Semantic Modeling for Data

Chapter 7. Bad Semantics

Words are wonderfully elastic. They can be mispronounced, misspelled, misused or mistranslated. Even the most precise technical term can be stretched into a verb or adjective, slang or idiom, stretched all the way around until it finds itself facing itself, a mirror image, the exact opposite of itself. Mark my words, it can.

Ron Brackin

In Chapter 2, we saw how many semantic modeling languages and frameworks provide certain predefined modeling elements with a specific meaning and behavior in mind (e.g., rdfs:subClassOf for class subsumption or skos:exactMatch for entity interlinking). Nevertheless, it is not always the case that modelers follow this meaning when using the language. Defining hyponyms as synonyms, instances as classes, or nontransitive relations as transitive ones, are all examples of common semantic mistakes that lead to problematic models.

This chapter brings together the most common of these mistakes that you should anticipate when building or using a semantic model, and provides guidelines and heuristics for avoiding them.

Bad Identity

Identity in semantic modeling refers to the problem of determining whether two elements have the same meaning. Depending on the domain and kinds of elements, this can be a pretty difficult task that, if we don’t carefully address it, can lead to inaccurate semantic models that might produce erroneous inferences. Let’s see why this is the case and how we can avoid some common pitfalls.

Bad Synonymy ...