Mass argumentation and the Semantic Web
Citation: Iyad Rahwan (2008) Mass argumentation and the Semantic Web. Web Semantics: Science, Services and Agents on the World Wide Web (RSS)
DOI (original publisher): 10.1016/j.websem.2007.11.007
Semantic Scholar (metadata): 10.1016/j.websem.2007.11.007
Sci-Hub (fulltext): 10.1016/j.websem.2007.11.007
Internet Archive Scholar (search for fulltext): Mass argumentation and the Semantic Web
Download: http://www.mit.edu/~irahwan/docs/JWS2008.pdf
Tagged: Computer Science
(RSS) WWW (RSS), Semantic Web (RSS), argumentation (RSS), Web 2.0 (RSS), AIF (RSS), ontologies (RSS)
Summary
The goal of this paper is to apply this background, particularly to "identify fallacies made by dialogue participants" on the WWW. The Semantic Web is seen as a way to do this.
This paper begins by presenting a definition of argumentation, reviewing Walton's and Toulmin's analyses of arguments, and showing basic argument diagrams. It also discusses the importance, for computer modelling, of classifications of arguments, citing Walton's 1996 book, Argumentation Schemes for Presumptive Reasoning, as the most influential of these. (ed: It seems likely that the 2008 book Walton co-authored, Argumentation Schemes, ISBN:978-0521897907, would now be preferred.) Then, it presents the notion of critical questions and criteria for argument acceptability, which form part of the core basis for this work. Finally, dialogue games are mentioned in passing.
To make the limitations of current systems clear, Rahwan next reviews four Web2.0 systems for "supporting mass argumentation":
- Truthmapping
- Debatebase
- Standpoint - gone, but check archive.org for http://www.standpoint.com/
- Standpedia - gone, but check archive.org for http://www.standpedia.com/
The limitations he finds are that
- there is limited or no integration between the different argument repositories
- argument structure is relatively shallow --distinguishing only between premises and conclusions, pro and con, or the like
- hyperlinks and cross-references carry no explicit semantics -- e.g. to tell what types of arguments there are -- limiting automated search and evaluation
The Semantic Web could improve the situation in several ways:
- a shared ontology or interchange format could allow aggregation or integration between different sites
- alternately, crosswalking or mapping between different ontologies could allow for automatic translation
- semantic markup could allow explicit annotation of an argument, or distinguishing between an argument's components
Rahwan imagines argumentation both being made explicit on individual blogs, and being collected and edited en masse, perhaps at semantic wikis. Discourse DB is an example of the latter, and journalists are asked to identify whether they are for, against, or mixed on an issue. For the former, he mentions Standpoint and semantic blogging a la (What Would It Mean to Blog on the Semantic Web?, where people could indicate whether a post or link is an endorsement or rebuttal, etc.
Rahwan next describes an infrastructure based on the Argument Interchange Format (AIF) (AIF was begun in Towards an argument interchange format) and Walton's argumentation schemes (referenced earlier).
AIF
- argument network - a directed graph of arguments with two (disjoint) types of nodes:
- information nodes - passive information (e.g. claim, premise, data, etc)
- scheme nodes - indicate the schemes used
- rule of inference application (RA) node (i.e. where deductive + noon-deductive inference rules are applied)
- conflict application (CA) node (i.e. applications of criteria)
- preference application (PA) node (i.e. "applications of (possibly abstract) criteria of preference among evaluated nodes")
These are instances of AIF's schemes; those schemes resembles inference rules for deductive logic but can "include non-inductive inference". There are three (disjoint) types of schemes, named as above but, appropriately, with 'scheme' instead of 'application node'.
Then a simple argument is a tuple of
- nodes denoting premises
- one RA node
- one conclusion node
A CA-node captures the type of conflict "from one information or scheme node to another information or scheme node".
Conflict schemes
The paper also introduces the notion of "conflict scheme", which captures the form of a conflict, for instance that A2's premise negates A1's conclusion, meaning that they are incompatible.
These seem to formalize Walton's critical questions, for instance, identifying when an argument is not admissable.
Implementation
AIF-RDF is the extended ontology, as implemented with RDF and RDFS.
ArgDF is a Semantic Web-based system that uses the AIF-RDF ontology and permits the following operations:
- Creating new arguments
- Attacking and supporting premises and conclusions already in the database - thus supporting chaining and identifying conflicts
- Searching (using RQL)
- Linking existing premises to a new argument
- Attacking arguments through implicit assumptions
- Creating new schemes - WITHOUT modifying hte ontology
As the author notes "One would not expect normal Web users to use the ontology directly to author complex argument structures. A more realistic approach is to attempt to create as much annotation as possible in the background without direct user intervention" - perhaps using NLP techniques, for instance.
The paper also mentions the following:
Markup languages
- Argument Markup Language (AML), an XML-based language used in Araucaria
Ontologies
- ScholOnto (used in ClaiMaker)
- Compendium's ontology, an extension of IBIS
- Assurance and Safety Case Environment (ASCE) - uses claims, arguments and evidence to support cases and documentation making arguments about safety
Databases
- AraucariaDB - an AML-formatted corpus of analyzed and annotated arguments which can be searched using XPath, and "filtered by text, analyst, date, argument scheme, etc" (ed: not working as of 2010-June, but see http://araucaria.computing.dundee.ac.uk/ )
Theoretical and Practical Relevance
Question answering with Semantic Web technology would be one possible application, as the paper mentions.
