Coca-Cola, Toucans and Charles Sanders Peirce

The crowning achievement of the semantc Web is the simple use of URIs to identify data. Further, if the URI identifier can resolve to a representation of that data, it now becomes an integral part of the HTTP access protocol of the Web while providing a unique identifier for the data. These innovations provide the basis for distributed data at global scale, all accessible via Web devices such as browsers and smartphones that are now a ubiquitous part of our daily lives.

Yet, despite these profound and simple innovations, the semantic Web’s designers and early practitioners and advocates have been mired in a muddled, metaphysical argument of at least a decade over what these URIs mean, what they reference, and what their actual true identity is. These muddles about naming and identity, it might be argued, are due to computer scientists and programmers trying to grapple with issues more properly the domain of philosophers and linguists. But that would be unfair. For philosophers and linguists themselves have for centuries also grappled with these same conundrums [1].

As I argue in this piece, part of the muddle results from attempting to do too much with URIs while another part results from not doing enough. I am also not trying to directly enter the fray of current standards deliberations. (Despite a decade of controversy, I optimistically believe that the messy process of argument and consensus building will work itself out [2].) What I am trying to do in this piece, however, is to look to one of America’s pre-eminent philosophers and logicians, Charles Sanders Peirce (pronounced “purse”), to inform how these controversies of naming, identity and meaning may be dissected and resolved.

‘Identity Crisis’, httpRange-14, and Issue 57

The Web began as a way to hyperlink between documents, generally Web pages expressed in the HTML markup language. These initial links were called URLs (uniform resource locators), and each pointed to various kinds of electronic resources (documents) that could be accessed and retrieved on the Web. These resources could be documents written in HTML or other encodings (PDFs, other electronic formats), images, streaming media like audio or videos, and the like [3].

All was well and good until the idea of the semantic Web, which postulated that information about the real world — concepts, people and things — could also be referenced and made available for reasoning and discussion on the Web. With this idea, the scope of the Web was massively expanded from electronic resources that could be downloaded and accessed via the Web to now include virtually any topic of human discourse. The rub, of course, was that ideas such as abstract concepts or people or things could not be “dereferenced” nor downloaded from the Web.

One of the first things that needed to change was to define a broader concept of a URI “identifier” above the more limited concept of a URL “locator”, since many of these new things that could be referenced on the Web went beyond electronic resources that could be accessed and viewed [3]. But, since what the referent of the URI now actually might be became uncertain — was it a concept or a Web page that could be viewed or something else? — a number of commentators began to note this uncertainty as the “identity crisis” of the Web [4]. The topic took on much fervor and metaphysical argument, such that by 2003, Sandro Hawke, a staffer of the standards-setting W3C (World Wide Web Consortium), was able to say, “This is an old issue, and people are tired of it” [5].

Yet, for many of the reasons described more fully below, the issue refused to go away. The Technical Architecture Group (TAG) of the W3C took up the issue, under a rubric that came to be known as httpRange-14 [6]. The issue was first raised in March 2002 by Tim Berners-Lee, accepted for TAG deliberations in February 2003, with then a resolution offered in June 2005 [7]. (Refer to the original resolution and other information [6] to understand the nuances of this resolution, since particular commentary on that approach is not the focus of this article.) Suffice it to say here, however, that this resolution posited an entirely new distinction of Web content into “information resources” and “non-information resources”, and also recommended the use of the HTTP 303 redirect code for when agents requesting a URI should be directed to concepts versus viewable documents.

This “resolution” has been anything but. Not only can no one clearly distinguish these de novo classes of “information resources” [19], but the whole approach felt arbitrary and kludgy.

Meanwhile, the confusions caused by the “identity crisis” and httpRange-14 continued to perpetuate themselves. In 2006, a major workshop on “Identity, Reference and the Web” (IRW 2006) was held in conjunction with the Web’s major WWW2006 conference in Edinburgh, Scotland, on May 23, 2006 [8]. The various presentations and its summary (by Harry Halpin) are very useful to understand these issues. What was starting to jell at this time was the understanding that the basis of identity and meaning on the Web posed new questions, and ones that philosophers, logicians and linguists needed to be consulted to help inform.

