SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.20052 · EMNLP · 2024

Understanding and Mitigating Language Confusion in LLMs

Sebastian Ruder, Kelly Marchisio, Wei-Yin Ko, Alexandre Bérard, Théo Dehaze

arXiv · PDF · Open in the Atlas

Code that ran

3 of 3 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 0 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2406.20052.json · badge: /api/ran/2406.20052.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
for-ai/language-confusion canonical (classifier label)
classifier basis: census
3 ran · 0 no recorded run
FunctionStatusWhere it lives
normalize Ran
ran; our contract draft was wrong, not the code
for-ai/language-confusion/compute_metrics.py
code served (permissive licence) · get_code("a996a61a010f6934")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import string

def normalize(text: str) -> str:
    text = text.split('\nQ:')[0].strip()
    text = text.translate(str.maketrans('', '', string.punctuation))
    text = text.replace('—', ' ')
    text = text.replace('،', '')
    return text


ARGS = {
    'text': 'Hello world! This is a test.\nQ: Some question here'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
tokenize Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
for-ai/language-confusion/compute_metrics.py
code served (permissive licence) · get_code("3f7e1abc53ea4130")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import functools

# Mock tokenizers to avoid actual downloads/network calls
class MockZhTokenizer:
    def initialize(self):
        pass
    
    def cut(self, line: str):
        return line.split()

class MockJaTokenizer:
    def parse(self, line: str):
        return ' '.join(line.split())

# Set up global tokenizers with mocks
ja_tokenizer = None
zh_tokenizer = None

@functools.lru_cache(maxsize=2 ** 20)
def tokenize(line: str, lang: str) -> list[str]:
    global ja_tokenizer, zh_tokenizer
    if lang == 'zh':
        if zh_tokenizer is None:
            zh_tokenizer = MockZhTokenizer()
            zh_tokenizer.initialize()
        return list(zh_tokenizer.cut(line))
    elif lang == 'ja':
        if ja_tokenizer is None:
            ja_tokenizer = MockJaTokenizer()
        return ja_tokenizer.parse(line).split()
    else:
        return line.split()

# Define arguments for the function
ARGS = {
    'line': 'hello world test',
    'lang': 'en'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
fugashi, jieba, unidic — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
turns_to_prompt Ran
ran; our contract draft was wrong, not the code
for-ai/language-confusion/few_shot_prompting.py
code served (permissive licence) · get_code("8ad8ac8b774652af")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

# Construct valid arguments for turns_to_prompt
turns = [
    {
        'role': 'User',
        'message': 'Hello, how are you?'
    },
    {
        'role': 'Chatbot',
        'message': 'I\'m fine and you?'
    },
    {
        'role': 'User',
        'message': 'How much is 1+1?'
    }
]

ARGS = {
    'turns': turns
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 3 samples lifted from this paper's linked repositories, 3 carry a function identity and 0 do not. Of the 3 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 3 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

We investigate a surprising limitation of LLMs: their inability to consistently generate text in a user's desired language. We create the Language Confusion Benchmark (LCB) to evaluate such failures, covering 15 typologically diverse languages with existing and newly-created English and multilingual prompts. We evaluate a range of LLMs on monolingual and crosslingual generation reflecting practical use cases, finding that Llama Instruct and Mistral models exhibit high degrees of language confusion and even the strongest models fail to consistently respond in the correct language. We observe that base and English-centric instruct models are more prone to language confusion, which is aggravated by complex prompts and high sampling temperatures. We find that language confusion can be partially mitigated via fewshot prompting, multilingual SFT and preference tuning. We release our language confusion benchmark, which serves as a first layer of efficient, scalable multilingual evaluation. 1

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2406.20052")
get_code_for_paper("2406.20052")
have("2406.20052")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2406.20052.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2406.20052.svg)](https://syntology.ai/paper/2406.20052)

Connect an agent — have() is free.