Earlier postings in this recent series traced the progress in climbing the data federation pyramid to today’s current emphasis on the semantic Web. Partially this series is aimed at disabusing the notion that data extensibility can arise simply by using the XML (eXtensible Markup Language) data representation protocol. As Stonebraker and Hellerstein correctly observe:
XML is sometimes marketed as the solution to the semantic heterogeneity problem . . . . Nothing could be further from the truth. Just because two people tag a data element as a salary does not mean that the two data elements are comparable. One could be salary after taxes in French francs including a lunch allowance, while the other could be salary before taxes in US dollars. Furthermore, if you call them “rubber gloves” and I call them “latex hand protectors”, then XML will be useless in deciding that they are the same concept. Hence, the role of XML will be limited to providing the vocabulary in which common schemas can be constructed.
This series also covers the ontologies and the OWL language (written in XML) that now give us the means to understand and process these different domains and “world views” by machine. According to Natalya Noy, one of the principal researchers behind the Protégé development environment for ontologies and knowledge-based systems:
How are ontologies and the Semantic Web different from other forms of structured and semi-structured data, from database schemas to XML? Perhaps one of the main differences lies in their explicit formalization. If we make more of our assumptions explicit and able to be processed by machines, automatically or semi-automatically integrating the data will be easier. Here is another way to look at this: ontology languages have formal semantics, which makes building software agents that process them much easier, in the sense that their behavior is much more predictable (assuming they follow the specified explicit semantics–but at least there is something to follow). 
Again, however, simply because OWL (or similar) languages now give us the means to represent an ontology, we still have the vexing challenge of how to resolve the differences between different “world views,” even within the same domain. According to Alon Halevy:
When independent parties develop database schemas for the same domain, they will almost always be quite different from each other. These differences are referred to as semantic heterogeneity, which also appears in the presence of multiple XML documents, Web services, and ontologies–or more broadly, whenever there is more than one way to structure a body of data. The presence of semi-structured data exacerbates semantic heterogeneity, because semi-structured schemas are much more flexible to start with. For multiple data systems to cooperate with each other, they must understand each other’s schemas. Without such understanding, the multitude of data sources amounts to a digital version of the Tower of Babel. 
In the sections below, we describe the sources for how this heterogeneity arises and classify the many different types of heterogeneity. I then describe some broad approaches to overcoming these heterogeneities, though a subsequent post looks at that topic in more detail.
Causes and Sources of Semantic Heterogeneity
There are many potential circumstances where semantic heterogeneity may arise (partially from Halevy ):
- Enterprise information integration
- Querying and indexing the deep Web (which is a classic data federation problem in that there are literally tens to hundreds of thousands of separate Web databases) 
- Merchant catalog mapping
- Schema v. data heterogeneity
- Schema heterogeneity and semi-structured data.
Naturally, there will always be differences in how differing authors or sponsors create their own particular “world view,” which, if transmitted in XML or expressed through an ontology language such as OWL may also result in differences based on expression or syntax. Indeed, the ease of conveying these schema as semi-structured XML, RDF or OWL is in and of itself a source of potential expression heterogeneities. There are also other sources in simple schema use and versioning that can create mismatches . Thus, possible drivers in semantic mismatches can occur from world view, perspective, syntax, structure and versioning and timing:
- One schema may express a similar “world view” with different syntax, grammar or structure
- One schema may be a new version of the other
- Two or more schemas may be evolutions of the same original schema
- There may be many sources modeling the same aspects of the underlying domain (“horizontal resolution” such as for competing trade associations or standards bodies), or
- There may be many sources that cover different domains but overlap at the seams (“vertical resolution” such as between pharmaceuticals and basic medicine).
Regardless, the needs for semantic mediation are manifest, as are the ways in which semantic heterogeneities may arise.
Classification of Semantic Heterogeneities
The first known classification scheme applied to data semantics that I am aware of is from William Kent nearly 20 years ago. (If you know of earlier ones, please send me a note.) Kent’s approach dealt more with structural mapping issues (see below) than differences in meaning, which he pointed to data dictionaries as potentially solving.
The most comprehensive schema I have yet encountered is from Pluempitiwiriyawej and Hammer, “A Classification Scheme for Semantic and Schematic Heterogeneities in XML Data Sources.”  They classify heterogeneities into three broad classes:
- Structural conflicts arise when the schema of the sources representing related or overlapping data exhibit discrepancies. Structural conflicts can be detected when comparing the underlying DTDs. The class of structural conflicts includes generalization conflicts, aggregation conflicts, internal path discrepancy, missing items, element ordering, constraint and type mismatch, and naming conflicts between the element types and attribute names.
