STEADY, LAH! A PREDICTIVE MODEL FOR THE SINGLISH TONGUE

Abstract

Singlish, or Singapore colloquial English, is a creole language largely based on standard English but mixes vocabulary and grammar from multiple local languages and dialects – a result of its multilingual and multicultural history. This makes modern Singlish considered unique to Singaporeans and relatively difficult to decipher for foreigners and newcomers. Despite this, existing research on the linguistics of Singlish has shown that its grammar does contain predictable aspects and patterns. Therefore, this study aims to make sense of Singlish syntax and grammar through natural language processing models – namely, the discrete time Markov chain, Recurrent Neural Network, and Long Short-term Memory models – which provides a probabilistic approach to understanding of how Singlish expressions are generated. Text data utilised in this study comprises over 20,000 unique sentences, obtained from text message and online forum posts which use predominantly Singlish three models. In addition to comparing the accuracy of the models used, their ability to predict certain grammatical aspects of Singlish was also evaluated. Following this, the trained models were also used to create a simple Singlish text generation program to validate their learning. As such, the insights of this study would be helpful towards creating chatbots or predictive text completion tools suited to the Singaporean context, bolstering existing attempts within the government and private sectors. This study additionally provides a starting framework towards predicting the discourse particles of Singlish (e.g. lah, leh), which are highly nuanced with respect to purpose and sentiments expressed – from there, further study could be done on this topic. The techniques used in modelling the Singlish language would also have applications towards many other creole languages around the world, which are similarly formed historically through mixing vocabulary and grammar from different languages.