self,
        *,
        model: Model | None = None,
        threshold: float = 0.1,
        use_onnx: bool = False,
    ) -> None: ...

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

Import

from llm_guard.output_scanners import LanguageSame

I/O Contract

Inputs

Name	Type	Required	Description
prompt	str	Yes	The input prompt (its language is detected for comparison)
output	str	Yes	The LLM output (its language is compared with the prompt's language)

Constructor Parameters

Name	Type	Required	Default	Description
model	None	No	None	Custom language detection model (defaults to papluca/xlm-roberta-base-language-detection)
threshold	float	No	0.1	Minimum confidence for a language to be included in the detected set
use_onnx	bool	No	False	Whether to use ONNX runtime for inference

Outputs

Name	Type	Description
sanitized_output	str	The output (unmodified)
is_valid	bool	Whether the output language matches the prompt language
risk_score	float	Risk score (-1.0 to 1.0)

Usage Examples

Basic Usage

from llm_guard.output_scanners import LanguageSame

scanner = LanguageSame(threshold=0.1)

prompt = "Cual es la capital de Francia?"
output = "La capital de Francia es Paris."

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

if is_valid:
    print("Output language matches prompt language")
else:
    print(f"Language mismatch detected (risk: {risk_score})")

Detecting Language Mismatch

from llm_guard.output_scanners import LanguageSame

scanner = LanguageSame(threshold=0.1)

prompt = "Wie geht es Ihnen?"  # German
output = "I am doing well, thank you!"  # English

sanitized_output, is_valid, risk_score = scanner.scan(prompt, output)
# is_valid will likely be False due to language mismatch
print(f"Languages match: {is_valid}, Risk: {risk_score}")

Related Pages

• Principle:Protectai_Llm_guard_Language_Consistency_Checking