| Summary: |
This study presents a dataset serving as a benchmark for the recognition of printed text in lexical lists using Chinese-IPA glossing. The paper provides an overview of the baseline model, transcription model, and PyLaia engines employed in the research. Furthermore, it elucidates the specific need for digitizing the aforementioned lexical lists, outlines the methodology employed for training the baseline model for layout analysis, and describes the training process of the transcription model using the ground truth data generated on Transkribus. This comprehensive approach encompasses both the images of the lexical list content and their corresponding transcriptions as input. Additionally, the study highlights the limitations of the model and identifies avenues for future development. By making this dataset openly accessible, it can be utilized by researchers seeking to digitize lexical lists using Chinese-IPA glossing. Moreover, since the model can recognize both Chinese characters and IPA symbols, it has the potential to contribute to linguistic analysis of languages documented in Chinese-IPA glossing.
|
