Manipuri Chunking: An Incremental Model with POS and RMWE

摘要

This paper records the work of Manipuri Chunking by using the commonly use tool of Support Vector Machine (SVM). Manipur being a very highly agglutinative language have to be careful in selecting the features for running the SVM. An experiment is being performed with 35,000 words to check whether the POS tagged and the Reduplicated Multiword Expression (RMWE) can improve the Chunk identification. With a linguistic expert the Newspaper corpus is maintained to the Gold standard. The experimental system is designed as an incremental model with notation and identification in each stage. The Chunks are identified with a list of selected features. The chunks are very much related with the Part of Speech (POS) thus in the second stage POS tagging is done with the identified Chunks as one of the features which is again followed by the chunking. The third stage is with the RMWE. An experiment is again conducted with a list of carefully selected features for the SVM in order to find the Chunk with POS and RMWE as other features. Comparisons and evaluations are performed in each phase and the final output is drawn with completely tagged chunk Manipuri text. The experiment also identifies the POS tagging with a Recall (R) of 71.97%, Precision (P) of 87.16% and F-measure (F) of 78.84%. Apart from POS it also identifies the RMWE with a Recall (R) of 89.39%, Precision (P) of 98.33% and F-measure (F) of 93.65%. The system shows a final chunking with a Recall (R) of 70.45%, Precision (P) of 86.11% and F-measure (F) of 77.50%.