Summary of IRW Workshop

The workshop on Identity, Reference, and the Web (IRW2006) was filled to the brim with members of the Web community eager to get to the heart of questions that had been previously considered too theoretical or philosophical for reasonable progress. All the papers, presentations, and podcasts are archived on the web-page http://www.ibiblio.org/hhalpin. Unlike many list-serv discussions on these topics, the mood was conciliatory and practical.

Presentations

Harry Halpin of Edinburgh University gave a brief overview of the problems within the scope of "Identity, Reference, Meaning, and the Web." Berners-Lee's "every thing must have a URI" slogan, where meaning is dictated by its owner of the URI, was shown to have a historical parallel in Russell and Kripke. This causal theory of names had meaning determined by direct reference through acquaintance and causal chains. In contrast, Frege posited the notion of "sense," as formalized by an interpretation in logic, where the meaning of a whole statement determined reference, and so reference was usually left underdetermined. Lewis's notion of convention was used to explain the rise of tagging, "since people will in general use the minimum amount of convention to solve their co-ordination problem." Members of the audience seemed most interested in how tagging could lead to bottom-up ontologies, and the take-home point was that all viewpoints in the current debate had a lineage in philosophy going back more than a century.

Dan Connolly of the W3C followed with an example-driven approach to explaining the W3C TAG's resolution of httpRange-14. First, he points out that Berners-Lee's original viewpoint was that any http URI not containing a "hash" should be taken to about a document and URIs with hashes were therefore safe to use with the Semantic Web. Since this principle would apply to HTTP POST the word document was replaced with the more general information resource. Finally, in order to use metadata vocabularies like DCMI, the hash requirement was dropped so that anything returning an 2xx HTTP code was an information resource, while anything returning a 303 or a 4xx could be used as a non-information resource identifier. So using a URI to denote a person that returned a 2xx code (such as web-page) was in violation of this rule. Since HTTP redirects are expensive to set-up, the recommendation is simply to use hash URIs. Since this could conflict with using a hash to identify document fragments, Dan recommended that such authors be allowed to "opt out" using hashes as document fragments in future versions of HTML using a "profile" attribute. This audience agreed this line of argument made much more clear the TAG's reasoning, although there was one concern there was no explicit means to identify resources beyond information resources, to which Dan reminded the audience that the TAG's role was to clarify consensus in Web architecture and not to prescribe solutions prematurely.

Valentina Presutti of the University of Bologna explained an ontology, the IRIE formalism, that distinguished the many possible types of resources beyond just "information resources." By segregating URIs by type such as people and web-pages, one makes much more clear Semantic Web statements that are easier for machines to process. If a URI for Aldo Gangemi is used both as web-page and a person then non-sensical statements such as "Aldo Gangemi is a male and non-well formed HTML" could result. Some of the main distinguishing types of resources were resources as a concept in virtual space, a computational object in physical space, and a proxy for something else. These patterns were then formally defined and based on the DOLCE upper ontology. There was widespread agreement that such a fine-grained analysis of the types of resources could be useful, and that only the category "Semantic Resource" was considered controversial. Dan Connolly noted that proxy-for was like foaf:topic and exact-proxy-for was like foaf:primaryTopic, showing that this design pattern as already in implicit use in Semantic Web vocabularies.

David Booth of Hewlett-Packard stated that URI identity is a myth, for what matters about a URI is what you can do with it. The obvious action with URIs that one can deference them to obtain authoritative, if partial, descriptive information. Since all descriptive information is inherently partial by nature, what matters is that the partial information be sufficient to identify the resource for a given application. Yet on the Web-scale, one loses the network effect unless one could provide information sufficient for identification for everybody. In an algorithm given, this follow your nose principle could use the TAG's resolution in order to follow 303s until partial information is given. However, even with this useful application the critical problem "information resources" is that they can not be precisely and unambiguously identified according to objective criteria. This problem is exemplified by a hypothetical URI that identified the combination of a human and their web-page. He brought up the question about whether or not a URI could be precisely identified with a particular information resource, since two differing URIs may return the same representations. He then proposed that an information resource be redefined as only a URI-named network source of representations. The audience was in agreement with Booth on the usefulness of partial information and the difficulty in guaranteeing sufficient partial information.

