Changing modes of scientific discourse analysis, changing perceptions of science

摘要

New information technologies from text extraction, to data visualisation and semantic technologies, introduce a knowledge representation that reflects the view of the actors building the tools of the knowledge they are trying to represent. In the case of technologies applied to scientific knowledge, the tools thus represent a view of the core tenets, goals, and results of science, which are embodied in the standards, models and tools built to manipulate scientific data, rhetoric, and knowledge. We are interested in identifying a few trends in recent data modelling developments that, we believe, represent an increasingly `human-centric' view of scientific discourse.The primary shift that we observe is a changing focus from the textual analysis of the content described by a paper, to the author's rhetorical and pragmatic intent. An underlying knowledge representation reflected in this shift is that scientific discourse is viewed as a purposeful, persuasive text, which is created by a human actor, who aims to influence and persuade human readers. This shift represents a greater focus on discourse analysis, as opposed to 'text mining': from a previous focus on identifying entity-relationships triplets within biological abstracts (e.g. [2;4]) we now see a growing body of work that is devoted to the analysis of the rhetorical and pragmatic goals of texts (e.g.[l;3;4;8]). The shift to discourse means the recognition of the force of contextual factors in determining the epistemic value of key statements in texts. For example, statements of hypotheses and related evidence can be directly claimed by the authors of an article, or they can be indirectly claimed by others who are cited in the article. Direct and indirect claims have a different epistemic value, which it is important for a knowledge representation to reflect. They indicate the author's acceptance of the claims of others, and their proposal to have their own claims accepted by others. Both the proposal and ac-ceptance of claims are contextually bound, and are dependent as much on the act of reading as on indicators in the text. To read a text as discourse means always to be involved in an act of interpretation: that is, one where the reader interprets what (for example) the epistemic value of a statement is, or its author, or for a typical (implied) reader, or for the actual reader. Parsing discourse itself is an act of interpretation, and implicitly asserts claims regarding the epistemic values of statements in a text. In this type of analysis, it makes a great difference who or what does the parsing. When this is done using a set of tools for discourse representation (hypotheses, arguments, and other discourse relationships) that is tailored to the user's knowledge environment, the outcome is a human-technical parser/interpreter. This new type of 'pragmatic' parser could influence the way that hypotheses and their relationships to evidence are represented. This means that a tool could deliver a different value for e.g. a summary of a corpus of papers, depending on the background, interests, and belief structure the reader holds. Once we start realising the importance that individuals' minds and backgrounds have for processing the discourse they encounter, it is clear that the optimal tool for analysing scientific knowledge will be one with an inherent bias but one in which the bias is acknowledged. The user, reader, processor of discourse plays a non-negligible role in the way in which knowledge is formulated, represented, and stored. This non-objective data representation runs counter to scientists' expectations and assumptions of their own processes of communication and knowledge transfer, yet it is an obvious next step in the development of tools and patterns of scientific information technologies.