RoMaPla: Using t-Test for Evaluating Robustness of Marathi Plagiarism

Abstract

Identifying plagiarism of the document is a mandatory task in the academic domain. Generally online available tools are used to check plagiarism. These tools calculate similarity between the documents using a sequence of the tokens/words present in the documents which are to be compared. A semantic relationship between the words for eg., word and its synonym are treated as different, while calculating the similarity between the documents. Few tools may be available for checking the similarity of English documents. But checking the plagiarism of Marathi documents is comparatively untouched field. Information present in the Marathi language is growing due to multilingual processing. The existing MaPla (Marathi Plagiarism checker) proved that Document synset matrix for Marathi (DSMM) similarity results are near to readings observed using cognitive ability of humans and it was performed on 4 documents. To further confirm robustness of MaPla, we experimented with 24 documents to calculate the similarity between all pairs of documents using cosine measure. Thus two, 24 × 24 matrices are formulated using DSMM and manual readings. Paired t-test, which was not carried out in MaPla, proves that there is no significant difference between two matrices and hence proves the robustness of the proposed technique.