Allen Ginsberg of MITRE brought up the informative example of "Is Pluto a planet?" Since multiple ontologies may both account for all evidence in their system of belief and can even be resilient to future evidence, ontologies are likely to enter irreconcilable conflict. This is because the non-linguistic world is not fully ontologically determinate, such that linguistic entities correspond with entities in the world in order to obtain analytic truth (the correspodence version of truth). The alternative is the holistic view of truth that posits that the meaning of language to be determined by its use and that language and belief structure the world we describe. Therefore, the world is ontologically indeterminate and open to differing terms like URIs having differing meanings, but as it stands the current Semantic Web forces us, when using a URI, to accept all beliefs about a URI. This becomes a real problem if a URI for Pluto forcing our reasoner to use the "fact" that Pluto is a planet even if our ontology wishes to not make a claim on this contentious fact. Should we be forced to use a URI that claims Pluto is not a planet, or mint a new URI, thus diminishing the network effect? Ginsberg explained the R-URI solution that would force the user to explicate exactly what ontologies fully or partially explicated the situation. The audience found the Pluto example extremely thought-provoking and a definite problem for current Semantic Web architecture, yet did not seem to believe the R-URI solution was the correct way to solve it. The audience noted that the Pluto problem was a fundamental one, since it allows us to have differing "local" Semantic Webs and threatened the monotonicity of the Semantic Web.

Steve Pepper gave a talk on PRIs (Public Resource Identifiers). In Pepper's viewpoint, the fact that URIs constitute a universal, globally unique way of naming things is one of the most powerful aspects of the Web. However, people are not using (and re-using) URIs as identifiers extensively enough to realise the great potential that this offers. Pepper cited three main reasons: the confusion caused by the many flavors of URIs (URNs, TDBs, XRIs, etc.); the lack of a discovery mechanism; and the difficulty of knowing what a given identifier actually identifies. He suggests uniting around the use of http: URIs (since they are already the most popular and have a widely-implemented resolution mechanism); ensure that such Public Resource Identifiers resolve to some kind of human-readable indication of the thing identified (called a Public Resource Descriptor); and agree a simple set of conventions for the contents of a PRD that would enable discovery through the use of search engines or repositories. This, it seems, avoids some of the messiness and disregard of current practice in approaches like the TAG httpRange-14 finding. PRIs are very similar to Published Subjects, except for the requirement to use the http scheme. Pepper suggests creating a XG with sponsors from the W3C, ISO SC34, and OASIS to fast-track a W3C Recommendation and ISO Report in order to get PRIs off the ground. Overall, the audience was in agreement with Pepper's practical approach to solving the problem, and their main concern was that XG might be too exclusionary. Pepper then noted that only by keeping it a small group of people would progress be likely.

Peter Patel-Schneider of Bell Labs presented a paper for a more small, yet fundamental, change in how the Semantic Web operates that would allow meanings to be local the Web. Under the intended meaning view, a URI means what it is intended to mean by its owner. However, this has two problems. First, it is almost impossible to tell what the owner intended. To avoid these issues, meaning on the Semantic Web must be constrained to focus on only the formal meaning. Second, the user of a URI may disagree with what the owner intends. For example, the URI http://www.whitehouse.gov#president denotes Bush, and Patel-Schneider disagrees that this is a true statement. Patel-Schneider put forward the solution that meaning on the Semantic Web should be local as to allow divergence of meaning by default and force ontologies from elsewhere to be explicitly imported. To avoid these issues, meaning on the Semantic Web must be constrained to focus on only the formal meaning. Within this domain of formalized meaning, acceptance of community norms and other opinions can be expressed explicitly by importing the corresponding ontologies, via the use of owl:imports. Patel-Schneider is firmly against way meaning being localized in a "follow your nose" fashion at the racine. By using only owl:imports, it may be only Bush that believes he is the President, but by refusing to import his statement a reasoner doens't have to "believe" it. should to be to assume that no statements about the racine should be considered unless an owl:imports is used to explicitly use other ontologies. The audience seemed sympathetic to the proposal, and Patel-Schneider, while disavowing any non-formal meaning, did not see the harm in a URI returning a PRI document.

Live via an audio connection, Pat Hayes of IHMC gave a presentation "in defense of ambiguity." As opposed to many of the other presentations that seemed to be committed to eliminating ambiguity, Hayes drilled home the point that ambiguity is inherent in the structure of the world and knowledge representation is no exception. There are two types of distinct relationships (that Hayes claims are muddled by the use of the word "identity" in the Web community) between names and things, that of access and that of reference. Names that refer to must refer to by description and are thus always using a partial description. Reference far outnumbers access, since only things like webcams and HTML web-pages can be accessed via the Web. All English words are ambiguous, so there is no way to guarantee the sender and the receiver have the same thing in mind. For example, is a "carpet in a an office" part of the office or not? "Saying more" by adding richer formal ontologies makes the problem worse, since it introduces finer distinctions. If URIs are global and eternal in scope, they cannot be unambiguous since there is always the possibility of another finer-level distinction being made in the future. So we should use the ambiguity of reference to our advantage by letting URIs for reference access documents to clarify their meaning, as the advantage of the global scope is for access rather than reference. We should have an explicit naming convention for things that can be accessed on the Web, like the names in named graphs or some profile in HTML. Hayes also disagreed with the TAG's resolution of httpRange-14, citing that "if a document can access something directly, then the URI must refer to what it accesses." Furthermore one is likely to presume, if there is no accessibile document at the URI, then that URI is being used for reference. Instead, Pat holds that uses a URI for both access and reference can actually be a good thing, and the use of excess redirection is not only pointless, breaking some current practice, and forces the "way a document to delivered" to disambiguate whether a URI is used for access or reference. Members of the audience disagreed that the Web was entirely like natural language, being a human argument, and that humans do even in natural language (as in a parent giving distinct to distinct children) attempt to reduce ambiguity.

