PDFAnno: a Web-based Linguistic Annotation Tool for PDF Documents

PDFAnno: a Web-based Linguistic Annotation Tool for PDF Documents Hiroyuki Shindo1, Yohei Munesada, Yuji Matsumoto1 1Graduate School of Information and Science Nara Institute of Science and Technology 8916-5, Takayama, Ikoma, Nara, 630-0192, Japan shindo@is.naist.jp, y.munesada@gmail.com, matsu@is.naist.jp Abstract We present PDFAnno, a web-based linguistic annotation tool for PDF documents. PDF has become widespread standard for various types of publications, however, current tools for linguistic annotation mostly focus on plain-text documents. PDFAnno offers functions for various types of linguistic annotations directly on PDF, including named entity, dependency relation, and coreference chain. Furthermore, for multi-user support, it allows simultaneous visualization of multi-user’s annotations on the single PDF, which is useful for checking inter-annotator agreement and resolving annotation conflicts. PDFAnno is freely available under open-source license at https://github.com/paperai/pdfanno. Keywords: text annotation, annotation tool, pdf 1. Introduction Gold standard annotations for texts are a prerequisite for training and evaluation of statistical models in Natural Lan- guage Processing (NLP). Since human annotation is known as one of the most costly and time-consuming tasks in NLP, an easy-to-use and easy-to-manage annotation tool is highly required for cost effective development of gold stan- dard data. Currently, general-purpose linguistic annotation tools such as BRAT (Stenetorp et al., 2012) and We- bAnno (Yimam et al., 2013) only support text documents. Some commercial software packages provide annotation functions for PDF, however, they lack a function of relation annotation suitable for dependency relation and coreference chain. Since PDF has become widespread standard for many publications, a linguistic annotation tool for PDF is strongly desired for knowledge extraction from PDF documents. For example, previous work has devel- oped an annotated corpus for coreference resolution on scientific papers (Panot et al., 2014; Schafer et al., 2012; Steven et al., 2008). In their work, PDF articles are con- verted to plain-text format using OCR software, then im- port them to a text annotation tool. As pointed out in the literature, OCR errors are present in the data and they need to clean up the text by viewing the associated PDF file. This motivates us to develop a new annotation tool that can di- rectly annotate on PDF. There are two types of annotation processes for creating an annotated text from a PDF file as shown in Figure 1. One is to convert the PDF into plain text or HTML for- mat, then annotate it using a text annotation tool, as in the previous work. Another one is to annotate the PDF di- rectly, then convert the annotated PDF into plain text for- mat. Even if annotated plain-text is eventually necessary, the latter one has at least two benefits. First, PDF is often much more readable for annotators than plain text since it is well-structured with sections and paragraphs. This helps us maintain annotation quality and consistency. Second, (a) (b) Figure 1: Annotation flows for PDF documents. (a) convert PDF into a plain text, then annotate it. (b) annotate PDF file, then convert it into a text file. the annotations become insensitive to OCR errors. That …

Content truncated to 500 words