Posted:December 12, 2011

State of SemWeb Tools - 2011Number of Semantic Web Tools Passes 1000 for First Time; Many Other Changes

We have been maintaining Sweet Tools, AI3‘s listing of semantic Web and -related tools, for a bit over five years now. Though we had switched to a structWSF-based framework that allows us to update it on a more regular, incremental schedule [1], like all databases, the listing needs to be reviewed and cleaned up on a periodic basis. We have just completed the most recent cleaning and update. We are also now committing to do so on an annual basis.

Thus, this is the inaugural ‘State of Tooling for Semantic Technologies‘ report, and, boy, is it a humdinger. There have been more changes — and more important changes — in this past year than in all four previous years combined. I think it fair to say that semantic technology tooling is now reaching a mature state, the trends of which likely point to future changes as well.

In this past year more tools have been added, more tools have been dropped (or abandoned), and more tools have taken on a professional, sophisticated nature. Further, for the first time, the number of semantic technology and -related tools has passed 1000. This is remarkable, given that more tools have been abandoned or retired than ever before.

Click here to browse the Sweet Tools listing. There is also a simple listing of URL links and categories only.

We first present our key findings and then overall statistics. We conclude with a discussion of observed trends and implications for the near term.

Key Findings

Some of the key findings from the 2011 State of Tooling for Semantic Technologies are:

  • As of the date of this article, there are 1010 tools in the Sweet Tools listing, the first it has passed 1000 total tools
  • A total of 158 new tools have been added to the listing in the last six months, an increase of 17%
  • 75 tools have been abandoned or retired, the most removed at any period over the past five years
  • A further 6%, or 55 tools, have been updated since the last listing
  • Though open source has always been an important component of the listing, it now constitutes more than 80% of all listings; with dual licenses, open source availability is about 83%. Online systems contribute another 9%
  • Key application areas for growth have been in SPARQL, ontology-related areas and linked data
  • Java continues to dominate as the most important language.

Many of these points are elaborated below.

The Statistical Picture

The updated Sweet Tools listing now includes nearly 50 different tools categories. The most prevalent categories, each with over 6% of the total, are information extraction, general RDF tools, ontology tools, browser tools (RDF, OWL), and parsers or converters. The relative share by category is shown in this diagram (click to expand):

Since the last listing, the fastest growing categories have been SPARQL, linked data, knowledge bases and all things related to ontologies. The relative changes by tools category are shown in this figure:

Though it is true that some of this growth is the result of discovery, based on our own tool needs and investigations, we have also been monitoring this space for some time and serendipity is not a compelling explanation alone. Rather, I think that we are seeing both an increase in practical tools (such as for querying), plus the trends of linked data growth matched with greater sophistication in areas such as ontologies and the OWL language.

The languages these tools are written in have also been pretty constant over the past couple of years, with Java remaining dominant. Java has represented half of all tools in this space, which continues with the most recent tools as well (see below). More than a dozen programming or scripting languages have at least some share of the semantic tooling space (click to expand):

Sweet Tools Languages

With only 160 new tools it is hard to draw firm trends, but it does appear that some languages (Haskell, XSLT) have fallen out of favor, while popularity has grown for Flash/Flex (from a small base), Python and Prolog (with the growth of logic tools):

PHP will likely continue to see some emphasis because of relations to many content management systems (WordPress, Drupal, etc.), though both Python and Ruby seem to be taking some market share in that area.

New Tools

The newest tools added to the listing show somewhat similar trends. Again, Java is the dominant language, but with much increased use of JavaScript and Python and Prolog:

Sweet Tools Languages

The higher incidence of Prolog is likely due to the parallel increase in reasoners and inference engines associated with ontology (OWL) tools.

The increase in comprehensive tool suites and use of Eclipse as a development environment would appear to secure Java’s dominance for some time to come.

Trends and Observations

These dry statistics tend to mask the feel one gets when looking at most of the individual tools across the board. Older academic and government-funded project tools are finally getting cleaned out and abandoned. Those tools that remain have tended to get some version upgrades and improved Web sites to accompany them.

The general feel one gets with regard to semantic technology tooling at the close of 2011 has these noticeable trends:

  • A three-tiered environment – the tools seem to segregate into: 1) a bottom tier of tools (largely) developed by individuals or small groups, now most often found on Google Code or Github; 2) a middle-tier of (largely) government-funded projects, sometimes with multiple developers, often older, but with no apparent driving force for ongoing improvements or commercialization; and 3) a top-tier of more professional and (often) commercially-oriented tools. The latter category is the most noticeable with respect to growth and impact
  • Professionalism – the tools in the apparent top tiers feel to have more professionalism and better (and more attractive) packaging. This professionalism is especially true for the frameworks and composite applications. But, it also applies to many of the EU-funded projects from Europe, which has always been a huge source of new tool developments
  • More complete toolsets – similarly, the upper levels of tools are oriented to pragmatic problems and problem-solving, which often means they embody multiple functions and more complete tooling environments. This category actually appears to be the most visible one exhibiting growth
  • Changing nature of academic releases – yet, even the academic releases seem to be increasing in professionalism and completeness. Though in the lowest tier it is still possible to see cursory or experimental tool releases, newer academic releases (often) seem to be more strategically oriented and parts of broader programmatic emphases. Programs like AKSW from the University of Leipzig or the Freie Universität Berlin or Finland’s Semantic Computing Research Group (SeCo), among many others, tend to be exemplars of this trend
  • Rise of commercial interests and enterprise adoption – the growing maturity of semantic technologies is also drawing commercial interest, and the incubation of new start-ups by academic and research institutions acts to reinforce the above trends. Promising projects and tools are now much more likely to be spun off as potential ventures, with accompanying better packaging, documentation and business models
  • Multiple languages and applications – with this growing complexity and sophistication has also come more complicated apps, combining multiple languages and functions. In fact, for some time the Sweet Tools listing has been justifiably criticized by some as overly “simplifying” the space by classifying tools under (largely) single applications or single languages. By the 2012 survey, it will likely be necessary to better classify the tools using multiple assignments
  • Google code over SourceForge for open source (and an increase in Github, as well) – virtually all projects on SourceForge now feel abandoned or less active. The largest source of open source projects in the semantic technology space is now clearly Google Code. Though of a smaller footprint today, we are also seeing many of the newer open source projects also gravitate to Github. Open source hosting environments are clearly in flux.

I have said this before, and been wrong about it before, but it is hard to see the tooling growth curve continue at its current slope into the future. I think we will see many individual tools spring up on the open source hosting sites like Google and Github, perhaps at relatively the same steady release rate. But, old projects I think will increasingly be abandoned and older projects will not tend to remain available for as long a time. While a relatively few established open source standards, like Solr and Jena, will be the exception, I think we will see shorter shelf lives for most open source tools moving forward. This will lead to a younger tools base than was the case five or more years ago.

I also think we will continue to see the dominance of open source. Proprietary software has increasingly been challenged in the enterprise space. And, especially in semantic technologies, we tend to see many open source tools that are as capable as proprietary ones, and generally more dynamic as well. The emphasis on open data in this environment also tends to favor open source.

Yet, despite the professionalism, sophistication and complexity trends, I do not yet see massive consolidation in the semantic technology space. While we are seeing a rapid maturation of tooling, I don’t think we have yet seen a similar maturation in revenue and business models. While notable semantic technology start-ups like Powerset and Siri have been acquired and are clear successes, these wins still remain much in the minority.

[1] Please use the comments section of this post for suggesting new or overlooked tools. We will incrementally add them to the Sweet Tools listing. Also, please see the About tab of the Sweet Tools results listing for prior releases and statistics.

Posted by AI3's author, Mike Bergman Posted on December 12, 2011 at 8:29 am in Open Source, Semantic Web Tools, Structured Web | Comments (3)
The URI link reference to this post is:
The URI to trackback this post is:
Posted:July 18, 2011

Photo courtesy of levelofhealth.comA Decade of Remarkable Advances in Ten Grand IT Challenges

I’ve been in the information theory and technology game for quite some time, but believe nothing has matched the pace of advances of the past ten years. As one example, it was a mere eight years ago that I was sitting in a room with language translation vendors contemplating automated translation techniques for US intelligence agencies. The prospects finally looked doable, but the success of large-scale translation was not assured.

At about that same time, and the years until just recently, a whole slew of Grand Challenges [1] in computing hung out there: tantalizing yet not proven. These areas ranged from information extraction and natural language understanding to speech recognition and automated reasoning.

But things have been changing fast, and with a subtle steadiness that has caused it to go largely unremarked. Sure, all of us have been aware of the huge changes on the Web and search engine ubiquity and social networking. But some of the fundamentally hard problems in computing have also gone through some remarkable (but largely unremarked) advances.

We now have smart phones that speak instructions to us while we instruct them by voice in turn. Virtually all information conceivable is now indexed and made available through the Web; structure is now rapidly characterizing that information, making it even more useful to discover and organize. We can translate documents online with acceptable accuracy into more than 60 languages [2]. We can get directions to or see satellite views of virtually any place on earth. We have in fact become accustomed to new technology magic on a nearly daily basis, so much so that the pace of these advances seems to be a constant, blunting our perspective of just how rapid these advances have been progressing.

