Structured Evidential Argumentation System


SEAS is well received by analysts because

SEAS Basic Concepts

Why is a new methodology needed?

Elements of Structured Argumentation

Our approach is based on the concept of a structured argument. A structured argument is based on a hierarchically organized set of questions that is used to assess whether an opportunity or threat of a given type is imminent. This hierarchy of questions is called the argument's template (as opposed to the argument, which answers the set of question posed by the template). The skeletal structure of this hierarchy is call the argument skeleton. Questions higher in the skeleton, called derivative questions, are answered by combining the answers to the questions immediately below them. This hierarchy of questions supporting questions may go a few levels deep before bottoming out in questions that must be directly assessed and answered; these are called primitive questions.

The figure below illustrates the skeleton of a seventeen-question argument template, with five derivative questions (1, 1.1, 1.2, 1.3, 1.4) and twelve primitive questions (1.1.1, 1.1.2, 1.1.3, 1.2.1, 1.2.2, 1.2.3, 1.3.1, 1.3.2, 1.3.3, 1.4.1, 1.4.2, 1.4.3). The links represent support relationships among the questions. A derivative question is supported by all the derivative and primitive questions immediately below it. For example, question 1 is answered based upon the answers to 1.1, 1.2, 1.3, and 1.4, and 1.2 is answered based upon the answers to 1.2.1, 1.2.2, and 1.2.3.


An inference method is used to automatically answer the derivative questions (light blue nodes, below) based upon the answers to primitive questions (darker blue nodes). The user answers the primitive questions in the question hierarchy, and the answers to the derivative questions are automatically calculated. An inference method pairs a fusion method with each derivative question. A fusion method combines the answers to the supporting questions to derive an answer to a derivative question. A typical fusion method might take the maximum answer as the conclusion when combining several answers assessed along a continuous scale. The same argument skeleton and fusion methods are typically used to support multiple argument templates over widely differing topics.

To complete an argument template given an argument skeleton and inference method, one associates a multiple-choice question with each node in the skeleton. To facilitate the rapid comprehension of arguments, we use a traffic light metaphor; relating choices to colored lights along a linear scale, from green at the low end to red at the high end. The questions in a template are typically yes/no or true/false; the multiple-choice answers for primitive questions partition this range, associating an answer with each colored light. Typically, a five-light scale is used (green, yellow-green, yellow, orange, red). Here green might correspond to false, red to true, and the other three to varying degrees of certainty (see below). No multiple-choice answers are associated with derivative questions; within arguments, their answers are strictly summarized by lights indicating their degree of certainty

The challenge in authoring an argument template is to break the problem down into a hierarchically structured set of questions (see the example below) that matches the selected argument skeleton and whose interrelationships among the answers follow the fusion methods. Therefore, it is critical that the author understands the structure of the argument skeleton and the effect of the fusion methods when fashioning the questions and multiple-choice answers that will be posed by the argument template.

1. POLITICAL: Is this country headed for a political crisis?

Arguments are formed by answering the questions posed by a template. Answers are chosen from the multiple choices given by the associated template. The rationale for each answer is recorded in text. Upon answering each question, the template's inference method is applied, deriving the answers to derivative questions. Using the traffic light metaphor, arguments can be displayed as a tree of colored nodes (see below). Nodes represent questions, and colors represent answers. The line of reasoning can be easily comprehended and the user is able to quickly determine which answers are driving the conclusion.

Information used as evidence to support the answers given in an argument are recorded as part of the argument. When information that is potentially relevant to answering a question posed is first found, it is entered as an exhibit. An exhibit assigns a unique identifier to the information, and records information for accessing it and a citation string for referring to it (typically consisting of some combination of title, author, and date). When the relevance of the information to the question at hand is determined, the exhibit is promoted to evidence. The relevance is recorded in two ways: as text explaining the significance and as the answer to the question that would be chosen if the answer were to be based solely upon this evidence. When evidence is present, the rationale typically explains how the collective evidence supports the answer chosen, explaining away that evidence that contradicts the answer and weaving together the supporting evidence to arrive at the stated conclusion.

