Evaluating Multilingual and Code-Switched Alignment in LLMs via Synthetic Natural Language Inference

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Abdaljalil, Samir, Serpedin, Erchin, Qaraqe, Khalid, Kurban, Hasan
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916910191345664
author Abdaljalil, Samir
Serpedin, Erchin
Qaraqe, Khalid
Kurban, Hasan
author_facet Abdaljalil, Samir
Serpedin, Erchin
Qaraqe, Khalid
Kurban, Hasan
contents Large language models (LLMs) are increasingly applied in multilingual contexts, yet their capacity for consistent, logically grounded alignment across languages remains underexplored. We present a controlled evaluation framework for multilingual natural language inference (NLI) that generates synthetic, logic-based premise-hypothesis pairs and translates them into a typologically diverse set of languages. This design enables precise control over semantic relations and allows testing in both monolingual and mixed-language (code-switched) conditions. Surprisingly, code-switching does not degrade, and can even improve, performance, suggesting that translation-induced lexical variation may serve as a regularization signal. We validate semantic preservation through embedding-based similarity analyses and cross-lingual alignment visualizations, confirming the fidelity of translated pairs. Our findings expose both the potential and the brittleness of current LLM cross-lingual reasoning, and identify code-switching as a promising lever for improving multilingual robustness. Code available at: https://github.com/KurbanIntelligenceLab/nli-stress-testing
format Preprint
id arxiv_https___arxiv_org_abs_2508_14735
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Evaluating Multilingual and Code-Switched Alignment in LLMs via Synthetic Natural Language Inference
Abdaljalil, Samir
Serpedin, Erchin
Qaraqe, Khalid
Kurban, Hasan
Computation and Language
Artificial Intelligence
Large language models (LLMs) are increasingly applied in multilingual contexts, yet their capacity for consistent, logically grounded alignment across languages remains underexplored. We present a controlled evaluation framework for multilingual natural language inference (NLI) that generates synthetic, logic-based premise-hypothesis pairs and translates them into a typologically diverse set of languages. This design enables precise control over semantic relations and allows testing in both monolingual and mixed-language (code-switched) conditions. Surprisingly, code-switching does not degrade, and can even improve, performance, suggesting that translation-induced lexical variation may serve as a regularization signal. We validate semantic preservation through embedding-based similarity analyses and cross-lingual alignment visualizations, confirming the fidelity of translated pairs. Our findings expose both the potential and the brittleness of current LLM cross-lingual reasoning, and identify code-switching as a promising lever for improving multilingual robustness. Code available at: https://github.com/KurbanIntelligenceLab/nli-stress-testing
title Evaluating Multilingual and Code-Switched Alignment in LLMs via Synthetic Natural Language Inference
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2508.14735