What remains problematic is that even if we were to have a perfect representation of phrases into triplets, this collection of sentences still do not answer the question ‘what does p53 activate?’ An important omission of this representation is that we get no grip on the validity or the epistemic value of each sentence: does it contain new experimental knowledge, created by the author; is it a citation of accepted knowledge, or is the statement purely hypothetical? In other words, what is the author intent behind the statement? [..] To be able to accept these statements and add them to a knowledgebase, a user needs to be convinced that, first of all, the author intends a statement to be a plausible claim (as opposed, for instance, to a hypothetical claim, or a disputed citation), and secondly, that there is adequate backing for this claim. So two steps are needed: first, the assignment of epistemic status to a sentence (e.g. ‘known fact’ or ‘experimental result’ or ‘hypothesis’), and secondly, a link to the evidence the author has to support her claim. [..] ..we need to know where new knowledge is presented in the text, and how this knowledge is supported by evidence, either through experiments, or through references. What we would like to have is a list of claims or hypotheses, made by specific authors, some presentation of evidence for the hypothesis, as well as relationships connecting them, concerning a) the nature of the evidence and b) the relationship to other hypotheses. [...] What is critical here is the identification of new knowledge, claimed by the authors, vs. the elements on which this knowledge is based, in terms of experimental results and references to other work, and the underlying relationships.

The shift to author intent means shifting our conceptualization of the text towards discourse: that is, a move from viewing the text as a collection of verbs and nouns, to a view of the contextualized pragmatic language used for science. [...] We call this conceptual approach ‘Hypotheses, Evidence and Relationships’ (HypER). We argue that this representation adds essential knowledge to fact extraction, by taking into account how scientific hypotheses are argued, supported by experimental findings, and how they are interconnected. We are thus arguing for the need to add the dimension of pragmatics [Schoop, 2006] to existing semantic representations. [...] To paraphrase [Hovy, 1993]: ‘As an initial assumption, we take it that scientific discourse is goal-oriented: scientists communicate for a reason.’ [...] However, discourse goals are rarely analyzed for biological texts, which is our topic of study. So what intent do biologists have? We argue that primary research articles should be treated, primarily, as persuasive texts.. [...] The author’s main goal is to persuade the reader of the validity of her claims. There are two aspects to this: the value for the author(s) and the value for the reader(s). The author puts a claim forward as having a certain value, but readers are not constrained to accept it that. The persuasiveness of the discourse lies in the authors’ attempt to persuade their readers to accept the epistemic values they put on claims. The predominant goal of scientific authors is to convince their peers of their claims, and share the epistemic values they have assigned to statements. To do this, they use rhetoric, typical to the narrative form, and supported by references and (experimental) data [Latour et al., 1997].

A. Hypothesis Creation/Identification tools – to manually or automatically create and/or extract hypotheses and relationships B. Argumentation Representation tools – to allow user interaction with the knowledge presented, and discussions between the authors/users C. Discourse Representations – for representing documents containing hypotheses and evidence D. Rhetorical relationships/argumentational schema’ s – for relations between hypotheses, and hypotheses and evidence E. Peer review tools – to validate the hypotheses, experimental descriptions and data F. System for Methodological modeling tools – to model and compare experimental methods G. Intellectual property rights management – for this disconnected set of content Combined, these elements could form the building blocks of a system that allows a user to explore the provenance of a specific claim, evaluate the data supporting it, and follow the trail of claims derived from or leading to the current claim.