From Big Data to Argument Analysis and Automated Extraction: A Selective Study of Argument in the Philosophy of Animal Psychology from the Volumes of the Hathi Trust Collection

摘要

The Digging by Debating (DbyD) project aimed to identify, extract, model, map and visualiseudphilosophical arguments in very large text repositories such as the Hathi Trust. The projectudhas: 1) developed a method for visualizing points of contact between philosophy and theudsciences; 2) used topic modeling to identify the volumes, and pages within those volumes,udwhich are ‘rich’ in a chosen topic; 3) used a semiformaluddiscourse analysis technique toudmanually identify key arguments in the selected pages; 4) used the OVA argument mappingudtool to represent and map the key identified arguments and provide a framework forudcomparative analysis; 5) devised and used a novel analysis framework applied to theudmapped arguments covering role, content and source of propositions, and the importance,udcontext and meaning of arguments; 6) created a prototype tool for identifying propositions,udusing naive Bayes classifiers, and for identifying argument structure in chosen texts, usingudpropositional similarity; 7) created tools to apply topic modeling to tasks of rating similarity ofudpapers in the PhilPapers repository. The methods from 1 to 5 above, have enabled us toudlocate and extract the key arguments from each text. It is significant that, in applying theudmethods, a nonexpertudwith limited or no domain knowledge of philosophy has both identifiedudthe volumes of interest from a key ‘Big Data Set’ (Hathi Trust) AND identified key argumentsudwithin these texts. This provided several key insights about the nature and form of argumentsudin historical texts, and is a proofofconceptuddesign for a tool that will be usable by scholars.udWe have further created a dataset with which to train and test prototype tools for bothudproposition and argument extraction. Though at an early stage, these preliminary results areudpromising given the complexity of the task.udSpecifically, we have prototyped a set of tools and methods that allow scholars to moveudbetween macroscale,udglobal views of the distributions of philosophical themes in suchudrepositories, and microscaleudanalyses of the arguments appearing on specific pages in textsudbelonging to the repository. Our approach spans bibliographic analysis, science mapping,udand LDA topic modeling conducted at Indiana University and machineassistedudargumentudmarkup into Argument Interchange Format (AIF) using the OVA (Online Visualization ofudArgument) tool from the University of Dundee, where the latter has been used to analyse andudrepresent arguments by the team based at the University of East London, who alsoudperformed a detailed empirical analysis of arguments in selected texts. This work has beenudarticulated as a proof of concept tool – linked to the repository PhilPapers – designed byudmembers linked to the University of London. This project is showing for the first time how biguddata text processing techniques can be combined with deep structural analysis to provideudresearchers and students with navigation and interaction tools for engaging with the large andudrich resources provided by datasets such as the Hathi Trust and PhilPapers. Ultimately ourudefforts show how the computational humanities can bridge the gulf between the “big data”udperspective of firstgenerationuddigital humanities and the close readings of text that are theud“bread and butter” of more traditional scholarship in the humanities.