Describing semantics using MultiNet.In discussing graphs in chapter 5, §6 I began to outline the logical basis of a semantic net and showed how it intended to connect semantics to the real world. Then in chapter 6, §4, ¶6 I gave a adaptation of its use to describe the syntax of a simple sentence. This was followed immediately in §5 and §6 with the use of the semantic net to characterize the semantics of the traditional syntactic types of sentences. When we discussed the complex sentence, we showed a MultiNet version in chapter 7 §3 ¶5 followed up with another one to illustrate the atomic structure of meaning §4 ¶3. There was also a summary of the MultiNet system in §8 ¶3. Finally I decided that it would be best to confine additional MultiNet models to this excursus. MultiNet is undergirded by a semantic net and proves to serve as a very useful and natural way to relate the semantic structure to the logical structure of the predicate calculus. Although Helbig’s book is not well organized for learning the system from scratch, sufficient details are enumerated, particularly in Part II and the appendixes, to allow a persistent student to construct a grammar for the semantics of English. It must, however, be pointed out that cognative scientists and theoretical linguists have not developed the representation of knowledge and semantic universals very well at this time. In our case it will be mainly the structure of English that will inform us of the semantic structure. In science the desire is to handle the structural and conceptual requirements of any and all languages. |
Semantic features.The most basic and important reference for MultiNet is the characterization of the nodes of the net. Each node has the specification of a certain feature-attribute selected in parallel with the syntactic categories of the basic vocabulary of English (and German). These are set out on Helbig’s figure 17.1, which I reproduce here in a rather different format in a series of figures. So as not to lose sight of the syntactic basis I have added an indication of the part of speech of each major class of entity. Each feature-attribute represented by MultiNet is indicated with the two-letter abbreviation (here in italics) next to it. Next to each leaf node of each ontology is a word from the vocabulary of English that is to be taken as typical of the interpretation of words so classified. In doing this I have discovered some very important exceptions that I must make to MultiNet. The quantity nouns and certain limiting adjectives have been made into a separate part of speech, but I make them a subcategory of noun. The student will also find that some limiting adjectives must be classified as part of the determiner system. Secondly, the author has a “formal entity,” which he makes a grammatical term. Apparently this category is for proper nouns, which are his only examples. A third exception is what I call a measurement. In English (and German) this single concept is actually syntactically complex, usually consisting of a number and a unit of measurement taken together. |
| It is possible to extend these classes further and further, presumably until there is only one word remaining in the most precisely defined class. The proper nouns in English are those that have a definite reference often without the help of a determiner. When the definite article appears, it is usually considered a part of the noun. Figure 3 shows a few of these as some of the sub-classes of nouns designating living (usually movable) and discrete objects. These concepts are often divided between those that designate unique individuals and those that designate collections, or sets of those objects. On the top of the figure I have indicated what features need to be on the personal pronouns to distinguish the concepts they normally refer to. |
| The ontologies in figures 4–6 extend some of the other branches of nouns that may be represented with proper nouns. |
MultiNet’s layer attributes.Apparently MultiNet cannot commit to localizing certain attributes to a single place in the intensional layer of description. The layer attributes are used to embed conceptual objects, situations, locations, and times into this layer. These six attributes are: 1) cardinality, 2) factuality (called “facticity”), 3) generality (degree of generalization), 4) quantification, 5) reference (or determination), 6) variability. I have taken the quantification of this layer, usually expressed using a limiting adjective or identifier, to be equivalently expressed with a noun quantifier (cf. figure 1). It may well be MultiNet’s usually avoidance of syntactic derivation that caused a side-stepping of this semantic generalization. It should also be easy to see that the first and last are closely related to mathematical modeling (pre-extensional layer). These tell us how an entity is best represented algebraically: 1) as an integer or the end points of a scale, or 6) as a constant or a variable. I have combined these together in figure 7 to be represented by the features of extensionality. I have also left generality to expressed by both intensional and extensional means. It seems that in Engilsh this attribute is usually expressed by the context of the phrase involved. We have already stated our position with respect to assertion, which appears to contain the expression of “facticity.” MultiNet wants to extend the represention of reality to conceptual objects, so that unicorns are marked as unreal. Our position is that the plausibility of a scientific model of the state of our current knowledge is an unreal expectation for a natural language model. |
| The final layer attribute remaining to be disintegrated into semantic features of the language is reference. These features are all concentrated in the use to which English (and German) put the determiner. Vocabulary items from the determiner system are laid out in figure 8 along with semantic features that might be used to distinguished them. Notice that we repeat the definite article the in two places. Notice also that the quantity noun (number and non-number of figures 1 and 7) appears again here as a quantifying adjective, a term of indeterminate reference. |
| It is important for the student of MultiNet to realize that I have redefined reference for use in describing the syntax of natural language when we make distinct assertions. Because MultiNet was designed to describe knowledge as an alternative to natural language, it must define as [+Indeterminate] certain terms that are linguistically [+Determinate]. |