These advances are perhaps not the realization of artificial intelligence as articulated in the 1950s to 1980s, but are contributing to a machine-based ability to do tasks useful to humans heretofore impossible and at scales unimaginable. As Google and IBM’s Watson are showing, statistics (among other techniques) applied to massive knowledge bases or text corpora are breaking down all of the Grand Challenges of symbolic computing. The image that is emerging is less one of intelligent machines working autonomously than it is of computers working interactively or semi-automatically with humans to address previously unsolvable problems.

By using a perspective of the decade past, we also demark the seminal paper on the semantic Web by Berners-Lee, Hendler and Lassila from May 2001 [3]. Yet, while this semantic Web vision has been a contributor to the success of the Grand Challenge advances of the past ten years, I think we can also say that it has not been the key or even a primary driver. That day may still yet come. Rather, I think we have to look to natural language and statistics surrounding large-scale corpora as the more telling drivers.

Ten Grand Challenge Advances

Over the past ten years there have been significant advances on at least ten Grand Challenges in symbolic computation. As the concluding section notes, these advances can be traced in most part to broader advances in natural language processing, the logical and semiotic bases for interoperability, and standards (nominally in the semantic Web) for embracing them. Here are these ten areas of advance, all achieved over the past ten years:

#1 Information Extraction

Information extraction (IE) uses various forms of natural language processing (NLP) to identify structured information within unstructured or semi-structured documents. These documents are presented in machine-readable form (including straight text, various document formats or HTML) with the various types of information “tagged” or prompted for inclusion. Information types that can be extracted with one of the various techniques include entities, relations, topics, categories, and so forth. Once tagged or extracted, the information in the documents can now be included and linked to standard structured information (as might come from conventional databases) or to structure in other documents.

Most recently, a large number of online services and open source systems have also become available with strengths in one or more of these extraction types [4]. Some current examples include Yahoo! Term Extraction, OpenCalais, BeliefNetworks, OpenAmplify, Alchemy API, Evri, Extractiv, Illinois Tagger, and about 80 others [4].

#2 Machine Translation

Machine translation is the automatic translation of machine-readable text from one human language to another. Accurate and acceptable machine translation requires applying different types of knowledge including grammar, semantics, facts about the real world, etc. Various approaches have been developed and refined over time.

Especially helpful has been the availability of huge corpora in multiple languages to which large-scale statistical analysis may be applied (as is the case of Google’s machine translation) or human editing and refinement (as is the case with the more than 280 language versions of Wikipedia).

While it is true none of these systems have 100% accuracy (even human translators show much variation), the more advanced ones are truly impressive with remaining ambiguities flagged for resolution by semi-automatic means.

#3 Sentiment Analysis

Though sentiment analysis is strictly speaking a subset of information extraction, it has the more demanding and useful task of extracting subjective information, often across a group of documents or texts. Sentiment analysis can be applied to online reviews to determine the “polarity” about specific objects, and it is especially useful for identifying public opinion trends or evaluating social media for ranking, polling or marketing purposes.

Because of its greater difficulty and potential high value, many of the leading sentiment analysis capabilities remain proprietary. Some capable open source versions are available nonetheleless. There is also an interesting online application using Twitter feeds.

#4 Disambiguation

Many words have more than one meaning. Word sense disambiguation uses either machine learning, dictionaries (gazetteers) of known entities and concepts, ontologies or linguistic databases such as WordNet, or combinations thereof to evaluate ambiguous terms or phrases and resolve them based on context. Some systems need to be “trained” or some work automatically or others are based on evaulation and prompting (semi-automatic) to complete the disambiguation process.

State-of-the-art systems have greater than 90% precision [5]. Most of the leading open source NLP toolkits have quite capable disambiguation modules, and even better proprietary systems exist.

#5 Speech Synthesis and Recognition

Speech synthesis is the conversion of text to spoken speech and has been around for quite some time. Speech recognition is a far more difficult task in that a given sound clip or real-time spoken speech of a person must be converted to a textual representation, which itself can then be acted upon such as navigating or making selections. Speech recognition is made difficult because of individual voice differences, the variations of human languages and speech patterns, and the need to segment speech into a sequence of words. (In most spoken languages, the sounds representing successive letters blend into each other, so the conversion of the modulated wave form to discrete characters or tokens can be a very difficult process.)

Crude systems of a decade ago required much training with a specific speaker’s voice to show much effectiveness. Today, the range and ability to use these systems without training has markedly improved.

Until recently, improvements largely were driven by military and intelligence requirements. Today, however, with the ubiquity of smart phones and speech interfaces, the consumer market is greatly accelerating progress.

#6 Image Recognition

Image recognition is the ability to determine whether or not an electronic image contains some specific object, feature, or activity, and then to extract the image data associated with it. Today, under specific circumstances and for specific tasks, this can be done by computer. However, for the general case of arbitrary objects in arbitrary situations this challenge has not yet been fully met. The systems of today work best for simple geometric objects (e.g., polyhedra), human faces, printed or hand-written characters, or vehicles, and in specific situations, typically described in terms of well-defined illumination, background, and orientation of the object relative to the camera.

Auto license recognition at intersections, face recognition by security cameras, and greatly expanded and improved character recognition systems (machine vision) represent some of the current state-of-the-art. Again, smart phone apps are helping to drive advances.

#7 Interoperability Standards and Methods

Rapid Progress in Climbing the Data Federation Pyramid

Most of the previous advances are related to extracting structured information or mapping or deriving additional structured information. Once obtained, of course, the next challenge is in how to relate that information together; that is, how to make it interoperate.

We have been steadily climbing a data federation pyramid [6] — and at an impressively accelerating rate since the adoption of the Internet and Web. These network innovations gave us a common basis and protocols for connecting distributed devices. That, in turn, has freed us to concentrate on the standards for data representation and interoperability.

XML first provided a means for a common data serialization that encouraged various communities and industries to devise exchange vocabularies. RDF provided a means for a common data model, one that was both simple and extensible at the same time [7]. OWL built upon that basis to enable us to build common domain models (see next).

There are alternatives to the semantic Web standards of RDF and OWL such as common logic and there are many competing data exchange formats to XML. None of these standards is essential on its own and all have their communities and advocates. However, because they are standards and they share common network bases, it has also been relatively easy to convert amongst the various available protocols. We are nearly at a global level where everything is connected, machine-readable, and in structured form.

#7 Common Domain Models

Semantics in machine-readable form means that we can more confidently link and combine available information. We are seeing a veritable explosion of domain models to represent various domains and viewpoints in consensual, interoperable form. What this means is that we are now gaining the computing vocabularies and grammars — along with shared community models (world views) — to get this stuff to work together.

Five years ago we called this phenomena mashups, but no one uses that term any longer because these information brewpots are everywhere, including in our very hands when we interact with the apps on our smart phones. This glue of domain models is generally as invisible to us as is the glue in laminates or the resin in plastics. But they are the strength and foundations nonetheless that enable much of the computing magic unfolding around us.

#9 Virtual Apps (Cloud Computing)

Once the tyranny of physical separation was shattered between data and machine by the network, the rationale for keeping the data with the app or even the user with the app disappeared. Cloud computing may seem mysterious or sound to have some high-octave hum, but it really is nothing more than saying that the Web enables us to treat all of our computing resources as virtual. Data can be anywhere; machines and hard drives can be anywhere; and applications can be anywhere.

And, virtualness brings benefits in and of itself. Whole computing environments can be installed or removed nearly instantaneously. Peak computing demands can be met with virtual headrooms. Backup and rollover and redundancy practices and strategies can change. Web services mean tailored capabilities can be invoked from anywhere and integrated for local needs. Massive computing resources and server farms can be as accessible to the individual as they are to prior computing behemoths. Combined with continued advances in underlying computing hardware and chips, the computing power available to any user is rising exponentially. There is now even more power in the power curve.

#10 Big Data

One hears stories of Google or the National Security Agency having access and managing servers measured in the hundreds of thousands. Entirely new operating systems and computing environments — many with roots in open source — such as virtual operating systems and MapReduce approaches like Hadoop have been innovated to deal with the current era of “big data”.

MapReduce is a framework for processing huge datasets using a large number of servers. The “map” step partitions the problem into tractable sub-problems, organized in a tree structure. The “reduce” step then takes the answers to all the sub-problems and combines them to produce the final output.

Such techniques enable analysis of datasets of a size impossible before. This has enabled the development of statistics and analytical techniques that have been able to make correlations and find patterns for some of the Grand Challenge tasks noted before that simply could not be addressed within previous limits. The “big data” approach is providing a brute force alternative to previously intractable problems.

Why Such Progress?

Declining hardware costs and increasing performance (such as from Moore’s Law), combined with the adoption of the Internet + Web network, set the fertile conditions for these unprecedented advances in computing’s Grand Challenges. But the adaptive radiation in innovations now occurring has its own dynamics. In computing terms, we are seeing the equivalent of the Cambrian explosion in evolutionary history.

The dynamics driving this computing explosion are based largely, I believe, on the statistics of information retrieval and extraction needed to cope with the scale of documents on the Web. That, in turn, has impelled innovations in big data and distributed architectures and designs that have pried open previously closed computing lockboxes. As data from everywhere and from every provenance pours into the system, means for handling and interoperating with it have become imperatives. These forces, in turn, have been channeled and are being met through the open and standards-based approaches that helped lead to the development of the Internet and its infrastructure in the first place.

