Development of a web-based system for the postcorrection of historical OCR'ed texts

该项目旨在开发一个基于网络的工具和系统，用于OCR识别的历史文本的后校正。由于任何OCR结果都有其错误的份额，OCR文本输出的有用性在人文学科的许多应用程序至关重要地依赖于一个postcorrection设施。CIS开发了一个名为PoCoTo（Post-Correction Tool）的独立开源Java（非基于Web）版本的Postcorrection工具，该工具采用先进的语言技术，可以在OCR输出和原始图像的一致视图中显示具有历史拼写的文档的整个错误系列。PoCoTo已经为广大公众所知，并正在德国的数字人文项目中用于Postcorrection目的。由于其用户的具体需求，我们希望开发和分发它作为一个开源的基于Web的，多用户系统。这套系统将发展成为一个以伺服器为基础的基础设施的一部分，以支援机构的光学字符识别工作流程，而无须安装在个人电脑上。用户更正将用于计算更好的统计误差 背景中假定误差系列的分布， 从而加速校正。 2.一些简单的扩充拉丁语完整形式的词汇将提供基础 使语言技术（统计错误分析）完全适用于 对拉丁文文本的OCR输出进行后期校正。 3.系统关于OCR引擎的灵活性将通过以下方式增加： 可以处理开源OCR系统OCRopus的输出， 艾比和宇宙魔方的输出如有需要，进一步改进系统 与其他小组的合作将在《公约》范围内的一项后续建议中加以规划。 即将推出的DFG历史OCR计划。

This projects aims to develop a web based tool and system for the postcorrection ofOCR recognized historical texts. As any OCR result has its share of errors,the usefulness of OCR text output for many applications in the humanities crucially dependson a postcorrection facility. A standalone open source Java (non-web-based) version of such a postcorrectiontool named PoCoTo (Post-Correction Tool) was developed at CIS featuringadvanced language technology by which whole error series of documents withhistorical spellings can be displayed in a concordance view of OCR output and original image.PoCoTo has already made known to a wide public and is being used in Digital Humanities projectsin Germany for postcorrection purposes. Due to specific demands from its users we want to develop and distribute it as an open source web based, multi-user system. Instead of needingto be locally installed on one's own computer, it will be developed into a component of a server basedinfrastructure to support an institutional OCR workflow.Apart from this infrastructural change, the following additional goals will be pursued:1. User corrections will get used to calculate ever better statistical error profiles of putative errors series in the background, thus speeding up the correction. 2. Some simple augmentations of a Latin full form lexicon will provide the foundations to make the language technology (statistical error profiling) fully applicable to the postcorrection of the OCR output of Latin texts. 3. The flexibility of the system with regard to OCR engines will be increased by making it possible to treat the output of the open source OCR system OCRopus in addition to Abbyy and Tesseract output. Should the need arise, further improvements of the system and cooperations with other groups will be planned in a follow-up proposal within the context of the forthcoming DFG initiative for historical OCR.