Balisage Paper: Discourse situations and markup interoperability

Introduction

Although context clearly influences how markup structures are deployed and how they are retrieved and processed, the ways that context determines the interpretation of markup are not well-understood or consistently conceptualized. The role of context has been acknowledged in a few narrow areas -- for example, in how it determines the grammatical mood of the assertions signaled by tags and the ontological status of XML documents (Renear 2003). Context is an obvious part of markup systems, since processing any XML document requires the interplay of the document with structures and systems that are not part of the document.

A developed understanding of the role of context in markup can illuminate well-known problems for information systems, particularly those associated with interoperability. While there is a general sense that interoperability is a sort of freedom from local context, interoperability itself is not well-theorized in terms of markup structures and interpretation. Preservation and portability of markup structures may be aided by approaches that operate in reference to the logical meaning of those structures (Dubin et al. 2003), but the reliance on the semantic meaning of structures presents a sort of boot-strapping challenge since these interpretations must be determined at some point. Context (both internal and external to a document) will play an essential role in mapping from syntax into semantics, so we need a clear understanding of how context shapes the assignment of meaning to markup structures.

Wrightson has demonstrated how situation semantics (Barwise and Perry 1983) can provide insights into the communication of information through markup structures (Wrightson 2001) and has presented a corresponding toolkit for the analysis of XML documents as utterances (Wrightson 2005). Applying situation semantics to markup is attractive since it is intended as a general account of communication and so holds the promise of integraing our understanding of markup meaning into a general theory of communication.

That much markup seems to function at least in some respects like a natural language, rather than a formal language, makes the connection with theories of natural lanaguage particularly promsing. Situation semantics is used here to support an account of the assignment of meaning to markup structures; first through the definition of a particular kind of interoperability for markup structures and then in an examination of the role of schema in interoperability and interpretation.

Situation semantics supports a view of communication of information through markup structures that gives explicit accounting for the context in which the record was created and in which it is interpreted. In what follows situation semantics is used to define interoperability and to demonstrate how specific parts of Dublin Core OAI-PMH records convey contextual information that supports the interpretation of those records.

Situation Semantics and XML Metadata

This section briefly introduces concepts from situation semantics and shows how they may be applied to XML metadata.

Barwise and Perry apply situation semantics to the interpretation of natural language utterances (Barwise and Perry 1983) by modeling the meaning of an indicative sentence as a relation between a situation in which the sentence is uttered and a situation which is described by the utterance of the sentence. The position here is that since metadata records are sequences of indicative sentences, situation semantics can usefully explain the ways in which they carry meaning. An utterance situation is made up of a discourse situation and the speaker's connections (discussed in the following section). Any situation which uniquely anchors the roles of a speaker, an addressee, a discourse location, and an expression is considered a discourse situation.

The technical notion of a situation here is very close to our intuitive one: a situation occurs at a space-time location and involves individuals participating in certain roles and standing in relations. It also closely corresponds to the notion of a state of affairs, especially since situations are abstract objects that may or may not obtain.

In the case of descriptive metadata the role of the speaker can be anchored to the metadata creator at the time when the record is created. ^[1]In contrast, the role of the addressee is left open until the record is retrieved and viewed by some consumer, and it is only at this point that all of the roles are anchored to form a complete discourse situation. The entire discourse situation will be extended in time and typically mediated through several computational environments. This extension in time is a primary motivator for interoperability efforts that are concerned with supporting the preservation (especially in the long term) of metadata records.

Since descriptive metadata records are created without specification of the addressee, the speaker and the addressee may be operating in very distinct environments. This kind of disparity is common in written communication, where an author and a reader may be separated by centuries or differences of language. Regardless, the differences between the environment of a speaker and an addressee of XML documents requires our attention in order to understand the contribution that elements of those environments make to the interpretation of documents.

The described situation for a metadata record is one in which a resource with each of the properties given in the record exists. The space-time location for this described situation will temporally precede and overlap the space-time location of the situation in which the record was created, since presumably a resource exists before it is described.^[2] Hopefully the discourse situation that captures the communication of metadata will fall entirely within the described situation that arises from that metadata, and consumers are able to access the described resource. But one can easily imagine cases where the resource indicated by the metadata record no longer exists at the time of the retrieval of the record, or it no longer has the properties indicated in the record (e.g. it is no longer accessible through a given URL). It is also possible to imagine cases where the described situation does not obtain at all, since the record is simply erroneous. In any case, a record won't facilitate communication and access to resources unless an addressee is able to interpret it. To ensure the operation of records and to support access, we need to understand what goes into the interpretation of descriptive metadata records.

Defining Descriptive Interoperability

While interoperability is frequently called upon as a motivating factor for improvement in information systems, it is not often given a systematic characterization within a general framework for description and communication. A working notion of interoperability for descriptive metadata in XML (inspired by (Sperberg-McQueen, et al. 2000), (Renear, et al. 2002) and in line with the OAIS Reference Model (Lavoie 2004)) is that an interoperable description is one that supports licensing the same set of assertions in any environment.