These powerful evolutionary forces in computing are clearly evident in the ten Grand Challenge advances above. But the challenges above are also silent on another factor, underpinning the interoperability initiatives, that is only now just becoming evident and exerting its own powerful force. That is the workable, intellectual foundations for interoperability itself.

Clearly, as the advances in the Grand Challenges show, we are seeing immense exposures of new structured information and impressive means for accessing and managing it on a global, distributed scale.  Yet all of this data and this structure begs the question of how to get the information to work together. Further, the sources and viewpoints and methods by which all of this data has been created also puts a huge premium on means to deal with the diversity. Though not evident, and perhaps not even known to many of the innovators and practitioners, there has been a growing intellectual force shaping our foundational views about the nature of things and their representations. This force has been, I believe, one of those root cause drivers helping to show the way to interoperability.

John Sowa, despite his unending criticism of the semantic Web in favor of common logic, has nonetheless been a very positive evangelist for the 19th century American logician and philosopher, Charles Sanders Peirce. Sowa points out that the entire 20th century largely neglected Peirce’s significant contributions in many areas and some philosophers appropriated Peircean insights without proper attribution [8]. Indeed, Peirce has only come to wider attention within the past decade or so. Much of his voluminous lifetime writings have still not yet been committed to publication.

Among many notable contributions, Peirce was passionate about signs and their triadic representations, in a field known as semiotics. The philosophical and logical basis of his triangle of signs deserves your attention, which can not be adequately treated here [9]. However, as summarized by Sowa [8], “A semiotic view of language and logic gets to the heart of the philosophical controversies and their practical implications for linguistics, artificial intelligence, and related subjects.”

In essence, Peirce’s triadic logic of semiotics helps clarify philosophical questions about things, how they are perceived and how they are named that has vexed philosophers at least since the time of Aristotle. What Peirce was able to put forward was a testable logic for how things and the names of things can be understood and related to one another, via logical statements or structures. These, in turn, can be symbolized and formalized into logical constructs that can capture the structure of natural language as well as more structured data.

The clarity of Peirce’s logic of signs is an underlying factor, I believe, for why we are finally seeing our way clear to how to capture, represent and relate information from a diversity of sources and viewpoints that is defensible and interoperable [10]. As we plumb Peircean logics further, I believe we will continue to gain additional insights and methods for combining and relating information. The next phase of our advances on these Grand Challenges is likely to be fueled more by connections and interoperability than in basic extraction or representation.

The Widening Explosion

We are not seeing the vision of artificial intelligence unfold as posed three decades ago. Nor are we seeing the AI-complete type of problems being solved in their entirety [11]. Rather, we are seeing impressive but incomplete approaches. Full automation and autonomy are not yet at hand, and may be so far in the future as to never be. But we are nevertheless seeing advances across the board in all Grand Challenge areas.

What is emerging is a practical achievement of the Grand Challenges, the scale and scope of which is unprecedented in symbolic computing. As we see Peircean logic continue to take hold and interoperability grow in usefulness and stature, I think it fair to say we can look back in ten years to describe where we stand today as having been in the midst of an evolutionary explosion.

[1] Grand Challenges were United States policy objectives for high-performance computing and communications research set in the late 1980s. According to “A Research and Development Strategy for High Performance Computing”, Executive Office of the President, Office of Science and Technology Policy, 29 pp., November 20, 1987, “A grand challenge is a fundamental problem in science or engineering, with broad applications, whose solution would be enabled by the application of high performance computing resources that could become available in the near future.”
[2] For example, as of July 17, 2011, Google offered 63 different source or target languages for translation.
[3] Tim Berners-Lee, James Hendler and Ora Lassila, 2001. “The Semantic Web”. Scientific American Magazine; see
[4] Go to Sweet Tools, and enter the search ‘information extraction’ to see a list of about 85 tools.
[5] See, for example, Roberto Navigli, 2009. “Word Sense Disambiguation: A Survey,” ACM Computing Surveys, 41(2), 2009, pp. 1–69. See
[6] M.K. Bergman, 2006. “Climbing the Data Federation Pyramid,” AI3:::Adaptive Information blog, May 25, 2006; see
[7] M. K. Bergman, 2009. “Advantages and Myths of RDF,” AI3:::Adaptive Information blog, April 8, 2009. See
[8] John Sowa, 2006. “Peirce’s Contributions to the 21st Century”, in H. Schärfe, P. Hitzler, & P. Øhrstrøm, eds., Conceptual Structures: Inspiration and Application, LNAI 4068, Springer, Berlin, 2006, pp. 54-69. See
[9] See, as a start, the Wikipedia article on Charles Sanders Peirce (pronounced “purse”), as well as the Arisbe collection of his assembled papers (to date). Also see John Sowa, 2010. “The Role of Logic and Ontology in Language and Reasoning,” from Chapter 11 of Theory and Applications of Ontology: Philosophical Perspectives, edited by R. Poli & J. Seibt, Berlin: Springer, 2010, pp. 231-263. See Sowa also says, “Although formal logic can be studied independently of natural language semantics, no formal ontology that has any practical application can ever be developed and used without acknowledging its intimate connection with NL semantics.”
[10] While Peirce’s logic and clarity of conceptual relationships is compelling, I find reading his writings quite demanding.
[11] In the field of artificial intelligence, the most difficult problems are informally known as AI-complete or AI-hard, meaning that the difficulty of these computational problems is equivalent to solving the central artificial intelligence problem of making computers as intelligent as people. Computer vision, autonomous robots and understanding natural language are amongst challenges recognized by consensus as being AI-complete. However, practical advances on the Grand Challenges were never defined as needing to meet the AI-complete criterion. Indeed, it is even questionable whether such a hurdle is even worthwhile or meaningful on its own.

Posted by AI3's author, Mike Bergman Posted on July 18, 2011 at 10:00 pm in Adaptive Innovation, Semantic Web, Structured Web | Comments (3)
The URI link reference to this post is:
The URI to trackback this post is:
Posted:June 2, 2011

Schema.orgContrary to Some Views, Google and Co.’s Microdata Effort will Also Boost RDF

In my opinion, perhaps the most important event for the structured Web since RDF was released a dozen years ago was today’s joint announcement by the search engine triumvirate of Google, Bing and Yahoo! releasing is a vendor specification for nearly 300 mini-schema (or structured record definitions) that can be used to tag information in Web pages. These schema are organized into a clean little hierarchy and cover many of the leading things — from organizations to people to products and creative works — that can be written about and characterized on the Web.

These schema specifications are based on the microdata standard presently under review as part of the pending HTML5 specification. Microdata are set record descriptions of key-value pair attributes that can be embedded into the HTML Web page language. These microdata schema are similar to microformats, but broader in coverage and extensible. Microdata is also simpler than RDFa, another W3C specification that the organizers call “. . . extensible and very expressive, but the substantial complexity of the language has contributed to slower adoption.”

Is the Initiative a Slap in RDF’s Face?

Various forums have been alive with howls and questions from many RDF and RDFa advocates that this initiative negates years of effort behind those formats. Yet I and my company, Structured Dynamics, which base our entire technology approach on semantics and RDF, do not see this announcement as a threat or rejection. What gives; what is the difference in perspective?

In our view, RDF and its triple representations in its data model, is the simplest and most expressive means to represent any data or any data relationship. As such, RDF, and its language extensions such as OWL and ontologies, provide a robust and flexible canonical data model for capturing any extant data or schema. No matter what the native form of the source information, we can boil it down to RDF and inter-relate it to any other information. It is for these reasons (and others) we have frequently termed RDF as the universal data solvent.

But, simple records and simple data need not be encumbered with the complexity of RDF. We have long argued for the importance of naive data structs. Many of these are simple key-value pairs where the subject is implied. The so-called little structured data records in Wikipedia, called infoboxes, are of this form. JSON and many simple data formats also have cleaner data formats.

The basic fact that RDF provides a universal data model for any kind of native data does not necessarily translate into its use as the actual data exchange format. Rather, winning data exchange formats are those that can be easily understood, easily expressed and therefore widely used. I think there is a real prospect that microdata, ready for ingest and expression by the Web’s leading search engines, may represent a real sea change in the availability and expression of structured data on the Web.

More structure — not less — is the real fuel that will promote greater adoption of RDF when it comes time to interoperate that data. The RDF community should rejoice that more structure will be coming to the Web from Google et al.’s announcement. We should also soon see an explosion of tools and utilities and services that make it easy to automatically add such structure to Web pages via single clicks. Then, once this structure is available, watch out!

So, while the backers of also announced their continued support for microformats and RDFa as they presently exist, I rather suspect today’s announcement represents a denouement for these alternative formats. Though these formats may be creatively destroyed, I think the effect on RDF itself will be a profound and significant boost. I foresee clarity coming to the marketplace regarding RDF’s role:  as a canonical means for expressing data of any form, and not necessarily as a data exchange format.

The Initiative is No Surprise

This initiative, led by Google, should be no surprise. Google is the registered agent for the Web site and has been the key proponent of microdata via its support of Ian Hickson in the WhatWG and HTML5 work groups. As I stated a couple of years back, Google has also not hidden its interests in structured data. Practically daily we see more structured data appear in Google search results and it has maintained a very active program in structured data extraction from text and tables for some years.

