Mutant Agents as Guides for Multiple Taxonomies
J. Alfredo Sánchez1, César A. Flores1, John L. Schnase2
1Laboratory of Interactive and Cooperative Technologies,
Universidad de las Américas-Puebla
2Center for Botanical Informatics, Missouri Botanical Garden
{alfredo, cesarf},


Mutant is an agent-based solution to the problem of dealing with multiple taxonomic perspectives for a vast data repository. We apply this approach to the design of a digital library environment in which agents called mutants serve as guides through multiple plant taxonomies for a distributed community of users interested in botany. We discuss the importance of multiple taxonomic perspectives and the potential of agents to assist users in dealing with their complexity. A web-accessible prototype is currently undergoing user testing and evaluation.


Classification is one of the most important processes on which we rely to deal with complexity. We classify objects occurring in nature as well as abstractions we have generated. Taxonomy is the study of the general principles of scientific classification. A taxonomy provides a particular way to organize objects in a given realm according to specific criteria (e.g. shape, color, purpose, etc.). Taxonomies are valuable resources in identifying and understanding newly discovered objects or concepts. Frequently, however, multiple taxonomies exist for the same entities. These arise because different viewpoints serve different purposes or simply because scientists do not always agree on how knowledge can best be organized. In this paper we focus on an agent-based solution to the problem of providing access to multiple taxonomic perspectives for a large collection of data. We instantiate this approach in a digital library in which agents serve as guides through multiple plant taxonomies for a distributed community of users interested in botany.

Multiple taxonomies are not foreign to computer science. For example, several taxonomies have been proposed to help researchers understand the very notion of agent. In this way,  Wooldridge and Jennings [1995] suggest an organization of existing work on agents according to three broad lines of research, namely, theories, languages and architectures. In contrast, Franklin and Graesser [1996] classify computational agents into software agents and artificial life agents, whereas Sánchez [1996, 1997] classifies agents into three major categories: programmer, network and user agents. When presented with an instance of an agent, not only could two researchers disagree on which taxonomy to use for reference, but their opinions could also differ regarding the significance of various attributes of the agent being classified. As a result, even if the same taxonomy is considered, a given agent instance can be placed in different categories by two different experts.

Although in the previous example differing taxonomic opinions may impact only a relatively small community and differences are likely to be reconciled in the long run, controversy lies at the very center of scientific disciplines such as systematic botany. Systematic botany (or plant taxonomy) is the study of plant identification, nomenclature and evolution [Jones and Luchsinger 1986]. Between 300,000 and 450,000 plant species have been estimated to exist on planet Earth, although it is likely that only as little as 25% of these species have been identified [Stace 1989]. Systematic botany provides a framework for understanding and discussing plant diversity in an orderly fashion. Plant identification, nomenclature and classification are the major aspects of plant taxonomy. Identification is the process of determining whether an entity is similar to (or different from) other known entities. Classification involves defining and organizing groups of existing elements, whereas nomenclature assigns scientific names to these groups so they can be easily referred to in any form of communication rather than providing a lengthy description [Sivarajan 1991]. The currently prevailing taxonomic system organizes plants in hierarchical groups, whereby every category or taxonomic group, referred to as a taxon (plural taxa), is contained in successively more general groups. Typical taxonomic levels include family, genus, species and variety.

Multiple Taxonomies in the Floristic Digital Library

We have been working on the design and construction of The Floristic Digital Library (FDL), an ambitious initiative aimed at enabling computer- and network-based cooperative work in botany and systematics. FDL will make it possible for the botanical community to conduct cooperative research and will provide interested users with a variety of services associated with a vast, up-to-date repository containing validated data about plants throughout the planet.

Traditionally, paper-based publications in systematic botany have relied on editorial committees to validate the information they print and to define a taxonomy that somehow reconciles differing taxonomic perspectives. This taxonomy is used consistently to provide a framework for plant descriptions, illustrations, and the discussion of their  properties. In the FDL, authors and editors can still follow the traditional, well-established procedures. However, a number of enhancements make the FDL a more malleable medium. For example, authors and editors accessing the FDL from different locations in the global network may work simultaneously on the same objects. Also, attentive agents may watch authors while they enter structured plant descriptions and offer suggestions to expedite the process [Sánchez et al. 1998a]. Agents also improve user awareness about other library patrons by providing recommendations about places that users with similar interests have visited and presenting related information or user annotations [Sánchez et al. 1998b].

In this paper, we focus on the problems and opportunities generated by the existence of multiple taxonomies in the FDL. In their digital library, not only can botanists have access to the taxonomic viewpoint agreed upon by an editorial committee, but they can also view plant information considering alternative taxonomic opinions. Indeed, taxonomic variations need not be discarded. They can be recorded and enrich the discussion among users exploring information spaces. A general example of alternative taxonomies is illustrated in Figure 1. In this figure, two taxonomic points of view are considered. The first taxonomy (call it T1, dashed lines) states that species s2, s4 and s5 belong to genus g1 and that species s4 has varieties v1 and v2. The second taxonomy (call it T2, continuous lines), however, considers that s4 and s5 under T1 actually form a single species s3, which belongs to a different genus g2. Moreover, species s3 has three varieties (v1, v2 and v3).

