Early attempts to describe the content of both digital and hard copy geospatial products generally consisted of lists of features and attributes, along with Boolean conditions (most often based on inherent properties of the feature) for including them. Enumeration is time-consuming and can require much trial and error. "New" features and attributes can be difficult to incorporate into a specification and into software for building or manipulating data sets. It is often difficult to express context-sensitive conditions for feature inclusion.

There is rhyme and reason underlying good cartography that would be useful to express in a systematic way. Maps are more than collections of features--they are models of how selected geophysical "systems" are conformed, and they must reflect to rules for integrity and completeness that are difficult to capture in lists of features, even with extensive descriptions of exceptions.

We are investigating whether we can describe the content of geospatial data sets using a common underlying set of rules, which can be invoked for generation of individual coverages as well as for thinning data.

There are reasons to believe that mapping is built on abstract underlying rules. Anecdotal evidence on how cartographers train new cartographers points to abstract principles rather than lists of required features as the way practitioners organize and check their work. Cartographers can make mistakes. While there are clearly differences that can be attributed to personal style, there are many generally agreed upon, albeit frequently unspoken, criteria for quality assurance checks, and map-making institutions often have well-defined approval and release procedures.

Additional evidence can be found in the fact that, having learned to read maps, we are able to understand maps that we have never seen before. Not only can we read maps that are a lot like the ones we have previously used, we are able to read maps that are quite different from what we have seen before.

There appear to be assumptions of completeness that pertain to many kinds of maps. For example, if a map-reader locates himself on a DMA map that shows a small stream, but no major river between his current position and his chosen destination, the reader can correctly infer that no major river is there. If a major river existed there, we could probably agree on the following statements

* The map was wrong in that a river was missing

* The map user read the map correctly

Finally, there appear to be anomalies that require explanation. For example, an interstate highway may be depicted with broken lines showing construction. A road leading to a river without a bridge may include an annotation about ferry service, or information concerning whether or not the river might be forded. A requirement to explain anomalies indicates that some underlying expectation applies.

We want to find a notation that permits formal expression of the kinds of rules that we need. We hope to find a small number of rules that will make many correct predictions, and we are using the following test for adequacy of a set of rules:

For any map or geospatial data set in the series described, the set of rules should produce all of the features that need to be included, and none of those that should not be there.

The rules we have been working with do not correspond to current procedures or software; we are looking for simpler explanations for the complicated guidance we now provide cartographers and software developers. Anecdotal examples lead us to suspect that "cartographic judgment" can be explained as multiple underlying rules that work together.

We have found rules that apply to content and rules for presentation. We have focused on rules for the content of DMA products. We have found that content rules are of the following two types:

* Rules to select or "qualify" relevant data

* Rules to winnow out the most important data:

* Rules for decluttering

* Reduction of selected features to symbols that convey information otherwise conveyed in a "to scale" delineation

We are looking for abstract rules that apply to classes of features. In our rule system features and icons are lexical items, represented by generic labels. All features and the attributes associated with them are assigned to one or more relevance categories. Assignment to a relevance category is based on how we treat features and attributes in our maps and spatial data sets, rather than on taxonomic decomposition. Features and attributes may be associated with more than one relevance category and may be subject to rules based on each of them. When a rule is invoked, it applies to all features and attributes that belong to the specified relevance category.

We have been working on a set of rules to describe the contents of coverages related to transportation. We find that different rules may apply to coverages intended to support use of a particular mode of transportation than to those intended primarily for orientation; we find that the differences are not explained by the scale or "level of detail" of the overall data set.

We will present a sample set of rules to describe the content of transportation coverages and some of the things we have learned along the way. We will also present candidate criteria for determining what the "home" coverage should be (for maintenance purposes) when a feature is relevant in more than thematic coverage.