Examining Multilingual Embedding Models Cross-Lingually Through LLM-Generated Adversarial Examples
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866914082987180032 |
|---|---|
| author | Michail, Andrianos Clematide, Simon Sennrich, Rico |
| author_facet | Michail, Andrianos Clematide, Simon Sennrich, Rico |
| contents | The evaluation of cross-lingual semantic search models is often limited to existing datasets from tasks such as information retrieval and semantic textual similarity. We introduce Cross-Lingual Semantic Discrimination (CLSD), a lightweight evaluation task that requires only parallel sentences and a Large Language Model (LLM) to generate adversarial distractors. CLSD measures an embedding model's ability to rank the true parallel sentence above semantically misleading but lexically similar alternatives. As a case study, we construct CLSD datasets for German--French in the news domain. Our experiments show that models fine-tuned for retrieval tasks benefit from pivoting through English, whereas bitext mining models perform best in direct cross-lingual settings. A fine-grained similarity analysis further reveals that embedding models differ in their sensitivity to linguistic perturbations. We release our code and datasets under AGPL-3.0: https://github.com/impresso/cross_lingual_semantic_discrimination |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_08638 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Examining Multilingual Embedding Models Cross-Lingually Through LLM-Generated Adversarial Examples Michail, Andrianos Clematide, Simon Sennrich, Rico Computation and Language The evaluation of cross-lingual semantic search models is often limited to existing datasets from tasks such as information retrieval and semantic textual similarity. We introduce Cross-Lingual Semantic Discrimination (CLSD), a lightweight evaluation task that requires only parallel sentences and a Large Language Model (LLM) to generate adversarial distractors. CLSD measures an embedding model's ability to rank the true parallel sentence above semantically misleading but lexically similar alternatives. As a case study, we construct CLSD datasets for German--French in the news domain. Our experiments show that models fine-tuned for retrieval tasks benefit from pivoting through English, whereas bitext mining models perform best in direct cross-lingual settings. A fine-grained similarity analysis further reveals that embedding models differ in their sensitivity to linguistic perturbations. We release our code and datasets under AGPL-3.0: https://github.com/impresso/cross_lingual_semantic_discrimination |
| title | Examining Multilingual Embedding Models Cross-Lingually Through LLM-Generated Adversarial Examples |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2502.08638 |