A key difference between an expert and novice analyst is that the expert knows where to look for relevant information. Discovery tools provide a means for recording where to look for relevant information. They are typically recorded as part of the argument template, but can also be added as part of an argument. In either case, a discovery tool is associated with a question. A typical discovery tool might invoke a query to a search engine (e.g., Google) or reference a periodical on the web. In either case, the resulting information is examined to determine what if anything should be turned into an exhibit or evidence.

All of the arguments and templates thus far discussed are uni-dimensional. That is, they each are designed to arrive at the answer to a single overall question, the one upper most in the hierarchy. Multi-dimensional arguments and template are made up of multiple uni-dimensional components, where each addresses a common topic from a different perspective. For example, the assessment of the stability of a nation state might best be addressed by several independent assessments of the leadership, social, political, military, external, and economic situations (see below).

Other elements include collections, signal flags and memos. Collections are named containers into which objects can be placed, including other collections. They are used as an organization tool, to group objects on common topics, making them easier to find when needed. Signal flags are annotations on exhibits or evidence that mark them for analyst attention. Typically these are used to signal the arrival of new information that has yet to be fully incorporated into the associated argument. Memos are annotations that can be placed on any object. Unlike signal flags, they include a textual subject and body through which a message is conveyed pertaining to the objects on which they are attached. They and signal flags are devices for communication among multiple contributing analysts.

Arguments, templates, and collections all have situation descriptors and publication information. Situation descriptors capture the situation an argument addresses, the type of situation a template is intended to address, or the situation or type of situation to which a collection pertains. They include both textual elements and elements selected from fixed taxonomies of terms that typically capture the who, what, where, and when of situations. Publication information determines who has access to an object and whether or not they can modify that object.


Collections are named containers into which one can place SEAS objects on a common theme. That theme is partially expressed by the name given a collection and the situation descriptor associated with it. The type of the collection can be used to further expresses this theme. A sequential collection indicates that the items in the collection are linearly ordered and constitute a series. One element in the series does not replace a previous element, but adds to it, by addressing a different aspect of the theme, usually a different time period. For example, a sequential collection is an ideal way to organize monthly arguments on a common topic, where each argument assesses the situation during a different month. On the other hand, each item in a versioning collection is meant to replace the previous item, typically correcting or enhancing it. Its items too are linearly ordered, but there is typically only one item in active use, the current item, while the items that came before it are retained to ensure the integrity of earlier assessments, and as an historical record. Besides an item being designated as current, other items can be designated as the previous or next item. The next item is the one in line to become the next current item, at which time the present current item will become the previous. A versioning collection is ideal for tracking improvements and enhancements to a template over time. The initial version is established as the current one while the next one is under development. When the next one is ready to replace the current, the role of the current is changed to previous, the role of next changed to current, and a new copy of the next (now current) template is added to the collection and designated the next item. In so doing, arguments developed on earlier versions of the template are still based upon the same versions, yet the versioning collection makes it clear that there are newer versions available and which is the best to build upon at the moment. An alternatives collection captures the idea that its items are in competition with one another to be designated the best; the order in which the items are listed is of no consequence. This type of collection can be used to organize arguments that represent differing opinions on a common topic. A miscellaneous collection indicates that there is no additional theme and that the order in which the items are listed is of no consequence. Such a collection might be used to collect exhibits on a common topic for later use in support of arguments.


Memos provide a means for annotating SEAS objects with the equivalent of sticky notes, formatted as memos. These can be used to record personal reminders or as a means of communicating with other users that have access to the objects to which they are attached, or the collections that they are in. Since memos include both Authors and Audience, access to them can be further restricted to specific individuals or groups (see Publishing, Collaboration, and Access Control below). Since they can be placed on published objects, they provide a way to mark-up what are otherwise unmodifiable objects. They include fields that are filled with text, including the Subject and Body, and the Type that is selected from a list. The memo Type indicates the purpose of the memo: they can be used to leave Instructions for others on how to use arguments/templates/collections/exhibits/evidence/discovery tools, to Critique any such objects, to record overriding Assumptions, to attach a Summary, to state the Context within which this object was/should be used, to indicate what is left To-Do, to indicate that an object is For-Review by others, or to attach a miscellaneous Comment. When viewing an argument, memos attached to its underlying template are visible, meaning that memos pertaining to instructions, assumptions, and context for the template's use are visible when arguments are created based upon them.

