Guidelines for Deciding What Concepts to Put In

It is very important in building an ontology to consider the ontological status of a symbol and decide whether the symbol should be a concept in the ontology. Often, a quick and easy solution to a problem in representing the meaning of a word or in getting the language processor to work on an input is to throw a symbol into the ontology as a concept. An ontology builder must resist this temptation and add only those symbols as concepts that deserve to be concepts in the ontology. In addition, an ontology developer is often under pressure from a lexicographer and sometimes also from an analyzer builder to provide a ``concept'' for a symbol that is not really a concept or not really different from another concept or other representational apparatus that is already in the system. Reckless addition of dubious ``concepts'' will inevitably lead to a database of poor ontological quality.

The following guidelines help in deciding whether a symbol is a concept or instance or belongs to one of the other three classes of symbols above:

1. Do not add instances as concepts in the ontology. For example, IBM and DEC (as CORPORATIONs) are not concepts in the ontologies. They should be added as instances of appropriate concepts. Rules of thumb for distinguishing an instance from a concept are listed below.
2. Do not decompose and add further concepts just because you can. If we do not need a concept (or some slot or facet or filler for a concept) for our immediate goals in language processing, we do not want to put that concept in. Because ontologies are largely unbounded, it is important for us to focus our efforts on building those parts that we need immediately.
3. For example, EVENTs like BUY or MARKETING can be decomposed to a great extent. However, unless we have an indication that we need the details of such decomposition for language processing, we do not decompose the EVENTs.
4. Do not add a concept if there is already one ``close'' to it or slightly more general than the one being considered. Consider the expressiveness of the representation provided by gradations (i.e., Attribute values) before adding separate concepts.
5. For example, we do not need separate concepts for suggest, urge, and order. They are all gradations of the same concept, a directive-act, with various degrees of force which can be captured in an appropriate Attribute.
6. Do not add specialized EVENTs with particular arguments as new concepts (unless they are significantly different from the existing concepts).
7. For example, we do not need separate concepts for walk-to-airport-terminal and walk-to-parking-lot.
8. Certain elements of text meaning such as aspect, Time-Relations, attitudes and so on that are instance-specific belong only in the TMRs. Such details do not have a place in the ontology (except in some cases of complex EVENTs and the Relations between their subevents).
9. For example, Breakfast is probably a concept in the ontology (and a subclass of Meal, say) but a meal that happened at 3 O'clock on a particular day is not a separate concept in the ontology.
10. There is another view of the ontology where all the primitives in text meaning representation must be part of the ontology. In this view, primitive symbols used in representing time, aspect, and attitude for example must be in the ontology. We consider such primitive symbols to be in the ontology implicitly but we do not add them as concepts in the ontology. In other words, we are taking the practical approach of not adding concepts that will never be used. The analyzer will never access the concepts for time and aspect from the ontology and therefore we do not care to put them in the ontology.
11. There is a catch to this however. In any language, there will be a word for time and a word for aspect. When the lexicon of that language wants to map such words to concepts in the ontology, we will have to have those concepts in the ontology. Does this mean we should be adding those concepts anyway? Or does it mean those words should be mapped to slots of other existing concepts? (See below for a discussion various ways of mapping the meaning of a word.) For example, the word for time in a language can be mapped to the time slot of some EVENT. If this is done, we do not need to add time, etc. as concepts in the ontology.
12. One must also remember that ontologies are supposed to be language independent. As such, if any part of a meaning representation is specific to a particular language (such as that a particular `EVENT' does not have a transitive form in English) that part does not belong in the ontology.
13. The analyzer and the TMR have a SET notation. Hence, there is no need to create ontological concepts for collections of different types of things in the world.
14. For example, there is no concept standing for a collection of chemical products, veterinary medicines, and cough drops, though such a collection may appear in a text.