Accounting for Context in Markup: Which Situation, Whose Semantics?
Copyright © 2015 by the author. Used by permission.
Table of Contents
Copyright © 2015 by the author. Used by permission.
Table of Contents
Markup systems are designed to control the assignment of meaning to particular structures that appear within texts. Controlling the assignment of meaning lets us create textual artifacts that are robust with respect to changes in technical environments and domains. The controlled assignment of meaning to linguistic structures means that the output of markup systems resemble artificial languages but they retain many of the features of natural languages. Markup tags frequently use human-readable labels, chunks of natural language text appear to system users, and much like in human conversation there is a back-and-forth where information is exchanged between agents via linguistic structures.
Despite the intentions of markup systems to build independence from particular technical environments for the interpretation, processing and use of their output, context is still a major factor. The analysis of context in information systems has received some attention in the last few years (Lee 2011), but we could still use more discussion of the interplay between contextual factors and the assignment of meaning in markup systems.
In (Wickett, 2010), situation semantics — formal machinery developed by Barwise and Perry in Situations and Attitudes (Barwise and Perry 1983) — was proposed to conceptualize markup semantics within a general theory of communication. Barwise and Perry describe their general linguistic theory as a “relational theory of meaning”, and propose that the meaning of an expression is a relation between an utterance of the expression and a described situation. Although the authors themselves have expressed some skepticism about the general usefulness of the formal machinery they propose, they argue that “situation semantics is particularly well-suited to the study of constraints and their central role in the flow of information.”
The following analysis focuses on markup systems for the creation, interchange, and display of descriptive metadata. In terms of their meaning, descriptive metadata records correlate roughly to singular noun phrases in natural language. A descriptive metadata record is a bounded sequence of statements all referring to some single entity. Therefore, the analysis here leans particularly on Barwise and Perry’s work on the meaning of singular noun phrases and on the use of discourse situations to model their meaning.
Wickett (2010) applied discourse situations to metadata records with a focus on the notion of interoperability. In order to prepare for a complete analysis of the assignment of meaning in descriptive metadata, the analysis here focuses the application of discourse situations to descriptive metadata by analyzing three scenarios in which a metadata record is used to convey information: a documentary scenario, a transport scenario and a discovery scenario. These scenarios highlight distinct aspects of the purpose that metadata records are designed to serve. Purpose and function within an information system are part of the overall context in which metadata records are created and interpreted. The goal is to focus on how the specific functional roles that metadata play shape the assignment of meaning to markup structures.
The next section describes these general metadata scenarios. In following section, an example is analyzed in terms of discourse situations and the relation theory of meaning. In addition to providing insight on the functional aspect of context for metadata semantics, this exercise will let us consider whether discourse situations are expressive enough for accounting for the assignment of meaning to markup structures.
In a documentary scenario, metadata is created to describe, or document, a resource. The resources here may be objects of any kind (intellectual, visual, physical) and may come from a variety of domains (cultural, scientific, business). The distinguishing factors of a documentary scenario reside in the purpose and scoping of the creation of the metadata.
In a documentary scenario, the purpose is to make assertions about an object in order to describe what it is. A metadata record may be embedded and stored as part of a complex information object that also contains the resource itself, or stored and managed separately from the resources being described. The claims made about a resource during the creation of documentary metadata may pertain to entities with varying relations to the resource, including the resource itself, logical or physical parts of the resource, the creator of a resource, and the social or commercial context of the resource. Assertions about the internal logical structure of an object might also be embedded within an encoded version of the resource, as is the case with part of speech tagging; or they might be recorded externally to the resource itself, as is typically the case with descriptions of physical artifacts such as an outfit consisting of a matching dress, hat and shoes.
Metadata in these scenarios align well with Gilliland’s definition of descriptive metadata: Metadata used to identify and describe collections and related information resources (Gilliland, 2008). The purpose of interacting with a markup system in a documentary scenario is to create a descriptive record that encodes the assertions about the resource in a consistent way that is accessible within an information system and can therefore, in some sense, “stand for” the resource. It has been argued that the creation of this kind of descriptive record is part of the process that makes objects from the world recognizable as documents, since any old thing (like my watch) might be fairly considered a document if it is described and arranged within a system that allows my watch to serve as evidence of (for example) early 21st century fashion (Buckland, 1997).
