OCR model for lexical lists in Chinese-IPA Glossing, Ground Truth
收藏资源简介：
The ground truth dataset for the OCR model consisted of jpg, pdf, and xml files. The training process was conducted using Transkribus, employing a PyLaia model constructed using ground truth data derived from lexical lists encompassing ten literary works that document Burmish languages. These languages include Achang, Bola, Chashan, Langsu, Leqi, and Zaiwa. The lexical lists utilized for the training phase were predominantly composed in both Chinese characters and International Phonetic Alphabet (IPA) symbols. The training was done on 311 pages and validation on 34 pages of ten lexical lists of Burmish languages on Transkribus with the default PyLaia model: (1) Achang<br> • adapted by Hill & Cooper (2020) from Dai & Cui (1985)<br> (2) Bola<br> • adapted by Hill & Cooper (2020) from He & Chen (2004)<br> (3) Bola<br> • adapted by Hill & Cooper (2020) from Dai et al. (2007)<br> (4) Chashan<br> • adapted by Hill & Cooper (2020) from Dai et al. (2010)<br> (5) Langsu<br> • adapted by Hill & Cooper (2020) from He & Chen (2004)<br> (6) Langsu<br> • adapted by Hill & Cooper (2020) from Dai (2005)<br> (7) Leqi<br> • adapted by Hill & Cooper (2020) from He & Chen (2004)<br> (8) Leqi<br> • adapted by Hill & Cooper (2020) from Dai & Li (2006)<br> (9) Leqi<br> • adapted by Hill & Cooper (2020) from Dai & Jie (2007)<br> (10) Zaiwa<br> • adapted by Hill & Cooper (2020) from Xu & Xu (1984) The material utilized for assessing the performance of the trained models is enclosed within the dataset, comprising a lexical list of the Tujia language as documented by Tian in 1986. The primary objective of the trained model was the recognition of printed lexical lists employing Chinese-IPA glossing.
