Posted:September 12, 2011

Judgment for Semantic TechnologiesFive Unique Advantages for the Enterprise

There have been some notable attempts of late to make elevator pitches [1] for semantic technologies, as well as Lee Feigenbaum’s recent series on Are We Asking the Wrong Question? about semantic technologies [2]. Some have attempted to downplay semantic Web connotations entirely and to replace the pitch with Linked Data (capitalized). These are part of a history of various ways to try to make a business case around semantic approaches [3].

What all of these attempts have in common is a view — an angst, if you will — that somehow semantic approaches have not fulfilled their promise. Marketing has failed semantic approaches. Killer apps have not appeared. The public has not embraced the semantic Web consonant with its destiny. Academics and researchers can not make the semantic argument like entrepreneurs can.

Such hand wringing, I believe, is misplaced on two grounds. First, if one looks to end user apps that solely distinguish themselves by the sizzle they offer, semantic technologies are clearly not essential. There are very effective mash-up and data-intensive sites such as many of the investment sites (Fidelity, TDAmeritrade, Morningstar, among many), real estate sites (Trulia, Zillow, among many), community data sites (American FactFinder, CensusScope,, among many), shopping sites (Amazon, Kayak, among many), data visualization sites (Tableau, Factual, among many), etc. , etc., that work well, are intuitive and integrate much disparate information. For the most part, these sites rely on conventional relational database backends and have little semantic grounding. Effective data-intensive sites do not require semantics per se [4].

Second, despite common perceptions, semantics are in fact becoming pervasive components of many common and conventional Web sites. We see natural language processing (NLP) and extraction technologies becoming common for most search services. Google and Bing sprinkle semantic results and characterizations across their standard search results. Recommendation engines and targeted ad technologies now routinely use semantic approaches. Ontologies are creeping into the commercial spaces once occupied by taxonomies and controlled vocabularies. Semantics-based suggestion systems are now the common technology used. A surprising number of smartphone apps have semantics at their core.

So, I agree with Lee Feigenbaum that we are asking the wrong question. But I would also add that we are not even looking in the right places when we try to understand the role and place of semantic technologies.

The unwise attempt to supplant the idea of semantic technologies with linked data is only furthering this confusion. Linked data is merely a means for publishing and exposing structured data. While linked data can lead to easier automatic consumption of data, it is not necessary to effective semantic approaches and is actually a burden on data publishers [5]. While that burden may be willingly taken by publishers because of its consumption advantages, linked data is by no means an essential precursor to semantic approaches. None of the unique advantages for semantic technologies noted below rely on or need to be preceded by linked data. In semantic speak, linked data is not the same as semantic technologies.

The essential thing to know about semantic technologies is that they are a conceptual and logical foundation to how information is modeled and interrelated. In these senses, semantic technologies are infrastructural and groundings, not applications per se. There is a mindset and worldview associated with the use of semantic technologies that is far more essential to understand than linked data techniques and is certainly more fundamental than elevator pitches or “killer apps.”

Five Unique Advantages

Thus, the argument for semantic technologies needs to be grounded in their foundations. It is within the five unique advantages of semantic technologies described below that the benefits to enterprises ultimately reside.

#1: Modern, Back-end Data Federation

The RDF data model — and its ability to represent the simplest of data up through complicated domain schema and vocabularies via the OWL ontology language — means that any existing schema or structure can be represented. Because of this expressiveness and flexibility, any extant data source or schema can be represented via RDF and its extensions. This breadth means that a common representation for any existing schema may be expressed. That expressiveness, in turn, means that any and all data representations can be described in a canonical way.

A shared, canonical representation of all existing schema and data types means that all of that information can now be federated and interrelated. The canonical means of federating information via the RDF data model is the foundational benefit of semantic technologies. Further, the practice of giving URIs as unique identifiers to all of the constituent items in this approach makes it perfectly suitable to today’s reality of distributed data accessible via the Web [6].

#2: Universal Solvent for Structure

I have stated many times that I have not met a form of structured data I did not like [7]. Any extant data structure or format can be represented as RDF. RDF can readily express information contained within structured (conventional databases), semi-structured (Web page or XML data streams), or unstructured (documents and images) information sources. Indeed, the use of ontologies and entity instance records in RDF is a powerful basis for driving the extraction systems now common for tagging unstructured sources.

(One of the disservices perpetuated by an insistence on linked data is to undercut this representational flexibility of RDF. Since most linked data is merely communicating value-attribute pairs for instance data, virtually any common data format can be used as the transmittal form.)

The ease of representing any existing data format or structure and the ability to extract meaningful structure from unstructured sources makes RDF a “universal solvent” for any and all information. Thus, with only minor conversion or extraction penalties, all information in its extant form can be staged and related together via RDF.