Google and its search engine partners recognize that search needs are evolving from keyword retrievals to structure, relationships, and filtered, targeted results. Those advances come from structure — as well as the semantic relationships between things that something like the begins to represent.

Many within the W3C and elsewhere questioned why Google was pushing microdata when there were competing options such as microformats or RDFa (or even earlier variants). Of course, like Microsoft of a decade earlier, some ascribed Google’s microdata advocacy as arising from commercial interests or clout in advertising alone. Of course Google has an economic interest in the growth and usefulness of the Web. But I do not believe its advocacy to be premised on clout or “my way or the highway.”

Google and the search engine triumvirate understand well — much better than many of the researchers and academics that dominate mailing list discussions — that use and adoption trump elegance and sophistication. When one deconstructs the design of microdata and the nearly 300 schema now released behind it, I think the pragmatic observer can only come to one conclusion: Job well done!

Why This is Exciting

I have been a fervent RDF advocate for nearly a decade and have also been a vocal proponent of the structured Web as a necessary stepping stone to the semantic Web. In fact, here is a repeat of a diagram I have used many times over the past 5 years:

Transition in Web Structure
Document Web Structured Web
Semantic Web
Linked Data
  • Document-centric
  • Document resources
  • Unstructured data and semi-structured data
  • HTML
  • URL-centric
  • circa 1993
  • Data-centric
  • Structured data
  • Semi-structured data and structured data
  • XML, JSON, RDF, etc
  • URI-centric
  • circa 2003
  • Data-centric
  • Linked data
  • Semi-structured data and structured data
  • RDF, RDF-S
  • URI-centric
  • circa 2007
  • Data-centric
  • Linked data
  • Semi-structured data and structured data
  • URI-centric
  • circa ???

When one looks at the schema of schema that accompany today’s announcement, it is really clear just how encompassing and important these instant standards will become:
























Organization (con’t)



Today’s announcement is the best news I have heard in years regarding the structured Web, RDF, and the semantic Web. This announcement is — I believe — the signal event of the structured Web. With regard to my longstanding diagram above, I can go to bed tonight knowing we have now crossed the threshold into the semantic Web.

Posted by AI3's author, Mike Bergman Posted on June 2, 2011 at 8:57 pm in Adaptive Information, Structured Web | Comments (7)
The URI link reference to this post is:
The URI to trackback this post is:
Posted:March 18, 2011

Writing and Sharing Data Can be Lightened Up Friday     Brown Bag Lunch

Ever since I first started to learn in earnest about ontology, something has been gnawing at me. The term seemed to be (shall I say?) an obtuse one whose obscurity was not the result of subtle precision or technicality, but rather one of fuzziness. As I introduced my Intrepid Guide to Ontology two years ago, I noted:

The root of the [ontology] term is the Greek ontos, or being or the nature of things. Literally and in classical philosophy, ontology was used in relation to the study of the nature of being or the world, the nature of existence. Tom Gruber, among others, made the term popular in relation to computer science and artificial intelligence about 15 years ago when he defined ontology as a “formal specification of a conceptualization.”

Simple Data StructsSince then, I have continued to find ontology one of the hardest concepts to communicate to clients and quite a muddled mess even as used by practitioners. I have come to the conclusion that this problem is not because I have failed to grasp some ephemeral nuance, but because the term as used in practice is indeed fuzzy and imprecise.

What Isn’t an Ontology?

Even two years ago, I noted more than 40 different types of information structure that have at one time or another been labelled as an example of an “ontology”:

Since then, I could add even more terms to this list.

Lack of precision as to what ontology means has meant that it has been sloppily defined. As I have harped upon many times regarding semantic Web terminology, this is a sad state of affairs for the semWeb endeavor that has meaning at the core of its purpose.

I’m pretty sure that the original intent in embracing the concept of ontology within the realm of knowledge representation was not to see this term so broadly misused or mis-applied. I suspect, as well, that if we could sharpen up our understanding and remove some of the fuzziness that we could improve communications with the lay public across many levels of the semWeb enterprise.

The Useful Distinction of the TBox and ABox

Recently, I have been looking to the semantic Web’s roots in description logics. One of my writings, Thinking ‘Inside the Box’ with Description Logics, looked at the conceptual distinctions between the so-called ‘TBox‘ and ‘ABox‘. That is, a knowledge base is a logical schema of roles and concepts and the relationships between them (the TBox), which is populated by the actual data (instances) asserting memberships and attributes (“facts”) (the ABox).

By analogy, in a conventional relational database system, the database or logical schema would correspond to the TBox; the actual data records or tables would correspond to the ABox. Often, the term ontology is used to cover both ABox and TBox statements (which, I argue, only makes the understanding of the ‘ontology’ concept more difficult).

My recent writing, Back to the Future with Description Logics, discussed at some length the advantages of keeping the TBox and ABox separate. This current article now expands on those thoughts, particularly with respect to the definition and understanding of ontology.

The starting point for this new mindset is to return to the ideas of data records or data tables v. the logical schema that is prevalent in relational databases.

So Many Structs, So Little Time

The last time I took a census, about a year ago, there were more than 100 converters of various record and data structure types to RDF [2]. These converters — also sometimes known as translators or ‘RDFizers’ — generally take some input data records with varying formats or serializations and convert them to a form of RDF serialization (such as RDF/XML or N3), often with some ontology matching or characterizations. That last census listed these converters:

  • RDF
    • Serialization formats:
      • RDF/XML
      • N3
      • Turtle
    • Automatically recognized ontologies:
      • SIOC
      • SKOS
      • FOAF
      • AtomOWL
      • Annotea
      • Music Ontology
      • Bibliographic Ontology
      • EXIF
      • vCard
      • Others
  • (X)HTML pages
  • HTML header metadata
    • Dublin Core
  • Embedded microformats
    • eRDF
    • RDFa
    • hCard
    • hCalendar
    • XFN
    • xFolk
  • Syndication Formats:
    • RSS 2.0
    • Atom
    • OPML
    • OCS
    • XBEL (for bookmarks)
  • GRDDL [1]
  • REST-style Web service APIs:
    • Google Base
    • Flickr
    • Ning
    • Amazon
    • eBay
    • Freebase
    • Facebook
    • raw HTTP
    • Etc.
  • Files (multitude of file formats and MIME types, including):
    • MS Office
    • OpenOffice
    • Open Document Format
    • images
    • audio
    • video
    • Etc.
  • Web services:
    • BPEL
    • WSDL
    • XBRL
    • XBEL
  • Data exchange formats
    • iCalendar
    • vCard
  • Virtuoso VADs
  • OpenLink license files
  • Third party metadata extraction frameworks:
Note that MIT’s SIMILE RDFizers also recognizes these formats:

There is a growing list of third-party RDFizers as well:

This wealth of formats shows the robustness of the RDF data model to capture structure and data relationships from virtually any input form. This is what makes RDF so exciting as a canonical target for getting data to interoperate.

Let’s Make this Elementary, Dr. Watson

However — and this is crucial — standard users for decades have preferred simple, text-based and human readable formats for writing and transferring their structured data.

These various forms, sometimes well specified with APIs and sometimes almost ad hoc as in spreadsheet listings, are what we call ‘structs‘. Structs can all be displayed as text and have, at minimum, explicit or inferrable key-value pairs to convey data relationships and attributes, with data types and values often noted by various white space, delimiter or other text conventions.

There is no doubt that the vast majority of extant data is found in such formats, including the results of data or information extraction from unstructured text. Indeed, even HTML and many markup languages with their angle bracket-delimited fields fall into this category.

There have literally been hundreds of various formats proposed over decades for conveying lightweight data structures. Most have been proprietary or limited to specific domains or users. Some, such as fielded text, structured text, simple declarative language (SDL), or more recently YAML or its simpler cousin JSON, have become more widely adopted and supported by formal specifications, tools or APIs. JSON, especially, is a preferred form for Web 2.0 applications.

Some, like microformats or this example BibTeX record below (with some non-standard extensions), rely less on syntax conventions and may use reserved keywords (such as AUTHOR or TITLE as shown) to signal the key type for the key-value pair:

ID_LOCAL arXiv:0711.3808
AUTHOR <a href="#Schramm_O">Oded Schramm</a>
ID arXiv:0711.3808
JOURNAL Electron. Res. Announc. Math. Sci.
PAGES 17--23
TITLE Hyperfinite graph limits
15, 2008 Journal Issue
YEAR 2008

Some of these simple formats have been more successful than others, though none have achieved market dominance. There also appear to be few universal principles that have emerged as to syntax or format. Nonetheless, any of these various struct forms are easy for casual readers to understand and easy for domain experts to write.

For modeling and interoperability purposes, many of these forms are patently inadequate. That is why many of these simpler forms might be called “naïve”: they achieve their immediate purpose of simple relationships and communication, but require understood or explicit context in order to be meaningfully (semantically) related to other forms or data.

Yet, if we have learned nothing else with the phenomenal success of the Web it is this: simplicity trumps elegance or expressivity.

RDF and the Skinny ABox

The RDF (Resource Description Framework) data model is expressed as simple subject-predicate-object “triple” statements. That sounds fancy, but just substitute verb for predicate and noun for subject and object. In other words: Dick sees Jane; or, the ball is round. It may sound like a kindergartner reader, but it is how data can be easily represented and built up into more complex structures and stories.