Choi Hee Chul of DERI Korea presented a work on trust management for personal identity that touched on topics of identity in general. Names are given their meaning by the relationships they have, such as links or trusted relationships. He mentioned the utility of having multiple identities, giving the example of the author who writes under a pseudonym and wishes to hide their identity from their work and family. On the Semantic Web this would entail separate URIs, so that this common practice violates the "One URI for a Thing" principle. Choi then describes how the idea of a name only being valid in a particular context, and a person may have multiple contexts.

John Black of Deltek then discussed the use of socially constructed web-sites, such as deli.cio.us, in grounding the naming process. He noted that for a name to be adopted one should separate the process of naming and accepting the name, and that these must involve at least two agents. Once this naming action, a performative, is complete the name then becomes common ground among a community of agents that all recognize the name. Black noted that this process is more akin to establishing a contract in eBay than using a search engine like Swoogle. However, socially constructed websites allow users to actively name a URI using tags and then for other users to endorse that URI and tag it as well, providing a model for social meaning.

Henry Thompson of the W3C and University of Edinburgh chaired the workshop and summarized the results and themes. The main theme he identified was whether a URI whose primary purpose was to reference or name a thing should someone wear its nature on its sleeve. If this Semantic Web URI can be dereferenced to retrieve a representation, what exactly should it return, and should this be standardized? He also noted there was a call for a separate naming convention. Lastly, he ended by noting Hayes had said that "the Semantic Web languages would operate exactly unchanged if the identifiers in them were not URIs at all, and if the Web did not exist." Thompson noted that if this was indeed the case, then the Semantic Web could not distinguish itself from the failure classical artificial intelligence. Indeed, the new and most interesting element of the Semantic Web that distinguished it from artificial intelligence was its use of the URI, and so we must be able to leverage URIs for more than their uniqueness. Whether we learn how to do so or not is fundamental for the success of the Semantic Web.

Conclusion

There was a surprising amount of agreement that the Web community should move forward on this issue. Indeed, it seemed that there was a demand that the nature of the various types of resources should be clarified, and some sort of best practice for using URIs on the Semantic Web (at least) be explicated. To synthesize the proposals, it appears that something similar to PRIs (Pepper's proposal) and a local theory of meaning (Peter Patel-Schneider's proposal) for the Semantic Web would solve many of the problems. Pepper's proposal would allow a URI for the Semantic Web that is meant as a name (as opposed to only for the access of documents) should be explicitly declared as a "name" for purposes of reference. This explicit acting of naming should also include a way to identify the "namer" who gave this URI the name. This type of resource should in turn leverage the "follow your nose" property of the Web, and return some representation that can tell both a human and a machine that this resource is supposed to be used as a name. Furthermore, following Peter Patel-Schneider's proposal, Semantic Web applications should not assume that all Semantic Web statements are compatible, but instead assume only statements that explicitly include other statements are compatible. There may need to be work on clarifying the role of owl:imports in RDF.

The word semantic resources was defined as a shorthand by Thompson for URIs being used for primarily for reference as opposed to access (PRIs, R-URIs, tdb, Subject Indicator, non-information resource). There are a number of options on how these semantic URIs could be given an objective criteria distinguishing them from other resources. One was the use profile attribute in normal HTML or even MIME type to distinguish these types of resources. It would be vital to return a representation that was explicitly human-readable for these resources and that would tell the user this resource was "special," such as a small graphic similar to "W3C Approved HTML" that could state "PRI" or "CoolURI." Also, one could incorporate further information, such as the namer and naming relationship, as RDF, possibly embedded as RDFa in the HTML itself.

One could easily incorporate the "follow your nose" principle using not only human-readable documentation but RDF (through explicit use of code>owl:imports, such that the semantic resource. The "namer" relationship and other distinguishing characteristics, as given for example by the IRIE formalism, could be incorporated in this RDF named graph if needs be. Lastly, this approach of serving descriptions in RDF and a specially demarcated HTML could be implemented in the Semantic Wikipedia with a minimum of effort, and could also be used with redirection if needed. In general, this seems to be the most pragmatic way forward for leveraging URIs as long as fundamental facts about the nature of ambiguity are taken into account. Indeed, a focus upon the role of URIs in the Web is key to making the Semantic Web effort succeed.

Valid XHTML 1.0!