The fiat of the TAG’s 2005 resolution has failed to take hold. Over the ensuing years, various eruptions have occurred on mailing lists and within the TAG itself (now expressed as Issue 57) to revisit these questions and bring the steps moving forward into some coherent new understanding. Though linked data has been premised on best-practice implementation of these resolutions [9], and has been a qualified success, many (myself included) would claim that the extra steps and inefficiencies required from the TAG’s httpRange-14 guidance have been hindrances, not facilitators, of the uptake of linked data (or the semantic Web).

Today, despite the efforts of some to claim the issue closed, it is not. Issue 57 and the periodic bursts from notable semantic Web advocates such as Ian Davis [10], Pat Hayes and Harry Halpin [11], Ed Summers [12], Xiaoshu Wang [13], David Booth [14] and TAG members themselves, such as Larry Masinter [15] and Jonathan Rees [16], point to continued irresolution and discontent within the advocate community. Issue 57 currently remains open. Meanwhile, I think, all of us interested in such matters can express concern that linked data, the semantic Web and interoperable structured data have seen less uptake than any of us had hoped or wanted over the past decade. As I have stated elsewhere, unclear semantics and muddled guidelines help to undercut potential use.

As each of the eruptions over these identity issues has occurred, the competing camps have often been characterized as “talking past one another”; that is, not communicating in such a way as to help resolve to consensus. While it is hardly my position to do so, I try to encapsulate below the various positions and prejudices as I see them in this decades-long debate. I also try to share my own learning that may help inform some common ground. Forgive me if I overly simplify these vexing issues by returning to what I see as some first principles . . . .

What’s in a Name?

One legacy of the initial document Web is the perception that Web addresses have meaning. We have all heard of the multi-million dollar purchasing of domains [17] and the adjudication that may occur when domains are hijacked from their known brands or trademark owners. This legacy has tended to imbue URIs with a perceived value. It is not by accident, I believe, that many within the semantic Web and linked data communities still refer to “minting” URIs. Some believe that ownership and control over URIs may be equivalent to grabbing up valuable real estate. It is also the case that many believe the “name” given to a URI acts to name the referent to which it refers.

This perception is partially true, partially false, but moreover incomplete in all cases. We can illustrate these points with the global icon, “Coca-Cola”.

As for the naming aspects, let’s dissect what we mean when we use the label “Coca-Cola” (in a URI or otherwise). Perhaps the first thing that comes to mind is “Coca-Cola,” the beverage (which has a description on Wikipedia, among other references). Because of its ubiquity, we may also recognize the image of the Coca-Cola bottle to the left as a symbol for this same beverage. (Though, in the hilarious movie, The Gods, They Must be Crazy, Kalahari Bushmen, who had no prior experience of Coca-Cola, took the bottle to be magical with evil powers [18].) Yet even as reference to the beverage, the naming aspects are a bit cloudy since we could also use the fully qualified synonyms of “Coke”, “Coca-cola” (small C), “Classic Coke” and the hundreds of language variants worldwide.

On the other hand, the label “Coca-Cola” could just as easily conjure The Coca-Cola Company itself. Indeed, the company web site is the location pointed to by the URI of http://www.thecoca-colacompany.com/. But, even that URI, which points to the home Web page of the company, does not do justice to conveying an understanding or description of the company. For that, additional URIs may need to be invoked, such as the description at Wikipedia, the company’s own company description page, plus perhaps the company’s similar heritage page.

Of course, even these links and references only begin to scratch the surface of what the company Coca-Cola actually is: headquarters, manufacturing facilities, 140,000 employees, shareholders, management, legal entities, patents and Coke recipe, and the like. Whether in human languages or URIs, in any attempt to signify something via symbols or words (themselves another form of symbol), we risk ambiguity and incompleteness.

URI shorteners also undercut the idea that a URI necessarily “names” something. Using the service bitly, we can shorten the link to the Wikipedia description of the Coke beverage to http://bit.ly/xnbA6 and we can shorten the link to The Coca-Cola Company Web site to http://bit.ly/9ojUpL. I think we can fairly say that neither of these shortened links “name” their referents. The most we can say about a URI is that it points to something. With the vagaries of meaning in human languages, we might also say that URIs refer to something, denote something or identify (but not in the sense of completely define) something.

