Argumentative Zoning (AZ) was developed by Teufel (1999) as the first attempt to
annotate rhetorical moves in research articles automatically. Rhetorical ‘move’ refers to “a
discoursal or rhetorical unit that performs a coherent communicative function in a written or spoken discourse” (Swales, 2004, p. 228). Accepting Myers (1992) definition of argument: “any proof, demonstration, or reason that is useful for persuading the audience of the validity of a statement”, AZ was developed on the premise that “arguing is an important part of presenting an idea” (Teufel, 1999). AZ is an analysis of document
structure based on the idea that various rhetorical moves (such as critiquing existing work of others, making a goal statement, etc.) in scientific text documents form a scientific argument.
AZ analysis assumes that rhetorical pieces within the text should be classified based on the ownership of the ideas in the paper (such as new contributions, citable ideas of others, background knowledge that everybody accepts, etc.), and the sentiment towards the cited
work. The ultimate aim of Teufel’s work was to provide an intelligent library search tool for researchers that can summarise single or multiple research papers, and display a visual relationship between papers through the use of citation maps.
Argumentative zoning was built on Swales (1990) model of argumentative moves. Swales’ model is based on the analysis of journal articles representing a variety of discipline-based writing practices. Swales examined the introductions to 48 articles in the natural and social sciences, and found that most of them contain a sequence of rhetorical ‘moves’, (Create a
Research Space, CARS), which have been used to analyse text in a three-move structure.
Figure 3.3 Swale's CARS Model (RA = Research Article)
Swales (1990) articulated the move analysis, as a representation of academic research articles in terms of hierarchically organised text made up of distinct sections; each section can be subdivided into moves, and each move can be broken down into steps. Based on the figure 3.3 above, the ‘introduction’ includes three basic moves: move 1 in the beginning,
followed by move 2 and concluded by move 3 (Berkenkotter, 1989).
Move 1: Establishing a territory (establish the field in which the author works) Move 2: Establishing a niche (justify the present study by indicating a gap in
Move 3: Occupying the niche (introduce and describe the present study, own study,
by indicating what the investigation that author is reporting will accomplish for the field).
Swales (1990) argued that each of these moves can be made through one or a series of ‘steps’. Teufel took Swales’ idea as a basis: “[the] argumentative status of a certain move is visible on the surface by linguistic cues”, which means authors introduce linguistic cues (meta-discourse signals) while writing (Teufel, 1999, p. 84). These can be identified to understand and interpret the argumentative and rhetorical status of authors’ writing and their stance.
In addition to Swales’ model, argumentative zoning was built on Hyland’s system of the description of meta-discourse. Meta-discourse refers to the features of text that provide linguistic cues which engage the readers, and explicitly convey the authors’ intended meaning, expressing their viewpoint, argument and claim, and signalling their stance (Hyland, 2005). Rather than simply defining meta-discourse as ‘discourse about
discourse’, Hyland (2005) defined the concept of ‘meta-discourse’ as an important element of the document, that is not only used to organise ideas but also to relate to readers. It is an umbrella term that helps to relate the text to its context, which glues the important parts of a text together but, more significantly, it helps readers to understand existing knowledge and strategies used by other members (authors/researchers) of the subject area, as well as the writer’s stance towards these.
According to Teufel, the definition of the argumentative zones is given by the single rhetorical act, which are salient sentences. These sentences are landmark sentences that include meta-discourse cues like ‘in this paper we develop a method for’ or ‘in contrast to REFERENCE, our approach uses...’. Teufel’s particular interest is in the rhetorical status
of these landmark sentences with respect to the communicative function of the whole paper.
“AZ is independent of writing style, subject matter, and, to a certain degree, subdomain, but relies on text type specific expectations (communicative acts)” (Teufel, 1999, p. 22). Teufel’s approach takes each research paper to be one rhetorical act. She defined seven
categories, argumentative zones (as given in the figure below), which cover an entire article. This model of scientific argumentation is based on the idea that scientific articles have typical argument structures regardless of their discipline, such as expressions of the author’s stance towards other work (Teufel, 1999). Therefore, the claim is that they are not specific to a domain, but are discipline-independent, since the theory and technique of AZ has been shown to be robust and operational (Teufel, 1999).
Figure 3.4 Argumentative zones
In her work, Teufel investigated, with a corpus of 200 papers, how humans perform the analysis, and how much they agree or disagree. She found that they agree to a great extent, and how an automatic, rather shallow process can apply the analysis, based on machine learning and features of sentences. The ultimate aim of her work was to provide an
intelligent library search tool for researchers that could include the summarisation of single or multiple articles and also improved citation indexes, by means of citation maps which could help people grasp relationships between papers.
Originally, argumentative zoning was proposed for automatic summarisation and information retrieval tasks. Later, it was also used for educational purposes (Feltrim, Teufel, das Nunes, & Aluísio, 2006) and citation indexing (Teufel, 2006). Since the theory
and technique of argumentative zoning are shown to be robust and operational, subsequent work consisted of annotation experiments in different disciplines, including chemistry (Teufel, Siddharthan, & Batchelor, 2009) and biology (Mizuta, Korhonen, Mullen, & Collier, 2006).
AZ has become an influential approach to the automated summarisation of scientific articles that has been built upon by the Xerox Incremental Parser, as explained in the following section.