Neural text normalization leveraging similarities of strings and sounds

TL;DR

This paper introduces neural models for text normalization that utilize string and sound similarities to effectively handle misspellings, abbreviations, and phonetic substitutions, outperforming baseline models.

Contribution

The paper presents novel neural models that incorporate string and sound similarities for improved text normalization, demonstrating superior performance over baseline approaches.

Findings

Models leveraging string similarity handle misspellings and abbreviations.

Sound similarity models address phonetic substitutions and emphasis.

Proposed models achieve higher F1 scores than baselines.

Abstract

We propose neural models that can normalize text by considering the similarities of word strings and sounds. We experimentally compared a model that considers the similarities of both word strings and sounds, a model that considers only the similarity of word strings or of sounds, and a model without the similarities as a baseline. Results showed that leveraging the word string similarity succeeded in dealing with misspellings and abbreviations, and taking into account the sound similarity succeeded in dealing with phonetic substitutions and emphasized characters. So that the proposed models achieved higher F$_1$ scores than the baseline.