RDF triples can be applied equally to all structured, semi-structured and unstructured content. RDF is clearly a most capable data model that — through its ability to be extended with further concepts and relationships (vocabulary) — can create elegant and logical structures to represent comprehensive domains and knowledge bases. Finding such a model has been a quest in my professional life; I believe we finally have a winner to facilitate data interoperability using RDF.

But RDF has not achieved the market acceptance that its suitability as a data representation model might suggest. I think there are three reasons for this:

  • First, RDF was first presented and “sold” as an XML serialization. This failing has been well understood for some time. This unfortunate early linkage of RDF caused confusion between data model and the XML syntax. The rather simple and incremental building blocks of triple RDF statements when presented in the nested XML syntax led to lengthy and hard-to-read specifications (for easier reading and use, see either the N3 or Turtle syntaxes)
  • Second, triples by definition are 50% more complicated than a key-value pair. While the basic RDF statement might be simple like a Dick-and-Jane reader, as a data specification format it is still more complex than my personal attributes of sex:Male and hair:Red and born:California. Those three “facts” can not be said nearly so quickly in RDF. And, if we also adhere to linked data, each one of these items requires a URI unique identifier to boot! It is important not to ignore the desire for simple and human readable data-specification formats
  • Third, as this entry began and as we will conclude, RDF and its fuzzy relationship to ontology has led to over-specification of what needs to be included in the data record. What could simply be a record specification of an object and its attributes presented as simple key-value pairs has become burdened with “ontology” and “conceptual” relationships.

Canonical forms embody all of the specification that the canon guiding them requires. What we may have failed to see in embracing RDF, however, is that getting useful data into the system need not carry all of this burden.

Lightening Up and Shifting Work to the TBox

So, what does all of this have to do with my starting diatribe about the term ontology?

Whether a single database or the federation across all information known to human kind, we have data records (structs of instances) and a logical schema (ontology of concepts and relationships) by which we try to relate this information. This is a natural and meaningful split: structure and relationships v. the instances that populate that structure.

Stated this way, particularly for anyone with a relational database background, the split between schema and data is clear and obvious. Yet, the RDF, semantic Web and linked data communities have done an abysmal job of recognizing this fundamental separation of concerns.

We create “ontologies” that mix instances and schema. We insist on simple data record conversions that are burdened with relationship specifications as well. We tout a “linked data” infrastructure that is based solely on the same identity of instances without respect or attention to structure or conceptual relationships. We dismiss communities that work to express their data with useful local structures. We insist on standards and practices up and down the data staging and preparation chain that turns off the general market and makes us seem arrogant and dismissive. Frankly, in so many ways, we just don’t get it [3].

What has struck me personally over the past few months as these realizations have unfolded has been how much our own mindsets and language may be trapping us.

  • Does existing structured data need to be expressed as RDF in order to be useful and integrated?
  • Exposing linked and instance data is great, but to what end; what are the conceptual or structural schema?
  • Why is our standards process so inward looking and parochial (often petty)? What purpose or who does this serve?

At least for this diatribe, my essential conclusion is that we need to shift the burden of the schema and conceptual relations and (yes) world views to the TBox. We need to skinny down the ABox and make it a warm and welcoming environment by which any structured data (including the most naïve) can join.

So, ultimately, the bottom line is this: the burden of the semantic Web rests on us, not the providers of structured data.

It is time to streamline the ABox to smooth data contributions, assume as publishers the responsibility for the TBox, and keep those concerns separate. As for instance-related stuff, I now intend to refer to them as structs governed by a controlled vocabulary (at most). I intend to reserve ontology as a means to describe a given world view, a TBox, the schema and its relations of the domain at hand. And, frankly, this definition of ontology brings it back in balance with its roots in ontos and the nature of the world.

It’s a good time to lighten up!

Friday      Brown Bag Lunch This Friday brown bag leftover was first placed into the AI3 refrigerator on January 22, 2009, and is one of the more popular historical posts of this blog.  This reprise is unchanged since its original posting, though we have continued to make progress on constructs such as irON to capture this idea. Microdata in HTML5 is also an important contribution, to which we will devote some attention in the near future.

[1] GRDDL (Gleaning Resource Descriptions from Dialects of Languages) is a W3C markup format for getting RDF data out of XML and XHTML documents using explicitly associated transformation algorithms, typically represented in XSLT GRDDL accomodates a wide variety of dialects (see one listing) and can be combined with arbitrary transformation mechanisms (though currently mostly based on XSLTs).
[2] Also see the listing of “dynamic” RDFizers at
[3] I don’t mean to imply that there are not those in the community interested in lightweight data structures or their conversion, just that they have been more of a minority to date. For example, the 5th Workshop on Scripting and Development for the Semantic Web is coming up this summer in conjunction with the 6th European Semantic Web Conference in Crete, Greece; this year’s organizers are Gunnar Aastrand Grimnes (DFKI Knowledge Management Lab), Chris Bizer (Freie Universität Berlin) and Sören Auer (Universität Leipzig). As other examples focusing on JSON, there are a couple of efforts to define representation conventions from Talis and GBV for serializing RDF; Jim Ley, Kanzaki Masahide and Dave Beckett (likely among others) have written simple and straightforward RDF and Turtle parsers and converters; there was a floated idea for an RDF version of JSON called RDFON that has now evolved into the TURF approach; and JDIL (JSON data integration layer) instructs how to add namespaces to JSON to enable encoding RDF. Still further examples are Beckett’s Triplr and Auer’s ASKW Triplify lightweight conversion services involving many different formats. These are all laudable efforts with good relevance to a lighter ABox approach, I think.

Posted by AI3's author, Mike Bergman Posted on March 18, 2011 at 2:08 am in Adaptive Information, Brown Bag Lunch, irON, Structured Web | Comments (2)
The URI link reference to this post is:
The URI to trackback this post is:
Posted:February 28, 2011

Photo courtesy goldonomic.comWikipedia + UMBEL + Friends May Offer One Approach

In the first part of this series we argued for the importance of reference structures to provide the structures and vocabularies to guide interoperability on the semantic Web. The argument was made that these reference structures are akin to human languages, requiring a sufficient richness and terminology to enable nuanced and meaningful communications of data across the Web and within the context of their applicable domains.

While the idea of such reference structures is great — and perhaps even intuitive when likened to human languages — the question is begged as to what is the basis for such structures? Just as in human languages we have dictionaries, thesauri, grammar and style books or encyclopedia, what are the analogous reference sources for the semantic Web?

In this piece, we tackle these questions from the perspective of the entire Web. Similar challenges and approaches occur, of course, for virtually every domain and specific community. But, by focusing on the entirety of the Web, perhaps we can discern the grain of sand at the center of the pearl.

Bootstrapping the Semantic Web

The idea of bootstrapping is common in computers, compilers or programming. Every computer action needs to start from a basic set of instructions from which further instructions or actions are derived. Even starting up a computer (“booting up”) reflects this bootstrapping basis. Bootstrapping is the answer to the classic chicken-or-egg dilemma by embedding a starting set of instructions that provides the premise at start up [1]. The embedded operand for simple addition, for example, is the basis for building up more complete mathematical operations.

So, what is the grain of sand at the core of the semantic Web that enables it to bootstrap meaning? We start with the basic semantics and “instructions” in the core RDF, RDFS and OWL languages. These are very much akin to the basic BIOS instructions for computer boot up or the instruction sets leveraged by compilers. But, where do we go from there? What is the analog to the compiler or the operating system that gives us more than these simple start up instructions? In a semantics sense, what are the vocabularies or languages that enable us to understand more things, connect more things, relate more things?

To date, the semantic Web has given us perhaps a few dozen commonly used vocabularies, most of which are quite limited and simple pidgin languages such as DC, FOAF, SKOS, SIOC, BIBO, etc. We also have an emerging catalog of “things” and concepts from Wikipedia (via DBpedia) and similar. (Recall, in this piece, we are trying to look Web-wide, so the many fine building blocks for domain purposes such as found in biology, medicine, finance, astronomy, etc., are excluded.) The purposes and scope of these vocabularies widely differ and attack quite different slices of the information space. SKOS, for example, deals with describing simple knowledge structures like taxonomies or thesauri; SIOC is for describing social media.

By virtue of adoption, each of these core languages has proved its usefulness and role. But, as skew lines in space, how do these vocabularies relate to one another? And, how can all of the specific domain vocabularies also relate to those and one another where there are points of intersection or overlap? In short, after we get beyond the starting instructions for the semantic Web, what is our language and vocabulary? How do we complete the bootstrap process?

Clearly, like human languages, we need rich enough vocabularies to describe the things in our world and a structure of the relationships amongst those things to give our communications meaning and coherence. That is precisely the role provided by reference structures.

The Use and Role of ‘Gold Standards’

To prevent reference structures from being rubber rulers, some fixity or grounding needs to establish the common understanding for its referents. Such fixed references are often called ‘gold standards‘. In money, of course, this used to be a fixed weight of gold, until that basis was abandoned in the 1970s. In the metric system, there are a variety of fixed weights and measures that are employed. In the English language, the Oxford English Dictionary (OED) is the accepted basis for the lexicon. And so on.

Yet, as these examples show, none of these gold standards is absolute. Money now floats; multiple systems of measurement compete; a variety of dictionaries are used for English; most languages have their own reference sets; etc. The key point in all gold standards, however, is that there is wide acceptance for a defined reference for determining alignments and arbitrating differences.

