Some Applications Are More Semantic than Others

An application is the automation of a business process or processes in software. Applications range in size from the very large to the very small. An example of a very small application is a Web Services system that performs only a single function, such as currency conversion or freight rating.

Semantic Precision

Semantic precision refers to the resolution or precision of the terms in the definition of the system. For example, a business system that has a field for "date package arrived at sorting station" has a higher degree of semantic precision than one labeled "date." In the latter case it is often possible to derive an equivalent degree of semantic precision from the context; whether this is something that the application can do or whether it is up to a human interpreter is discussed later in this chapter.

Figure 3.5 shows a spreadsheet field with weak semantic precision. All that is known about field B3 is that it is currency, which could be U.S., Canadian, or Australian dollars. We infer that this number represents some form of income by its alignment and proximity to the label on A3, but we have no idea what kind of income this is (gross income, net income, etc.), nor do we know whose income it is (this is information you might glean from the file name, or worse, from the directory you found this spreadsheet in). We don't know which year this concerns, or when fiscal March starts and ends.

Figure 3.5: Spreadsheet.

Figure 3.6 shows a section of a tax return that contains high semantic precision. The form identifies itself, the individual to whom it refers, and each category of income.

Figure 3.6: A small portion of a tax return.

Semantic Referent

A referent is a person or thing to which a linguistic expression refers. An extensional referent is one where the exact physical instance is referred to directly. So "Joe Jones" is an extensional referent, as is the person with Social Security number 123-45-6789 whose tax return is excerpted in Figure 3.6.

An intensional referent is one where the final referent is not known until some rule or other indirection is resolved. Examples of intensional referents are "the president of the United States" and "the claim adjudicator for back injuries." In each case the expression must be evaluated at some time after it was originally written, with some new context (in the case of the president, with the current date). Generally, we construct these things such that they will be resolvable at the time they need to be evaluated, but that is part of the design skill.

Semantic Precision—Range

Semantic precision typically ranges from low to high along a gradient:

• General type (text, number, date, etc.)

• Type with contextual qualifier (invoice date, return reason, etc.)

• Type with context and subcontext qualifier (invoice prepared date, package cleared customs date, etc.)

• Intensional referent (admitting physician, user default location, etc.)

• Extensional referent (Dept 47, St. Anthony's Hospital, etc.)

Semantic Veracity

Semantic veracity refers to how closely represented data agree with their referents. Although they might coincide, veracity is not the same as precision. Generally, you cannot determine the veracity of data from only the data; you must also examine the system that collected and verified the data and determine how the data are used (context).

A news Web site that provides a field for address when subscribing to its mailing list has very low veracity. The site does not verify the address data, and there are no obviously negative or positive implications for providing correct information. On the other hand, a pharmaceutical Web form has higher veracity because medications must be ordered for the right person and sent to the correct address. Some effort is made to verify information, and there are obvious negative and positive consequences to providing correct information. Federal Express has a high level of veracity on address information. It delivers packages to addresses and requires a person to sign for the package, indicating validity and verification of the address.

Typical levels of semantic veracity include the following:

• None

• Syntactic validation (only numbers in a numeric field)

• Validated to controlled vocabulary (against a list of state abbreviations, American Medical Association [AMA] procedure codes, etc.)

• Validated to other internal information (part number is in our inventory, quantity on hand greater than 0)

• Corroborative validity checks (zip code is in state, procedure corresponds with diagnosis)

• Closed loop with intended originator (email verified by round trip)

• Audit—third party verified information through independent means (letters sent to creditors, transcripts verified from university, etc.)

Semanticness of Applications

The preceding was to establish what we mean by a system (or application) being "more semantically aware" than another. If a system has a high degree of semantic precision (the information in it is semantically tagged to a specific level of discernment) and the system has gone through procedures to ensure that the information is valid, the system is considered to have a high level of semantic awareness.

In Figure 3.7, a low-precision and low-veracity application is semantically unaware. This isn't necessarily bad; in fact, it is much more flexible.

Figure 3.7: Semanticness of applications.

Trusted documents have a high degree of veracity, but it is left to a human to interpret the data. A classic example is the county real estate recording office.

The computer industry got a lot of its bad reputation from systems that allowed garbage in and provided garbage out (GIGO). Often this was at a high degree of precision. This is where we have gathered a lot of detailed information with very little validation.

In summary, an application is semantically aware if it has high precision and high veracity.