An Equipment Model and Its Role in the Interpretation of Nominal Compounds

An Equipment Model and Its Role in the Interpretation of Nominal Compounds Tomasz Ksiezyk and Ralph Grishman Department of Computer Science Courant Institute of Mathematical Sciences - New York University 251 Mercer St. New York, New York 10012 (212) 460-7446, (212) 460-7492 Abstract For natural language understanding systems designed for domains including relatively complex equipment, it is not sufficient to use general knowledge about this equipment. We show problems which can be solved only if the system has access to a detailed equipment model. We discuss features of such models, in particular, their ability to simulate the equipment's behavior. As an illustration, we describe a simulation model for an air compressor. Finally, we demonstrate how to find referents in ~is model for nominal compounds. 1. Introduction The work presented here is part of the PROTEUS (-PROtotype TExt Understanding System) system currently under development at the Courant Institute of Mathematical Sciences, New York University. l The objective of our research is to understand short natural language texts about equipment. Our texts at present are CASualty REPorts (CASREPs) which describe failures of equipment installed on Navy ships. Our initial domain is the starting air system for propulsion gas turbines. A typical CASREP consists of several sentences, for example: Unable to maintain lube oil pressure to SAC [Starting Air Compressor]. Disengaged immediately after alarm. Metal particles in oil sample and strainer. It is widely accepted among researchers that in order to achieve natural language understanding systems robust enough for practical application, it is necessary to provide them with a lot of common-sense and domain-specific knowledge. However, so far, there is no consensus as to what is the best way of choosing, organizing and using such knowledge. The novelty of the approach presented here is that, besides general knowledge about equipment, we also use a quite extensive simulation model for the specific piece of equipment which the texts deal with. We found that for understanding purposes it is more appropriate to make the simulation qualitative rather than quantative. Thus, for example, we are not interested in the precise value of oil pressure, but only whether it is too low or too high. The model is built from instances of prototypes which contain the bulk of general knowledge. It exists in the system permanently. In this situation the analysis of a piece of text consists of two stages: (1) locating in the model the objects mentioned in text; (2) interpreting the text using both the specific information residing in the model and the general knowledge which is accessible from the model. There is no clear-cut distinction between these two stages (see discussion of the examples in the next section). i An overview of the system is given in [Grishman 1986], submitted to the AAAI-86. 81 We see the following merits of having a simulation model: (a) the model provides us with a reliable background against which we can check the correctness of the understanding process on several levels: finding referents of noun phrases, assigning semantic …

Content truncated to 500 words