Gold standards or reference standards play the role of referees or arbiters. What is the meaning of this? What is the definition of that? How can we tell the difference between this and that? What is the common way to refer to some thing?

Let’s provide one example in a semantic Web context. Let’s say we have a dataset and its schema A that we are aligning with another dataset with schema B. If I say two concepts align exactly across these datasets and you say differently, how do we resolve this difference? On one extreme, each of us can say our own interpretation is correct, and to heck with the other. On the other extreme, we can say both interpretations are correct, in which case both assertions are meaningless. Perhaps papering over these extremes is OK when only two competing views are in play, but what happens when real problems with many actors are at stake? Shall we propose majority rule, chaos, or the strongest prevails?

These same types of questions have governed human interaction from time immemorial. One of the reasons to liken the problem of operability on the semantic Web to human languages, as argued in Part I, is to seek lessons and guidance for how our languages have evolved. The importance of finding common ground in our syntax and vocabularies — and, also, critically, in how we accept changes to those — is the basis for communication. Each of these understandings needs to be codified and documented so that they can be referenced, and so that we can have some confidence of what the heck it is we are trying to convey.

For reference structures to play their role in plugging this gap — that is, to be much more than rubber rulers — they need to have such grounding. Naturally, these groundings may themselves change with new information or learning inherent to the process of human understanding, but they still should retain their character as references. Grounded references for these things — ‘gold standards’ — are key to this consensual process of communicating (interoperating).

Some ‘Gold Standards’ for the Semantic Web

The need for gold standards for the semantic Web is particularly acute. First, by definition, the scope of the semantic Web is all things and all concepts and all entities. Second, because it embraces human knowledge, it also embraces all human languages with the nuances and varieties thereof. There is an immense gulf in referenceability from the starting languages of the semantic Web in RDF, RDFS and OWL to this full scope. This gulf is chiefly one of vocabulary (or lack thereof). We know how to construct our grammars, but we have few words with understood relationships between them to put in the slots.

The types of gold standards useful to the semantic Web are similar to those useful to our analogy of human languages. We need guidance on structure (syntax and grammar), plus reference vocabularies that encompass the scope of the semantic Web (that is, everything). Like human languages, the vocabulary references should have analogs to dictionaries, thesauri and encyclopedias. We want our references to deal with the specific demands of the semantic Web in capturing the lexical basis of human languages and the connectedness (or not) of things. We also want bases by which all of this information can be related to different human languages.

To capture these criteria, then, I submit we should consider a basic starting set of gold standards:

  • RDF/RDFS/OWL — the data model and basic building blocks for the languages
  • Wikipedia — the standard reference vocabulary of things, concepts and entities, plus other structural guidances
  • WordNet — lexical language references as an aid to natural language processing, and
  • UMBEL — the structural reference for the connectedness of things for basic coherence and inference, plus a vocabulary for mapping amongst reference structures and things.

Each of these potential gold standards is next discussed in turn. The majority of discussion centers on Wikipedia and UMBEL.

RDF/RDFS/OWL: The Language

Naturally, the first suggested gold standard for the semantic Web are the RDF/RDFS/OWL language components. Other writings have covered their uses and roles [2]. In relation to their use as a gold standard, two documents, one on RDF semantics [3] and the other an OWL [4] primer, are two great starting points. Since these languages are now in place and are accepted bases of the semantic Web, we will concentrate on the remaining members of the standard reference set.

Wikipedia: The Vocabulary (and More)

The second suggested gold standard for the semantic Web is Wikipedia, principally as a sort of canonical vocabulary base or lexicon, but also for some structural aspects. Wikipedia now contains about 3.5 million English articles, by far larger than any other knowledge base, and has more than 250 language versions. Each Wikipedia article acts as more or less a reference for the thing it represents. In addition, the size, scope and structure of Wikipedia make it an unprecedented resource for researchers engaged in natural language processing (NLP), information extraction (IE) and semantic Web-related tasks.

For some time I have been maintaining a listing called SWEETpedia of academic and research articles focused on the use of Wikipedia for these tasks. The latest version tracks some 250 articles [5], which I guess to be about one half or more of all such research extant. This research shows a broad variety of potential roles and contributions from Wikipedia as a gold standard for the semantic Web, some of which is detailed in the tables below.

An excellent report by Olena Medelyan et al. from the University of Waikato in New Zealand, Mining Meaning from Wikipedia, organized this research up through 2008 and provided detailed commentary and analysis of the role of Wikipedia [6]. They noted, for example, that Wikipedia has potential use as an encyclopedia (its intended use), a corpus for testing and modeling NLP tasks, as a thesaurus, a database, an ontology or a network structure. The Intelligent Wikipedia project from the University of Washington has also done much innovative work on “automatically learned systems [that] can render much of Wikipedia into high-quality semantic data, which provides a solid base to bootstrap toward the general Web” [7].

However, as we proceed through the next discussions, we’ll see that the weakest aspect of Wikipedia is its category structure. Thus, while Wikipedia is unparalleled as the gold standard for a reference vocabulary for the Web, and has other structural uses as well, we will need to look elsewhere for how that content is organized.

Major Wikipedia Initiatives

Many groups have recognized these advantages for Wikipedia, and have built knowledge bases around it. Also, many of these groups have also recognized the category (schema) weaknesses in Wikipedia and have proposed alternatives. Some of these major initiatives, which also collectively represent a large number of the research articles in SWEETpedia, include:

Project Schema Basis Comments
DBpedia Wikipedia Infoboxes excellent source for URI identifiers; structure extraction basis used by many other projects
Freebase User Generated schema are for domains based on types and properties; at one time had a key dependence on Wikipedia; has since grown much from user-generated data and structure; now owned by Google
Intelligent Wikipedia Wikipedia Infoboxes a broad program and a general set of extractors for obtaining structure and relationships from Wikipedia; was formerly known as KOG; from Univ of Washington
SIGWP Wikipedia Ontology the Special Interest Group of Wikipedia (Research or Mining); a general group doing research on Wikipedia structure and mining; schema basis is mostly from a thesaurus; group has not published in two years
UMBEL UMBEL Reference Concepts RefConcepts based on the Cyc knowledge base; provides a tested, coherent concept schema, but one with gaps regarding Wikipedia content; has 28,000 concepts mapped to Wikipedia
WikiNet Extracted Wikipedia Ontology part of a long-standing structure extraction effort from Wikipedia leading to an ontology; formerly known as WikiRelate; from the Heidelberg Institute for Theoretical Studies (HITS)
Wikipedia Miner N/A generalized structure extractor; part of a wider basis of Wikipedia research at the Univ of Waikato in New Zealand
Wikitology Wikipedia Ontology general RDF and ontology-oriented project utilizing Wikipedia; effort now concluded; from the Ebiquity Group at the Univ of Maryland
YAGO WordNet maps Wordnet to Wikipedia, with structured extraction of relations for characterizing entities


It is interesting to note that none of the efforts above uses the Wikipedia category structure “as is” for its schema.

Structural Sources within Wikipedia

The surface view of Wikipedia is topic articles placed into one or more categories. Some of these pages also include structured data tables (or templates) for the kind of thing the article is; these are called infoboxes. An infobox is a fixed-format table placed at the top right of articles to consistently present a summary of some unifying aspect that the articles share. For example, see the listing for my home town, Iowa City, which has a city infobox.

However, this cursory look at Wikipedia in fact masks much additional and valuable structure. Some early researchers noted this [8]. The recognition of structure has also been a key driver for the interest in Wikipedia as a knowledge base (in addition to its global content scope). The following table is a fairly complete listing of structure possibilities within Wikipedia (see Endnotes for any notes):

Wikipedia Structure Potential Applications Note
Entire Corpus
knowledge base; graph structure; corpus for n-grams, other constructions [9]
category suggestion; semantic relatedness; query expansion; potential parent category
Contained Articles
semantically-related terms (siblings)
hyponymic and meronymic relations between terms
Listing Pages/Categories
semantically-related terms (siblings)
Patterned Categories
functional metadata [9]
Infobox Templates
synonyms; key-value pairs
units of measure; fact extraction [9]
category suggestion; entity suggestion
coordinates; places; geolocational; (may also appear in full article text)
Issue Templates
Multiple Types
exclusion candidates; other structural analysis; examples include Stub, Message Boxes, Multiple Issues [9]
Category Templates [13]
Category Name
disambiguation; relatedness
Category Links
semantic relatedness
First Paragraph
definition; abstract
Full Text
complete discussion; related terms; context; translations; NLP analysis basis; relationships; sentiment
synonymy; spelling variations, misspellings; abbreviations; query expansion
named entities; domain specific terms or senses
category suggestion (phrase marked in bold in first paragraph)
Section Heading(s)
category suggestion; semantic relatedness [9]
See Also
related concepts; query expansion [9]
Further Reading
related concepts [9,10]
External Links
related concepts; external harvest points
Article Links
related terms; co-occurrences
synonyms; spelling variations; related terms; query expansion
link graph; related terms
category suggestion; functional metadata
category suggestion; functional metadata
external harvest points [9,10]
thumbnails; image recognition for disambiguation; controversy (edit/upload frequency) [11]
related concepts; related terms; functional metadata [9]
Disambiguation Pages
Article Links
sense inventory
Discussion Pages
Discussion Content
Redux for Article Structure
see Articles for uses
History Pages
Edit Frequency
topicalness; controversy (diversity of editors, reversions)
Edit Basis
lexical errors [9]
instances; named entity candidates
Alternate Language Versions
Redux for All Structures
see all items above; translation; multilingual alignment; entity disambiguation [12]

The potential for Wikipedia to provide structural understandings is evident from this table. However, it should be noted that, aside from some stray research initiatives, most effort to date has focused on the major initiatives noted earlier or from analyzing linking and infoboxes. There is much additional research that could be powered by the Wikipedia structure as it presently exists.

