Abstract
Various argumentation analysis tools permit the analyst to represent functional components of an argument (e.g., data, claim, warrant, backing), how arguments are composed of subarguments and defenses against potential counterarguments, and argumentation schemes. In order to facilitate a study of argument presentation in a biomedical corpus, we have developed a hybrid scheme that enables an analyst to encode argumentation analysis within the framework of Rhetorical Structure Theory (RST), which can be used to represent the discourse structure of a text. This paper describes the hybrid representation scheme and illustrates its use for investigation of contexts that license omission of elements of an argument. The analyses given in the paper involve reconstruction of enthymemes. Defeasible argumentation schemes serve as a constraint on reconstruction. In addition, the examples illustrate several other types of contextual constraints on reconstruction of enthymemes.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
Reed and Long (1997) propose ordering heuristics designed to maximize persuasiveness. However, we suspect that maximizing persuasiveness may conflict with a goal of transparency in some cases.
While use of a corpus may provide guidance for design of an argument generation system, it would of course be advisable to evaluate the transparency of generated arguments empirically, i.e., via controlled studies with readers.
For more detail, see Green et al. (in preparation).
Each of a person’s cells contains two alleles (copies) of each gene. One allele is inherited from the egg cell of the mother, and the other allele is inherited from the sperm cell of the father. A genetic condition resulting from one/two mutated alleles is called autosomal dominant/recessive. This distinction is encoded in the domain model.
Whether one FGFR3 allele is mutated (in all cells).
An example of a type of arc that does not have a causal interpretation is an arc from a risk factor (e.g., ethnicity) to a genotype.
The computer system is not designed to draw these conclusions, i.e., its task is not to automate the reasoning of a medical provider. Its task is to (re)generate the arguments that a genetic counselor would include in a patient letter.
In addition to the components of argumentation schemes described here, the specification of argumentation schemes in Green et al. (in preparation) includes restrictions on their applicability similar in function to certain types of critical questions of presumptive argumentation schemes (Walton 1996).
From a letter in the corpus of genetic counseling letters.
The text is analyzed as two instances of the argumentation scheme, one about each parent, that have been combined for conciseness of presentation.
The argumentation schemes are formally defined in Green et al. (in preparation). They are paraphrased in this article for readability.
The intended effect of Evidence is to increase the reader’s (R’s) belief in N; the constraint on N and S is that “comprehending S increases R’s belief of N” (Mann 2005). (In this and subsequent definitions of RST relations, R refers to the reader and N and S refer to the nucleus and satellite, respectively.)
Since the backing’s function is to provide support for a warrant, an alternative approach would be to describe the backing in ArgRST terms as the satellite of a relation (e.g., Evidence) whose nucleus is the warrant. Since this would complicate the analyses, and since there are no cases in the corpus where both a backing and warrant are given explicitly in the same argument, we have decided to ignore this functional distinction in ArgRST for now.
The intended effect of Background is to increase R’s ability to comprehend N; the constraint on N and S is that S increases R’s ability “to comprehend an element in N” (Mann 2005). (This RST relation reflects the role of the Warrant in providing a link from Data to Claim. As far as we know, ArgRST is the first tool to encode Toulmin’s distinction between data and warrant in RST terms, in particular, to use RST’s Background relation to represent warrants. Kibble (2006) uses Motivation, Evidence, Background and Concession to analyze arguments, but does not use Background in the same way as in ArgRST.)
RST analyses are diagrammed as trees also in Power et al. (2003).
For a discussion of the issue of whether RST’s relations hold between spans of text or meanings see Power et al. (2003).
For our purposes, for a computational system to be practical the knowledge resources available to it must be feasible to acquire and it must be possible to make use of the knowledge in a computationally tractable representation and reasoning system.