#3: Adaptive, Resilient Schema

A singular difference between semantic technologies (as we practice them) and conventional relational data systems is the use of an open world approach [8]. The relational model is a paradigm where the information must be complete and it must be described by a schema defined in advance. The relational model assumes that the only objects and relationships that exist in the domain are those that are explicitly represented in the database. This makes the closed world of relational systems a very poor choice when attempting to combine information from multiple sources, to deal with uncertainty or incompleteness in the world, or to try to integrate internal, proprietary information with external data.

Semantic technologies, on the other hand, allow domains to be captured and modeled in an incremental manner. As new knowledge is gained or new integrations occur, the underlying schema can be added to and modified without affecting the information that already exists in the system. This adaptability is generally the biggest source of economic benefits to the enterprise from semantic technologies. It is also a benefit that enables experimentation and lowers risk.

#4: Unmatched Productivity

Having all information in a canonical form means that generic tools and applications can be designed to work against that form. That, in turn, leads to user productivity and developer productivity. New datasets, structure and relationships can be added at any time to the system, but how the tools that manipulate that information behave remains unchanged.

User productivity arises from only needing to learn and master a limited number of toolsets. The relationships in the constituent datasets are modeled at the schema (that is, ontology) level. Since manipulation of the information at the user interface level consists of generic paradigms regarding the selection, view or modification of the simple constructs of datasets, types and instances, adding or changing out new data does not change the interface behavior whatsoever. The same bases for manipulating information can be applied no matter the datasets, the types of things within them, or the relationships between things. The behavior of semantic technology applications is very much akin to having generic mashups.

Developer productivity results from leveraging generic interfaces and APIs and not bespoke ones that change every time a new dataset is added to the system. In this regard, ontology-driven applications [9] arising from a properly designed semantic technology framework also work on the simple constructs of datasets, types and instances. The resulting generalization enables the developer to focus on creating logical “packages” of functionality (mapping, viewing, editing, filtering, etc.) designed to operate at the construct level, and not the level of the atomic data.

#5: Natural, Connected Knowledge Systems

All of these factors combine to enable more and disparate information to be assembled and related to one another. That, in turn, supports the idea of capturing entire knowledge domains, which can then be expanded and shifted in direction and emphasis at will. These combinations begin to finally achieve knowledge capture and representation in its desired form.

Any kind of information, any relationship between information, and any perspective on that information can be captured and modeled. When done, the information remains amenable to inspection and manipulation through a set of generic tools. Rather simple and direct converters can move that canonical information to other external forms for use by existing external tools. Similarly, external information in its various forms can be readily converted to the internal canonical form.

These capabilities are the direct opposite to today’s information silos. From its very foundations, semantic technologies are perfectly suited to capture the natural connections and nature of relevant knowledge systems.

A Summary of Advantages Greater than the Parts

There are no other IT approaches available to the enterprise that can come close to matching these unique advantages. The ideal of total information integration, both public and private, with the potential for incremental changes to how that information is captured, manipulated and combined, is exciting. And, it is achievable today.

With semantic technologies, more can be done with less and done faster. It can be done with less risk. And, it can be implemented on a pay-as-you-benefit basis [10] responsive to the current economic climate.

But awareness of this reality is not yet widespread. This lack of awareness is the result of a couple of factors. One factor is that semantic technologies are relatively new and embody a different mindset. Enterprises are only beginning to get acquainted with these potentials. Semantic technologies require both new concepts to be learned, and old prejudices and practices to be questioned.

A second factor is the semantic community itself. The early idea of autonomic agents and the heavy AI emphasis of the initial semantic Web advocacy now feels dated and premature at best. Then, the community hardly improved matters with its shift in emphasis to linked data, which is merely a technique and which completely overlooks the advantages noted above.

However, none of this likely matters. The five unique advantages for enterprises from semantic technologies are real and demonstrable today. While my crystal ball is cloudy as to how fast these realities will become understood and widely embraced, I have no question they will be. The foundational benefits of semantic technologies are compelling.

I think I’ll take this to the bank while others ride the elevator.

[1] This series was called for by Eric Franzon of Contributions to date have been provided by Sandro Hawke, David Wood, and Mark Montgomery.
[2] See Lee Feigenbaum, 2011. “Why Semantic Web Technologies: Are We Asking the Wrong Question?,” TechnicaLee Speaking blog, August 22, 2011; see, and its follow up on “The Magic Crank,” August 29, 2011; see For a further perspective on this issue from Lee’s firm, Cambridge Semantics, see Sean Martin, 2010. “Taking the Tech Out of SemTech,” presentation at the 2010 Semantic Technology Conference, June 23, 2010. See
[3] See, for example, Jeff Pollock, 2008. “A Semantic Web Business Case,” Oracle Corporation; see
[4] Indeed, many semantics-based sites are disappointingly ugly with data and triples and URIs shoved in the user’s face rather than sizzle.
[5] Linked data and its linking predicates are also all too often misused or misapplied, leading to poor quality of integrations. See, for example, M.K. Bergman and F. Giasson, 2009. “When Linked Data Rules Fail,” AI3:::Adaptive Innovation blog, November 16, 2009. See
[6] Greater elaboration on all of these advantages is provided in M. K. Bergman, 2009. “Advantages and Myths of RDF,” AI3:::Adaptive Innovation blog, April 8, 2009. See
[7] See M.K. Bergman, 2009. “‘Structs’: Naïve Data Formats and the ABox,” AI3:::Adaptive Innovation blog, January 22, 2009. See
[8] A considerable expansion on this theme is provided in M.K. Bergman, 2009. “‘The Open World Assumption: Elephant in the Room,” AI3:::Adaptive Innovation blog, December 21, 2009. See
[9] For a full expansion on this topic, see M.K. Bergman, 2011. “Ontology-driven Apps Using Generic Applications,” AI3:::Adaptive Innovation blog, March 7, 2011. See
[10] See M.K. Bergman, 2010. “‘Pay as You Benefit’: A New Enterprise IT Strategy,” AI3:::Adaptive Innovation blog, July 12, 2010. See