- Domain conflicts arise when the semantic of the data sources that will be integrated exhibit discrepancies. Domain conflicts can be detected by looking at the information contained in the DTDs and using knowledge about the underlying data domains. The class of domain conflicts includes schematic discrepancy, scale or unit, precision, and data representation conflicts.
- Data conflicts refer to discrepancies among similar or related data values across multiple sources. Data conflicts can only be detected by comparing the underlying DOCs. The class of data conflicts includes ID-value, missing data, incorrect spelling, and naming conflicts between the element contents and the attribute values.
Moreover, mismatches or conflicts can occur between set elements (a “population” mismatch) or attributes (a “description” mismatch).
The table below builds on Pluempitiwiriyawej and Hammer’s schema by adding the fourth major explicit category of language, leading to about 40 distinct potential sources of semantic heterogeneities:
|Generalization / Specialization|
|Internal Path Discrepancy|
|Missing Item||Content Discrepancy|
|Attribute List Discrepancy|
|DOMAIN||Schematic Discrepancy||Element-value to Element-label Mapping|
|Attribute-value to Element-label Mapping|
|Element-value to Attribute-label Mapping|
|Attribute-value to Attribute-label Mapping|
|Scale or Units|
|Data Representation||Primitive Data Type|
|ID Mismatch or Missing ID|
|LANGUAGE||Encoding||Ingest Encoding Mismatch|
|Ingest Encoding Lacking|
|Query Encoding Mismatch|
|Query Encoding Lacking|
|Parsing / Morphological Analysis Errors (many)|
|Syntactical Errors (many)|
|Semantic Errors (many)|
Most of these line items are self-explanatory, but a few may not be:
- Homonyms refer to the same name referring to more than one concept, such as Name referring to a person v. Name referring to a book
- A generalization/specialization mismatch can occur when single items in one schema are related to multiple items in another schema, or vice versa. For example, one schema may refer to “phone” but the other schema has multiple elements such as “home phone,” “work phone” and “cell phone”
- Intra-aggregation mismatches come when the same population is divided differently (Census v. Federal regions for states, or full person names v. first-middle-last, for examples) by schema, whereas inter-aggregation mismatches can come from sums or counts as added values
- Internal path discrepancies can arise from different source-target retrieval paths in two different schema (for example, hierarchical structures where the elements are different levels of remove)
- The four sub-types of schematic discrepancy refer to where attribute and element names may be interchanged between schema
- Under languages, encoding mismatches can occur when either the import or export of data to XML assumes the wrong encoding type. While XML is based on Unicode, it is important that source retrievals and issued queries be in the proper encoding of the source. For Web retrievals this is very important, because only about 4% of all documents are in Unicode, and earlier BrightPlanet provided estimates there may be on the order of 25,000 language-encoding pairs presently on the Internet
- Even should the correct encoding be detected, there are significant differences in different language sources in parsing (white space, for example), syntax and semantics that can also lead to many error types.
It should be noted that a different take on classifying semantics and integration approaches is taken by Sheth et al.  Under their concept, they split semantics into three forms: implicit, formal and powerful. Implicit semantics are what is either largely present or can easily be extracted; formal languages, though relatively scarce, occur in the form of ontologies or other descriptive logics; and powerful (soft) semantics are fuzzy and not limited to rigid set-based assignments. Sheth et al.’s main point is that first-order logic (FOL) or descriptive logic is inadequate alone to properly capture the needed semantics.
From my viewpoint, Pluempitiwiriyawej and Hammer’s  classification better lends itself to pragmatic tools and approaches, though the Sheth et al. approach also helps indicate what can be processed in situ from input data v. inferred or probabalistic matches.
Importance of Reference Standards
An attractive and compelling vision — perhaps even a likely one — is that standard reference ontologies become increasingly prevalent as time moves on and semantic mediation is seen as more of a mainstream problem. Certainly, a start on this has been seen with the use of the Dublin Core metadata initiative, and increasingly other associations, organizations, and major buyers are busy developing “standardized” or reference ontologies. Indeed, there are now more than 10,000 ontologies available on the Web. Insofar as these gain acceptance, semantic mediation can become an effort mostly at the periphery and not the core.
But, such is not the case today. Standards only have limited success and in targeted domains where incentives are strong. That acceptance and benefit threshold has yet to be reached on the Web. Until such time, a multiplicity of automated methods, semi-automated methods and gazetteers will all be required to help resolve these potential heterogeneities.