After an opening (sentences 1–2), the letter covers the diagnosis that the child has achondroplasia (3–8), general information about typical effects of this condition on health (9–25), the probable genetic source of the child’s condition (26–31), recurrence risks of having another child with this condition (32–37), recommendations for managing the child’s health (38–44), and a closing (45–48).
By convention, since the corpus does not contain the names of actual persons, P refers to the patient.
We interpret this as argumentation, rather than explanation, since it supports a position that could be challenged by the audience, i.e., either the primary audience, the parents of P, or persons reviewing the letter to evaluate the quality of service provided by the clinic.
Conjunction is a multinuclear RST relation where the conjuncts “form a unit in which each item plays a comparable role” (Mann 2005).
The text is analyzed as multiple instances of the argumentation scheme, one for each test, that have been combined for conciseness of presentation.
According to Mann, the intended effect of Antithesis is to increase “R’s positive regard for N”; the constraint on N is that “W has positive regard for N”; the constraint on N and S is that “N and S are in contrast …; because of the incompatibility that arises from the contrast, one cannot have positive regard for both of those situations; comprehending S and the incompatibility between the situations increases R’s positive regard for N” (Mann 2005). In the following example from (Mann 2005), (12) is analyzed as the satellite and (13) as the nucleus: “(12) But I do not think endorsing a specific nuclear freeze proposal is appropriate for CCC. (13) We should limit our involvement in defense and weaponry to matters of process, such as exposing the weapons industry’s influence on the political process.”
Antithesis is defined in RST in terms of fixed roles of the reader (R) and writer (W). However, in our use of Antithesis, we have replaced the fixed roles of R and W with the roles of arguing for or against a position. Since Antithesis may be embedded in Antithesis, as shown in Fig. 4, this enables the notation to correctly represent to which party a position has been attributed. Furthermore, since in this genre, a letter may be intended for multiple audiences, we describe the party in terms of social role (e.g., Parent) for clarity.
According to Mann, the intended effect of Concession is to increase “R’s positive regard for N”; the constraint on N is that “W has positive regard for N”; the constraint on S is that “W is not claiming that S does not hold”; the constraint on N and S is that “W acknowledges a potential or apparent incompatibility between N and S; recognizing the compatibility between N and S increases R’s positive regard for N” (Mann 2005). In the following example from (Mann 2005), (2) is analyzed as the satellite and (3) as the nucleus: “(2) Tempting as it may be, (3) we should not embrace every popular issue that comes along.”
The constraint on N and S is that “S presents a restatement of the content of N, that is shorter in bulk” (Mann 2005). The definition of Preparation includes the constraint on order that “S precedes N in the text” but does not require N and S to be semantically related (Mann 2005). Although in the example shown in Fig. 6a, b S precedes N, the relation of Summary seems a better fit semantically than the relation of Preparation. An alternative to describing our example with the Summary relation is to use Elaboration, which Mann defines as follows: “S presents additional detail about the situation or some element of subject matter which is presented in N or inferentially accessible in N” (Mann 2005). In one type of Elaboration described by Mann, N presents a generalization, followed by a more specific instance in S. In our example, (3.2)–(3.4) would be analyzed as N and (5)–(7) as S of that type of Elaboration relation. While selecting the appropriate RST relation from these candidates may matter for purposes of discourse analysis, it would not change our analysis of the pair of related arguments.
We interpret these as arguments, embedded in a narrative describing the patient’s visit to the clinic, for two reasons. In this genre a patient letter serves as documentation for the medical record (Baker et al. 2002). One apparent function of (5)–(7) is to record the argument for the initial diagnosis (i.e., for the initially hypothesized cause of the patient’s symptoms) in case there are questions in the future as to the quality of care provided to the patient. Although in this case the test result confirmed the initial diagnosis, the letter from the corpus discussed in the previous section describes a case in which the initial diagnosis was not confirmed by the test results. In such a situation it more easily can be seen that the argument for the initial diagnosis provides justification for actions taken to attempt to confirm that diagnosis. Second, the argument for the initial diagnosis in (5)–(7) strengthens the argument given in (8)–(9) for the final (same) diagnosis: “(8) The most common change in this gene causing deafness is called 35delG … (9) The change found in P’s GJB2 gene was a rarer change …”. Since (3.2)–(3.4) expresses a more general form of the argument conveyed in (5)–(7), it can be construed as a summarized argument.