Posted by AI3's author, Mike Bergman Posted on September 12, 2011 at 3:11 am in Linked Data, Semantic Enterprise, Semantic Web | Comments (4)
The URI link reference to this post is:
The URI to trackback this post is:
Posted:December 6, 2010

Reference Concepts Provide DirectionAnd, Seven Guidelines for this Second of Two Semantic ‘Gaps’

I have been writing and speaking of late about next priorities to promote the interoperability of linked data and the semantic Web. In a talk a few weeks back to the Dublin Core (DCMI) annual conference, I summarized these priorities as the need to address two aspects of the semantic “gap”:

  1. One aspect is the need for vetted reference sources that provide the entities and concepts for aligning disparate content sources on the Web, and
  2. A second aspect is the need for accurate mapping predicates that can represent the often approximate matches and overlaps of this heterogeneous content.

I discussed the second aspect in an earlier post [1]. In today’s installment, we now focus on the “gap” relating to reference concepts.

The Web Increases the Need for Organization

Interoperability comes down to the nature of things and how we describe those things or quite similar things from different sources. Given the robust nature of semantic heterogeneities in diverse sources and datasets on the Web (or anywhere else, for that matter!) [2], how do we bring similar or related things into alignment? And, then, how can we describe the nature or basis of that alignment?

Of course, classifiers since Aristotle and librarians for time immemorial have been putting forward various classification schemes, controlled vocabularies and subject headings. When one wants to find related books, it is convenient to go to a central location where books about the same or related topics are clustered. And, if the book can be categorized in more than one way — as all are — then something like a card catalog is helpful to find additional cross-references. Every domain of human endeavor makes similar attempts to categorize things.

On the Web we have none of the limitations of physical books and physical libraries; locations are virtual and copies can be replicated or split apart endlessly because of the essentially zero cost of another electron. But, we still need to find things and we still want to gather related things together. According to Svenonius, “Organizing information if it means nothing else means bringing all the same information together” [3]. This sentiment and need remains unchanged whether we are talking about books, Web documents, chemical elements or linked data on the Web.

Like words or terms in human language that help us communicate about things, how we organize things on the Web needs to have an understood and definable meaning, hopefully bounded with some degree of precision, that enables us to have some confidence we are really communicating about the same something with one another. However, when applied to the Web and machine communications, the demands for how these definitions and precisions apply need to change. This makes the notion of a Web basis for organization both easier and harder than traditional approaches to classification.

It is easier because everything is virtual: we can apply multiple classification schema and can change those schema at will. We are not locked into historical anomalies like huge subject areas reserved for arcane or now historically less important topics, such as the Boer Wars or phrenology. We need not move physical books around on shelves in order to accommodate new or expanded classification schemes. We can add new branches to our classification of, say, nanotechnology as rapidly as the science advances.

Yet it is harder because we can no longer rely on the understanding of human language as a basis for naming and classifying things. Actually, of course, language has always been ambiguous, but it is manifestly more so when put through the grinder of machine processing and understanding. Machine processing of related information adds the new hurdles of no longer being able to rely on text labels (“names”) alone as the identifier of things and requires we be more explicit about our concept relationships and connections. Fortunately, here, too, much has been done in helping to organize human language through such lexical frameworks as WordNet and similar.

The Idea and Role of Reference Concepts

Many groups and individuals have been grappling with these questions of how to organize and describe information to aid interoperability in an Internet context. Among many, let me simply mention two because of the diversity their approaches show.

Bernard Vatant, for one, has with his colleagues been an advocate for some time for the need for what he calls “hubjects.” With an intellectual legacy from the Topic Maps community, the idea of “hubjects” is to have a flat space of reference subjects to which related information can link and refer. Each hubject is the hub of a spoked wheel of representations by which the same subject matter from different contexts may be linked. The idea of the flat space or neutrality in the system is to place the “hubject” identifier (referent) outside of other systems that attempt to organize and provide “meta-frameworks” of knowledge organization. In other words, there are no inherent suggested relationships in the reference “hubject” structure: just a large bin of defined subjects to which external systems may link.