From this discussion, we can assert with respect to the use of URIs as “names” that:

1. In all cases, URIs are pointers to a particular referent
2. In some cases, URIs do act to “name” some things
3. Yet, even when used as “names,” there can be ambiguity as to what exactly the referent is that is denoted by the name
4. Resolving what such “names” mean is a matter of context and reference to further information or links, and
5. Because URIs may act as “names”, it is appropriate to consider social conventions and contracts (e.g., trademarks, brands, legal status) in adjudicating who can own the URI.

In summary, I think we can say that URIs may act as names, but not in all or most cases, and when used as such are often ambiguous. Absolutely associating URIs as names is way too heavy a burden, and incorrect in most cases.

What is a Resource?

The “name” discussion above masks that in some cases we are talking about a readable Web document or image (such as the Wikipedia description of the Coke beverage or its image) versus the “actual” thing in the real world (the Coke beverage itself or even the company). This distinction is what led to the so-called “identity crisis”, for which Ian Davis has used a toucan as his illustrative thing [10].

As I note in the conclusion, I like Davis’ approach to the identity conundrum insofar as Web architecture and linked data guidance are concerned. But here my purpose is more subtle: I want to tease apart still further the apparent distinction between an electronic description of something on the Web and the “actual” something. Like Davis, let’s use the toucan.

In our strawman case, we too use a description of the toucan (on Wikipedia) to represent our “information resource” (the accessible, downloadable electronic document). We contrast to that a URI that we mean to convey the actual physical bird (a “non-information resource” in the jumbled jargon of httpRange-14), which we will designate via the URI of http://example.com/toucan.

Despite the tortured (and newly conjured) distinction between “information resource” and “non-information resource”, the first blush reaction is that, sure, there is a difference between an electronic representation that can be accessed and viewed on the Web and its true, “actual” thing. Of course people can not actually be rendered and downloaded on the Web, but their bios and descriptions and portrait images may. While in the abstract such distinctions appear true and obvious, in the specifics that get presented to experts, there is surprising disagreement as to what is actually an “information resource” v. a “non-information resource” [19]. Moreover, as we inspect the real toucan further, even that distinction is quite ambiguous.

When we inspect what might be a definitive description of “toucan” on Wikipedia, we see that the term more broadly represents the family of Ramphastidae, which contains five genera and forty different species. The picture we are showing to the right is but of one of those forty species, that of the keel-billed toucan (Ramphastos sulfuratus). Viewing the images of the full list of toucan species shows just how divergent these various “physical birds” are from one another. Across all species, average sizes vary by more than a factor of three with great variation in bill sizes, coloration and range. Further, if I assert that the picture to the right is actually that of my pet keel-billed toucan, Pretty Bird, then we can also understand that this representation is for a specific individual bird, and not the physical keel-billed toucan species as a whole.

The point of this diversion is not a lecture on toucans, but an affirmation that distinctions between “resources” occur at multiple levels and dimensions. Just as there is no self-evident criteria as to what constitutes an “information resource”, there is also not a self-evident and fully defining set of criteria as to what is the physical “toucan” bird. The meaning of what we call a “toucan” bird is not embodied in its label or even its name, but in the context and accompanying referential information that place the given referent into a context that can be communicated and understood. A URI points to (“refers to”) something that causes us to conjure up an understanding of that thing, be it a general description of a toucan, a picture of a toucan, an understanding of a species of toucan, or a specific toucan bird. Our understanding or interpretation results from the context and surrounding information accompanying the reference.

In other words, a “resource” may be anything, which is just the way the W3C has defined it. There is not a single dimension which, magically, like “information” and “non-information,” can cleanly and definitely place a referent into some state of absolute understanding. To assert that such magic distinctions exist is a flaw of Cartesian logic, which can only be reconciled by looking to more defensible bases in logic [20].

