Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Empowering Multi-step Reasoning across Languages via Tree-of-Thoughts

Leonardo Ranaldi, Giulia Pucci, Federico Ranaldi, Elena Sofia Ruzzetti, Fabio Massimo Zanzotto

TL;DR

The proposed Cross-lingual Tree-of-Thoughts (Cross-ToT) method, through a self-consistent cross-lingual prompting mechanism inspired by the Tree-of-Thoughts approach, provides multi-step reasoning paths in different languages that, during the steps, lead to the final solution.

Abstract

Reasoning methods, best exemplified by the well-known Chain-of-Thought (CoT), empower the reasoning abilities of Large Language Models (LLMs) by eliciting them to solve complex tasks in a step-by-step manner. Although they are achieving significant success, the ability to deliver multi-step reasoning remains limited to English because of the imbalance in the distribution of pre-training data, which makes other languages a barrier. In this paper, we propose Cross-lingual Tree-of-Thoughts (Cross-ToT), a method for aligning Cross-lingual CoT reasoning across languages. The proposed method, through a self-consistent cross-lingual prompting mechanism inspired by the Tree-of-Thoughts approach, provides multi-step reasoning paths in different languages that, during the steps, lead to the final solution. Experimental evaluations show that our method significantly outperforms existing prompting methods by reducing the number of interactions and achieving state-of-the-art performance.

Empowering Multi-step Reasoning across Languages via Tree-of-Thoughts

TL;DR

The proposed Cross-lingual Tree-of-Thoughts (Cross-ToT) method, through a self-consistent cross-lingual prompting mechanism inspired by the Tree-of-Thoughts approach, provides multi-step reasoning paths in different languages that, during the steps, lead to the final solution.

Abstract

Reasoning methods, best exemplified by the well-known Chain-of-Thought (CoT), empower the reasoning abilities of Large Language Models (LLMs) by eliciting them to solve complex tasks in a step-by-step manner. Although they are achieving significant success, the ability to deliver multi-step reasoning remains limited to English because of the imbalance in the distribution of pre-training data, which makes other languages a barrier. In this paper, we propose Cross-lingual Tree-of-Thoughts (Cross-ToT), a method for aligning Cross-lingual CoT reasoning across languages. The proposed method, through a self-consistent cross-lingual prompting mechanism inspired by the Tree-of-Thoughts approach, provides multi-step reasoning paths in different languages that, during the steps, lead to the final solution. Experimental evaluations show that our method significantly outperforms existing prompting methods by reducing the number of interactions and achieving state-of-the-art performance.
Paper Structure (37 sections, 5 equations, 5 figures, 12 tables)

This paper contains 37 sections, 5 equations, 5 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Cross-lingual Multi-step Reasoning
  3. Chain-of-Thought Across Languages
  4. Self-consistent Parallel Chain-of-Thougts
  5. Experiments
  6. Data
  7. Understanding tasks
  8. Commonsense Reasoning task
  9. Arithmetic Reasoning task
  10. Evaluated Languages
  11. Experimental Setup
  12. Main Results
  13. Analysis
  14. The English Matter
  15. The Impact of the Languages
  16. ...and 22 more sections

Figures (5)

  • Figure 1: Our Cross-ToT elicits the LLM to generate step-by-step Cross-lingual reasoning. Furthermore, different pathways are developed during these reasoning steps. This mechanism develops the Chains-of-Thoughts in a Self-consistent way, streaming with the different pathways.
  • Figure 2: Accuracies (%) on MGSM using "Cross-ToT", "Cross-ToT + English" and in binary version "Cross-ToT ( English + Target Language".
  • Figure 3: Accuracies (%) on Language Understanding benchmarks XNLI and PAWS-X introduced in Section \ref{['sec:data']}
  • Figure 4: The impact of integrating languages in our Cross-ToT on the final performance. Following Table \ref{['tab:language_distribution']}, we integrate languages from low-resources to high-resources and vice versa. We also propose the same experiments with the addition of English.
  • Figure 5: The analysis of reasoning quality between GPT-3.5 (Native-CoT) and CLP in qin2023crosslingual and our Cross-ToT