New Developments in Ontological Semantics

In this paper we discuss ongoing activity within the approach to natural language processing known as ontological semantics, as defined in Nirenburg and Raskin (forthcoming). After a brief discussion of the principal tenets on which this approach is built, and a revision of extant implementations that have led toward its present form, we concentrate on some specific aspects that are key to the development of this approach, such as the acquisition of the semantics of lexical items and, intimately connected with this, the ontology, the central resource in this approach. Although we review the fundamentals of the approach, the focus is on practical aspects of implementation, such as the automation of static knowledge acquisition and the acquisition of scripts to enrich the ontology further. 1. Fundamentals of ontological semantics Ontological semantics is defined as “an integrated set of complex theories, methodologies, descriptions and implementations” (Nirenburg and Raskin, forthcoming), where a theory is the set of statements that determine the format of descriptions -obtained by applying certain methodologiesof the phenomena that the theory deals with. From this definition it follows that ontological semantics places strong emphasis on content rather than formalism. This characteristic has a strong impact on all four aspects of the approach, which is highly eclectic with regard to both representation and processing of content, linguistic or otherwise. It also implies, however, that a substantial effort in terms of acquisition of the resources is required prior to the successful implementation of the various processors that the approach requires. Historically, a number of research projects have contributed to bring ontological semantics into its current state. These projects have been aimed at producing robust large-scale natural language processing systems to be used in machine translation, information retrieval and extraction, text summarization, and other such tasks. Perhaps the most relevant ones are Dionysus (Monarch et al., 1989), Pangloss (Nirenburg, 1994), Mikrokosmos (Nirenburg et al., 1995), CAMBIO (Nirenburg, 2000a), and CREST (Nirenburg, 2000b). 1.1. Theory and methodologies We define theory as a set of statements that determine the format of descriptions of phenomena in the purview of the theory. A theory is effective if it comes with an explicit methodology for acquiring these descriptions. A theory associated with an application is interested in descriptions that support the work of an application. Figure 1 specifies how that schema applies to ontological semantics (the general notions are listed as headers in the four boxes; their interpretation for ontological semantics is given in the rest of the text in the boxes). Figure 1: Theory, methodologies, and applications The theory of ontological semantics includes the format and the semantics of dynamic knowledge sources, i.e., text meaning representations as well as static knowledge sources: the ontology, the fact database, the lexicons and the onomasticons as well as the generic processing architecture for analysis of meaning and its manipulation, including the generation of text off of it. The description part in ontological semantics includes all 1 The Text Meaning Representation (TMR) is the format used for representing text meaning in ontological semantics. The TMR is constructed compositionally from the meaning of the individual elements, defined either ontologically or procedurally, that constitute texts: words, bound morphemes, syntactic structures, and word, phrase, and clause order in the input text. A proper description of their construction process and format falls beyond the scope of this paper. See Nirenburg and Raskin (forthcoming) for an in-depth discussion. the knowledge sources, both static and dynamic (generic procedures for extraction, representation and manipulation of meaning), implemented to provide full coverage for a language (or languages) and the world. In practice, an ontological semantic description is always partial, covering only a subset of subject domains and sublanguages, and constantly under development, through the process of acquisition and as a side effect of the operation of any applications based on ontological semantics. The methodology of ontological semantics consists of acquisition of the static knowledge sources, discussed in section 3, and of the procedures for producing and manipulating dynamic knowledge structures.