# The Multilingual Curse at the Retrieval Layer: Evidence from Amharic

arXiv: https://arxiv.org/abs/2605.24556 · HTML: https://mlsanity.com/paper/2605.24556/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Generation and context integration, one of 15 directions in Retrieval-augmented generation.
- Builds on 18 of its 32 references that are in the map.
- First posted on arXiv 23 May 2026.

## Placed in

- Retrieval-augmented generation / [Generation and context integration](https://mlsanity.com/rag/generation-and-context-integration/)

## Builds on (most cited first, 18 of 18)

- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT](https://mlsanity.com/paper/2004.12832/) (2020)
- [Teaching Machines to Read and Comprehend](https://mlsanity.com/paper/1506.03340/) (2015)
- [Passage Re-ranking with BERT](https://mlsanity.com/paper/1901.04085/) (2019)
- [XL-Sum: Large-Scale Multilingual Abstractive Summarization for 44 Languages](https://mlsanity.com/paper/2106.13822/) (2021)
- [SPLADE: Sparse Lexical and Expansion Model for First Stage Ranking](https://mlsanity.com/paper/2107.05720/) (2021)
- [mGTE: Generalized Long-Context Text Representation and Reranking Models for Multilingual Text Retrieval](https://mlsanity.com/paper/2407.19669/) (2024)
- [Efficient Natural Language Response Suggestion for Smart Reply](https://mlsanity.com/paper/1705.00652/) (2017)
- [EmbeddingGemma: Powerful and Lightweight Text Representations](https://mlsanity.com/paper/2509.20354/) (2025)
- [XOR QA: Cross-lingual Open-Retrieval Question Answering](https://mlsanity.com/paper/2010.11856/) (2020)
- [mMARCO: A Multilingual Version of the MS MARCO Passage Ranking Dataset](https://mlsanity.com/paper/2108.13897/) (2021)
- [From Distillation to Hard Negative Sampling: Making Sparse Neural IR Models More Effective](https://mlsanity.com/paper/2205.04733/) (2022)
- [Arctic-Embed 2.0: Multilingual Retrieval Without Compromise](https://mlsanity.com/paper/2412.04506/) (2024)
- [Towards Best Practices for Training Multilingual Dense Retrieval Models](https://mlsanity.com/paper/2204.02363/) (2022)
- [PyLate: Flexible Training and Retrieval for Late Interaction Models](https://mlsanity.com/paper/2508.03555/) (2025)
- [An Amharic News Text classification Dataset](https://mlsanity.com/paper/2103.05639/) (2021)
- [Milco: Learned Sparse Retrieval Across Languages via a Multilingual Connector](https://mlsanity.com/paper/2510.00671/) (2025)
- [Optimized Text Embedding Models and Benchmarks for Amharic Passage Retrieval](https://mlsanity.com/paper/2505.19356/) (2025)

## Abstract (arXiv)

Multilingual retrieval increasingly underpins cross-lingual question answering and retrieval-augmented generation. Strong zero-shot scores on multilingual benchmarks are often taken as evidence that current encoders transfer reliably across many languages. We argue that this assumption breaks down for underrepresented, morphologically rich languages, and use Amharic as a diagnostic case. Under a shared passage retrieval protocol covering dense, late-interaction, learned sparse, and cross-encoder paradigms, we compare zero-shot multilingual retrievers, Amharic-fine-tuned multilingual retrievers, and monolingual Amharic retrievers. The strongest zero-shot multilingual retriever underperforms the strongest monolingual Amharic first-stage retriever by 23% relative MRR@10. Fine-tuning two recent multilingual embedding models on the same Amharic supervision yields 32-60% relative MRR@10 gains over zero-shot, but the best Amharic-fine-tuned multilingual model remains below the strongest monolingual Amharic retriever. These findings indicate that zero-shot multilingual retrieval is not a sufficient proxy for equitable information access in the LLM era: for underrepresented languages, retrieval must be evaluated and adapted in-language rather than inferred from aggregate multilingual benchmarks. To foster future research, we publicly release the dataset, codebase, and trained models at https://github.com/rasyosef/amharic-neural-ir.