You know that feeling that you’ve come home? I recently joined a 2-day workshop at Elsevier’s Disruptive Technology Labs, Amsterdam, where an exciting group of people shared what is clearly a harmonic convergence that’s been waiting to happen for a long time. HypER: Hypotheses, Evidence & Relationships is the name we gave ourselves. From the nascent wiki homepage:
The next step in the evolution of the digital research infrastructure is the detection, navigation and analysis of Hypotheses, Evidence & Relationships in the literature: the building blocks of knowledge-level claims.
This community brings together researchers in argumentation, computational linguistics, sociology of science, hypermedia, semiotics, semantic and pragmatic web.
What makes this particularly exciting from the perspective of our Hypermedia Discourse group is that we’ve been working since 1998 on tools and theory for human-annotation of discourse relations, while meantime, the machines have been getting smarter at extracting certain forms of these from traditional scientific texts. Now, as I argued in my presentation, there’s potentially a marriage made in heaven for us to really move the knowledge infrastructure forward, out of the shadow of the printing press, into a network-native environment that takes seriously the power of the social/semantic web. Moreover, it takes seriously the pragmatic level of communication by recognising the rhetorical role that different statements play in scientific communications.
My specific argument on the human+machine annotation synergy is that no matter how smart the machines get, we will always need human annotation of hypotheses, evidence, and discourse relationships, and very high quality user interfaces (Slide 10):
Researchers read meanings into texts that are not there, and with which the author might disagree
- so we will always require manual annotation tools
- we need ways to make connections to connections
- extremely complex connections may remain the province of human sensemaking (e.g. is analogous to)
Good user interfaces will be needed
- to view, edit and navigate HypERnets, whether manually or automatically constructed
Scientific discourse is a social process
- we take huge care in our writing about how we position ourselves in relation to our peers — will we trust unsupervised machines to extract and position our more complex claims?