Dictionary Insertion Prompting for Multilingual Reasoning on Multilingual Large Language Models

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Advanced, quick

Summary

Dictionary Insertion Prompting (DIP) is a novel method designed to enhance multilingual reasoning in Large Language Models, which are often English-centric and struggle with non-English tasks. DIP addresses this by looking up non-English words in a prompt and inserting their English equivalents directly into the prompt's middle. This technique aims to improve the LLM's internal translation to English and leverage its stronger English-based reasoning capabilities, leading to significantly better results. The method was evaluated across 10 to 200 languages using synthetic multilingual benchmarks created from existing English reasoning datasets like GSM8K and AQuA, translated via NLLB. A critical finding revealed that interleaving the dictionary words within the original prompt significantly outperforms prepending or appending them, highlighting the importance of insertion placement.

Key takeaway

For NLP Engineers developing multilingual LLM applications, consider implementing Dictionary Insertion Prompting (DIP) to enhance non-English reasoning. By strategically interleaving English word equivalents directly into your non-English prompts, you can leverage the LLM's stronger English-centric reasoning capabilities. This approach, particularly with interleaved insertion, offers a concrete method to improve performance on diverse language tasks without retraining, directly addressing limitations in current multilingual models.

Key insights

Dictionary Insertion Prompting improves multilingual LLM reasoning by interleaving English word equivalents into non-English prompts.

Principles

• Multilingual LLMs benefit from English-centric reasoning.
• Prompt structure impacts external knowledge utility.
• Interleaving dictionary words is optimal.

Method

DIP identifies non-English words in a prompt, looks up their English counterparts in a dictionary, and inserts these English words into the middle of the original prompt to enhance LLM reasoning.

In practice

• Create synthetic multilingual benchmarks.
• Interleave English terms in non-English prompts.
• Test dictionary insertion placement.

Topics

• Multilingual LLMs
• Prompt Engineering
• Dictionary Insertion Prompting
• Cross-lingual Reasoning
• NLLB Translator
• FLORES-200

Best for: AI Engineer, Machine Learning Engineer, Research Scientist, AI Scientist, NLP Engineer, Prompt Engineer

Related on AIssential

• LANG: Reinforcement Learning for Multilingual Reasoning with Language-Adaptive Hint Guidance — LANG enhances multilingual LLM reasoning by using language-adaptive hints with a progressive decay schedule, preventing language drift.
• llm 0.32a2 — OpenAI's GPT-5 class models now use `/v1/responses` for interleaved reasoning, enhancing tool call visibility.
• Can LLMs Be Constrained to the Past? Improving Knowledge Cutoff through Recall-Based Prompting — LLMs can better adhere to knowledge cutoffs using recall-based prompting, especially for causally related information.
• Are you speaking my languages? On spoken language adherence in multimodal LLMs — Multimodal LLMs require soft prompting and specific strategies to ensure spoken language adherence in ASR, preventing transcription errors.
• English is Not All You Need: Systematically Exploring the Role of Multilinguality in LLM Post-Training — Multilingual post-training significantly improves LLM performance across languages and tasks, outperforming English-only approaches.
• Cross-Lingual Token Arbitrage: Optimizing Code Agent Context Windows via Local LLM Preprocessing — Proactive, edge-side prompt rewriting with a local LLM significantly reduces token costs for coding agents.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.