Peirce and the Logic of Signs

The logic behind these distinctions and nuances leads us to Charles Sanders Peirce (1839 – 1914). Peirce (pronounced “purse”) was an American logician, philosopher and polymath of the first rank. Along with Frege, he is acknowledged as the father of predicate calculus and the notation system that formed the basis of first-order logic. His symbology and approach arguably provide the logical basis for description logics and other aspects underlying the semantic Web building blocks of the RDF data model and, eventually, the OWL language. Peirce is the acknowledged founder of pragmatism, the philosophy of linking practice and theory in a process akin to the scientific method. He was also the first formulator of existential graphs, an essential basis to the whole field now known as model theory. Though often overlooked in the 20th century, Peirce has lately been enjoying a renaissance with his voluminous writings still being deciphered and published.

The core of Peirce’s world view is based in semiotics, the study and logic of signs. In his seminal writing on this, “What is in a Sign?” [21], he wrote that “every intellectual operation involves a triad of symbols” and “all reasoning is an interpretation of signs of some kind”. Peirce had a predilection for expressing his ideas in “threes” throughout his writings.

Semiotics is often split into three branches: 1) syntactics – relations among signs in formal structures; 2) semantics – relations between signs and the things to which they refer; and 3) pragmatics – relations between signs and the effects they have on the people or agents who use them.

Peirce’s logic of signs in fact is a taxonomy of sign relations, in which signs get reified and expanded via still further signs, ultimately leading to communication, understanding and an approximation of “canonical” truth. Peirce saw the scientific method as itself an example of this process.

A given sign is a representation amongst the triad of the sign itself (which Peirce called a representamen, the actual signifying item that stands in a well-defined kind of relation to the two other things), its object and its interpretant. The object is the actual thing itself. The interpretant is how the agent or the perceiver of the sign understands and interprets the sign. Depending on the context and use, a sign (or representamen) may be either an icon (a likeness), an indicator or index (a pointer or physical linkage to the object) or a symbol (understood convention that represents the object, such as a word or other meaningful signifier).

An interpretant in its barest form is a sign’s meaning, implication, or ramification. For a sign to be effective, it must represent an object in such a way that it is understood and used again. This makes the assignment and use of signs a community process of understanding and acceptance [20], as well as a truth-verifying exercise of testing and confirming accepted associations.

John Sowa has done much to help make some of Peirce’s obscure language and terminology more accessible to lay readers [22]. He has expressed Peirce’s basic triad of sign relations as follows, based around the Yojo animist cat figure used by the character Queequeg in Herman Melville’s Moby-Dick:

In this figure, object and symbol are the same as the Peirce triad; concept is the interpretant in this case. The use of the word ‘Yojo’ conjures the concept of cat.

This basic triad representation has been used in many contexts, with various replacements or terms at the nodes. Its basic form is known as the Meaning Triangle, as was popularized by Ogden and Richards in 1923 [23].

The key aspect of signs for Peirce, though, is the ongoing process of interpretation and reference to further signs, a process he called semiosis. A sign of an object leads to interpretants, which, as signs, then lead to further interpretants. In the Sowa example below, we show how meaning triangles can be linked to one another, in this case by abstracting that the triangles themselves are concepts of representation; we can abstract the ideas of both concept and symbol:

We can apply this same cascade of interpretation to the idea of the sign (or representamen), which in this case shows that a name can be related to a word symbol, which in itself is a combination of characters in a string called ‘Yojo’:

According to Sowa [22]:

Catherine Legg [20] notes that the semiotic process is really one of community involvement and consensus. Each understanding of a sign and each subsequent interpretation helps come to a consensus of what a sign means. It is a way of building a shared understanding that aids communication and effective interpretation. In Peirce’s own writings, the process of interpretation can lead to validation and an eventual “canonical” or normative interpretation. The scientific method itself is an extreme form of the semiotic process, leading ultimately to what might be called accepted “truths”.

Peircean Semiotics of URIs

So, how do Peircean semiotics help inform us about the role and use of URIs? Does this logic help provide guidance on the “identity crisis”?

The Peircean taxonomy of signs has three levels with three possible sign roles at each level, leading to a possible 27 combinations of sign representations. However, because not all sign roles are applicable at all levels, Peirce actually postulated only ten distinct sign representations.

Common to all roles, the URI “sign” is best seen as an index: the URI is a pointer to a representation of some form, be it electronic or otherwise. This representation bears a relation to the actual thing that this referent represents, as is true for all triadic sign relationships. However, in some contexts, again in keeping with additional signs interpreting signs in other roles, the URI “sign” may also play the role of a symbolic “name” or even as a signal that the resource can be downloaded or accessed in electronic form. In other words, by virtue of the conventions that we choose to assign to our signs, we can supply additional information that augments our understanding of what the URI is, what it means, and how it is accessed.

