1 / 51

The Toulmin Argument Model in Artificial Intelligence Or: how semi-formal, defeasible argumentation schemes creep into

The Toulmin Argument Model in Artificial Intelligence Or: how semi-formal, defeasible argumentation schemes creep into logic. Bart Verheij Artificial Intelligence, University of Groningen, The Netherlands. The Uses of Argument. Original aim:

albert
Download Presentation

The Toulmin Argument Model in Artificial Intelligence Or: how semi-formal, defeasible argumentation schemes creep into

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. The Toulmin Argument Model in Artificial IntelligenceOr: how semi-formal, defeasible argumentation schemes creep into logic Bart Verheij Artificial Intelligence, University of Groningen, The Netherlands

  2. The Uses of Argument Original aim: ‘to criticize the assumption, made by most Anglo-American academic philosophers, that any significant argument can be put in formal terms: not just as a syllogism, since for Aristotle himself any inference can be called a ‘syllogism’ or ‘linking of statements’, but a rigidly demonstrative deduction of the kind to be found in Euclidean geometry.’

  3. The Uses of Argument ‘In no way had I set out to expound a theory of rhetoric or argumentation: my concern was with twentieth-century epistemology, not informal logic.’

  4. Toulmin’s model Hitchcock, D., & B. Verheij (eds.) (2006). Arguing on the Toulmin Model. New Essays in Argument Analysis and Evaluation.Argumentation Library, Vol. 10. Springer, Dordrecht. Hitchcock, D. & B. Verheij (2005). The Toulmin model today: Introduction to special issue of Argumentation on contemporary work using Stephen Edelston Toulmin's layout of arguments. Argumentation, Vol. 19, No. 3, pp. 255-258.

  5. Toulmin’s model

  6. Main themes of Toulmin (1958) • Argument analysis involves half a dozen distinct elements, not just two.

  7. (1) Anne is one of Jack’s sisters; All Jack’s sisters have red hair; So, Anne has red hair. (2) P(t) (x) (P(x) → Q(x)) ----------------------- Q(t)

  8. (1, backing version) Anne is one of Jack’s sisters; Each one of Jack’s sisters has (been checked individually to have) red hair; So, Anne has red hair. (1, warrant version) Anne is one of Jack’s sisters; Any sister of Jack’s will (i.e. may be taken to) have red hair; So, Anne has red hair.

  9. Toulmin's adaptation of (1): Datum: Anne is one of Jack’s sisters. Claim: Anne has red hair. Warrant: Any sister of Jack’s will (i.e. may be taken to) have red hair. Backing: All his sisters have previously been observed to have red hair. Qualifier: Presumably Rebuttal: Anne has dyed/gone white/lost her hair ...

  10. Main themes of Toulmin (1958) • Argument analysis involves half a dozen distinct elements, not just two. • Many, if not most, arguments are substantial, hence defeasible.

  11. Three warrants: A whale will be a mammal A Bermudan will be a Briton A Saudi Arabian will be a Muslim Point to similar inferential connections: Infer that a particular whale is a mammal Infer that a particular Bermudan is a Briton Infer that a particular Saudi Arabian is a Muslim

  12. But are based on different kinds of standards: A whale will be (i.e. is classifiable as) a mammal A Bermudan will be (in the eyes of the law) a Briton A Saudi Arabian will be (found to be) a Muslim Backings use: A system of taxonomical classification Statutes governing the nationality of people born in the British colonies Statistics on the distribution of religious beliefs among nationalities

  13. Main themes of Toulmin (1958) • Argument analysis involves half a dozen distinct elements, not just two. • Many, if not most, arguments are substantial, hence defeasible. • Standards of good reasoning and argument assessment are non-universal.

