Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fast, Not Fancy: Rethinking G2P with Rich Data and Rule-Based Models

Mahta Fetrat Qharabagh, Zahra Dehghanian, Hamid R. Rabiee

TL;DR

This work tackles homograph disambiguation in G2P for low-resource languages by building HomoRich, a large, balanced Persian homograph dataset via a semi-automated pipeline that combines human input and LLM-assisted labeling. Leveraging HomoRich, the authors improve a neural G2P model (Homo-GE2PE) by approximately 29.72% in homograph accuracy and, crucially, introduce HomoFast eSpeak, a fast, rule-based variant that gains about 30.66% in homograph accuracy without sacrificing latency. They also demonstrate that fine-tuning a T5-based G2P model on their dataset (Homo-T5) yields competitive performance, underscoring the data quality's value across architectures. By releasing HomoRich and the enhanced tools under permissive licenses, the paper highlights a practical path toward accurate, real-time G2P for screen readers and other accessibility technologies in low-resource languages.

Abstract

Homograph disambiguation remains a significant challenge in grapheme-to-phoneme (G2P) conversion, especially for low-resource languages. This challenge is twofold: (1) creating balanced and comprehensive homograph datasets is labor-intensive and costly, and (2) specific disambiguation strategies introduce additional latency, making them unsuitable for real-time applications such as screen readers and other accessibility tools. In this paper, we address both issues. First, we propose a semi-automated pipeline for constructing homograph-focused datasets, introduce the HomoRich dataset generated through this pipeline, and demonstrate its effectiveness by applying it to enhance a state-of-the-art deep learning-based G2P system for Persian. Second, we advocate for a paradigm shift - utilizing rich offline datasets to inform the development of fast, rule-based methods suitable for latency-sensitive accessibility applications like screen readers. To this end, we improve one of the most well-known rule-based G2P systems, eSpeak, into a fast homograph-aware version, HomoFast eSpeak. Our results show an approximate 30% improvement in homograph disambiguation accuracy for the deep learning-based and eSpeak systems.

Fast, Not Fancy: Rethinking G2P with Rich Data and Rule-Based Models

TL;DR

This work tackles homograph disambiguation in G2P for low-resource languages by building HomoRich, a large, balanced Persian homograph dataset via a semi-automated pipeline that combines human input and LLM-assisted labeling. Leveraging HomoRich, the authors improve a neural G2P model (Homo-GE2PE) by approximately 29.72% in homograph accuracy and, crucially, introduce HomoFast eSpeak, a fast, rule-based variant that gains about 30.66% in homograph accuracy without sacrificing latency. They also demonstrate that fine-tuning a T5-based G2P model on their dataset (Homo-T5) yields competitive performance, underscoring the data quality's value across architectures. By releasing HomoRich and the enhanced tools under permissive licenses, the paper highlights a practical path toward accurate, real-time G2P for screen readers and other accessibility technologies in low-resource languages.

Abstract

Homograph disambiguation remains a significant challenge in grapheme-to-phoneme (G2P) conversion, especially for low-resource languages. This challenge is twofold: (1) creating balanced and comprehensive homograph datasets is labor-intensive and costly, and (2) specific disambiguation strategies introduce additional latency, making them unsuitable for real-time applications such as screen readers and other accessibility tools. In this paper, we address both issues. First, we propose a semi-automated pipeline for constructing homograph-focused datasets, introduce the HomoRich dataset generated through this pipeline, and demonstrate its effectiveness by applying it to enhance a state-of-the-art deep learning-based G2P system for Persian. Second, we advocate for a paradigm shift - utilizing rich offline datasets to inform the development of fast, rule-based methods suitable for latency-sensitive accessibility applications like screen readers. To this end, we improve one of the most well-known rule-based G2P systems, eSpeak, into a fast homograph-aware version, HomoFast eSpeak. Our results show an approximate 30% improvement in homograph disambiguation accuracy for the deep learning-based and eSpeak systems.
Paper Structure (159 sections, 15 figures, 3 tables)

This paper contains 159 sections, 15 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Approaches
  4. Rule-based approaches
  5. Neural approaches
  6. LLM-based approaches
  7. Datasets
  8. Methodology
  9. Data Preparation
  10. Data Statistics
  11. Data Augmentation
  12. Proposed G2P Tools
  13. Homo-GE2PE (Neural)
  14. HomoFast eSpeak (Rule-based)
  15. Results
  16. ...and 144 more sections

Figures (15)

  • Figure 1: Prompt for generating homograph sentences.
  • Figure 2: LLM-powered G2P workflow qharabagh2025llm
  • Figure 3: Dataset structure with example entry.
  • Figure 4: Data source distribution in HomoRich dataset.
  • Figure 5: Sample counts per pronunciation.
  • ...and 10 more figures