Character augmentation substitutes or injects characters into the text. In other words, it creates typing errors. Therefore, the method seems counterintuitive. Still, just like noise injection in image augmentation, scholarly published papers illustrate the benefit of character augmentation in improving AI forecasting accuracy, such as Effective Character-Augmented Word Embedding for Machine Reading Comprehension by Zhuosheng Zhang, Yafang Huang, Pengfei Zhu, and Hai Zhao, from the 2018 CCF International Conference on Natural Language Processing.
The three standard methods for character augmentation are listed as follows:
- The Optical Character Recognition (OCR) augmenting function substitutes frequent errors in OCR by converting images into text, such as the letter o into the number 0 or the capital letter I into the number 1.
- The Keyboard augmenting method replaces a character with other characters that are adjacent to it. For example, a typical...