Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Template-Driven LLM-Paraphrased Framework for Tabular Math Word Problem Generation

Xiaoqiang Kang, Zimu Wang, Xiaobo Jin, Wei Wang, Kaizhu Huang, Qiufeng Wang

TL;DR

This work tackles data scarcity in tabular math word problems by introducing TeLL, a template-driven LLM-paraphrased framework that couples extracted templates with paraphrasing to produce correct and diverse TMWP samples. It constructs TabMWP-TeLL, a high-quality dataset with illustrative step-by-step solutions, and demonstrates that training with TabMWP+TabMWP-TeLL substantially improves TMWP-solving performance across several LLMs, especially on harder problems. The approach blends template abstraction, augmentation, and context-rich paraphrasing to capture realistic backgrounds while preserving core problem structure. The results highlight the practical impact of principled data augmentation on robust mathematical reasoning in LLMs and point to future work in extending to commonsense and multi-hop reasoning tasks.

Abstract

Solving tabular math word problems (TMWPs) has become a critical role in evaluating the mathematical reasoning ability of large language models (LLMs), where large-scale TMWP samples are commonly required for LLM fine-tuning. Since the collection of high-quality TMWP datasets is costly and time-consuming, recent research has concentrated on automatic TMWP generation. However, current generated samples usually suffer from issues of either correctness or diversity. In this paper, we propose a Template-driven LLM-paraphrased (TeLL) framework for generating high-quality TMWP samples with diverse backgrounds and accurate tables, questions, answers, and solutions. To this end, we first extract templates from existing real samples to generate initial problems, ensuring correctness. Then, we adopt an LLM to extend templates and paraphrase problems, obtaining diverse TMWP samples. Furthermore, we find the reasoning annotation is important for solving TMWPs. Therefore, we propose to enrich each solution with illustrative reasoning steps. Through the proposed framework, we construct a high-quality dataset TabMWP-TeLL by adhering to the question types in the TabMWP dataset, and we conduct extensive experiments on a variety of LLMs to demonstrate the effectiveness of TabMWP-TeLL in improving TMWP solving performance. The code and data of this paper are available at: https://github.com/Jason8Kang/TELL.

Template-Driven LLM-Paraphrased Framework for Tabular Math Word Problem Generation

TL;DR

This work tackles data scarcity in tabular math word problems by introducing TeLL, a template-driven LLM-paraphrased framework that couples extracted templates with paraphrasing to produce correct and diverse TMWP samples. It constructs TabMWP-TeLL, a high-quality dataset with illustrative step-by-step solutions, and demonstrates that training with TabMWP+TabMWP-TeLL substantially improves TMWP-solving performance across several LLMs, especially on harder problems. The approach blends template abstraction, augmentation, and context-rich paraphrasing to capture realistic backgrounds while preserving core problem structure. The results highlight the practical impact of principled data augmentation on robust mathematical reasoning in LLMs and point to future work in extending to commonsense and multi-hop reasoning tasks.

Abstract

Solving tabular math word problems (TMWPs) has become a critical role in evaluating the mathematical reasoning ability of large language models (LLMs), where large-scale TMWP samples are commonly required for LLM fine-tuning. Since the collection of high-quality TMWP datasets is costly and time-consuming, recent research has concentrated on automatic TMWP generation. However, current generated samples usually suffer from issues of either correctness or diversity. In this paper, we propose a Template-driven LLM-paraphrased (TeLL) framework for generating high-quality TMWP samples with diverse backgrounds and accurate tables, questions, answers, and solutions. To this end, we first extract templates from existing real samples to generate initial problems, ensuring correctness. Then, we adopt an LLM to extend templates and paraphrase problems, obtaining diverse TMWP samples. Furthermore, we find the reasoning annotation is important for solving TMWPs. Therefore, we propose to enrich each solution with illustrative reasoning steps. Through the proposed framework, we construct a high-quality dataset TabMWP-TeLL by adhering to the question types in the TabMWP dataset, and we conduct extensive experiments on a variety of LLMs to demonstrate the effectiveness of TabMWP-TeLL in improving TMWP solving performance. The code and data of this paper are available at: https://github.com/Jason8Kang/TELL.

Paper Structure

This paper contains 27 sections, 3 equations, 6 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Math Word Problems.
  4. Tabular Math Word Problems.
  5. Math Word Problem Generation.
  6. Methodology
  7. Problem Definition
  8. TeLL for TMWP Generation
  9. Overview.
  10. Template Abstraction.
  11. Template Augmentation.
  12. Template Selection and Instantiation.
  13. Problem Paraphrasing.
  14. Experiments
  15. Baselines
  16. ...and 12 more sections

Figures (6)

  • Figure 1: Illustration of an original TMWP sample and the generated samples with LLMs, templates, and ours with an illustrative solution. (Notes: "Stem" means the first digit, and "Leaf" means the last digit in stem-leaf plots).
  • Figure 2: Overall framework of the proposed TeLL method to generate TMWPs with correctness and diversity, consisting of five steps: 1) template abstraction, 2) template augmentation, 3) template selection, 4) template instantiation, and 5) problem paraphrasing.
  • Figure 3: Prompt for template augmentation.
  • Figure 4: Prompt for paraphrasing template-based problems to natural problems.
  • Figure 5: Question distribution of the TabMWP dataset.
  • ...and 1 more figures