A different and more formalized approach has been put forward by the FRSAD working group [4], dealing with subject authority data. Subject authority data is the type of classificatory information that deals with the subjects of various works, such as their concepts, objects, events, or places. As the group stated, the scope of this effort pertains to the “aboutness” of various conceptual works. The framework for this effort, as with the broader FRBR effort, are new standards and approaches appropriate to classifying electronic bibliographic records.

Besides one of the better summaries and introductions to the general problems of subject classification in general, the FRSAD approach makes its main contribution in clearly distinguishing the idea of something (which it calls a thema, or entity used as the subject of a work) from the name or label of something (which it calls nomen). For many in the logic community, steeped in the Peirce triad of sign-object-interpretant [5], this distinction seems rather obvious and straightforward. But, in library science, labels have been used interchangeably as identifiers, and making this distinction clean is a real contribution. The FRSAD effort does not itself really address how the thema are actually found or organized.

The notion of a reference concept used herein combines elements from both of these approaches. A reference concept is the idea of something, or a thema in the FRSAD sense. It is also a reference hub of sorts, similar to the idea of a “hubject”. But it is also much more and more fully defined.

So, let’s first begin by representing a reference concept in relation to its referers and possible linking predicates as follows:

A referer needs to link appropriately to its reference concept, with some illustrative examples shown on the arrows in the diagram. These links are the predicates, ranging from the exact to the approximate, discussed in the first semantic “gap” posting. (Note: see that earlier post for a longer listing of existing, candidate linking predicates. No further comment is made in this present article as to whether those in that earlier posting or the example ones above are “correct” or not; see the first post for that discussion.)

If properly constructed and used, a reference concept thus becomes a fixed point in an information space. As one or more external sources link to these fixed points, it is then possible to gather similar content together and to begin to organize the information space, in the sense of Svenonius. Further, and this is a key difference from the “hubject” approach, if the reference concept is itself part of a coherent structure, then additional value can be derived from these assignments, such as inference, consistence testing, and alignments. (More on this latter point is discussed below.)

Seven Guidelines for a Reference Concept

If the right factors are present, it should be possible to relate and interoperate multiple datasets and knowledge representations. If present, these factors can result in a series of fixed reference points to which external information can be linked. In turn, these reference nodes can form constellations to guide the traversal to desired information destinations on the Web.

Let’s look at the seven factors as to what constitutes guidelines for best practices.

Guideline #1: Persistent URI

By definition, a Web-based reference concept should adhere to linked data principles and should have a URI as its address and identifier. Also, by definition as a “reference”, the vocabulary or ontology in which the concept is a member should be given a permanent and persistent address. Steps should be taken to ensure 24×7 access to the reference concept’s URI, since external sources will be depending on it.

As a general rule, the concepts should also be stated as single nouns and use CamelCase notation (that is, class names should start with a capital letter and not contain any spaces, such as MyNewConcept).

Guideline #2: Preferred Label

Provide a preferred label annotation property that is used for human readable purposes and in user interfaces. For this purpose, a construct such as the SKOS property of skos:prefLabel works well. Note, this label is not the basis for deciding and making linkages, but it is essential for mouseovers, tooltips, interface labels, and other human use factors.

Guideline #3: Definition

Give all concepts and properties a definition. The matching and alignment of things is done on the basis of concepts (not simply labels), which means each concept must be defined [6]. Providing clear definitions (along with the coherency of its structure) gives an ontology its semantics. Remember not to confuse the label for a concept with its meaning. For this purpose, a property such as skos:definition works well, though others such as rdfs:comment or dc:description are also commonly used.

The definition is the most critical guideline for setting the concept’s meaning. Adequate text and content also aid semantic alignment or matching tasks.

Guideline #4: Tagset

Include explicit consideration for the idea of a “semset” or “tagset”, which means a series of alternate labels and terms to describe the concept. These alternatives include true synonyms, but may also be more expansive and include jargon, slang, acronyms or alternative terms that usage suggests refers to the same concept. The semset construct is similar to the “synsets” in Wordnet, but with a broader use understanding. Included in the semset construct is the single (per language) preferred (human-readable) label for the concept, the prefLabel, an embracing listing of alternative phrase and terms for the concept (including acronyms, synonyms, and matching jargon), the altLabels, and a listing of prominent or common misspellings for the concept or its alternatives, the hiddenLabels.

This tagset is an essential basis for tagging unstructured text documents with reference concepts, and for search not limited to keywords. The tagset, in combination with the definition, is also the basis for feeding many NLP-driven methods for concept or ontology alignment.

Guideline #5: Language Independent

The practice of using an identifier separate from label, and language qualified entries for definition, preferred label and tagset (alternative labels) means that multi-lingual versions can be prepared for each concept. Though this is a somewhat complicated best practice in its own right (for example, being attentive to the xml:lang=”en” tag for English), adhering to this practice provides language independence for reference concepts.