The creation of documentary records of objects has been the focus of the subfield of information organization within library and information science for much of the 20th century. The goal behind much of the research and development in information organization was to create documentary systems and practices that would allow descriptive metadata to be created consistently and in ways that would maximize their usefulness. For example, Elaine Svenonius presents a number Principles of Description that outline core motivating factors in the creation of descriptive metadata Svenonius, 2000. Documentary scenarios are particularly bound by the principle of representation and accuracy, which direct record creators to represent resources as they represent themselves and emphasize the importance of accurate recording in the process of metadata creation.
In a transport scenario, attention is on the encapsulation and portability of some metadata. In contrast to a documentary scenario, the primary object of attention in a transport scenario is a metadata record, not the resource that the record describes. The overall purpose in a transport scenario is to maintain the meaning of metadata across a change in institutional and technical environments.
Transport scenarios for metadata have become more prevalent as the resources available for creating new descriptions of resources have become more limited. Transport scenarios have received much attention in the digital age, with the creation of protocols and metadata packaging standards for the transmission of records via the Hypertext Transfer Protocol (HTTP). But enabling record sharing and transmission while maintaining the fidelity of encodings and the semantics of metadata were driving factors in the development of the MARC standards in the 1960s. Many of the same general approaches and concerns are focal points of both current and earlier efforts, such as the need for standardized record formatting and for self-describing record structures.
The temporal scope of a transport scenario is determined by a transport event such as a system migration or ingest of records from an institution into a federated portal. There are efficiencies to be gained from making the protocols and procedures involved in a record transport event as general as possible, but the essence of record transport scenarios center around these exchange events. While system designers will naturally hope that an exchange format can be accepted by many systems, the critical issue is that each individual exchange event is successful in the sense that there is no loss of information between a record in the originating system and the transported record in the receiving system.
One of the more prevalent strategies for transporting metadata records is to specify a data structure that is a container (or “wrapper”) for metadata records. Generally this kind of container format will consist of a prologue or header portion, along with one or more metadata records. The header typically contains information that pertains directly to the transport event or supplies information relevant for interpreting and processing the metadata records. The Open Archives Initiative - Protocol for Metadata Harvesting (OAI-PMH) exemplifies this strategy. This record format serves as metadata wrapper as well as a message protocol for requesting and receiving metadata records. OAI-PMH uses HTTP as a transport layer and relies on pre-defined HTTP methods for requesting data from a web server (Lagoze, et al. 2002).
The response to an OAI-PMH request is an XML document with three top-level elements: a responseDate element that serves as a timestamp for the transaction, a request element that contains details of the protocol request, and an element with the same name as the verb of the request that generated the response (or an error element). In responses to record requests (as opposed to requests to list identifiers, for example), this final element will contain metadata records. Each record occurs with a “record” element and the metadata is contained in a “metadata” element with a child element that indicates a data standard that the record conforms to. OAI-PMH responses are required to include Dublin Core metadata, but may also include other formats if they are available for the requested records on the server.
In a discovery scenario, metadata records are used as the basis for information system interactions where the user's goal is to locate a resource that is suitable for some purpose. These system interactions are typically realized through search functions that operate through keyword or field-based searching, or through the construction of browsing interfaces that select and filter items based on a set of fields. Records are retrieved and displayed to a user based on criteria provided by the user.
Enabling the discovery and selection of information resources is one of the primary goals of library and information science, and a primary motivator in information organization and information retrieval practice and research. The human agent interacting with a system is seen as a critical element of a discovery scenario, and the system interaction is driven by the user’s “information need”, or an “anomalous state of knowledge” Belkin, Oddy and Brooks, 1982. Library and information science has traditionally been oriented around document management and retrieval, so the strategy for addressing an information need is to locate and retrieve a document that has a good chance of addressing the information need. To be more precise, a system will typically provide a user with a list of document descriptions (metadata records) for documents that have a good chance of addressing the underlying information need.
