Action Language BC+: Preliminary Report

Joseph Babb, Joohyung LeeAAAI 2015 · DOI

Abstract: This report introduces $\mathcal{BC}+$, a powerful action language designed to bridge the gap between high-level descriptions of physical systems and modern logic programming. By combining the intuitive syntax of causal modeling with the full power of modern Answer Set Programming (ASP), $\mathcal{BC}+$ addresses limitations in previous languages regarding recursion, counting aggregates, and multi-valued action attributes.

Introduction

In Artificial Intelligence, action languages are formalisms used to describe transition systems—how actions change the state of a world. While languages like $\mathcal{B}$, $\mathcal{C}$, and $\mathcal{C}+$ have been influential, they suffer from a "capability gap." They often cannot directly utilize modern ASP features like aggregates and choice rules. Furthermore, there is a trade-off between handling recursion (as in $\mathcal{B}$) and complex causal dependencies (as in $\mathcal{C}+$). $\mathcal{BC}+$ is designed to resolve these issues by providing a unified framework based on stable model semantics.

Methodology

$\mathcal{BC}+$ is defined as a high-level notation for propositional formulas under stable model semantics. This foundation allows the language to naturally handle minimality and recursion, features that were previously difficult to reconcile with complex causal defaults.

A key feature of $\mathcal{BC}+$ is the use of multi-valued constants and action attributes. This enhances "elaboration tolerance," making it easier to add details to a model without extensive rewriting. For example, fluents and actions can be defined with explicit domains rather than being restricted to simple boolean values.

Figure 2: Definition of regular fluent and action constants with their domains, illustrating the language's support for multi-valued attributes.

The language implementation extends CPLUS2ASP, translating $\mathcal{BC}+$ causal laws into standard logic programs compatible with modern solvers like clingo.

Evaluation and Benchmarks

The language was tested against several commonsense reasoning benchmarks. In the "Blocks World" domain, $\mathcal{BC}+$ demonstrated succinctness, representing complex constraints (e.g., preventing floating blocks) in a few intuitive lines. In the "Two Switches" problem, $\mathcal{BC}+$ correctly modeled transitions where previous languages like $\mathcal{C}+$ struggled with cyclic causality.

Figure 1: A transition system diagram showing state transitions (p=f to p=t) mediated by action assignments (a=t), a fundamental concept in BC+ modeling.

Results and Discussion

Mathematically, $\mathcal{BC}+$ is a proper generalization of both $\mathcal{BC}$ and the definite fragment of $\mathcal{C}+$. It allows for recursive concepts (like InTower(B)) and the direct use of aggregates within causal laws. This makes it a highly expressive tool for modeling complex physical systems without forcing users to write raw ASP code.

Conclusion

$\mathcal{BC}+$ represents a significant step forward in action languages. While currently focused on the propositional level, it provides a robust framework for commonsense reasoning. Future work will focus on extending the language to the first-order level and exploring the integration of "knowledge-states" as seen in languages like $\mathcal{K}$.

Summary updated

BibTeX
@inproceedings{babb2015action,
  author = {Joseph Babb and Joohyung Lee},
  title = {{Action Language BC+: Preliminary Report}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2015},
  doi = {10.1609/aaai.v29i1.9405},
}
Figure