A statistical-based approach to word alignment involving automatically projecting part-of-speech (POS) tags is presented. The approach is referred to as the "lazy man's way" because it improves POS assignment for a resource-poor language by exploiting its similarity to a resource-rich one. This unsupervised learning method combines the N-gram and Dice Coefficient similarity functions in order to align English texts with Malay texts thus projecting the POS tags from English to Malay. It is a quick method that does not require the laborious effort needed to annotate the Malay dataset. A case study, an experiment done on 25 terrorism news articles written in Malay, has shown that leveraging pre-existing resources from a resource-rich language, i.e. English, to supplement a resource-poor language, i.e. Malay, is feasible and avoids building new text-processing tools from scratch. The system was tested on the Malay corpus, consisting of 5413 word tokens. The results reached values of 86.87% for precision, 72.56% for recall and 79.07% for F1-Score. This shows that the "lazy man's way", where a resource-poor language just exploits the rich linguistic information available in English, increases bitext projection accuracy significantly.
展开▼