The functions that metadata serve in these scenarios emphasize particular semantic features. For example, metadata classes are assumed to be mutually exclusive in order to facilitate searching. This means that users can assume that if they are only shown three items relevant to the subject of beekeeping, then there are only three relevant items in the collection being searched. There is also an assumption of monosemy between the user and the search index wherein it is assumed that users are assigning meaning to terms they submit to a search engine in the same way that meaning was assigned to terms in the construction of the index. This assumption is a departure from the way the way natural language generally functions — as Svenonius notes, “it is only in constructed languages that an isomorphism exists between terms and their referents” — and motivated the development of vocabulary control mechanisms and thesauri (Svenonius, 2000).
Many of the practices and principles for metadata creation are designed to enable discovery. This is particularly evident in the practices for bibliographic metadata, which arose to supply data for systems that required labor-intensive human indexing and searching. Although modern computational approaches have made data from any fields in a record accessible for searching (instead of requiring the selection of a limited set of entry points), they still rely on basically on string-matching. This means that search functions are constrained to operate at a syntactic rather than a semantic level, and the assignment of meaning to terms is, to a degree, opaque from the perspective of a retrieval system. The concern over the potential mismatch of semantics between a user and the information system can observed in the principles of use warrant and literary warrant, which are designed to align descriptive metadata with the vocabulary of users of an information system, and authors within a domain, respectivelySvenonius, 2000. The challenge of polysemy is not limited to a potential gap between users and systems, but extends to differences in term usage between authors and users, as demonstrated by the articulation of two separate principles.
This section introduces some core concepts from situation semantics, as discussed in Barwise and Perry’s Situations and Attitudes. Examples of metadata operating in the three contexts described in the previous section are then modeled in terms of these constructs.
The core motivating position of Situations and Attitudes is encapsulated by what Barwise and Perry refer to as The Relational Theory of Meaning:
The meaning of an expression φ is conceived as a relation between
situations, namely, between an utterance u and
a described situation s, written u [φ] s.
The appeal of situation semantics for conceptualizing the meaning of descriptive metadata lies in the observation that metadata records are expressions that, intuitively, describe situations. The situation that a metadata record describes is one in which the resource being described exists, and has the properties ascribed to it by the record. Since situation semantics is developed and employed by Barwise and Perry to handle the semantics of spoken natural language, it’s reasonable to assume that some adaptations will need to be made to use the theoretical apparatus for accounting for the semantics of descriptive metadata. The purpose of this modeling exercise to expose these differences, highlight the unique aspects of descriptive metadata in terms of assigning meaning, and explore some approaches to adapting situation semantics to metadata.
In the case of spoken natural language, an expression is uttered in a specific space-time location by an agent, to some audience. This is a state of affairs constituted by a space-time location, an individual in the role of the speaker, an individual in the role of the addressee, and an expression that is uttered by the speaker. Barwise and Perry have a general machinery for characterizing situations using individuals, relations, and space-time locations (on which basic ordering and inclusion operations are defined) as primitives. Relations, individuals and locations can be combined together into a constituent sequence (of primitives), which is associated with one of two truth values (which they typically express as “yes” and “no”). This structure allows the expression of any number facts about about a situation by constructing a constituent sequence out of an n-ary relation and n individuals and then associating that sequence with one of the two truth values.
Situations that can be usefully generalized over (for example, situations in which expressions are uttered) are classified with event types. The classification of situations into event types uses basic indeterminates and roles. When all indeterminates from an event type are anchored to specific individuals, relations, or locations, the result is a course of events. The basic indeterminates come in the same three flavors as the primitives: location indeterminates, relation indeterminates, and individual indeterminates. Roles are complex indeterminates that are defined for specific event types. The aspects of an event type that are common across the courses of events that realize an event type are typically specified directly, while the aspects that vary are represented with indeterminates.
The central event type for Barwise and Perry is the discourse situation. A discourse situation is an event-type DU with:
DU := at l : speaking, a ; yes
addressing, a, b ; yes
saying, a, [α] ; yes
The speaker’s connections and the setting provided by other parts of an utterance
are both critical elements of Barwise and Perry’s account of the meaning of an
Thus we can think of the meaning of α as a
relation d, c,
[α] σ, e, between discourse
situations, connections, a setting σ provided by other parts of the
utterance, and a described situation. The modeling exercise here focuses on the
elements of the discourse situation that can be identified from metadata
records. The setting and the speaker's connections discussed in the final
The next section discuss the discourse situations for descriptive metadata records in the three scenarios for descriptive metadata. In each case, the discussion is centered around a colloquial metadata records — information objects that follow specified delimiter conventions and use defined vocabularies of attributes and values, but do not have a specified formal semantics.
