Modular action language ALM for dynamic domain representation



Journal Title

Journal ISSN

Volume Title



The goal of this dissertation is to define a modular action language, ALM, for the elaboration tolerant representation of knowledge about medium-size dynamic domains. Discrete dynamic domains can be theoretically captured by transition diagrams. The problem of concisely and precisely specifying such diagrams was the subject of study for several decades. Action languages were introduced as a solution to this problem. However, traditional action languages are not suitable for the specification of medium and large size domains because they lack the means for structuring knowledge and do not thoroughly address the problem of representing objects of the domain. Language ALM addresses these issues.

The task of designing a modular action language is a relevant one in the field of AI: it is the first step towards the creation of intelligent agents capable of reasoning about, or acting in a wide variety of dynamic environments. More importantly, the design of ALM allows us to better understand how to represent knowledge. Other modular action languages exist (e.g., MAD, TAL-C), but they correspond to different intuitions and knowledge representation styles. ALM is an alternative to these languages.

The design of ALM was not a trivial task. ALM is intended to be a simple but powerful language that would allow for elegant representations of a variety of dynamic domains. Hence, our main design and evaluation criteria were its expressive power and its simplicity and elegance. Generally, there is a tension between the two, which makes it challenging to design a modular action language. Our language is intended to facilitate the creation of knowledge representation libraries, as well as the stepwise development, testing, and readability of a knowledge base. This is achieved by separating the description of a domain into two parts: a general theory, which consists of several modules that can be reused in modeling other domains, and an interpretation of this theory, which is particular to the specific domain. As well, ALM introduces classes of objects that can be described in terms of previously defined ones. Classes have attributes that are optional, a feature that contributes to the elaboration tolerance of the language. We tested that these and other features of our language accomplish our goal by modeling several domains, both commonsensical (e.g., motion) and specialized (e.g., cell division), in ALM.

We were satisfied with the formalizations that resulted and with the capabilities of our language. In this dissertation, we formally present language ALM, illustrate the methodology of its use for knowledge representation and problem solving, report on an application of ALM to question answering, and compare our language with the closest existing modular action language, MAD.



Artificial intelligence, Knowledge representation, Dynamic domains, Action languages