Pattern Matching Refinements to Dictionary-Based Code-Switching Point Detection

This study presents the development and evaluation of pattern matching refinements (PMRs) to automatic code switching point (CSP) detection. With all PMRs, evaluation showed an accuracy of 94.51%. This is an improvement to reported accuracy rates of dictionary-based approaches, which are in the range of 75.22%-76.26% (Yeong and Tan, 2010). In our experiments, a 100sentence Tagalog-English corpus was used as test bed. Analyses showed that the dictionary-based approach using part-ofspeech checking yielded an accuracy of 79.76% only, and two notable linguistic phenomena, (1) intra-word code-switching and (2) common words, were shown to have caused the low accuracy. The devised PMRs, namely: (1) common word exclusion, (2) common word identification, and (3) common n-gram pruning address this and showed improved accuracy. The work can be extended using audio files and machine learning with larger language resources.