In terms of situation semantics then, an interoperable description is one for which, given any addressee, the the set of connections that link expressions to their referents remains fixed. Interoperability then can be viewed as a kind of independence from particular features of the discourse situation. The time or location associated with the utterance of an interoperable description can vary without a change in the meaning of the utterance. In order to achieve this kind of interoperability, metadata creators must either convey these connections explicitly, or supply enough information to allow a user to discover the intended connections. Unless connections are given in (or pointed to by) a record, someone trying infer information from the document is left to determine connections on her own.

Of course, such a reader will not be entirely on her own. She can still use her background knowledge, and given a record that is human-readable, she might have a fair degree of success. But she would have to rely on a larger set of resource situations to establish connections for the expressions in the record than in a case where connections are explicitly given.

In fact, resource situations that supply the information necessary to process or interpret XML metadata will always be a requirement for determining the assertions licensed by a document. Grasping the intended connections for records requires a certain set of knowledge about the conventions in play, so it is unrealistic to expect a descriptive record to be fully interoperable as given above. This suggests an adjustment to the definition of interoperability, and characterizing it as an effort to reduce the burden of gaining access to the resource situations required to fully interpret a record.

Example: Analyzing a Dublin Core OAI-PMH Record

Figure 1: A partial OAI-PMH record

Figure 1 shows part of an XML record retrieved from an OAI server. As discussed above, since this is an XML document, the primary setting for the record is given by the XML processing instruction. This expression indicates connections that establish how the document is to be processed or interpreted. A secondary setting can be identified by examining the attributes within the root "OAI-PMH" element for the record.

The schema location attribute within the "OAI-PMH" tag acts as a pointer to documentation that specifies a certain kind of connections for child elements that follow this element. These connections provide the vocabulary and structural constraints for tags and attributes, and they may be overridden by further schema declarations within a child element. In this case, while the OAI-PMH.xsd document does give instructions for the construction of valid OAI-PMH documents by giving structural and datatype restrictions, it does not give the full intended semantics for tags that conform to it.

A third setting is given in this example by the namespace and schema attributes in the "oai_dc" element. As above, the schema location attribute points to documentation that specifies syntactic connections for the child elements. The connections given here override the those given in the OAI-PMH element. This means that the connections for "identifier" within the "oai_dc" element correspond to the Dublin Core Element Set, whereas the occurrences of "identifier" elements higher up in the record have the connections established by OAI-PMH.

The oai_dc.xsd documentation imports simpledc20021212.xsd, and therefore does provide access to the resource situation necessary to make the correct connections for element names in the core element set for Dublin Core. Each of the element names that appear within the third setting are defined in this documentation through marked-up annotations. In contrast, OAI-PMH.xsd only gives syntactic information and processing instructions. The intended semantics for OAI element names such as "identifier", "record", or "metadata" are not provided in the documentation itself, nor is there any reference (network locatable or otherwise) to a source for definitions of these terms. These terms are defined in the protocol documentation, but it is not clear how an addressee (reader/consumer) of the record could reach that documentation without additional information or direction.

Discussion

The meaning of markup structures can be modeled by representing the set of inferences that are licensed from a document (Marcoux et al. 2009), and the OAI markup vocabulary specifically has been given such an interpretation (Sperberg-McQueen 2005). However, such an approach assumes an unambiguous mapping from tags to the logical predicates associated with those tags. The formal tag set description approach does not directly address how meaning is constructed by a reader who encounters an XML document "in the wild", especially when that reader was not privy to the development of the vocabulary in use by the document. An overarching theory of communicative meaning like situation semantics can help us grasp the role that a formal tag set description plays when one is available, and see what is missing when that information is not available.

The use of natural-language identifiers for element names allows readers to interpret markup by exploiting the everyday resource situations that constantly support language-based communication (Wrightson 2005). But, as we demonstrated above with the case of the OAI identifier, the name of a tag alone does not always convey everything necessary to properly interpret the logical meaning of corresponding markup. The problems that arise for this kind of mis-interpretation become obvious when the markup is used to derive RDF that makes incorrect logical statements.

The lack of access to documentation that explains the intended semantics for tags is an aspect of what has been called the documentation problem (Sperberg-McQueen and Miller 2004). The issue could be addressed by sufficient access to documentation (prose or computational) that describes the resource situation against which markup was created, to allow readers of the documents to establish connections to interpret the document correctly. Without such support, interoperability across time and systems is an unlikely prospect.

Acknowledgements: I would like to thank Allen Renear, Ingbert Floyd, and members of the Conceptual Foundation Group at GSLIS for comments and support in developing this paper, and Richard Urban for bringing the example of the OAI identifier to my attention.