Date of Award


Degree Type


Degree Name

Doctor of Philosophy




David Heap

Second Advisor

Robert Mercer


Recent advances in the biomedical sciences have led to an enormous increase in the amount of research literature being published, most of it in electronic form; researchers are finding it difficult to keep up-to-date on all of the new developments in their fields. As a result there is a need to develop automated Text Mining tools to filter and organize data in a way which is useful to researchers. Human-annotated data are often used as the ‘gold standard’ to train such systems via machine learning methods.

This thesis reports on a project where three annotators applied two Models of rhetoric (argument) to a corpus of on-line biomedical research texts. How authors structure their argumentation and which rhetorical strategies they employ are key to how researchers present their experimental results; thus rhetorical analysis of a text could allow for the extraction of information which is pertinent for a particular researcher’s purpose. The first Model stems from previous work in Computational Linguistics; it focuses on differentiating ‘new’ from ‘old’ information, and results from analysis of results. The second Model is based on Toulmin’s argument structure (1958/2003); its main focus is to identify ‘Claims’ being made by the authors, but it also differentiates between internal and external evidence, as well as categories of explanation and implications of the current experiment.

In order to properly train automated systems, and as a gauge of the shared understanding of the argument scheme being applied, inter-annotator agreement should be relatively high. The results of this study show complete (three-way) inter-annotator agreement on

an average of 60.5% of the 400 sentences in the final corpus under Model 1, and 39.3% under Model 2. Analyses of the inter-annotator variation are done in order to examine in detail all of the factors involved; these include particular Model categories, individual annotator preferences, errors, and the corpus data itself. In order to reduce this inter­ annotator variation, revisions to both Models are suggested; also it is recommended that in the future biomedical domain experts, possibly in tandem with experts in rhetoric, be used as annotators.



To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.