In documentary metadata scenarios, descriptive metadata is created to document a resource. This process results in a record that represents the resource within the context of an information system. The record may be co-located with or embedded in the resource, or stored separately.
The figure below shows a portion of a metadata record that describes a digitized map that is included in the David Rumsey Historical Map Collection.
Figure 1: A record from the David Rumsey Map Collection
A discourse situation is constituted by the expression being uttered, the time-space location of the utterance, the speaker and the addressee. The expression being uttered (α) can be taken to be the descriptive metadata record in its entirety. What is shown in Figure 1 is a styled version of the record as shown in a web browser, but the source for the displayed record is an HTML table generated by a script that pulls the attributes and values from the record database, along with an image of the scanned map. Each attribute name is populated into a row, followed by a spacing character, and then the value for the attribute. The sequence of HTML elements is the expression α.
There are a number of time-space locations that are relevant for analyzing this metadata record. When viewing the record online, the serialization event that generated the HTML table from the database is an interesting temporal location. However, from the perspective of documentary metadata generation, the more relevant events are in the creation of the data objects that are the source for the record as displayed. That is the point at which some agent is recording assertions about the nature, function, and social, commercial, or historical context of the resource at hand.
There is a similar debate concerning the speaker of the utterance for this metadata record. There are many agents that can be identified as having, in some sense, uttered this record. But, by viewing this record as the output of a documentary process, we are guided toward the original cataloging event in which this record was created. The speaker, then, will be the agent who recorded the assertions about the object.
However, there are a few issues with identifying the speaker as the original record creator in a documentary metadata context. The goal of documentary metadata creation is often to construct a record that is useful for scholarly or scientific purposes. This means that records in documentary contexts may be updated as more information relevant to the interpretation of the object comes to light. This may include information about the history of the object itself, about agents connected to the object, or about the context in which the object was created or used. The changes to the metadata record may involve adding fields or attributes, removing attributes from a record, or modifying attribute values. If the same cataloger who initiated the record makes the changes, we can still identify the single individual as the speaker and consider the temporal location of the utterance to be dispersed across editing events.
The case of a single individual creating and updating a descriptive metadata record may occur in scholarly research environments when a scholar is documenting an object for some project or research program. But cataloging and metadata creation practices in library science typically de-emphasize any individual and focus on institutions as the producers and managers of metadata records. The preference for identifying institutions as record creators can be observed in the definitions and usages of MARC bibliographic fields (MARC 21), where the field for ‘cataloging source’ refers to the “MARC code for or name of the organization(s) that created” a metadata record. On the other hand, archival and museum paradigms for metadata creation have typically given more emphasis to a metadata creator. For example, the Encoded Archival Description (EAD) tag library (EAD, 2002) includes a “author” tag, which is intended to record the “name(s) of institution(s) or individual(s) responsible for compiling the intellectual content of the finding aid”. The metadata published by the David Rumsey Map Collection on their website does not include a specific attribute to record information about the creation of the records.
The most natural approach to understanding the relevant discourse situation is to take the map collection as an organization and to identify it as the speaker for the discourse situation in which this documentary record was uttered. The discourse location is most interesting in terms of the temporal location (as opposed to a physical location), and we can identify the temporal location as starting when the record for this map was initiated. This may have happened in anticipation of the scanning of a physical map, or as consequence of a scanning event. The temporal location for the record creation may be dispersed across many editing and updating events.
Identifying the addressee of a documentary metadata record is also challenging, and is connected to the identification of the complete discourse location. One approach, advocated by Wickett (2010), is to consider the discourse situation in which a metadata record is uttered to be initiated with the creation of the record and concluded with the retrieval and viewing of the record. On this account, the temporal discourse location for the map record is concluded at the point when the HTML table is generated. The addressee for the discourse situation is the user whose system interactions triggered the table generation.
While identifying the addressee as a user retrieving the record and extending the discourse location to that point in time satisfies the technical requirements of Barwise and Perry’s theoretical apparatus (the roles must be uniquely anchored to have a discourse situation), this solution does not align well with the documentary perspective on metadata creation, where one of the implicit goals is to construct a record that is useful and meaningful to a diverse set of users. Identifying an end-user as the addressee for a metadata record would imply that records are not meaningful until they are accessed by a user. Perhaps this is the correct view, and any counter-intuitive results can be handled by focusing on discourse situations defined around narrower windows of time in the lifecycle of record. An approach that aligns more closely to the purposes of documentary metadata creation views the events during which a cataloger is creating and revising a record as discourse situations, with the cataloger as the speaker and the system database as the addressee. If a cataloger views the record during its construction, then the discourse flips direction: the system is the speaker and the cataloger becomes the addressee.
Below is an OAI-PMH document issued by the Library of Congress that holds a metadata record for a very similar historical map. This is an example of metadata in the context of a transport scenario.
Figure 2: An OAI-PMH document
The expression in this case is the OAI-PMH document, which is an XML document. This XML document contains namespace information, a “responseDate” element, a “request” element, and a “GetRecord” element. In general, the “GetRecord” will hold as many “record” elements as correspond to the request issued to the server. In this case, there is a single record, which consists of a “header” element and a “metadata” element that contains Dublin Core metadata that describes the historical map.
Although this particular record was generated and displayed in response to a user action of loading a URL in a web browser, the more typical case is for OAI-PMH responses to be generated and received via a programmatic procedures. An aggregator that harvests content from the OAI-PMH server will request and retrieve records without direct interaction from a human agent and process those records on receipt. In terms of the relevant discourse situation, this means that in the typical transport scenario, both the speaker and the addressee are information systems.
Focusing on the OAI-PMH response as shown here (as opposed to a request that initiated the response), the speaker is indicated within the “request” element, which shows what request verb was issued to what OAI_PMH server. Therefore, the speaker for this discourse situation is the server OAI server at http://memory.loc.gov. The temporal aspect of the discourse location is also indicated in the OAI-PMH document itself, as the element content of the “responseDate” element. This is a timestamp for the issuing of the response, so it does not directly give us information about the receipt of the OAI-PMH document by the requesting server. It is reasonable to suppose that the document is received after the response is issued, and within the scope of some record ingest or harvesting event.
There is also a “datestamp” element within the record header, but this does not indicate the issuing of the response, rather the creation of the OAI-PMH record. This is information that is specific to the functions of the OAI-PMH server, intended for harvesters to be able to retrieve records added to the server before or after some point in time. Therefore it does not necessary indicate the creation of the original metadata record, which would be a documentary metadata scenario. Instead it seems to point to a transport-oriented event earlier in the lifecycle of the metadata record.
In a discovery scenario, metadata is used to aid in search or browsing by an individual user of an information system. Below is a screenshot of a search portal, showing the historical map from the David Rumsey collection aggregated into the Digital Public Library of America (DPLA).
Figure 3: Search results from the DPLA portal.
The expression here is the results list, which was populated in response to a query from an individual user. There is a subset of the results list that is generated from a metadata record for the object. The information that appears in the results list is typically a limited subset of the attribute values that are available. In the case of DPLA, the entry in the results list consists of information about type (e.g. “image”), title, a snippet of descriptive text, a thumbnail image of the object (when available), and a link to the object as it is represented by the institution who provided the data to the aggregation. Other information from the record also contributes to the counts in the left-hand "Refine Search" pane, including contributor, language and location information. This pane can also be considered a part of the expression.
When our attention is on the generation of the results list, the speaker for the discourse situation is the search portal, and the addressee is the individual user. The temporal location of the discourse situation is bounded by the system interactions that led to the generation of the results list. The event begins shortly after the user issues the query, and ends when the results list has been displayed to the user.
By analyzing the three metadata scenarios in terms of discourse situations, we can begin to make some observations about the differences between them and how that might shape the interpretation of metadata. One primary distinction to be made focuses on the speakers and addressees in each scenario. In a documentary scenario, the speaker is a human agent recording assertions about an object, and the addressee is an information system. In a transport scenario, both the speaker and addressee are information systems. In a discovery scenario, the speaker is an information system and the addressee is a human agent. The temporal extent of the relevant discourse situation is also tied to the nature of the metadata scenario. In a discovery scenario the discourse situation is bounded by a end-user system interactions and in a transport scenario it is bounded by a metadata harvesting event. Documentary scenarios are more complex, requiring consideration of a much longer period of time that is undetermined at the time of creation of the record, or shifting the focus to the interaction between a metadata creator and an information system.
A discourse situation is any situation in which a speaker utters some expression and lets us begin accounting for the context of use of an expression in a metadata scenario. But the roles of speaker, addressee, location and expression do not, by themselves, completely constrain the way an expression is used in an utterance. Fully address meaning requires going beyond who said what, when, and to whom. There must be a way to directly account for the referential aspect of language, and to specify what things are referred to by expressions in the context of a particular utterance. Barwise and Perry supplement the discourse situation to create a referential event-type with an additional role that accounts for the reference relation between noun phrases and unique individuals. When all of the referring phrases within an expression are linked to the objects to which they refer, the result is a partial function, called the speaker’s connections, from words in the expression to individuals. The speaker’s connections provides the link between the spoken utterance and the described situation.
Additionally, the assignment of referents in a phrase within an expression may be influenced by other parts of an utterance. Expressions that are part of a larger utterance may not have a link to a described situation on their own. This phenomenon can be seen clearly in the case of referring pronouns. At one point in a conversation, I might refer to my friend by her name “Molly”. At a later point in the conversation, I refer to the same person with the pronoun “she”. In isolation, a sentence with “she” cannot be connected to a described situation. But in the context of the entire utterance where the earlier expression was connected to an individual, “she” can be correctly interpreted as pointing to the same individual. Barwise and Perry group situational elements that come from a broader context of utterance for an expression and contribute systematically to the interpretation of later elements into what they call the setting.
Accounting for discourse situations is only a starting point for this analysis, but it does seem promising. Identifying the correct scope in terms of events and artifacts involved in the creation, management and use of metadata records is critical for precisely characterizing the assignment of meaning. Completing the analysis will require an account for the speaker’s connections and the settings. Analyzing the speaker's connections in these scenarios will require accounting for the schemas and data structures referred to by metadata creators and system designers. Furthermore, the analysis here uses the concept of a metadata record, which is in many ways an oversimplification of the construction, sharing and use of metadata. A complete anaylsis will need to more carefully account for the levels of representation and encoding involved in metadata creation and use.
[Barwise and Perry 1983] Barwise, J. and Perry, J.Situations and Attitudes (1983). MIT Press, Cambridge, MA.
[Belkin, Oddy and Brooks, 1982] Belkin, Nicholas J., Robert N. Oddy, and Helen M. Brooks. "ASK for information retrieval: Part I. Background and theory." Journal of documentation 38.2 (1982): 61-71. doi:https://doi.org/10.1108/eb026722.
[Buckland, 1997] Buckland, Michael. (1997) What is a "document"? Journal of the American Society of Information Science 48 (9): 804-809. doi:https://doi.org/10.1002/(SICI)1097-4571(199709)48:9<804::AID-ASI5>3.0.CO;2-V.
[Gilliland, 2008] Gilliland, Anne. Setting the stage. In Introduction to Metadata (2008). 3rd ed (online edition). Edited by Murtha Baca. http://www.getty.edu/research/publications/electronic_publications/intrometadata/setting.pdf
[Lagoze, et al. 2002] Lagoze, Carl, Herbert Van de Sompel, Michael Nelson, and Simeon Warner, ed. 2002. The Open Archives Initiative Protocol for Metadata Harvesting. Protocol Version 2.0 of 2002-06-14. Document Version 2015-01-08. Open Archives Initiative, 2002. http://www.openarchives.org/OAI/2.0/openarchivesprotocol.htm
[MARC 21] Library of Congress Network Development and MARC Standards Office. MARC 21 Format for Bibliographic Data. 1999 Edition Update No. 1 (October 2000) through Update No. 20 (April 2015). http://www.loc.gov/marc/bibliographic/
[Svenonius, 2000] Svenonius, Elaine. The Intellectual Foundation of Information Organization(2000). MIT Press, Cambridge, MA.
[Wickett, 2010] Wickett, Karen M. Discourse situations and markup interoperability: An application of situation semantics to descriptive metadata. In Proceedings of Balisage: The Markup Conference 2010. Balisage Series on Markup Technologies, vol. 5 (2010). doi:https://doi.org/10.4242/BalisageVol5.Wickett01.