As represented in the domain model, hearing loss due to GJB2 mutation follows an autosomal recessive inheritance pattern, i.e., hearing loss is expected when two mutated alleles are inherited, but not when only one mutated allele is inherited.
As frequently noted, a statement such as P(A|B) = p is often confused by people without expertise in probability theory with the non-equivalent statement P(B|A) = p. In both (5) and (7) of the letter analyzed in Fig. 6b, statements of the first type are used as backing for warrants of positive influence from A to B. However, a statement of the second type would have provided more information to the reader about the strength of each of the influence relations.
References
Azar, M. 1999. Argumentative text as rhetorical structure: An application of rhetorical structure theory. Argumentation 13: 97–114.
Baker, D.L., T. Eash, J.L. Schuette, and W.R. Uhlmann. 2002. Guidelines for writing letters to patients. Journal of Genetic Counseling 11 (5): 399–418.
Druzdzel, M.J., and M. Henrion. 1993. Efficient reasoning in qualitative probabilistic networks. In Proceedings of the 11th national conference on AI (AAAI-93), 548–553.
Green, N. 2005. A Bayesian network coding scheme for annotating biomedical information presented to genetic counseling clients. Journal of Biomedical Informatics 38: 130–144.
Green, N., R. Dwight, K. Navoraphan, and B. Stadler. Natural language generation of transparent arguments for lay audiences (in preparation).
Kibble, R. 2006. Dialectical text planning. In CMNA06. 6th Workshop on computational models of natural argument. Riva del Garda (Italy).
Mann, W.C. 2005. RST web site. http://www.sfu.ca/rst/. Accessed 21 Jan 2008.
Mann, W.C., and S.A. Thompson. 1988. Rhetorical structure theory: Toward a functional theory of text organization. Text 8: 243–281.
Power, R., D. Scott, and N. Bouayad-Agha. 2003. Document structure. Computational Linguistics 29 (2): 211–260.
Reed, C., and D. Long. 1997. Content ordering in the generation of persuasive discourse. In Proceedings of 15th international joint conference on artificial intelligence. Nagoya, Japan, 1022–1027.
Reed, C., and G. Rowe. 2004. Araucaria: Software for argument analysis, diagramming and representation. International Journal of AI Tools 14 (3–4): 961–980.
Reed, C., D. Walton, and F. Macagno. 2007. Argument diagramming in logic, law and artificial intelligence. The Knowledge Engineering Review 22 (1): 87–109.
Reiter, E., and R. Dale. 2000. Building natural language generation systems. Cambridge: Cambridge University Press.
Toulmin, S.E. 1958. The uses of argument. Cambridge: Cambridge University Press.
van Eemeren, F.H., R. Grootendorst, F. Snoek Henkemans, et al. 1996. Fundamentals of argumentation theory: A handbook of historical backgrounds and contemporary developments. Mahwah: Lawrence Erlbaum Associates.
Verheij, B. 2005. Evaluating arguments based on Toulmin’s scheme. Argumentation 19: 347–371.
Walton, D. 1996. Argumentation schemes for presumptive reasoning. Mahwah: Erlbaum.
Walton, D., and C.A. Reed. 2005. Argumentation schemes and enthymemes. Synthese 145: 339–370.
Acknowledgments
This work is supported by the National Science Foundation under CAREER Award No. 0132821.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Green, N.L. Representation of Argumentation in Text with Rhetorical Structure Theory. Argumentation 24, 181–196 (2010). https://doi.org/10.1007/s10503-009-9169-4
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10503-009-9169-4