Both signal flags and memos are used as a means of alerting, but they differ in several significant respects. Signal flags are meant to signal things that need to be addressed by someone (anyone); as soon as it is addressed, the flag is lowered; once lowered by one, it is lowered for all. On the other hand, memos can be used to alert a group to things that they all must do. For example, if a For-Review memo is created with a group as its Audience, then deletion of the memo by any member of the group does not delete it for the others; each member of the group must individually address it. Of course, if the group is included as Authors of the memo, deletion by any member of the group will delete it for all. Another difference is that a signal flag is visible to everyone that has access to the object to which it is attached; a memo can be further restricted to any subset of those that have access to the object. Signal flags have no type or other content and cannot be selectively filtered like memos. Memos also resemble email messages in some ways. However they differ in that they are attached to objects and they can be modified or retracted by their Authors after they are issued.

Publishing, Collaboration, and Access Control

Since SEAS is meant to be used by a community of analysts, it must address issues of privacy. When an analyst is in the early stages of argument development, they might not want their work to be accessible by others. During development, they might want certain individuals or groups to aid the process by reviewing or contributing to the process. Even when an argument is complete, they will want to control who it is that will be allowed to see the results. Further, when an argument is used as evidence in support of another argument, then that argument serving as evidence must be guaranteed to persist in its current state to guarantee the integrity of the argument it supports

To address these issues of access control and referencing, SEAS incorporates the concept of publishing. The concept is summarized in the following table. There are four key attributes that are related to two states of publishing: unpublished and published. The first, unique ID, is actually common to both, but it is so fundamental that we wanted to list it explicitly. All arguments, templates, and collections, no matter their publishing state, have an ID through which they can be uniquely identified. If there are multiple versions of an argument, template, or collection, each version has its own ID. Published arguments, templates, and collection are guaranteed to persist, that is, they will continue to exist; no such guarantee is made for unpublished arguments, templates, collections. As a consequence, only published objects can be reliably cited, much as only published works are (typically) included in bibliographies so that the reader has a real opportunity to obtain and read them. Unpublished arguments, templates, and collections are distinguished from published ones in that they are unstable i.e., likely to change in content. Published arguments, templates, and collections will not change. Finally, unpublished objects are distinguished from published ones in that their authors are given write access, while published ones restrict access by both their authors and audiences to reading.

All arguments, templates, and collections originate as unpublished works with a single author. While they remain unpublished, the author can add additional authors. Only the authors have access and they are free to make modifications as they see fit. Should more than one author attempt to change the same information at the same time in an unpublished argument or template, when the second author attempts to save their changes, they will be presented with a dialog that displays the version saved by the other author and their version, with an option to choose either one or to develop a new version by cutting and pasting between the two. When SEAS detects two authors simultaneously browsing the same unpublished argument or template, it warns the authors by displaying the collaboration warning symbol (see below). Once their draft argument, template, or collection is ready for limited external review, they might add people or organizations to the audience. It is dangerous for this audience to cite this unpublished work since it might go away or be substantially changed in the future. When an author decides that an argument, template, or collection is ready for external release, they publish it giving read access to a specified audience in addition to the authors. However, an argument can only be published if its underlying template is published. Once published, arguments, templates, and collections can be reliably cited and referenced in other arguments and collections since they are guaranteed to persist unchanged.

SEAS, High SEAS, Structured Evidential Argumentation System, and SRI Early Alert System, are trademarks of SRI International.
Copyright © 2000-2008 SRI International, 333 Ravenswood Ave., Menlo Park, CA 94025 USA. All rights reserved.
Patent Pending and Unpublished Copyright © 1998-2008, SRI International.