Sources such as Wikipedia, with its richness of concepts and multiple language versions, can then be a basis for creation of alternative language versions.

Guideline #6: Range and Domain

Use of domains and ranges assists testing, helps in disambiguation, and helps in external concept alignments. Domains apply to the subject (the left hand side of a triple); ranges to the object (the right hand side of the triple). Domains and ranges should not be understood as actual constraints, but as axioms to be used by reasoners. In general, domain for a property is the range for its inverse and the range for a property is the domain of its inverse.

Example of a Coherent Structure

Guideline #7: Part of Coherent Structure

When reference concepts, properly constructed as above, are also themselves part of a coherent structure, further benefits may be gained. These benefits include inferencing, consistency testing, discovery and navigation. For example, the sample at right shows that a retrieval for Saab cars can also inform that these are automobiles, a brand of automobile, and a Swedish kind of car.

To gain these advantages, the coherent structure need not be complicated. RDFS and SKOS-based lightweight vocabularies can meet this test. Properly constructed OWL ontologies can also provide these benefits.

When best practices are combined with being part of a coherent structure, we can refer to these structures as reference ontologies or domain ontologies.

The State of Reference Concepts

In part, these best practices are met to a greater or lesser extent by many current vocabularies. But few provide complete coverage, and across a broad swath of domain needs, major gaps remain. This unfortunate observation applies to upper-level ontologies, reference vocabularies, and domain ontologies alike.

Upper-level ontologies include the Suggested Upper Merged Ontology (SUMO), the Descriptive Ontology for Linguistic and Cognitive Engineering (DOLCE), PROTON, Cyc and BFO (Basic Formal Ontology). While these have a coherency of construction, they are most often incomplete with respect to reference concept construction. With the exception of SUMO and Cyc, domain coverage is also very general.

Our own UMBEL reference ontology [7] is closest to meeting all criteria. The reference concepts are constructed to standard. But coverage is fairly general, and not directly applicable to most domains (though it can help to orient specific vocabularies).

Wikipedia, as accessed via the DBpedia expression, has good persistent URIs, labels, altLabels and proxy definitions (via the first sentences abstract). As a repository of reference concepts, it is extremely rich. But the organizational structure is weak and provides very few of the benefits for coherent structures noted above.

Going back to 1997, DCMI has been involved in putting forward possible vocabularies that may act as “qualifiers” to dc:subject [8]. Such reference vocabularies can extend from the global or comprehensive, such as the Universal Decimal Classification or Library of Congress Subject Headings, to the domain specific such as MeSH in medicine or Agrovoc in agriculture [9]. One or more concepts in such reference vocabularies can be the object of a dc:subject assertion, for example. While these vocabularies are also a rich source of reference concepts, they are not constructed to standards and at most provide hierarchical structures.

In the area of domain vocabularies, we are seeing some good pockets of practice, especially in the biomedical and life sciences arena [10].  Promising initiatives are also underway in library applications [11] and perhaps other areas unknown to the author.

In summary, I take the state of the art to be quite promising. We know what to do, and it is being done in some pockets. What is needed now is to more broadly operationalize these practices and to extend them across more domains. If we can bring attention to and publicize exemplar vocabularies, we can start to realize the benefits of actual data interoperability on the Web.

[1] See M. K. Bergman, 2010. “The Nature of Connectedness on the Web,” AI3:::Adaptive Information blog, November 22, 2010. See
[2] See M. K. Bergman, 2006. “Sources and Classification of Semantic Heterogeneities,” AI3:::Adaptive Information blog, June 6, 2006. See
[3] From a quote on page 10 by Elaine Svenonius, 2000. The Intellectual Foundation of Information Organization, MIT Press, 2000, 255pp. I’d like to thank Karen Coyle for recently posting this quote on the Linked Library Data (LLD) mailing list.
[4] Marcia Lei Zeng, Maja Žumer, Athena Salaba, eds., 2010. Functional Requirements for Subject Authority Data (FRSAD): A Conceptual Model, prepared by the IFLA Working Group on the Functional Requirements for Subject Authority Records (FRSAR), June 2010, 75 pp. See This effort is part of the broader and well-known FRBR (Functional Requirements of Bibliographic Records) initiative.
[5] C.S. Peirce’s sign relations are covered under the discussion about Semiotic Elements under the Sign section on Peirce in Wikipedia. In the the context of this discussion, the sign corresponds to any of the labels or identifiers associated with the (reference concept) object, the meaning of which is provided by its interpretant definition and useful language labels. See also John Sowa, 2000. “Ontology, Metadata, and Semiotics,” presented at ICCS’2000 in Darmstadt, Germany, on August 14, 2000; see
[6] As another commentary on the importance of definitions, see
[7] UMBEL (Upper Mapping and Binding Exchange Layer) is an ontology of about 20,000 subject concepts that acts as a reference structure for inter-relating disparate datasets. It is also a general vocabulary of classes and predicates designed for the creation of domain-specific ontologies.
[8] Rebecca Guenther, 1997. Dublin Core Qualifiers/Substructure, October 15, 1997. See
[10] For example, see the Open Biological and Biomedical Ontologies (OBO) initiative and the W3C‘s  Semantic Web Health Care and Life Sciences Interest Group.
[11] See the W3C’s Linked Library Data initiative, with particular attention to topics and use cases.