Figure 1. Example of multiple taxonomies.

Clearly, whereas the introduction of multiple taxonomies adds enormous flexibility to the organization of an information space, it also adds considerable complexity and poses problems for the design of interfaces to allow for structure visualization and navigation. We have designed an environment, termed MUTANT (MUltiple TAxonomies ageNTs) in which agents act as guides for users, alerting them about the existence of alternative taxonomies and assisting them in the process of navigating through a complex information space.

The Design of Mutant 

The Floristic Digital Library instantiates the architecture for digital libraries described by Sánchez et. al [1997], in which user agents constitute one of the multiple services provided to patrons. In this architecture, users can select from various available agent classes defined by librarians and generate agent instances that will perform some useful task on the user's behalf.

As can be seen in Figure 2, the FDL includes a server consisting of a series of service components built on top of a distributed data repository, and multiple clients connected to the server (only one client is shown). Among other services, we have included user authentication, information retrieval, recommendation, and taxonomic navigation services. On the client side, users have access to services via specialized interfaces, such as tools for editing taxonomic treatments [Sánchez et al. 1998a], collaboration environments [Sánchez et al. 1998b] and taxonomic browsing tools. For the specific case of FDL, we have introduced a number of agent classes to address specific needs of floristic library users.

Agents assisting users to cope with the problems of multiple taxonomies have been added as a new class of agents we have termed mutants. Each mutant agent is designed to act as an autonomous entity representing a taxonomic point of view for the data available in the library. Agents in the Mutant class work in conjunction with a taxonomic browser for the FDL. Users may start browsing a Flora in the library by following the default (or preferred) taxonomy. When a user reaches an item that can be regarded from more than one taxonomic viewpoint, the Taxonomic Navigation component notifies an agent coordinator which invokes a mutant agents representing each of the existing alternative points of view. Each activated mutant offers to guide the user through the data using its particular perspective. The user may follow any of the mutants' alternatives and switch from one taxonomy to another, which might be slightly or completely different, thus changing the lens under which some portion of the floristic library is viewed and organized.  Figure 2 also illustrates the relationship of the newly introduced browser and agents with the other library components.

Figure 2. Mutant agents in the Floristic Digital Library.

The main asset for the FDL's Taxonomic Browser and Mutant agents is the underlying representation of the abstractions that are important for taxonomists and systematists when using the library. Multiple taxonomies can be thought of as a tree for which branches are labeled according to the taxonomy which defines each branch. Nodes in a taxonomic tree correspond to taxonomic names (e.g. Anemiaceae), which are commonly associated with each of the instances within a given rank (e.g. family). For certain portions of the floristic library, the taxonomic browser follows those branches labeled with the name of a default taxonomy. However, when a node is reached for which the ancestors or descendants exhibit multi-taxonomic features (e.g. number of elements different from the current taxonomy's)  an instance of the Mutant Agents class is invoked for each alternative label value. Each instance presents the user with an alternative path to continue browsing the library. If the user opts for one of the alternatives being offered by the Mutant guides that have been activated, the taxonomy represented by the selected agent becomes the taxonomy that the browser will follow from that point on, whereas the current taxonomy becomes an alternative path represented by a new Mutant agent instance.

Mutants can be thought of in terms of the model proposed by Rosenschein and Kaelbling [1995]. According to this model, an agent resides in an environment from which it constantly receives input through sensors and produces outputs or actions through effectors. Mutant agents get their input from both the users' requests and alternative taxonomies detected by the taxonomic navigation services. Output is produced in the form of representations for alternative taxonomies in the user interface.

Prototypical Implementation

We have developed a prototype for Mutant to demonstrate its feasibility and to study its applicability. The prototype builds upon technologies used by other FDL components, such as the Illustra database management system [IUG 1995], Web-accessible C programs using the Common Gateway Interface specification, and HTML and JavaScript elements.

Figure 3 is a snapshot of the taxonomic browser's interface presented to the user. Three major areas can be distinguished in this interface. On the top left area, the user is presented with the current taxon and an identifier for the taxonomy under consideration. In the example, the user is browsing at the kingdom level and the default (Flora of North America, or FNA) taxonomy is active. The top right area presents a stylized flower representing the mutant agents coordinator. When animated, it indicates that alternative taxonomies have been detected and that mutant agents will appear. Finally, the bottom area displays the names of the sub-taxa comprised by the taxonomic group indicated above according to the current taxonomy. In the example, only two sub-taxa (Dicotyledons and Magnoliophyta) have been retrieved from the database. At any point, the user may choose to see the taxa above or below the current level or retrieve associated data, such as illustrations, distribution maps and morphologic descriptions (not shown).

Figure 3. Mutant and the Taxonomic Browser.