Of course, in these regards, a URI is no different than any other sign in the Peircean world view: it must reside in a triadic relationship to its actual object and an interpretation of that object, with further understanding only coming about by the addition of further signs and interpretations.

In shortened form, this means that a URI, acting alone, can at most play the role of a pointer between an object and its referent. A URI alone, without further signs (information), can not inform us well about names or even what type of resource may be at hand. For these interpretations to be reliable, more information must be layered on, either by accepted convention of the current signs or the addition of still further signs and their interpretations. Since the attempts to deal with the nature of a URI resource by fiat as stipulated by httpRange-14 neither meet the standards of consensus nor empirical validity, the attempt can not by definition become “canonical”. This does not mean that httpRange-14 and its recommended practices can not help in providing more information and aiding interpretation for what the nature of a resource may be. But it does mean that httpRange-14 acting alone is insufficient to resolve ambiguity.

Moreover, what we see in the general nature of Peirce’s logic of signs is the usefulness of adding more “triads” of representation as the process to increase understanding through further interpretation. Kind of sounds like adding on more RDF triples, does it not?

Global is Neither Indiscriminate Nor Unambiguous

Names, references, identity and meaning are not absolutes. They are not philosophically, and they are not in human language. To expect machine communications to hold to different standards and laws than human communications is naive. To effect machine communications our challenge is not to devise new rules, but to observe and apply the best rules and practices that human communications instruct.

There has been an unstated hope at the heart of the semantic Web enterprise that simply expressing statements in the right way (syntax) and in the right form (RDF) is sufficient to facilitate machine communications. But this hope, too, is naive and silly. Just as we do not accept all human utterances as truth, neither will we accept all machine transmissions as reliable. Some of the information will be posted in error; some will be wrong or ill-fitting to our world view; some will be malicious or intended to deceive. Spam and occasionally lousy search results on the Web tell us that Web documents are subject to these sources of unsuitability, why is not the same true of data?

Thus, global data access via the semantic Web is not — and can never be — indiscriminate nor unambiguous. We need to understand and come to trust sources and provenance; we need interpretation and context to decide appropriateness and validity; and we need testing and validation to ensure messages as received are indeed correct. Humans need to do these things in their normal courses of interaction and communication; our machine systems will need to do the same.

These confirmations and decisions as to whether the information we receive is actionable or not will come about via still more information. Some of this information may come about via shared conventions. But most will come about because we choose to provide more context and interpretation for the core messages we hope to communicate.

A Go-Forward Approach

Nearly five years ago Hayes and Halpin put forth a proposal to add ex:refersTo and ex:describedBy to the standard RDF vocabulary as a way for authors to provide context and explanation for what constituted a specific RDF resource [11]. In various ways, many of the other individuals cited in this article have come to similar conclusions. The simple redirect suggestions of both Ian Davis [10] and Ed Summers [12] appear particularly helpful.

Over time, we will likely need further representations about resources regarding such things as source, provenance, context and other interpretations that would help remove ambiguities as to how the information provided by that resource should be consumed or used. These additional interpretations can mechanically be provided via referenced ontologies or embedded RDFa (or similar). These additional interpretations can also be aided by judicious, limited additions of new predicates to basic language specifications for RDF (such as the Hayes and Halpin suggestions).

In the end, of course, any frameworks that achieve consensus and become widely adopted will be simple to use, easy to understand, and straightforward to deploy. The beauty of best practices in predicates and annotations is that failures to provide are easy to test. Parties that wish to have their data consumed have incentive to provide sufficient information so as to enable interpretation.

There is absolutely no reason that these additions can not co-exist with the current httpRange-14 approach. By adding a few other options and making clear the optional use of httpRange-14, we would be very Peirce-like in our go-forward approach: We are being both pragmatic while we add more means to improve our interpretations for what a Web resource is and is meant to be.

Five 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, City-Data.com, 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.

And, 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.

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.

The 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]:

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.

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]:

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.

Jennifer Zaino of SemanticWeb.com 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.