From the standpoint of the broader semantic Web, the potential of Wikipedia in the areas of metadata enhancement and mapping to multiple human languages [12] are particularly strong. We are only now at the very beginning phases of tapping this potential.

Structural Weaknesses

The three main weaknesses with Wikipedia are its category structure [14], inconsistencies and incompleteness. The first weakness means Wikipedia is not a suitable organizational basis for the semantic Web; the next two weaknesses, due to the nature of Wikipedia’s user-generated content, are constantly improving.

Our recent effort to map between UMBEL and Wikipedia, undertaken as part of the recent UMBEL v 1.00 release, spent considerable time analyzing the Wikipedia category structure [15]. Of the roughly half million categories in Wikipedia, only about 85,000 were found to be suitable candidates to participate in an actual schema structure. Further breakdowns are shown by this table resulting from our analysis:

Wikipedia Category Breakdowns
Removals 20.7%
Administrative 15.7%
Misc Cleaning 5.0%
Functional (not schema) 61.8%
Fn Dates 10.1%
Fn Nationalities 9.6%
Fn Listings, related 0.8%
Fn Occupations 1.0%
Fn Prepositions 40.4%
Candidates 17.4%
SuperTypes 1.7%
General Structure 15.7%
TOTAL 100.0%

Fully 1/5 of the categories are administrative or internal in nature. The large majority of categories are, in fact, not structural at all, but what we term functional categories, which means the category contains faceting information (such as subclassifying musicians into British musicians) [16]. Functional categories can be a rich source of supplementary metadata for its assigned articles — though, no one has yet processed Wikipedia in this manner — but are not a useful basis for structural conceptual relationships or inferencing.

This weakness in the Wikipedia category system has been known for some time [17], but researchers and others still attempt to do mappings on mostly uncleaned categories. Though most researchers recognize and remove internal or administrative categories in their efforts, using the indiscriminate remainder of categories still leads to poor precision in resulting mappings. In fact, in comparison to one of the more rigorous assessments to date [18], our analysis still showed a 6.8% error rate in hand inspected categories.

Other notable category problems include circular references, skipped intermediate categories, misassigned categories and incomplete assignments.

Nonetheless, Wikipedia categories do have a valuable use in the analysis of local relationships (one degree of relatedness) and for finding missing category candidates. And, as noted, the functional categories are also a rich and untapped source of additional article metadata.

Like any knowledge base, Wikipedia also has inconsistent and incomplete coverage of topics [19]. However, as more communities accept Wikipedia as a central resource deserving completeness, we should see these gaps continue to get filled.

The DBpedia Implementation

One of the first database versions of Wikipedia built for semantic Web purposes is DBpedia. DBpedia has an incipient ontology useful for some classification purposes. Its major structural organization is built around the Wikipedia infoboxes, which are applied to about a third of Wikipedia articles. DBpedia also has multiple language versions.

DBpedia is a core hub of Linked Open Data (LOD), which now has about 300 linked datasets; has canonical URIs used by many other applications; has extracted versions and tools very useful for further processing; and has recently moved to incorporate live updates from the source Wikipedia [20]. For these reasons, the DBpedia version of Wikipedia is the suggested implementation version.

WordNet: Language Relationships

The third suggested gold standard for the Semantic Web is WordNet, a lexical database for the English language. It groups English words into sets of synonyms called synsets, provides short, general definitions, and records the various semantic relations between these synonym sets. The purpose is twofold: to produce a combination of dictionary and thesaurus that is more intuitively usable, and to support automatic text analysis and artificial intelligence applications. There are over 50 languages covered by wordnet approaches, most mapped to this English WordNet [21].

Though it has been used in many ontologies [22], WordNet is most often mapped for its natural language purposes and not used as a structure of conceptual relationships per se. This is because it is designed for words and not concepts. It contains hundreds of basic semantic inconsistencies and also lacks much domain applicability. Entities, of course, are also lacking. In those cases where WordNet has been embraced as a schema basis, much work is generally expended to transform it into an ontology suitable for knowledge representation.

Nonetheless, for word sense disambiguation and other natural language processing tasks, as well as for aiding multi-lingual mappings, WordNet and its various other language variants is a language reference gold standard.

UMBEL: A Coherent Structure

So, with these prior gold standards we gain a basic language and grammar; a base (canonical) vocabulary and some structure guidance; and a reference means for processing and extracting information from input text. Yet two needed standards remain.

One needed standard is a conceptual organizing structure (or schema) by which the canonical vocabulary of concepts and instances can be related. This core structure should be constructed in a coherent [23] manner and expressly designed to support inferencing and (some) reasoning. This core structure should be sufficiently large to embrace the scope of the semantic Web, but not so detailed as to make it computationally inefficient. Thus, the core structure should be a framework that allows more focused and purposeful vocabularies to be “plugged in”, depending on the domain and task at hand. Unfortunately, the candidate category structures from our other gold standards in Wikipedia and WordNet do not meet these criteria.

A second needed standard is a bit of additional vocabulary “glue” specifically designed for the purposes of the semantic Web and ontology and domain incorporation. We have multiple and disparate world views and contexts, as well as the things described by them [24]. To get them to interoperate — and to acknowledge differences in alignment or context — we need a set of relational predicates (vocabulary) that can capture a range of mappings from the exact to the approximate [25]. Unlike other reference vocabularies that attempt to capture canonical definitions within defined domains, this vocabulary is expressly required by the semantic Web and its goal to federate different data and schema.

UMBEL has been expressly designed to address both of these two main needs [26]. UMBEL is a coherent categorization structure for the semantic Web and a mapping vocabulary designed for dataset and conceptual interoperability. UMBEL’s 28,000 reference concepts (RefConcepts) are based on the Cyc knowledge base [27], which itself is expressly designed as a common sense representation of the world with express variations in context supported via its 1000 or so microtheories. Cyc, and UMBEL upon which it is based, are by no means the “correct” or “only” representations of the world, but they are coherent ones and thus internally consistent.

UMBEL’s role to allow datasets to be “plugged in” and related through some fixed referents was expressed by this early diagram [28]:

Lightweight Binding to an Upper Subject Structure Can Bring Order
[Click on image for full-size pop-up]

The idea — which is still central to this kind of reference structure — is that a set of reference concepts can be used by multiple datasets to connect and then inter-relate. These are shown by the nested subjects (concepts) in the umbrella structure.

UMBEL, of course, is not the only coherent structure for such interoperability purposes. Other major vocabularies (such as LCSH; see below) or upper-level ontologies (such as SUMO, DOLCE, BFO or PROTON, etc.) can fulfill portions of these roles, as well. In fact, the ultimate desire is for multiple reference structures to emerge that are mapped to one another, similar to how human languages can inter-relate. Yet, even in that desired vision, there is still a need for a bootstrapped grounding. UMBEL is the first such structure expressly designed for the two needed standards.

Mappings to the Other Standards

UMBEL is already based on the central semantic Web languages of RDF, RDFS, SKOS, and OWL 2. The recent version 1.00 now maps 60% of UMBEL to Wikipedia, with efforts for the remaining in process. UMBEL provides mappings to WordNet, via its Cyc relationships. More of this is in process and will be exposed. And the mappings between UMBEL and GeoNames [29] for locational purposes is also nearly complete.

The Gold Resides in Combining These Standards

Each of these reference structures — RDF/OWL, Wikipedia, WordNet, UMBEL — is itself coherent and recognized or used by multiple parties for potential reference purposes on the semantic Web. The advocacy of them as standards is hardly radical.

However, the gold lies in the combination of these components. It is in this combination that we can see a grounded knowledge base emerge that is sufficient for bootstrapping the semantic Web.

The challenge in creating this reference knowledge base is in the mapping between the components. Fortunately, all of the components are already available in RDF/OWL. WordNet already has significant mappings to Wikipedia and UMBEL. And 60% of UMBEL is already mapped to Wikipedia. The remaining steps for completing these mappings are very near at hand. Other vocabularies, such as GeoNames [29], would also beneficially contribute to such a reference base.

Yet to truly achieve a role as a gold standard, these mappings should be fully vetted and accurate. Automated techniques that embed errors are unacceptable. Gold standards should not themselves be a source for propagation of errors. Like dictionaries or thesauri, we need reference structures that are quality and deserving of reference. We need canonical structures and canonical vocabularies.

But, once done, these gold standards themselves become reference sources that can aid automatic and semi-automatic mappings of other vocabularies and structures. Thus, the real payoff is not that these gold standards themselves get actually embedded in specific domain uses or whatever, but that they can act as reference referees for helping align and ground other structures.

Like the bootstrap condition, more and more reference structures may be brought into this system. A reference structure does not mean reliance; it need not even have more than minimal use. As new structures and vocabularies are brought into the mix, appropriate to specific domains or purposes, reference to other grounding structures will enable the structures and vocabularies to continue to expand. So, not only are reference concepts necessary for grounding the semantic Web, but we also need to pick good mapping predicates for properly linking these structures together.