Posted by AI3's author, Mike Bergman Posted on December 6, 2010 at 2:32 am in Linked Data, Ontologies, Semantic Web, UMBEL | Comments (0)
The URI link reference to this post is:
The URI to trackback this post is:
Posted:November 22, 2010

Horse by RuthThe Reality is: Most Connections are Proximate

What does it mean to interoperate information on the Web? With linked data and other structured data now in abundance, why don’t we see more information effectively combined? Why express your information as linked data if no one is going to use it?

Interoperability comes down to the nature of things and how we describe those things or quite similar things from different sources. This was the major thrust of my recent keynote presentation to the Dublin Core annual conference. In that talk I described two aspects of the semantic “gap”:

  1. One aspect is the need for vetted reference sources that provide the entities and concepts for aligning disparate content sources on the Web, and
  2. A second aspect is the need for accurate mapping predicates that can represent the often approximate matches and overlaps of this heterogeneous content.

I’ll discuss the first “gap” in a later post. What we’ll discuss here is the fact that most relationships between putatively same things on the Web are rarely exact, and are most often approximate in nature.

“It Ain’t the Label, Mabel”

The use of labels for matching or descriptive purposes was the accepted practice in early libraries and library science. However, with the move to electronic records and machine bases for matching, appreciation for ambiguities and semantics have come to the fore. Labels are no longer an adequate — let alone a sufficient — basis for matching references.

The ambiguity point is pretty straightforward.  Refer to Jimmy Johnson by his name, and you might be referring to a former football coach, a NASCAR driver, a former boxing champ, a blues guitarist, or perhaps even a plumber in your home town. Or perhaps none of these individuals. Clearly, the label “Jimmy Johnson” is insufficient to establish identity.

Of course, not all things are named entities such as a person’s name. Some are general things or concepts. But, here, semantic heterogeneities can also lead to confusion and mismatches. It is always helpful to revisit the sources and classification of semantic heterogeneities, which I first discussed at length nearly five years ago. Here is a schema classifying more than 40 categories of potential semantic mismatches [1]:

Class Category Subcategory
STRUCTURAL Naming Case Sensitivity
Generalization / Specialization
Aggregation Intra-aggregation
Internal Path Discrepancy
Missing Item Content Discrepancy
Attribute List Discrepancy
Missing Attribute
Missing Content
Element Ordering
Constraint Mismatch
Type Mismatch
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
Data Format
DATA Naming Case Sensitivity
ID Mismatch or Missing ID
Missing Data
Incorrect Spelling
LANGUAGE Encoding Ingest Encoding Mismatch
Ingest Encoding Lacking
Query Encoding Mismatch
Query Encoding Lacking
Languages Script Mismatches
Parsing / Morphological Analysis Errors (many)
Syntactical Errors (many)
Semantic Errors (many)

Even with the same label, two items in different information sources can refer generally to the same thing, but may not be the same thing or may define it with a different scope and content. In broad terms, these mismatches can be due to structure, domain, data or language, with many nuances within each type.

The sameAs approach used by many of the inter-dataset linkages in linked data ignores these heterogeneities. In a machine and reasoning sense, indeed even in a linking sense, these assertions can make as little or nonsensical sense as talking about the plumber with the facts about the blues guitarist.

Cats, Paul Newman and Great Britain

Let’s take three examples where putatively we are talking about the same thing and linking disparate sources on the Web.Great Britain Usages

The first example is the seemingly simple idea of “cats”. In one source, the focus might be on house cats, in another domestic cats, and in a third, cats as pets. Are these ideas the same thing? Now, let’s bring in some taxonomic information about the cat family, the Felidae. Now, the idea of “cats” includes lynx, tigers, lions, cougars and many other kinds of cats, domestic and wild (and, also extinct!). Clearly, the “cat” label used alone fails us miserably here.

Another example is one that Fred Giasson and I brought up one year ago in When Linked Data Rules Fail [2]. That piece discussed many poor practices within linked data, and used as one case the treatment of articles in the New York Times about the (deceased) actor Paul Newman. The NYT dataset is about various articles written about people historically in the newspaper. Their record about Paul Newman was about their pool of articles with attributes such as first published and so forth, with no direct attribute information about Paul Newman the person. Then, they asserted a sameAs relationship with external records in Freebase and DBpedia, which acts to commingle person attributes like birth, death and marriage with article attributes such as first and last published. Clearly, the NYT has confused the topic ( Paul Newman) of a record with the nature of that record (articles about topics). This misunderstanding of the “thing” at hand makes the entailed assertions from the multiple sources illogical and useless [3].

