Paper Info
Title
A Novel Weighted Edit Distance-Based Spelling Correction Approach for Improving the Reliability of Devanagari Script-Based P300 Speller System.
Abstract
P300 speller-based brain computer interface is a direct communication from human-brain to computer machine without any muscular movements. In conventional P300 speller, a display paradigm is used to present alphanumeric characters to users and a classification system is used to detect the target character from the acquired electroencephalographic signals. In this paper, we present an 8 x 8 matrix consisting of Devanagari characters, digits, and special characters as Devanagari script (DS)-based display paradigm. The larger size of the display paradigm as compared with conventional 6 x 6 English row/column (RC) paradigm, involvement of matras and ardha-aksharas and similar looking characters in DS increase the adjacency problem, crowding effect, fatigue, and task difficulty. This results in deteriorated performance at the classification stage. Binary differential evolution algorithm was employed for optimal channel selection and support vector machine was used to classify target verses non target stimuli for the data set collected from ten healthy subjects using the DS-based paradigm. In order to further improve the system reliability in terms of higher accuracy at word prediction level, this paper proposes a novel spelling correction approach based on weighted edit distance (WED). A custom-built dictionary was incorporated and each misspelled word was replaced by a correct word of minimum WED from it. The proposed work is based on the validation of hypothesis that most of the target-error pairs lie in the same RC. Using the proposed spelling correction approach with optimal channel selection, an average accuracy of 99% was achieved at the word prediction level. The statistical analysis carried out in this paper shows that the proposed WED-based method improves the system reliability by significantly increasing in the accuracy of word prediction. This paper also validates that the proposed method performs better as compared to the conventional edit distance-based spelling correction approach.
Year
DOI
Venue
2016
10.1109/ACCESS.2016.2614494
IEEE ACCESS
Keywords
Field
DocType
P300 speller,brain-computer interface,edit distance,EEG,SVM,spelling correction,binary DE,optimization,Devanagari,channel selection
Adjacency list,Edit distance,Alphanumeric,Devanagari,Pattern recognition,Computer science,Support vector machine,Brain–computer interface,Speech recognition,Spelling,Artificial intelligence,Statistical classification
Journal
Volume
ISSN
Citations 
4
2169-3536
0
PageRank 
References 
Authors
0.34
0
3
Name
Order
Citations
PageRank
Rahul Kumar Chaurasiya110.69
Narendra D. Londhe29813.85
Subhojit Ghosh3249.71