This section is a review of some major works centered on the task of giving argu- mentation a structure, beyond the classical formalism of premises and conclusions. Toulmin (1958), in his very influential early work, describes the practical roles of arguments in persuasive texts. Initially based on an analysis of courtroom ar- guments, Toulmin’s layout identifies six different components of an argumentative text, the first three of which are always encountered:
• Conclusion: the core claim of the argument, supported by the rest of the text;
• Ground: a base fact (which doesn’t need backing), evidence that supports a claim ;
• Warrant: a statement describing how the Conclusion can be inferred from the Grounds;
• Backing: additional support for Warrants, in case they’re not convincing enough;
• Qualifier: giving additional information about the force of the claim. This separation has been discussed extensively in following works. Some of the main criticisms expansions of the model are:
• a Warrant can be viewed as the Conclusion of its own argument, creating a nested argument structure;
• the separation between Ground and Warrant isn’t always clear. Both are claims supporting the conclusion, but ground evidence itself, being an inter- pretation of reality, can also require backing;
• only the Rebuttal component may contain claims undermining the Conclu- sion. In general the role of an eventual opponent is not properly represented, which limits the usefulness of the framework in debates.
Van Eemeren and Grootendorst (1992) initiate the pragma-dialectical theory of argumentation, treating it as a complex speech act that obeys to informal rules to remain focused and non-fallacious. A debate is viewed as an exchange of stand- points, i.e. positive or negative stance toward a particular claim, and arguments to support one of the two. When Gricean maxims (Grice, 1978) attempt to describe the underlying assumptions of the participants in a conversation, pragma-dialectics define for debaters the following rules for a critical discussion, here informally abridged:
• Freedom: Advancing standpoints or casting doubt on them is always permit- ted;
• Burden of proof: A party must defend its own standpoints on request; • Standpoint: An attack must target a standpoint actually advanced by the
other party;
• Relevance: Defense of a standpoint must relate to the standpoint;
• Unexpressed premise: If a party leaves a premise implicit, they cannot deny it; nor they can force a premise onto the other party;
• Starting point: The premises accepted as the starting point of the debate shall not be altered;
• Argument scheme: A standpoint cannot be conclusively defended without appropriate argumentation;
• Validity: Arguments should be valid, provided potentially unexpressed premises; • Closure: A failed defense should result in the retraction of a standpoint; a
conclusive defense should result in the retraction of the attack;
• Usage: Parties should remain clear and unambiguous in their arguments, and conversely be accurate in the interpretation of the opponent.
Breaking one or several of those rules induce fallacious reasoning, of which the authors describe many varieties. Threatening the opponent, for example, is a basic violation of the first rule, as putting pressure on the other party hinders their freedom to cast doubt.
The structure implied by this framework is a tree of standpoints and arguments directly linked by attacks and supports.
Freeman (1991), updated by Freeman (2011), proposes a generic structure of argumentation, where claims are separated by their stance with respect to the core claim of the argument. The text can thus be viewed as an exchange between a proponent and an opponent view, attacking or undercutting the moves of the other side and supporting their own, creating the structure of a graph.
Peldszus and Stede (2013) synthesize this view in the formalism which basic concept were presented above, enabling the annotation of argumentative texts in a lightweight fashion.
In more detail, the formalism posits that the argumentative text has a central claim, which the author can back up with statements that are in a Support re- lation to it; this is a transitive relation, leading to “serial support” in Freeman’s terms. A statement can also have multiple Supports; these can be independent (each Support works on its own) or linked (only the combination of two state- ments provides the support). Also, the scheme distinguishes between “standard” and “example” support, whose function originates from providing an illustration, or anecdotal evidence.
When the text mentions a potential objection, this segment is labeled as bear- ing the role of “opponent’s voice”; this goes back to Freeman’s insight that any argumentation, even if monological, is inherently dialectical. The segment will be in an Attack relation to another one (which represents the proponent’s voice), and the scheme distinguishes between Rebuttal (denying the validity of a claim) and Undercut (denying the relevance of a premise for a claim). When the au- thor proceeds to refute the attack, the attacking segment itself is subject to a Rebuttal or Undercut relation.
The atomic components of such an analysis are Argumentative Discourse Units, which often are larger than edus: multiple discourse segments can play a common argumentative role. In such cases, the edus are linked together by a meta-relation called Join.
For illustration, here is a short sample text, with its analysis shown in Fig- ure 3.1.
Health insurance companies should naturally cover alternative medical treatments. Not all practices and approaches that are lumped together under this term may have been proven in clinical trials, yet it’s precisely their positive effect when accompanying conventional ’western’ medical therapies that’s been demonstrated as beneficial. Besides many general practitioners offer such counselling and treatments in parallel anyway - and who would want to question their broad expertise?
[e1] Health insurance companies should naturally cover alternative medical treatments.
[e2] Not all practices and approaches that
are lumped together under
this term may have been
proven in clinical
trials,
1
[e3] yet it's precisely their positive effect when accompanying conventional 'western' medical therapies that's been demonstrated as beneficial. 2 [e4] Besides many general practitioners offer such counselling and treatments in parallel anyway - 3
[e5] and who would want to question their broad expertise? 4 5 c9 c7 c6
Figure 3.1: Argumentation structure of the example text
From a completely different background, Dung (1995) proposes a logical frame- work for simplified abstract argument structures, consisting only of claims and at- tacks. An argumentation framework (AF) is formally defined as a pair (AR, attacks)
where AR is a set of claims, and attacks : AR 7→ AR is a function describing which claims attacks which.2
Dung then defines various properties of AFs and associated lemmas, the first of which are:
• A set of claims S ∈ AR is conflict-free iff its components don’t attack each other, i.e.
∀a, b ∈ S2 ¬attacks(a, b)
• A set of claims S ∈ AR is admissible iff its components defend each other from attacks, i.e.
∀a ∈ S, (∃b ∈ AR attacks(b, a)) =⇒ (∃c ∈ S attacks(c, b))
• Any AF has at least one preferred extension, a maximal (wrt set inclusion) admissible set of claims.
The rationale behind admissible sets of claims is that a debater who leaves counter- arguments unanswered is vulnerable, and admissible sets represents a well-rounded and believable argumentation. Dung introduces a high number of descriptive prop- erties of argumentation frameworks, tying his formalism to applications in game theory, non-monotonic reasoning and logic programming. The structures remain abstract and quickly drift away from linguistic concerns.