Our third example is the concept or idea or named entity of Great Britain. Depending on usage and context, Great Britain can refer to quite different scopes and things. In one sense, Great Britain is an island. In a political sense, Great Britain can comprise the territory of England, Scotland and Wales. But, even more precise understandings of that political grouping may include a number of outlying islands such as the Isle of Wight, Anglesey, the Isles of Scilly, the Hebrides, and the island groups of Orkney and Shetland. Sometimes the Isle of Man and the Channel Islands, which are not part of the United Kingdom, are fallaciously included in that political grouping. And, then, in a sporting context, Great Britain may also include Northern Ireland. Clearly, these, plus other confusions, can mean quite different things when referring to “Great Britain.” So, without definition, a seemingly simple question such as what the population of Great Britain is could legitimately return quite disparate values (not to mention the time dimension and how that has changed boundaries as well!).

These cases are quite usual for what “things” mean when provided from different sources with different perspectives and with different contexts. If we are to get meaningful interoperation or linkage of these things, we clearly need some different linking predicates.

Some Attempts at ‘Approximateness’

The realization that many connections across datasets on the Web need to be “approximate” is growing. Here is the result of an informal survey for leading predicates in this regard [4]:

  • skos:broadMatch
  • skos:related
  • ore:similarTo
  • dul:associatedWith
  • umbel:isAbout
  • skos:narrowMatch
  • vmf:isInVocabulary
  • skos:closeMatch
  • owl:equivalentClass
  • skos:mappingRelation
  • ov:similarTo
  • umbel:hasMapping
  • doape:similarThing
  • lvont:nearlySameAs
  • umbel:isRelatedTo
  • umbel:isLike
  • skos:exactMatch
  • sswap:hasMapping
  • umbel:hasCharacteristic
  • lvont:somewhatSameAs
  • dul:isAbout
  • skos:semanticRelation
  • rdfs:seeAlso
  • ore:describes
  • skos:narrowerTransitive
  • map:narrowerThan
  • dul:isConceptualizedBy
  • skos:narrower
  • umbel:isCharacteristicOf
  • prowl:defineUncertaintyOf
  • dc:subject
  • sumo:entails
  • link:uri
  • foaf:isPrimaryTopicOf
  • skos:broaderTransitive
  • dul:isComponentOf
  • foaf:focus
  • skos:relatedMatch
  • map:broaderThan
  • owl:sameAs
  • skos:broader
  • dul:isAssignedTo
  • wn:similarTo
  • sumo:refers
  • rdfs:subClassOf

Besides the standard OWL and RDFS predicates, SKOS, UMBEL and DOLCE [5] provide the largest number of choices above. In combination, these predicates probably provide a good scoping of “approximateness” in mappings.

Rationality and Reasoners

It is time for some leadership to emerge to provide a more canonical set of linking predicates for these real-world connection requirements. It would also be extremely useful to have such a canonical set adopted by some leading reasoners such that useful work could be done against these properties.

[1] See M. K. Bergman, 2006. “Sources and Classification of Semantic Heterogeneities,” AI3:::Adaptive Information blog, June 6, 2006. See
[2] See M. K. Bergman and F. Giasson, 2009. “When Linked Data Rules Fail,” AI3:::Adaptive Information blog, November 16, 2009. See
[3] On a different disappointing note, the critical errors that we noted a year ago and the NYT’s own acknowledgement on its site that:
“An RDFS description and English language documentation for the NYT namespace will be provided soon. Thanks for your patience.”
has still not been corrected, now a year later. Poor performance like this by a professional publisher gives linked data a bad name.
[4] These predicates have been obtained from personal knowledge and directed searches using the Falcons ontology search service. Simple Web searches on the namespace plus predicate name will provide more detail on any given predicate.
[5] UMBEL (Upper Mapping and Binding Exchange Layer) is an ontology of about 20,000 subject concepts that acts as a reference structure for inter-relating disparate datasets. It is also a general vocabulary of classes and predicates designed for the creation of domain-specific ontologies. For SKOS, see Alistair Miles and Sean Bechhofer, eds., 2009. SKOS Simple Knowledge Organization System Reference, W3C Recommendation, 18 August 2009; The Descriptive Ontology for Linguistic and Cognitive Engineering (DOLCE) is one of the more popular upper ontologies.
Posted:November 1, 2010

Jennifer Zaino of has just published an interview with me regarding our recently announced partnership with Ontotext and its relation to linked data. Thanks, Jenny, for a fair and accurate representation of our conversation!

Some of the questions related to reference vocabularies and linking predicates are somewhat hard to convey. Jenny did a very nice job capturing some nuanced concepts. I invite you to read the article yourself and judge.

Posted:October 25, 2010

Objective is to Tackle the ‘Semantics’ Gap in the Semantic Web

OntotextStructured Dynamics I’m pleased to announce that our company, Structured Dynamics, has formed a strategic partnership with Ontotext, a leading semantic technology company for the past 10 years.

