Scholars are using the Web every day to search, read, collaborate, and ultimately do their research. While some of the basic activities that the scholars do, such as reading and writing papers, are already well supported in the digital world, some essential scholarly primitives, such as annotation, augmentation, contextualization, and externalization, do not yet have clear support in terms of software tools. What scholars ultimately do during their research activity is to iteratively and collaboratively create new knowledge. With the advent of the Digital Humanities, we now have the opportunity—and technology—to capture at least a part of this knowledge and make it available as machine-processable data so to be better explorable and discoverable. In this paper, we present and discuss Pundit: a novel semantic annotation tool that enables scholars to collect, annotate, and contextualize Web resources. Deep-linking is used in conjunction with an RDF-based data model to allow granular selection of content (e.g. text excerpts, image fragments). Pundit aims at enabling scholars to produce meaningful machine-readable data that captures the semantics of their annotations. By providing a customizable annotation environment, where domain specific vocabularies can be loaded, and easy ways of integrating with existing Web archives or libraries, Pundit enables users to publish their annotations and collaboratively build a semantic graph. Such a graph can be consumed via HTTP APIs and standard SPARQL, thus allowing existing Linked Data applications to easily work with the data and Web clients in general to build specific visualizations.