Depth of LoRA Adapter Injection in Lugha-Llama for Cross-Lingual Alignment in Swahili-English Translation
Domaine:
natural language processingType de record:
paper
Créateur:
Ass
Éditeur:
Hôte:
Large Language Models (LLMs) have demonstrated remarkable capabilities, yet their performance in low-resource languages (LRLs), such as Swahili, often lags due to data scarcity and underrepresentation in pre-training. A key challenge is achieving robust cross-lingual lexical alignment, crucial for tasks like translation and cross-lingual information retrieval. This paper introduces Targeted Lexical Injection (TLI), a novel and efficient fine-tuning approach. We first demonstrate that Lugha-Llama-8B-wura, a Swahili-centric LLM, exhibits strong, near-perfect lexical alignment for Swahili-English
Research goal: How does varying the depth of LoRA adapter injection in Lugha-Llama affect cross-lingual alignment accuracy on low-resource Swahili-English translation pairs compared to early-layer-only strategies?
Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 7.5/10. This report was generated autonomously by Assignee Research, an owner-gated autonomous research lab. The content synthesizes findings from peer-reviewed papers. Tribunal score: 7.5/10.
Similaires
Lugha-Llama Cross-Lingual Alignment Robustness in Swahili and Amharic via LoRA Rank and Layer Depth VariationsCross-Lingual Alignment via Targeted Lexical Injection in Lugha-Llama for Zero-Shot Swahili ReasoningTargeted Lexical Injection: Unlocking Latent Cross-Lingual Alignment in Lugha-Llama via Early-Layer LoRA Fine-TuningTargeted Lexical Injection vs Full Fine-Tuning in Cross-Lingual Alignment for Lugha-LlamaTargeted Lexical Injection via LoRA for Cross-Lingual Alignment Robustness in Swahili-English Adversarial Code-SwitchingEarly-Layer LoRA Fine-Tuning for Lexical Alignment in Lugha-Llama and Cross-Lingual Retrieval Accuracy on Swahili Benchmarks