Working with Defaults in a Controlled Natural Language

Working with Defaults in a Controlled Natural Language Rolf Schwitter Department of Computing Macquarie University Sydney NSW 2109, Australia Rolf.Schwitter@mq.edu.au Abstract In this paper, we discuss how statements about defaults and various forms of excep- tions to them can be incorporated into an existing controlled natural language. We show how these defaults and exceptions are translated and represented in the an- swer set programming paradigm in order to support automated reasoning. 1 Introduction Defaults are statements in natural language that contain words such as generally, normally, or typ- ically and generalise over what a particular kind of objects does. These kinds of statements are very useful in human communication, since we often do not have complete information about the world, but must be able to draw conclusions based on incomplete information. These conclusions are preliminary, and we may be forced to withdraw them later when new information becomes avail- able. In this paper, we investigate how statements about defaults and exceptions to them can be in- corporated into an existing controlled natural lan- guage (White and Schwitter, 2009) and what kind of formal machinery is required to process these defaults and to reason with them in the answer set programming paradigm (Gelfond and Lifschitz, 1988; Lifschitz, 2008; Gebser et al., 2012). An- swer set programming (ASP) has its roots in logic programming and non-monotonic reasoning and is well suited for solving problems which involve commonsense reasoning (Eiter et al., 2009). It is important to note that we are working here with a controlled natural language (for a survey see (Kuhn, 2013)). Our controlled natural lan- guage (CNL) consists of a well defined subset of English and has been designed to serve as a knowl- edge representation language with automated rea- soning support (White and Schwitter, 2009). The CNL allows domain specialists to write a textual specification using the vocabulary of the applica- tion domain. The writing process of the CNL is guided by an intelligent authoring tool, and there is no need for the human author to formally en- code the knowledge since the language processor takes care of this process. 2 CNL and Answer Set Programming Our CNL processor translates a specification writ- ten in CNL with the help of a discourse representa- tion structure (DRS) (Schwitter, 2012) in the spirit of (Kamp and Reyle, 1993; van Eijck and Kamp, 2011) into an executable ASP program. An ASP program looks similar to a Prolog pro- gram but relies on a completely different compu- tational mechanism. Instead of deriving a solu- tion from a program specification using resolution like Prolog does, finding a solution in the ASP paradigm corresponds to computing one or more stable models (Gelfond and Lifschitz, 1988) that in principle always terminate. Stable models are also known as answer sets (Lifschitz, 2008). The building blocks of an ASP program are atomic formulas (atoms), literals and rules. A rule is an expression of the following form: L0 or ... or Lk ← Lk+1, ..., Lm, not Lm+1, ..., …

Content truncated to 500 words