Abstract:Large language models (LLMs) are at the forefront of transforming numerous domains globally. However, their inclusivity and effectiveness remain limited for non-Latin scripts and low-resource languages. This paper tackles the imperative challenge of enhancing the multilingual performance of LLMs without extensive training or fine-tuning. Through systematic investigation and evaluation of diverse languages using popular question-answering (QA) datasets, we present novel techniques that unlock the true potential of LLMs in a polyglot landscape. Our approach encompasses three key strategies that yield significant improvements in multilingual proficiency. First, by meticulously optimizing prompts tailored for polyglot LLMs, we unlock their latent capabilities, resulting in substantial performance boosts across languages. Second, we introduce a new hybrid approach that synergizes LLM Retrieval Augmented Generation (RAG) with multilingual embeddings and achieves improved multilingual task performance. Finally, we introduce a novel learning approach that dynamically selects the optimal prompt strategy, LLM model, and embedding model per query at run-time. This dynamic adaptation maximizes the efficacy of LLMs across languages, outperforming best static and random strategies. Additionally, our approach adapts configurations in both offline and online settings, and can seamlessly adapt to new languages and datasets, leading to substantial advancements in multilingual understanding and generation across diverse languages.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve multilingual performance in large - language models (LLMs), especially in non - Latin - alphabet and low - resource languages. Although current LLMs perform well in many fields, their effectiveness and inclusiveness in dealing with non - English and low - resource languages are still limited. The paper proposes a method without extensive training or fine - tuning to improve the multilingual performance of LLMs. Specifically, by systematically researching and evaluating the performance of different languages on popular question - answering (QA) datasets, the paper proposes three key strategies to significantly improve multilingual ability: 1. **Optimizing Prompt Words**: Carefully design prompt words for multilingual LLMs to unlock their potential capabilities, thereby achieving significant performance improvements in multiple languages. 2. **Hybrid Method**: Combine LLM retrieval - augmented generation (RAG) with multilingual embedding techniques, and retrieve relevant information through an improved text - embedding model, thereby improving the coherence and context - relevance of tasks. 3. **Dynamic Learning Method**: Dynamically select the optimal prompt - word strategy, LLM model, and embedding model at runtime to maximize the LLM effectiveness under different languages and tasks. These strategies aim to improve the performance of LLMs in multilingual environments without additional training or fine - tuning, thereby narrowing the gap with the state - of - the - art multilingual models. The paper also introduces how to adapt to new languages and datasets in offline and online environments to ensure robust performance in diverse language environments.

Bridging the Gap: Dynamic Learning Strategies for Improving Multilingual Performance in LLMs

Breaking Language Barriers with a LEAP: Learning Strategies for Polyglot LLMs

Improving Bilingual Capabilities of Language Models to Support Diverse Linguistic Practices in Education

Leveraging Large Language Models for Multiple Choice Question Answering

Bridging the Language Gap: Enhancing Multilingual Prompt-Based Code Generation in LLMs via Zero-Shot Cross-Lingual Transfer

Bridging the Gap: Enhancing LLM Performance for Low-Resource African Languages with New Benchmarks, Fine-Tuning, and Cultural Adjustments

Democratizing LLMs for Low-Resource Languages by Leveraging their English Dominant Abilities with Linguistically-Diverse Prompts

Not All Languages Are Created Equal in LLMs: Improving Multilingual Capability by Cross-Lingual-Thought Prompting

Getting More from Less: Large Language Models are Good Spontaneous Multilingual Learners

LLM for Everyone: Representing the Underrepresented in Large Language Models

Exploring Design Choices for Building Language-Specific LLMs

Optimizing Large Language Models for Dynamic Constraints through Human-in-the-Loop Discriminators

Crosslingual Capabilities and Knowledge Barriers in Multilingual Large Language Models

A Three-Pronged Approach to Cross-Lingual Adaptation with Multilingual LLMs

Refining Translations with LLMs: A Constraint-Aware Iterative Prompting Approach

Understanding and Mitigating Language Confusion in LLMs

Teaching LLMs to Abstain across Languages via Multilingual Feedback

RLHF Can Speak Many Languages: Unlocking Multilingual Preference Optimization for LLMs

Multi-OphthaLingua: A Multilingual Benchmark for Assessing and Debiasing LLM Ophthalmological QA in LMICs

Are Large Language Model-based Evaluators the Solution to Scaling Up Multilingual Evaluation?