Figure 4 shows an example of an actual situation in which the species perrieri is regarded as a member of the genus Bubbia. At this point, the taxonomic navigation services detect that an alternative taxonomy considers this species a member of a different genus. The agent coordinator animates its interface and invokes a mutant agent representing the alternative viewpoint (Flora of Russia, or FOR), which informs the user and offers to guide her through the alternative path. If the user selects the alternative taxonomy, then FOR becomes current and FNA is represented by a mutant agent offering this viewpoint as an alternative, as shown in Figure 5.

Figure 4. A mutant agent offers an alternative taxonomic opinion.
Figure 5. An alternative taxonomic opinion becomes current.

Related Work

The concept of agents as user guides has been used in educational hypermedia systems as a way to supplement navigational metaphors. A notable example of this approach is Apple Guides [Oren et al. 1990], in which video characters interact with users learning about U.S. history. Guides offer accounts of the same events in history described from diverse points of view. Other related projects include Coach [Selker 1994] and Letizia [Lieberman 1995]. Coach (Cognitive Adaptive Computer Help) monitors the actions of students learning a computer language and builds an adaptive user model. This project is built upon the assumption that unsolicited yet opportune assistance can improve teaching effectiveness and encourage the exploration of large knowledge domains. Letizia is an agent that assists web browsing by offering a breadth-first, simultaneous view of alternative pages to complement typically depth-first user browsing. Mutant agents incorporate notions of user guides that  volunteer potentially interesting or useful information into a setting in which alternative opinions occur naturally.

Ongoing and Future Work

The prototype of Mutant implements all the functionality described in this paper. However, the entire Floristic Digital Library project in which Mutant is immersed is a long-term ongoing initiative. Several of its components are being developed simultaneously. For this reason, much work still needs to be done to evaluate and refine the prototype.  Some issues already have become evident and need to be addressed. User interface issues are particularly important. Mechanisms are needed to deal with large number of alternative taxonomies (in the current scheme, this causes the screen to become cluttered with an agents popping up for each alternative). An ongoing project is exploring a 3D representation of multiple taxonomies which would make visualization of differences and similarities more straightforward. Finally, encoding and entering data for multiple taxonomies is still a slow and error-prone manual process. Automation of this task would speed up the construction of the environment and would facilitate the study of Mutant in an actual-use setting.


Franklin, S., and Graesser, A. 1996. Is it an agent, or just a program?: A taxonomy of autonomous agents. In Working Notes of the Third International Workshop on Agent Theories, Architectures and Languages (Budapest, Hungary, August), 193-206.

IUG. 1995. Illustra User's Guide. Release 3.2. Illustra Information Technologies, Oakland, Calif.

Jones, S. B., and Luchsinger, A. E. 1986. Plant Systematics, Second Edition. McGraw-Hill.

Lieberman, H. 1995. Letizia: An agent that assists in web browsing. In AI Applications in Knowledge Navigation and Retrieval: Papers from the 1995 AAAI Fall Symposium. AAAI Press, Menlo Park, Calif. 97-102.

Oren, T., Salomon, G., Kreitman, K., and Don, A. 1990. Guides: Characterizing the interface. In The Art of Human-Computer Interface Design. B. Laurel, Ed., Addison-Wesley, Reading, Mass. 367-381.

Rosenschein, S., and Kaelbling, L. 1995. A situated view of representation and control. Artificial Intelligence 73, 1, 149-173.

Sánchez, J. A., López, C. A., and Schnase, J. L. 1998a. An agent-based approach to the construction of floristic digital libraries. In Proceedings of the Third ACM International Conference on Digital Libraries (DL'98, Pittsburgh, PA., July). 210-216.

Sánchez, J. A., Fernández, L., and Schnase, J. L. 1998b. Agora: Enhancing awareness and collaboration in floristic digital libraries. In Proceedings of the Fourth CYTED-RITOS International Workshop on Groupware (CRIWG'98, Rio de Janeiro). 85-95.

Sánchez, J. A. 1997. A Taxonomy of Agents. Tech.  Rep.  ICT-97-1.  Laboratory of Interactive and Cooperative Technologies.  Department of Computer Systems Engineering, Universidad de las Américas-Puebla, Cholula, Pue.  72820, México. (

Sánchez, J. A. 1996. Agent Services. Ph. D. Dissertation. Department of Computer Science, Texas A&M University, College Station, Tex., August.

Sánchez, J. A., Leggett, J. J. and Schnase, J. L. 1997.  AGS: Introducing agents as services provided by digital libraries.  In Proceedings of the 2nd ACM International Conference on Digital Libraries (DL'97, Philadelphia, July), pp. 75-82.

Selker, T. 1994. Coach: A teaching agent that learns. Communications of the ACM 37, 7, 92-99.

Sivarajan, V. V. 1991. Introduction to the Principles of Plant Taxonomy. Cambridge University Press, Cambridge, England.

Stace, C. 1989. Plant Taxonomy and Biosystematics, 2nd. Edition. Cambridge University Press, Cambridge, England.

Wooldridge, M., and Jennings, N. 1995. Agent theories, architectures and languages: A survey. In Intelligent Agents, M. Wooldridge and N. Jennings, Eds. Springer-Verlag, New York. 1-39.