Defeasible Logic Programming (DeLP)

DeLP distinguishes between two fundamental rule types that form the backbone of its knowledge representation:

• Strict Rules: Represent non-defeasible, deductive knowledge. If the premises hold, the conclusion is irrefutable. Example: "All valid contracts require consideration."
• Defeasible Rules: Represent tentative, presumptive knowledge that admits exceptions. These are the engine of non-monotonic reasoning. Example: "A signed agreement presumptively indicates a valid contract."

This dual-rule architecture allows the system to model both the rigid axiomatic structure of legal codes and the rebuttable presumptions that characterize legal argumentation.

Arguments in DeLP are not merely conclusions but structured proof trees built from the program's rules:

• An argument for a literal L is a minimal, consistent set of defeasible rules that, together with strict rules and facts, derives L.
• Minimality ensures no extraneous rules are included, preventing bloated argument structures.
• Consistency requires that the argument's supporting set does not contradict the strict knowledge base, ensuring arguments are coherent with established legal axioms.

This formal structure enables precise computational representation of legal briefs, where each claim is supported by a traceable chain of authority.

Conflict resolution is not a static priority ordering but a dynamic dialectical process:

• Counter-arguments can attack either the conclusion (rebuttal) or a supporting premise (undercut) of another argument.
• Argumentation Lines are sequences of alternating pro and con arguments, where each argument defeats its predecessor according to a defined defeat criterion.
• Dialectical Trees organize all possible argumentation lines for a given claim, with nodes representing arguments and edges representing defeat relationships.

The marking procedure then evaluates the dialectical tree bottom-up, marking nodes as defeated or undefeated based on the availability of successful counter-arguments, ultimately determining warrant.

DeLP formalizes two distinct defeat mechanisms that govern when one argument prevails over another:

• Proper Defeat: Argument A properly defeats argument B when A is strictly more specific than B on the contested point. Specificity is computed syntactically by comparing the activation sets of the conflicting rules, ensuring the more contextually-grounded argument prevails.
• Blocking Defeat: Occurs when two arguments are incomparable in specificity, resulting in a stalemate that blocks warrant for either conclusion.

This specificity-based criterion mirrors the legal principle of lex specialis derogat legi generali (the specific law derogates from the general law), providing a computationally tractable analog to judicial conflict resolution.

The ultimate output of a DeLP system is the set of warranted literals—those conclusions that survive dialectical scrutiny:

• A literal L is warranted if there exists an undefeated argument for L in the dialectical tree.
• A literal is defeated if all arguments supporting it are defeated.
• A literal is undecided if the dialectical analysis results in a blocking defeat, indicating a genuine normative ambiguity that requires external resolution.

This tripartite outcome—warranted, defeated, undecided—provides a nuanced epistemic status for each conclusion, distinguishing between proven obligations, refuted claims, and genuine legal ambiguities that may require judicial interpretation.

The DeLP reasoning cycle is implemented through an interpreter that executes the following computational loop:

• Query Input: The system receives a literal L to evaluate.
• Argument Construction: All admissible arguments for and against L are constructed from the program's strict and defeasible rules.
• Dialectical Tree Construction: The interpreter builds the complete dialectical tree rooted at the query, recursively generating counter-arguments.
• Marking and Warrant: The tree is marked according to the dialectical procedure, and the warrant status is returned.

This interpreter architecture cleanly separates the knowledge base (rules and facts) from the inference engine (argument construction and dialectical evaluation), enabling modular updates to legal rules without modifying the reasoning machinery.

A mode of reasoning where conclusions are tentative and subject to retraction when new evidence emerges. Unlike deductive reasoning, which guarantees truth preservation, defeasible reasoning models how rational agents draw prima facie conclusions that can be defeated by counterarguments.

• Key Property: Conclusions are not monotonic; adding premises can invalidate prior conclusions
• Legal Application: A contract clause may establish an obligation that is later overridden by a force majeure provision
• Contrast: Classical logic treats all conclusions as irrevocable, making it unsuitable for legal domains where rules admit exceptions

A formal structure for representing and resolving disputes between conflicting arguments. Popularized by Dung's abstract framework, it models arguments as nodes and attack relations as directed edges, enabling the computation of acceptable argument sets through semantics such as grounded, preferred, and stable extensions.

• Core Components: Arguments, attack relations, and acceptability semantics
• DeLP Integration: DeLP instantiates abstract frameworks with concrete, logic-programming-based arguments
• Dialectical Trees: Arguments are evaluated through recursive structures that track pro and con reasoning chains

A family of formal logics where the set of entailed conclusions does not necessarily grow monotonically with the addition of new premises. This property is critical for modeling common-sense and legal reasoning, where default rules admit exceptions.

• Core Formalisms: Default logic (Reiter), circumscription (McCarthy), and autoepistemic logic (Moore)
• DeLP Relationship: DeLP is a concrete non-monotonic system built on logic programming with defeasible rules
• Canonical Example: 'Birds fly' is a default rule; learning that a specific bird is a penguin retracts the flying conclusion

The process of evaluating arguments through a structured dialogue game between a proponent and an opponent. In DeLP, dialectical analysis constructs a dialectical tree where each node is an argument, and children are counterarguments that defeat their parent.

• Warranted Conclusions: A conclusion is warranted if the proponent's argument survives all defeating counterarguments without entering an infinite loop
• Marking Procedure: Leaves of the dialectical tree are marked as undefeated (U) or defeated (D), and markings propagate upward
• Computational Property: The procedure is decidable for finite programs, making it suitable for automated legal reasoning engines

The algorithmic detection and reconciliation of contradictory legal rules. When two norms prescribe incompatible actions—such as a contractual duty to deliver goods and a statutory prohibition on export—resolution strategies determine which norm prevails.

• Resolution Principles: Lex superior (higher authority prevails), lex specialis (more specific rule prevails), and lex posterior (later rule prevails)
• DeLP Mechanism: Conflicts are resolved through the specificity criterion, where more specific arguments defeat more general ones
• Contrast with SDL: Standard Deontic Logic cannot represent conflicting obligations without deriving logical contradictions; DeLP handles them natively

The formal modeling of fallback obligations that activate when a primary duty is violated. For example, a contract may specify that late delivery incurs a penalty rather than voiding the agreement entirely.

• Chisholm's Paradox: A classic puzzle demonstrating that Standard Deontic Logic cannot consistently represent CTD scenarios
• DeLP Advantage: Defeasible rules naturally encode CTD structures as defeaters that modify rather than annihilate the original obligation
• Practical Import: Essential for modeling real-world contracts, which are drafted to handle non-ideal compliance rather than assume perfect performance