ViLexNorm: A Lexical Normalization Corpus for Vietnamese Social Media Text
This addresses the problem of processing noisy Vietnamese social media text for NLP researchers, but it is incremental as it applies an existing task to a new language.
The authors tackled lexical normalization for Vietnamese social media text by creating ViLexNorm, the first corpus for this task, comprising over 10,000 annotated sentence pairs, with the best system achieving a 57.74% Error Reduction Rate.
Lexical normalization, a fundamental task in Natural Language Processing (NLP), involves the transformation of words into their canonical forms. This process has been proven to benefit various downstream NLP tasks greatly. In this work, we introduce Vietnamese Lexical Normalization (ViLexNorm), the first-ever corpus developed for the Vietnamese lexical normalization task. The corpus comprises over 10,000 pairs of sentences meticulously annotated by human annotators, sourced from public comments on Vietnam's most popular social media platforms. Various methods were used to evaluate our corpus, and the best-performing system achieved a result of 57.74% using the Error Reduction Rate (ERR) metric (van der Goot, 2019a) with the Leave-As-Is (LAI) baseline. For extrinsic evaluation, employing the model trained on ViLexNorm demonstrates the positive impact of the Vietnamese lexical normalization task on other NLP tasks. Our corpus is publicly available exclusively for research purposes.