Zebo Xu, Prerit S. Mittal, Mohd. Mohsin Ahmed, Chandranath Adak, Zhenguang G. Cai
{"title":"Assessing penmanship of Chinese handwriting: a deep learning-based approach","authors":"Zebo Xu, Prerit S. Mittal, Mohd. Mohsin Ahmed, Chandranath Adak, Zhenguang G. Cai","doi":"10.1007/s11145-024-10531-w","DOIUrl":null,"url":null,"abstract":"<p>The rise of the digital era has led to a decline in handwriting as the primary mode of communication, resulting in negative effects on handwriting literacy, particularly in complex writing systems such as Chinese. The marginalization of handwriting has contributed to the deterioration of penmanship, defined as the ability to write aesthetically and legibly. Despite penmanship being widely acknowledged as a crucial factor in predicting language literacy, research on its evaluation remains limited, with existing assessments primarily dependent on expert subjective ratings. Recent initiatives have started to explore the application of convolutional neural networks (CNN) for automated penmanship assessment. In this study, we adopted a similar approach, developing a CNN-based automatic assessment system for penmanship in traditional Chinese handwriting. Utilizing an existing database of 39,207 accurately handwritten characters (penscripts) from 40 handwriters, we had three human raters evaluate each penscript’s penmanship on a 10-point scale and calculated an average penmanship score. We trained a CNN on 90% of the penscripts and their corresponding penmanship scores. Upon testing the CNN model on the remaining 10% of penscripts, it achieved a remarkable performance (overall 9.82% normalized Mean Absolute Percentage Error) in predicting human penmanship scores, illustrating its potential for assessing handwriters’ penmanship. To enhance accessibility, we developed a mobile application based on the CNN model, allowing users to conveniently evaluate their penmanship.</p>","PeriodicalId":48204,"journal":{"name":"Reading and Writing","volume":null,"pages":null},"PeriodicalIF":2.0000,"publicationDate":"2024-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Reading and Writing","FirstCategoryId":"95","ListUrlMain":"https://doi.org/10.1007/s11145-024-10531-w","RegionNum":2,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"EDUCATION & EDUCATIONAL RESEARCH","Score":null,"Total":0}
引用次数: 0
Abstract
The rise of the digital era has led to a decline in handwriting as the primary mode of communication, resulting in negative effects on handwriting literacy, particularly in complex writing systems such as Chinese. The marginalization of handwriting has contributed to the deterioration of penmanship, defined as the ability to write aesthetically and legibly. Despite penmanship being widely acknowledged as a crucial factor in predicting language literacy, research on its evaluation remains limited, with existing assessments primarily dependent on expert subjective ratings. Recent initiatives have started to explore the application of convolutional neural networks (CNN) for automated penmanship assessment. In this study, we adopted a similar approach, developing a CNN-based automatic assessment system for penmanship in traditional Chinese handwriting. Utilizing an existing database of 39,207 accurately handwritten characters (penscripts) from 40 handwriters, we had three human raters evaluate each penscript’s penmanship on a 10-point scale and calculated an average penmanship score. We trained a CNN on 90% of the penscripts and their corresponding penmanship scores. Upon testing the CNN model on the remaining 10% of penscripts, it achieved a remarkable performance (overall 9.82% normalized Mean Absolute Percentage Error) in predicting human penmanship scores, illustrating its potential for assessing handwriters’ penmanship. To enhance accessibility, we developed a mobile application based on the CNN model, allowing users to conveniently evaluate their penmanship.
期刊介绍:
Reading and writing skills are fundamental to literacy. Consequently, the processes involved in reading and writing and the failure to acquire these skills, as well as the loss of once well-developed reading and writing abilities have been the targets of intense research activity involving professionals from a variety of disciplines, such as neuropsychology, cognitive psychology, psycholinguistics and education. The findings that have emanated from this research are most often written up in a lingua that is specific to the particular discipline involved, and are published in specialized journals. This generally leaves the expert in one area almost totally unaware of what may be taking place in any area other than their own. Reading and Writing cuts through this fog of jargon, breaking down the artificial boundaries between disciplines. The journal focuses on the interaction among various fields, such as linguistics, information processing, neuropsychology, cognitive psychology, speech and hearing science and education. Reading and Writing publishes high-quality, scientific articles pertaining to the processes, acquisition, and loss of reading and writing skills. The journal fully represents the necessarily interdisciplinary nature of research in the field, focusing on the interaction among various disciplines, such as linguistics, information processing, neuropsychology, cognitive psychology, speech and hearing science and education. Coverage in Reading and Writing includes models of reading, writing and spelling at all age levels; orthography and its relation to reading and writing; computer literacy; cross-cultural studies; and developmental and acquired disorders of reading and writing. It publishes research articles, critical reviews, theoretical papers, and case studies. Reading and Writing is one of the most highly cited journals in Education, Educational Research, and Educational Psychology.