Ontotext is the developer of OWLIM, a highly scalable semantic database engine, and KIM, a popular semantic annotation and search platform. Its FactForge and LinkedLifeData services provide the largest curated and interoperable linked data platforms over which inferencing and reasoning may be applied. Some of Ontotext’s major clients include AstraZeneca, BBC and Korea Telecom. Major professional services include its own technologies, plus text mining and semantic annotation. Ontotext has notable and longstanding technical partnerships, such as with the GATE team and many of the other leading technologies and companies in the semantic Web space. We are very pleased to join forces with them.

Semantic ‘Gap’ is Basis of Partnership

Our partnership was formed to address some of the key semantic ‘gaps’ in the semantic Web. The partnership will focus on development of the next generation of the UMBEL and PROTON ontologies, as well as tools and applications based on them.

Volumes of linked data on the Web are growing. This growth is exposing three key weaknesses:

  1. inadequate semantics for how to link disparate information together that recognizes inherently different contexts and viewpoints and (often) approximate mappings
  2. misapplication of many linking predicates, such as owl:sameAs, and
  3. a lack of coherent reference concepts by which to aggregate and organize this linkable content.

Thanks to the efforts of the W3C (World Wide Web Consortium), we now have the techniques, languages and standards to deliver the “web” portion of the semantic Web. But, the practical “semantics” for actually effecting the semantic Web have heretofore been lacking. Early experience with linked data has exposed many poor practices. The lack of approximate linking predicates and reference concepts undercuts our ability to achieve meaningful semantic interoperability.

In forming our partnership, Ontotext and SD will shine attention on this semantics “gap”. We will also be aggressively seeking additional partners and players to join with us on this challenge. My recent outreach to DCMI (the Dublin Core Metadata Initiative) is one example of this commitment; we will be talking with others in the coming weeks.

Linked data and the prospects of the semantic Web are at a critical juncture. While we have seen much growth in the release of linked data, we are still not seeing much uptake (other than some curated pockets). Linkages between datasets are still disappointingly low, and quality of linkages is an issue. The time has come to stop simply shoveling more triples over the fence.

Building Blocks

The combination of UMBEL and PROTON offers a powerful blend to address these weaknesses. Our partnership will first provide a logical mapping and consolidated framework based on the two core ontologies. These will be made available as standard ontologies and via open source semantic annotation tools.

UMBEL PROTONUMBEL (Upper Mapping and Binding Exchange Layer) is both a vocabulary for building domain ontologies and a framework of more than 20,000 reference concepts. The UMBEL reference ontology is used to tag information and map existing schema in order to help link content and promote interoperability. UMBEL’s reference concepts and structure are a direct subset extraction of the Cyc knowledge base.

The PROTON ontology (PROTo ONtology) is a basic upper-level ontology that contains about 300 classes and 100 properties, providing coverage of the general concepts necessary for a wide range of tasks, including semantic annotation, indexing, and retrieval of documents. It is domain independent with coverage suitable to encompass any domain or named entity.

This consolidated framework will then be applied to organize and provide a coherent categorization of the Wikipedia online encyclopedia. One expression of this result will be a new version of Ontotext’s FactForge, already the largest and best performing reasoning engine leveraging linked data. This new version will allow easy access to the most central Linking Open Data (LOD) datasets such as DBpedia, Freebase, and Geonames, through the vocabularies of UMBEL and PROTON. Additional applications in linked data mining and general tagging of standard Web content are also contemplated by the partnership.

Ontotext’s proven reasoning technologies and ability to host extremely large knowledge bases with great performance are tremendous boons to the next iteration of UMBEL. We have been seeking large-scale coherency testing of UMBEL for some time and Ontotext is the perfect answer.

Ontotext’s CEO, Atanas Kiryakov, indicated their interest in UMBEL stemmed from what they saw as some stumbling blocks with linked data while developing FactForge. “The growth and maturation of linked data will require credible ways to orient and annotate the data,” said Kiryakov. “UMBEL is the right scope of comprehensiveness and size to use as one foundation for this,” he said. Ontotext is also the original developer and current maintainer of PROTON, which will also contribute in this role.

What is to Come?

The efforts of the partnership will first be seen with release of UMBEL v. 0.80 in the next couple of weeks. This update revises many aspects of the ontology based on two years of applied experience and updates it to OWL 2. Then, this basis will be used for broader mappings and linkages to Wikipedia. Those next mappings are earmarked for UMBEL version 1.00, slated for release by the end of the year. All of these planned efforts will be released as open source.

Among other intended uses, PROTON, UMBEL and FactForge form a layered reference data structure that will be used for data integration within the European Union research project RENDER. The large-scale RENDER project aims to integrate diverse methods in the ways Web information is selected, ranked, aggregated, presented and used.

Beyond that, further relationships and partnerships are being actively sought with players serious about interoperable, high-quality data on the semantic Web. We welcome inquiries or outreach.