Beyond One-Size-Fits-All Summarization: Customizing Summaries for Diverse Users

TL;DR

This work addresses the problem of readability-controlled abstractive summarization for Turkish by incorporating a language-specific readability metric (YOD) into a transformer-based framework. The authors develop a VBART-based multi-task model conditioned on readability levels via special tokens and dual heads for continuous and discrete YOD prediction, trained with a weighted, multi-objective loss that balances summarization accuracy and readability control. A custom dataset augmented with paraphrase-based synthetic data expands YOD coverage, enabling evaluation across 16 readability levels; results show parity with a supervised fine-tuning baseline in semantic metrics while offering superior readability targeting, especially in mid-to-high readability ranges. The study demonstrates the practicality and potential impact of readability-aware summarization for Turkish, with implications for accessibility in education and professional domains, and outlines data and resource-focused limitations and avenues for future work.

Abstract

In recent years, automatic text summarization has witnessed significant advancement, particularly with the development of transformer-based models. However, the challenge of controlling the readability level of generated summaries remains an under-explored area, especially for languages with complex linguistic features like Turkish. This gap has the effect of impeding effective communication and also limits the accessibility of information. Controlling readability of textual data is an important element for creating summaries for different audiences with varying literacy and education levels, such as students ranging from primary school to graduate level, as well as individuals with diverse educational backgrounds. Summaries that align with the needs of specific reader groups can improve comprehension and engagement, ensuring that the intended message is effectively communicated. Furthermore, readability adjustment is essential to expand the usability of summarization models in educational and professional domains. Current summarization models often don't have the mechanisms to adjust the complexity of their outputs, resulting in summaries that may be too simplistic or overly complex for certain types of reader groups. Developing adaptive models that can tailor content to specific readability levels is therefore crucial. To address this problem, we create our own custom dataset and train a model with our custom architecture. Our method ensures that readability levels are effectively controlled while maintaining accuracy and coherence. We rigorously compare our model to a supervised fine-tuned baseline, demonstrating its superiority in generating readability-aware summaries.

Figures (9)

• Figure 1: Synthetic Data Generation
• Figure 2: Average Token Length
• Figure 3: Special Token Conditioning
• Figure 4: Custom Model Architecture
• Figure 5: Loss curves comparison – supervised fine-tuned model vs. custom architecture.