Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Simplifying Translations for Children: Iterative Simplification Considering Age of Acquisition with LLMs

Masashi Oshika, Makoto Morishita, Tsutomu Hirao, Ryohei Sasano, Koichi Takeda

TL;DR

The paper tackles the problem of adjusting translation difficulty to a child’s reading level by introducing an iterative LLM-based post-editing approach guided by word-level Age of Acquisition (AoA). It constructs a back-translation–derived benchmark from Simple English Wikipedia to evaluate AoA-based simplification and demonstrates that iteratively replacing high-AoA words preserves MT quality while markedly reducing word complexity. The method achieves state-of-the-art BLEU and COMET scores on the benchmark and simplifies a large majority of translations after multiple iterations, highlighting its practical potential for child-friendly MT. The work enables per-word control of translation complexity, offering a path toward customizable, accessible machine translation across languages, with room for efficiency and cross-linguistic extensions.

Abstract

In recent years, neural machine translation (NMT) has been widely used in everyday life. However, the current NMT lacks a mechanism to adjust the difficulty level of translations to match the user's language level. Additionally, due to the bias in the training data for NMT, translations of simple source sentences are often produced with complex words. In particular, this could pose a problem for children, who may not be able to understand the meaning of the translations correctly. In this study, we propose a method that replaces words with high Age of Acquisitions (AoA) in translations with simpler words to match the translations to the user's level. We achieve this by using large language models (LLMs), providing a triple of a source sentence, a translation, and a target word to be replaced. We create a benchmark dataset using back-translation on Simple English Wikipedia. The experimental results obtained from the dataset show that our method effectively replaces high-AoA words with lower-AoA words and, moreover, can iteratively replace most of the high-AoA words while still maintaining high BLEU and COMET scores.

Simplifying Translations for Children: Iterative Simplification Considering Age of Acquisition with LLMs

TL;DR

The paper tackles the problem of adjusting translation difficulty to a child’s reading level by introducing an iterative LLM-based post-editing approach guided by word-level Age of Acquisition (AoA). It constructs a back-translation–derived benchmark from Simple English Wikipedia to evaluate AoA-based simplification and demonstrates that iteratively replacing high-AoA words preserves MT quality while markedly reducing word complexity. The method achieves state-of-the-art BLEU and COMET scores on the benchmark and simplifies a large majority of translations after multiple iterations, highlighting its practical potential for child-friendly MT. The work enables per-word control of translation complexity, offering a path toward customizable, accessible machine translation across languages, with room for efficiency and cross-linguistic extensions.

Abstract

In recent years, neural machine translation (NMT) has been widely used in everyday life. However, the current NMT lacks a mechanism to adjust the difficulty level of translations to match the user's language level. Additionally, due to the bias in the training data for NMT, translations of simple source sentences are often produced with complex words. In particular, this could pose a problem for children, who may not be able to understand the meaning of the translations correctly. In this study, we propose a method that replaces words with high Age of Acquisitions (AoA) in translations with simpler words to match the translations to the user's level. We achieve this by using large language models (LLMs), providing a triple of a source sentence, a translation, and a target word to be replaced. We create a benchmark dataset using back-translation on Simple English Wikipedia. The experimental results obtained from the dataset show that our method effectively replaces high-AoA words with lower-AoA words and, moreover, can iteratively replace most of the high-AoA words while still maintaining high BLEU and COMET scores.
Paper Structure (18 sections, 4 figures, 7 tables)

This paper contains 18 sections, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Text Simplification
  4. Automatic Post-Editing
  5. Complexity-control Method
  6. Dataset Creation
  7. Procedure using Back-Translation
  8. Creating Dataset from Simple English Wikipedia
  9. Proposed Method
  10. Experiments
  11. Settings
  12. Evaluation Metrics
  13. Compared Methods
  14. Results and Discussion
  15. Main Results
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Overview of the proposed method, which generates a simplified translation given the source sentence, initial translation, and target word. If the highest AoA in the output sentence is higher than the target age, the model will iteratively revise the sentence. The target word is defined as the word with the highest AoA in the initial translation.
  • Figure 2: Procedure for creating a data set and the corresponding input to the LLM.
  • Figure 3: Distribution of the AoA difference in the created dataset, showing only the sentences where the AoA difference is greater than 0. AoA difference was 0 for 1,164,870 sentence pairs for 66% of the dataset.
  • Figure 4: Statistical plots of highest AoA of the generated sentence by each method.