  14. Logic as psychology Describe an individual thinker’s thinking Logic as sociology Describe general habits and practices Logic as technology Provide recipes for rationality Logic as mathematics Find truths about logical relations Logic as jurisprudence Emphasize the cases we make for our claims

  15. Main themes of Toulmin (1958) • Argument analysis involves half a dozen distinct elements, not just two. • Many, if not most, arguments are substantial, hence defeasible. • Standards of good reasoning and argument assessment are non-universal. • Logic is to be regarded as generalised jurisprudence.

  16. The reception and refinement of Toulmin’s ideas in Artificial Intelligence 3.1 Reiter’s default rules 3.2 Pollock’s undercutting and rebutting defeaters 3.3 Prakken, Sartor & Hage on reasoning with legal rules 3.4 Dung’s admissible sets 3.5 Walton’s argumentation schemes 3.6 Reed & Rowe’s argument analysis software 3.7 Verheij’s formal reconstruction of Toulmin’s scheme

  17. The attack relation as a directed graph (Dung 1995)

  18.       Admissible sets Admissible, e.g.: {, }, {, , , , } Not admissible, e.g.: {, }, {}

  19. Dung’s types of extensions (1995) A conflict-free set of arguments is a stable extension if all arguments that are not in the set are attacked by an argument in the set. An admissible set of arguments is a preferred extension if it is an admissible set that is maximal with respect to set inclusion. A set of arguments is a complete extension if it is an admissible set that contains all arguments of which all attackers are attacked by the set. A set of arguments is a (the) grounded extension if it is a minimal complete extension.

  20. From sets to labelings (1996) A stage extension is a is a conflict free set of arguments, for which the union of the set with the set of arguments attacked by it is maximal. A set of arguments is a admissible stage extension if it is an admissible set, for which the union of the set with the set of arguments attacked by it is maximal.

  21. From sets to labelings (1996) A stage extension is a is a conflict free set of arguments, for which the union of the set with the set of arguments attacked by it is maximal. A set of arguments is a admissible stage extension if it is an admissible set, for which the union of the set with the set of arguments attacked by it is maximal. semi-stable extension (2006)

  22. Compatibility types Dialectical justification types

  23. A forest of extension types Compatibility types Dialectical justification types

  24. A forest of extension types … :-( Compatibility types Dialectical justification types

  25. From sets to labelings • A forest of extension types • Don't forget about support

  26. Pros & cons Peter has assaulted Jack Peter has assaulted Jack Police officer Jim testifies that he saw Peter assaulting Jack Police officer Anne testifies that she saw Peter not assaulting Jack

  27. Toulmin’s 1958 warrants Peter has assaulted Jack The warrant Police officers normally are right Police officer Jim testifies that he saw Peter assaulting Jack

  28. Pollock’s 1987 undercutting defeaters Peter has assaulted Jack The undercutter Jim is lying Police officer Jim testifies that he saw Peter assaulting Jack

  29. Preferred and stable extensions The notions of admissibility and preferred and stable extensions can be generalized to this setting. E.g., direct translation of admissibility: Require defense against all attacking subsets of Δ Subtle difference (admissible*): Require defense against all incompatible subsets of Δ For attack graphs: admissible = admissible*.

  30. The ‘gluing’ theorem Theorem. There is a stable extension of Δ if and only if there is a conflict-free set of sentences DA  Δ (the disambiguation) such that there is a DA-compatible admissible proof or an admissible refutation (and not both) for each element of Δ. Without *: holds for attack graphs, but not for attack-support graphs with nesting (entangled dialectical arguments) With*: holds for both. * *

  31. Example All 3-element subsets are admissible. All sentences are admissibly provable, and none is admissibly refutable1. Still there is no stable extension. But no sentence is admissible*. p2 p1 ~> q p2 ~> (q ~> q)} p1, p2 p1 q 1 Admissible refutation is here defined as attack by an admissible, and not as non-membership of the union of admissibles.