In this manner, many alternative vocabularies can be bootstrapped and mapped and then used as the dominant vocabularies for specific purposes. For example, at the level of general knowledge categorization, vocabularies such as LCSH, the Dewey Decimal Classification, UDC, etc., can be preferentially chosen. Other specific vocabularies are at the ready, with many already used for domain purposes. Once grounded, these various vocabularies can also interoperate.

Grounding in gold standards enables the freedom to switch vocabularies at will. Establishing fixed reference points via such gold standards will power a virtuous circle of more vocabularies, more mappings, and, ultimately, functional interoperability no matter the need, domain or world view.

This is the last of a two-part series on the importance and choice of reference structures (Part I) and gold standards (Part II) on the semantic Web.

[1] For example, according to the Wikipedia entry on Machine code, “A machine code instruction set may have all instructions of the same length, or it may have variable-length instructions. How the patterns are organized varies strongly with the particular architecture and often also with the type of instruction. Most instructions have one or more opcode fields which specifies the basic instruction type (such as arithmetic, logical, jump, etc) and the actual operation (such as add or compare) and other fields that may give the type of the operand(s), the addressing mode(s), the addressing offset(s) or index, or the actual value itself.”
[2] See, for example, M.K. Bergman, 2009. “Advantages and Myths of RDF,” AI3:::Adaptive Information blog, April 8, 2009; see and M.K. Bergman, 2010. “Ontology Tutorial Series,” AI3:::Adaptive Information blog, September 27, 2010; see
[3] Patrick Hayes, ed., 2004. RDF Semantics, W3C Recommendation 10 February 2004. See
[4] Pascal Hitzler et al., eds., 2009. OWL 2 Web Ontology Language Primer, a W3C Recommendation, 27 October 2009; see
[5] See SWEETpedia from the AI3:::Adaptive Information blog, which currently lists about 250 articles and citations.
[6] Olena Medelyan, Catherine Legg, David Milne and Ian H. Witten, 2008. Mining Meaning from Wikipedia, Working Paper Series ISSN 1177-777X, Department of Computer Science, The University of Waikato (New Zealand), September 2008, 82 pp. See This paper and its findings is discussed more in M.K. Bergman, 2008. “Research Shows Natural Fit between Wikipedia and Semantic Web,” AI3:::Adaptive Information blog, October 15, 2008; see
[7] For a comprehensive treatment, see Fei Wu, 2010. Machine Reading: from Wikipedia to the Web, a doctoral thesis to the Department of Computer Science, University of Washington, 154 pp; see To my knowledge, this paper also was the first to use the “bootstrapping” metaphor.
[8] Quite a few research papers have characterized various aspects of the Wikipedia structure. One of the first and most comprehensive was Torsten Zesch, Iryna Gurevych, Max Mühlhäuser, 2007b. Analyzing and Accessing Wikipedia as a Lexical Semantic Resource, and the longer technical report. See Also, 2008. In Proceedings of the Biannual Conference of the Society for Computational Linguistics and Language Technology, pp. 213221. Also, for another early discussion, see Linyun Fu, Haofen Wang, Haiping Zhu, Huajie Zhang, Yang Wang and Yong Yu, 2007. Making More Wikipedians: Facilitating Semantics Reuse for Wikipedia Authoring. See
[9] This structural basis in Wikipedia is largely untapped.
[10] Citations and references appear to be highly selective (biased) in Wikipedia; nonetheless, those available are useful seeding points for more suitable harvests.
[11] Images have been used a thumbnails and linked references to the articles they are hosted in, but have not been analyzed much for semantics or file names.
[12] There are a variety of efforts underway to use Wikipedia as a multi-language cross-reference based on its 250 language versions; search, for example, on “multiple language” in SWEETpedia. Both named entity and concept matches can be used to correlate in multiple languages. This is greatly aided by inter-language links.
[13] When present, these appear at the bottom of an article and have many related categories; see this one for the semantic Web.
[14] See further and for a discussion of use and guidelines for Wikipedia categories.
[15] For the release notice, see Annex H to the UMBEL Specifications provides a description of the mapping methodologies and results.
[16] Functional categories combine two or more facets in order to split or provide more structured characterization of a category. For example, Category:English cricketers of 1890 to 1918, has as its core concept the idea of a cricketer, a sports person. But, this is also further characterized by nationality and time period. Functional categories tend to have a A x B x C construct, with prepositions denoting the facets. From a proper characterization standpoint, the items in this category should be classified as a Person –> Sports Person –> Cricketer, with additional facets (metadata) of being English and having the period 1890 to 1981 assigned.
[17] See, for example, Massimo Poesio et al., 2008. ELERFED: Final Report, see, wherein they state, “We discovered that in the meantime information about categories in Wikipedia had grown so much and become so unwieldy as to limit its usefulness.” Additional criticisms of the category structure may be found in S. Chernov, T. Iofciu, W. Nejdl and X. Zhou, 2006. “Extracting Semantic Relationships between Wikipedia Categories,” in Proceedings of the 1st International Workshop: SemWiki’06—From Wiki to Semantics., co-located with the 3rd Annual European Semantic Web Conference ESWC’06 in Budva, Montenegro, June 12, 2006; and L Muchnik, R. Itzhack, S. Solomon and Y. Louzon, 2007. “Self-emergence of Knowledge Trees: Extraction of the Wikipedia Hierarchies,” in Physical Review E 76(1). Also, this blog post from Bob Bater at KOnnect, “Wikipedia’s Approach to Categorization,” September 22, 2008, provides useful comments on category issues; see
[18] Olena Medelyan and Cathy Legg, 2008. Integrating Cyc and Wikipedia: Folksonomy Meets Rigorously Defined Common-Sense, in Proceedings of the WIKI-AI: Wikipedia and AI Workshop at the AAAI08 Conference, Chicago, US. See
[19] As two references among many, see A. Halavais and D. Lackaff, 2008. “An Analysis of Topical Coverage of Wikipedia,” in Journal of Computer-Mediated Communication 13 (2): 429–440; and A. Kittur, E. H. Chi and B. Suh, 2009. “What’s in Wikipedia? Mapping Topics and Conflict using Socially Annotated Category Structure,” in Proceedings of the 27th Annual CHI Conference on Human Factors in Computing Systems, pp 4–9.
[20] See, especially DBpedia reference.
[21] See for a listing of known wordnets by language.
[22] For example, see this listing in Wikipedia.
[23] M.K. Bergman, 2008. “When is Content Coherent?,” AI3:::Adaptive Information blog, July 25, 2008; see
[24] For a couple of useful references on this topic, first see this discussion regarding contexts (and the possible relation to Cyc microtheories): Ramanathan V. Guha, Rob McCool, and Richard Fikes, 2004. “Contexts for the Semantic Web,” in Sheila A. McIlraith, Dimitris Plexousakis, and Frank van Harmelen, eds., International Semantic Web Conference, volume 3298 of Lecture Notes in Computer Science, pp. 32-46. Springer, 2004. See For another discussion about local differences and contexts and the difficulty of reliance on “common” understandings, see: Krzysztof Janowicz, 2010. “The Role of Space and Time for Knowledge Organization on the Semantic Web,” in Semantic Web 1: 25–32; see×307213/fulltext.pdf.
[25] OWL already provides the exact predicates; see further M.K. Bergman, 2010. “The Nature of Connectedness on the Web,” AI3:::Adaptive Information blog, November 22, 2010, 2008; see and the UMBEL mapping predicates in this vocabulary listing.
[26] UMBEL is a reference of 28,000 concepts (classes and relationships) derived from the Cyc knowledge base. The reference concepts of UMBEL are mapped to Wikipedia, DBpedia ontology classes, GeoNames and PROTON. UMBEL is designed to facilitate the organization, linkage and presentation of heterogeneous datasets and information. It is meant to lower the time, effort and complexity of developing, maintaining and using ontologies, and aligning them to other content. See further the UMBEL Specifications (including Annexes A – H), Vocabulary and RefConcepts.
[27] Cyc is an artificial intelligence project that has assembled a comprehensive ontology and knowledge base of everyday common sense knowledge, with the goal to provide human-like reasoning. The OpenCyc version 3.0 contains nearly 200,000 terms and millions of relationship assertions. Started in 1984, by 2010 an estimated 1000 person years had been invested in its development.
[28] This image and more related to the general question of interoperability in relation to a reference structure is provided in M.K. Bergman, 2007, “Where are the Road Signs for the Structured Web?,” AI3:::Adaptive Information blog, May 29, 2007; see
[29] GeoNames is a geographical database available for free download under a Creative Commons Attribution license. It contains over 10 million geographical names and consists of 7.5 million unique features, of which 2.8 million are populated places. All features are categorized into one out of nine feature classes and further subcategorized into one out of 645 feature codes. Given the importance of locational information, GeoNames is a natural complement to the gold standards mentioned herein. See further its Web site, which also showcases a nifty browser of mappings to Wikipedia.

Posted by AI3's author, Mike Bergman Posted on February 28, 2011 at 12:07 am in Semantic Web, Structured Web, UMBEL | Comments (2)
The URI link reference to this post is:
The URI to trackback this post is: