How to Incorporate Collaborative Discourse in Cultural Digital Libraries Holger Brocks, Adelheit Stein, Ulrich Thiel, Ingo Frommholz, Andrea Dirsch-Weigand Fraunhofer Institute Integrated Publication and Information Systems Dolivostr. 15 64293 Darmstadt, Germany {brocks, stein, thiel, frommhol, dirsch}@ipsi.fraunhofer.de ABSTRACT The COLLATE Project Discourse structure relations can be used to establish As an underlying platform-independent and cross- collaborative discourses within a cultural digital library. organizational infrastructure the World Wide Web (WWW) Ranging from factual to more interpersonal levels they can support scholarly work across the Internet in various describe the intended relations between data and metadata ways. The European-funded COLLATE project (IST– objects, especially annotations. Pragmatics aspects are 1999–20882) is set out to design and implement a covered by communicative acts, which complement the collaboratory for professional communities in the Arts and discourse structure relations and provide means for meta- Humanities working with digitized historic-cultural communication. The resulting interrelations between the material. various domain objects can be employed to perform Historical film documentation serves as example domain advanced context-based retrieval. for the setup of a technical environment for shared access Keywords to a digital repository comprising digitized multi-format task-guided collaboration, discourse structure relations, documents on several thousand European films of the early communicative acts, context-based retrieval 20th century. An extended description of the project can be INTRODUCTION found in [6] and on the project website Content-based access to cultural heritage content is a (http://www.collate.de). fundamental requirement for Digital Libraries in Arts and In our view a cultural Digital Library has to offer support Humanities. In order to offer adequate access functionality for task-guided collaboration, i.e. it should provide an we must go beyond current practices of merely providing environment for structured scientific discourses by defining digital reproductions of the documents in the collection. and maintaining a model of its users’ tasks and goals as Instead, results from current and previous scholarly work well as of the potential interrelations between these like evaluating and indexing these sources must be discourses and the objects in its domain. continuously integrated into the digital data repository in The main contributions of COLLATE will include: the form of various kinds of metadata like annotations or keywords. This, in turn, enables improved content-based • A Web-based collaboratory that provides a access to these value-added sources. comfortable working environment and user interfaces for supporting end-users in their annotation, indexing Digital library systems offer new ways of working with and retrieval of multi-format, multimedia historic source material, i.e. scholarly users actively contribute to archive material; and collections rather than simply access the data stored in the repository ([11]). To support collaborative aspects of • A comprehensive digital multimedia collection on cultural work, a digital library system has to reflect the European historic films and film documentation, complex interactions involved in group activities. annotated and interpreted by a multi-national team of experts using the COLLATE system. Collaboration can be regarded as a process in which two or more participants coordinate their actions toward achieving TASK-GUIDED COLLABORATION shared goals. Collaborative discourses describe extended Digital Libraries offer new opportunities for collaboration communication between two or more participants in a and communication that were unfeasible in traditional shared context ([12]). Annotations represent our core libraries ([11]). Our goal is to develop a cultural concept for establishing a discourse context. collaboratory, supporting interpretative work on mostly textual material. The starting point for collaborative work In the remainder of this paper we will describe in detail comprises already existing data in form of binary image how discourse structure relations and communicative acts representations of the digitized source documents. can be combined to introduce collaborative discourses to a Therefore, we do not focus on cooperative data acquisition cultural digital library system. but on collaborative content-based indexing, i.e. supporting user groups in their joint work with documents, allowing meta-level (e.g., request for clarification) are part of the additions, modifications and explicit negotiation of the COLLATE collaboration model. interpretation of documents and document passages. This kind of system-internal collaboration can, of course, Annotations are notes and comments added to a document be complemented by external communication mechanisms, (or a document region) to explain and interpret it. But to e.g., by email or discussion forums. serve as building blocks in a formal representation of a DISCOURSE STRUCTURE RELATIONS (DSR) discourse, an annotation has to consist of more than According to Rich and Sidner (see [12]) collaboration can unstructured and uncontrolled text which comments on a be regarded as a process in which two or more participants document. coordinate their actions toward achieving shared goals. A cultural digital library should not only provide means for Most collaboration between humans involves simple access to the data stored in the underlying communication. Discourses represent extended repository, but should also support its scholarly users in communication between two or more participants in a actively contributing to the collections (see, e.g., [11]). For shared context, such as collaboration. this reason we go beyond the mere replication of traditional It becomes obvious that the users’ individual tasks and domain-specific workflows by providing a comprehensive goals have to be taken into account for modeling a model of the various types of COLLATE domain objects collaborative system. Content-based indexing of a specific (e.g., documents, annotations), and their potential document – in this sense – can be considered as a global interrelations. Our notion of task-guided collaboration task, which can be decomposed into partial tasks. In the includes the recognition of structures as well as relations COLLATE context, the result of these partial tasks, which between different types of annotations. By taking the users’ are to be performed by various users, is value-added roles, tasks, and goals into account we aim to provide information in form of metadata objects associated with the comprehensive support for the various levels of indexing original document. But these partial tasks are only rarely and interpretation. performed in isolation. On the contrary, in most cases a In a collaborative environment each user maintains her own specific annotation will be part of a thematic thread, e.g., view on the digital repository, which might be dependent some newsgroup-like discussion about a certain topic. on her current role in the indexing process as well as on the Digital signatures are employed to ensure authenticity of actual task she is trying to perform with the system. For this the individual contributions as well as the chronological reason, the heterogeneous roles users can take up as well as order of the annotations. the social aspects of collaborative work have to be taken To capture the coherent aspects of the discussions, we into account. In order to cope with spontaneously emerging employ a model of discourse structure relations between a) profiles of interest and information needs an adequate binary image versions of the original document and collaborative environment has to maintain a model of the annotations and b) discussion threads realized as roles its users can take up as well as of the tasks they are annotations on annotations. supposed to perform (cf. [3]). elaboration The current task the user is trying to accomplish plays in fact a dominant role as it specifies the view on the domain analogy objects in the repository and the relevant retrieval comparison operations on them. Satisfying the various kinds of difference (spontaneously emerging) information needs forms the cause basis for maintaining a continuous flow of information between the system and its users by allowing them to background information contribute new knowledge derived from material accessed before. interpretation interpersonal The COLLATE system supports asynchronous support argument collaboration in indexing for non-technical users. The argumentation scanned documents refer, e.g., to films, plots, or censorship counterargument cases. Together with their associated metadata objects, in particular annotations, they represent the main focus of Figure 1: Discourse Structure Relations collaborative work, i.e. collaboration is performed through Our document-centered discourse model is loosely based annotating the various types of domain objects. on concepts derived from discourse theory. In particular we The interrelations between the various domain objects can adopted the concept of discourse structure relations (or either be unspecified or it can be modeled in a more rhetorical relations as defined by Mann & Thompson [10]). explicit way by defining specific types of admissible Even though it has originally been developed for relations. In addition, certain communicative acts on the monologues in the linguistic context of text cohesion we think that discourse structure relations can be used to describe admissible relations between various data and • Interpretation – (Subjective) interpretation of a metadata objects in the COLLATE context, especially statement being referred to (e.g., “…the author actually between annotations. means…”). Figure 1 shows the specific subset of relevant relations we • Argumentation – The statement or argument of the employed for COLLATE, ranging from factual to more other author is either supported, or a interpersonal levels (i.e. focusing on certain qualities of the counterargument/antithesis is formulated here. participants of the discourse). The seamless transition from factual to interpersonal For detailed reviews on other existing approaches and a discourse structure relations depicted in Figure 1 also meta-taxonomy of discourse structure relations see ([8], corresponds to the illocutionary aspects of an annotation, [9]). In the following we just briefly paraphrase the i.e. the specific communicative intention its author had in discourse structure relations used in COLLATE: mind at the time of creation (e.g., from stating factual • Elaboration – Providing additional, more detailed information towards active participation in a discussion information (e.g., “…it’s Paris in USA, and not in thread). France…”). Even though discourse structure relations proved adequate • Comparison – Comparative relations can be further for modeling the interrelations between annotations it sub-structured to emphasize semantic similarities or turned out, however, that there are some relevant pragmatic contrasts between two elements of a discourse. aspects of collaborative indexing work, which are not yet covered. In the next section we describe how discourse • Cause – To state a specific cause for a certain structure relations can be complemented by communicative circumstance. acts to introduce meta-communication, i.e. explicit • Background information – Using information about communication about domain objects, in a seamless way. the background of the author of the other annotation Figure 2 displays a fictious example discourse about the (e.g., “… As a lay-person the author does not take partial ban of the movie “Kuhle Wampe” by Berthold psychological aspects into account…”). Brecht. Figure 2: Example Discourse COMMUNICATIVE ACTS (CA) The speech-act oriented COR (COnversational Roles) has Scientific source analysis and interpretation can be been designed to model cooperative information-seeking regarded as a highly incremental process, which requires dialogues in a domain-independent way (cf. [14], [15]). extensive domain-specific knowledge. Within a Focusing on the pragmatics COR covers the basic collaborative environment, a user can explicitly perform illocutionary aspects and expectations of specific behavior communicative acts in order to request assistance for associated to a role of a dialogue. In essence it provides a accomplishing a particular task (cf. [1]). set of basic communicative acts (e.g., expectations, commitments, retractions, corrections, and clarifications) that are categorized on basis of their communicative In essence, we regard discourses as results of “cooperative purpose and role assignment. negotiations” where the discourse participants have certain For the COLLATE context only a subset of COR context-related obligations as well as specific expectations. primitives had to be adapted in order to cover the pragmatic A concrete example for the interrelation between aspects of collaborative indexing work. From the five communicative acts and discourse structure relations would global categories postulated by Searle ([13]) and used in be, for instance, if a user had some questions with respect COR we identified only three classes of communicative to a particular annotation, she can request some acts to be of most relevance for our purposes: explanations from its author. The system can support this • Assertives – Assertions represent a certain class of kind of meta-communication by providing certain means communicative acts that can be characterized as being for interaction, e.g., a “Request explanation” button located either true or false. Comments represent an important in each annotation dialog. Internally, the specific type of class of qualified assertives. Comments can either be discourse structure relations chosen to describe the requested or not. The relationship between non- interrelation between the original annotation and its requested comments and the discourse relations explanation (realized as an annotation on the original described in the previous section will be discussed annotation) has to reflect the difference between requested later. and non-requested annotations. CONTEXT-BASED RETRIEVAL • Directives – Attempts to get some other person to do Appropriate search and retrieval functionality represents a something are classified as directives. Requests, e.g., a fundamental requirement for enabling the user community search for documents or specific questions (“Could to access a cultural Digital Library in a reasonable way. To you please explain…”) directed to other users would allow for advanced content- and context-based search the represent the most prominent examples of this class of documents in the digital collection must be indexed by communicative acts. content and subject matter (cf. [3]). • Commissives – The illocutionary point of In context-based retrieval, a document does not stand for its commissives is to commit oneself to some future own, but also the actual context of this document is course of action. A promise to provide additional considered. For the COLLATE case, this means that we are information with respect to a certain film would be a dealing with the discourse context. The RDF statements, good example from the COLLATE domain. which are used to interrelate the various domain objects, The main purpose of these three classes of communicative are typed according to the discourse structure relation they acts is to allow for explicit communication on the meta- represent. With this information, we then know the specific level, i.e. meta-communication about the various indexing type of an annotation with respect to its context, e.g. an tasks supported by the system. elaboration or an example. INTERRELATION BETWEEN DSR AND CA Having a second look at Figure 2 one can see that only the Conceptually, discourse structure relations and inclusion of the associated annotations would yield it as communicative acts can be considered as complementary: relevant for a query like “political background”. But it also Communicative acts focus on illocutionary aspects of a becomes evident that the annotations within a certain specific dialogue situation, whereas discourse structure discourse context cannot be treated in isolation, e.g., the relations describe characteristic relationships between second counterargument weakens the statement it assertive acts, e.g., annotations or comments ([4]). comments upon in this context. But on closer inspection it becomes evident that some The introduction of discourse structure relations allows for communicative acts might invoke certain types of discourse novel retrieval options with respect to the discourse structure relations between the corresponding annotations. context. They can be used to create a ranking of relevant In our view, the set of discourse structure relations adopted documents according to the users’ queries. Depending on for COLLATE can be considered as specific instances of the specific type of its connecting relation, an annotation comments, i.e. they are treated as assertive communicative can possibly raise or lower the overall relevance weight of acts. its discourse context. From this perspective, we can regard explicit collaboration THE COLLATE SYSTEM in the context of the COLLATE project as the combination The COLLATE system is under development. At its of specified relation types between annotations, i.e. current stage of development, the COLLATE prototype discourse structure relations, which are complemented by a system already supports various tasks like cataloguing, certain set of admissible COR acts for meta-communication indexing (structured, free), and allows for unstructured (on the dialogue level) referring to the various types of discourses based on free comments. Furthermore, it offers COLLATE domain objects (e.g., annotations, cataloguing simple search options based on filmographic information as information). well as on transient information recorded as part of the scanning process. Figure 3: COLLATE System Prototype Figure 3 illustrates how part of the example discourse By employing the Resource Description Framework we are (from Figure 2) would be realized within the system. able to interrelate the various data and metadata objects in a After having accessed a document (via “Document dynamic way, i.e. depending on users’ information needs Search”) the user is presented a tree-list view of our system is able to generate corresponding views on the annotations already associated to the document. In this case documents and their associated metadata in a distributed a specific passage of the second document page, the repository. justification for the ban of the film “Kuhle Wampe”, has The technical details of the implementation of the already been marked-up and commented upon. In our COLLATE system are beyond the scope of this paper and example, the current user disagrees with the reasons stated will be described elsewhere. in the initial annotation. Therefore, she provides a RELATED WORK counterargument in which she states her reasons for the BSCW provides web-based support for collaborative disagreement environments, which represent directories of shared objects SEMANTIC WEB INTEGRATION and are associated with group memberships ([2]). The design of an adequate logical representation of the 1. Lotus Notes represents a commercial groupware product COLLATE domain model was a challenging task. for protected information sharing with limited In order to be able to identify and define the heterogeneous collaboration support via messaging data and metadata objects in our domain the different types (http://www.lotus.com/home.nsf/welcome/notes). of cultural contents had to be studied and classified into Collaboratories, unlike CSCW systems, support more certain categories. The various COLLATE domain objects thoroughly studies of the source material, which is stored as (e.g., cataloguing information, keywords, annotations) are a distributed set of digitized source documents (e.g., [7]). represented as instances of appropriate XML schema, uniquely identified by their URIs. DEBORA aims to provide digital access to a set of digitized printed works. It supports annotation sharing and Internally, the structured, document-centered discourses are personal annotation functionality, but without any explicit implemented using RDF descriptions, which relate the model for discourses ([11]). annotations to the objects they describe according to the COLLATE RDF Schema. Annotea is a web-based shared annotation system based on an open RDF infrastructure. Its implementation in the Amaya browser treats annotations as statements about web documents. Even though nested annotations have been 6. Kahan, J., Koivunen, M.R., Prud’Hommeaux, E. & introduced in recent versions, discussion threads are Swick, R.R. (2001). Annotea: An Open RDF realized as unstructured sequences of metadata ([5]). Infrastructure for Shared Web Annotations. In: CONCLUSIONS Proceedings of the WWW10 International Conference, For COLLATE, content-based indexing denotes Hong Kong, May2001. interpretative work on binary image representations of 7. Keiper, Jürgen; Brocks, Holger; Dirsch-Weigand, mostly textual material. In a collaborative environment, Andrea; Stein, Adelheit & Thiel, Ulrich (2001). several users cooperate in order achieve shared goals, e.g., COLLATE – A Web-Based Collaboratory for Content- to understand the rationale why a certain movie has been Based Access to and Work with Digitized Cultural banned. In our model of collaboration, annotations serve as Material. In: Bearman, D. & Garzotti, F. (Eds.), the primary means for semantic indexing. Discourse Proceedings of the International Cultural Heritage structure relations allow us to represent the various types of Informatics Meeting (ICHIM '01). Milano: Politecnico admissible interrelations between domain objects. di Milano, 2001, 495-511. Explicit communicative acts are used to cover specific 8. Kouzes, R.T.; Myers, J.D. & Wulf, W.A. (1996). illocutionary aspects, which are necessary to allow for Collaboratories: Doing science on the Internet. In: collaborative meta-communication about the indexing tasks IEEE Computer, Vol. 29, No. 8. to be performed. 9. Maier, E. & Hovy, E.H. (1991). A Metafunctionally Exploiting the internally represented knowledge about the Motivated Taxonomy for Discourse Structure Relations. discourse context of the users’ input (current task, role), i.e. In: Proceedings of the 3rd European Workshop on the interrelation between the annotations, allows for far Language Generation. Innsbruck, Austria, 1991. more specific and pragmatically relevant search and 10. Maier, E. & Hovy, E.H. (1993). Organising Discourse retrieval methods. RDF and XML Schema guarantee Structure Relations Using Metafunctions. In: Horacek, seamless integration of our idea of object-centered H. & Zock, M. (Eds.), New Concepts in Natural interaction between users with other Web-based services. Language Processing, London, Pinter, 1933, 69-86. The incorporation of document-centered discourses within 11. Mann, W.C., & Thompson, S.A. (1987a). Rhetorical a cultural digital library usability concerns must be taken Structure Theory: A Theory of Text Organization. In: into account. It has yet to be evaluated whether our user Polanyi, L. (Ed.), Discourse Structure, Norwood/NJ: community accepts our approach of task-guided Ablex, 1987, 85-96. collaboration. We hope that the provision of situation- adequate interfaces helps to perform “traditional” tasks as 12. Nichols, D.M., Pemberton, D., Dalhoumi, S. , Larouk, well as offer support for new working methods, which were O., Belisle, C. & Twindale M.B. (2000). DEBORA: unfeasible in traditional environments. Developing and Interface to Support Collaboration in a Digital Library. In: Borbinha, J.L. & Baker, T. (Eds.), REFERENCES Proceedings of the 4th European Conference on 2. Allen, J., Ferguson, G. & Stent, A. (2001). An Research and Advanced Technology for Digital Architecture For More Realistic Conversational Libraries (ECDL2000), Springer: Berlin, 2000, 239- Systems. In: Proceedings of Intelligent User Interfaces 248. (IUI ’01), ACM Press New York, 2001. 13. Rich, C., and Sidner, C.L. (1998). COLLAGEN: A 3. Bentley, R., Horstmann, T., Sikkel, K. & Trevor, J. Collaboration Manager for Software Interface Agents, (1995). Supporting collaborative information sharing User Modeling and User-Adapted Interaction, 1998, 8: with the World-Wide Web: The BSCW Shared 315-350. Workspace system, 4th International WWW Conference, Boston, 1995. 14. Searle, J.R. (1979). A Taxonomy of Illocutionary Acts. In: Searle, J.R. Expression and Meaning. Studies in the 4. Brocks, H.; Thiel, U.; Stein, A. & Dirsch-Weigand, A. Theory of Speech Acts. Cambridge/MA: Cambridge (2001). Customizable Retrieval Functions Based on University Press, 1979, 1-29. User Tasks in the Cultural Heritage Domain. In: P. Constantopoulos & I.T. Sølvberg (Eds.), Proceedings of 15. Sitter, S. & Stein, A. (1992). Modeling the illocutionary the 5th European Conference on Research and aspects of information-seeking dialogues. Information Advanced Technology for Digital Libraries (ECDL '01), Processing & Management, 1992, 28(2): 165–180. Berlin: Springer, 2001, 37-48. 16. Stein, A.; Gulla, J.A & Thiel, U. (1999). User-Tailored 5. Fischer, M.; Maier, E. & Stein, A. (1994). Generating Planning of Mixed Initiative Information-Seeking Cooperative System Responses in Information Retrieval Dialogues. User Modeling and User-Adapted Dialogues. In: Proceedings of the International Interaction, 1999, 9(1-2): 133-166. Workshop on Natural Language Generation (INLGW ’94), Kennbunkport, Maine, 1994, 207-216.