Text normalization for low-resource languages: the case of Ligurian
This addresses text normalization for endangered low-resource languages like Ligurian, but it is incremental as it applies existing neural methods to a new language.
The paper tackled text normalization for the low-resource Ligurian language by collecting 4,394 sentence pairs and the first open-source monolingual corpus, and showed that a compact transformer model with backtranslation and tokenization achieves very low error rates.
Text normalization is a crucial technology for low-resource languages which lack rigid spelling conventions or that have undergone multiple spelling reforms. Low-resource text normalization has so far relied upon hand-crafted rules, which are perceived to be more data efficient than neural methods. In this paper we examine the case of text normalization for Ligurian, an endangered Romance language. We collect 4,394 Ligurian sentences paired with their normalized versions, as well as the first open source monolingual corpus for Ligurian. We show that, in spite of the small amounts of data available, a compact transformer-based model can be trained to achieve very low error rates by the use of backtranslation and appropriate tokenization.