  32. From sets to labelings • A forest of extension types • Don't forget about support • Finding warrants is a knowledge engineering task

  33. Walton on argumentation schemes Generic AH a is a bad person. Therefore, a’s argument A should not be accepted. -> a semi-formal rule of inference

  34. Walton on argumentation schemes Argumentation schemes come with critical questions, e.g., for Generic AH: CQ1 Is the premise true (or well supported) that a is a bad person? CQ2 Is the allegation that a is a bad person relevant to judging a’s argument A? CQ3 Is the conclusion of the argument that A should be (absolutely) rejected even if other evidence to support A has been presented, or is the conclusion merely (the relative claim) that a should be assigned a reduced weight of credibility, relative to the total body of evidence available?

  35. Finding warrants is a knowledge engineering task 1. Determine the relevant types of sentences 2. Determine the conditional relations, i.e., the antecedents and consequents of the argumentation schemes • Determine the exceptions, i.e, the arguments against the use of the argumentation schemes • Determine the conditions of use for the argumentation schemes (Not necessarily in this order and perhaps sometimes going back to earlier steps)

  36. From sets to labelings • A forest of extension types • Don't forget about support • Finding warrants is a knowledge engineering task • How much logic helps?

  37. Argument assistants are computer programs that support argumentative tasks Analogy: Text writing assistants (aka word processing software) are computer programs that support text writing tasks Argumentation software

  38. Underlying defeasible logic • Automatic evaluation • Argument construction • Natural moves • Arguing about rules and exceptions 1999 ICAIL conference, 2003 Artificial Intelligence journal, 2005 Virtual Arguments book

  39. (RDF/OWL based)

  40. From sets to labelings • A forest of extension types • Don't forget about support • Finding warrants is a knowledge engineering task • How much logic helps? • Stories and/or arguments?

  41. A 1931 Wigmore chart Umilian was accused of murdering Jedrusik.

  42. Ten universal rules of evidence 1. The prosecution must present at least one well-shaped narrative. 2. The prosecution must present a limited set of well-shaped narratives. 3. Essential components of the narrative must be anchored. 4. Anchors for different components of the charge should be independent of each other. 5. The trier of fact should give reasons for the decision by specifying the narrative and the accompanying anchoring. 6. A fact-finder's decision as to the level of analysis of the evidence should be explained through an articulation of the general beliefs used as anchors. 7. There should be no competing story with equally good or better anchoring. 8. There should be no falsifications of the indictment's narrative and nested sub-narratives. 9. There should be no anchoring onto obviously false beliefs. 10. The indictment and the verdict should contain the same narrative. Wagenaar, W.A., van Koppen, P.J., and Crombag, H.F.M. (1993), Anchored Narratives. The Psychology of Criminal Evidence (London: Harvester Wheatsheaf).

  43. Ten universal rules of evidence 1. The prosecution must present at least one well-shaped narrative. 2. The prosecution must present a limited set of well-shaped narratives. 3. Essential components of the narrative must be anchored. 4. Anchors for different components of the charge should be independent of each other. 5. The trier of fact should give reasons for the decision by specifying the narrative and the accompanying anchoring. 6. A fact-finder's decision as to the level of analysis of the evidence should be explained through an articulation of the general beliefs used as anchors. 7. There should be no competing storywith equally good or better anchoring. 8. There should be no falsifications of the indictment's narrative and nested sub-narratives. 9. There should be no anchoring onto obviously false beliefs. 10. The indictment and the verdict should contain the same narrative. Wagenaar, W.A., van Koppen, P.J., and Crombag, H.F.M. (1993), Anchored Narratives. The Psychology of Criminal Evidence (London: Harvester Wheatsheaf).

  44. Events Chronology Evidence A mixed argumentative-narrative perspective Floris Bex (2009). Evidence for a Good Story. A Hybrid Theory of Arguments, Stories and Criminal Evidence. Dissertation, University of Groningen.

  45. From sets to labelings • A forest of extension types • Don't forget about support • Finding warrants is a knowledge engineering task • How much logic helps? • Stories and/or arguments?

More Related