History of FACSAID

The Facial Action Coding System Affect Interpretation Dictionary (FACSAID) has its roots in the work of the laboratory of Paul Ekman and Wallace V. Friesen during in the 1980s. They struggled with the dual problems of how to organize separate, isolated FACS Action Units (AUs) into relevant scores for chunks of behavior, and then, how to interpret what these chunks meant in terms of relevant psychological concepts. In regard to the latter issue, Wallace V. Friesen began writing a program in BASIC that sorted FACS scores into emotion categories, along with indications about the intensity of the emotion, and whether it was felt (part of an emotional response) or unfelt, and other characteristics. This program consisted of a long list of if "IF ... THEN" statements that evaluated whether the FACS score met certain criteria for the emotion categories. These decisions were organized around the concept of "core AUs" which were the essential muscular actions and combinations of actions identified with an emotional expression. Additional AUs in the FACS score were also considered in so far as they were consistent or inconsistent with the core, and how they affected the interpretation. Since the actions important for emotion interpretation were known, researchers interested only in the occurrence of emotion expression needed to look only for the "core" action units (AUs or muscular actions) or their combinations when scoring the record, significantly reducing FACS scoring time. This approach to shortening FACS coding time by selectively attending to certain AUs was called "Emotion FACS" or "EMFACS." The BASIC program for interpreting the resulting FACS scores was called, somewhat misleadingly, the EMFACS "Emotion Dictionary." The first version of the Emotion Dictionary ran on Northstar S-100 microcomputers in the early 1980s, then on a mutant interpreter called BAZIC, and finally under more popular BASIC implementations on standard PC desktop computers. After almost 15 years of discussions between Ekman and Friesen about how to interpret facial behaviors according to these kinds of rules, and fine-tuning the program and scoring procedure, Paul Ekman came to the conclusion that a more fine-grained approach was needed to capture the subtleties of emotion interpretation.

Paul Ekman decided that one needed to examine each particular FACS score generated by coders and make a decision about its meaning. Presumably, he did not think that the rules of a computer program could be complex enough to capture all the decisions one needed to make about classifying every FACS score. The disadvantage to this alternative approach was that it required a new decision by experts about every FACS score that had not been previously interpreted, and a method to look up the interpretations that were previously made. He recruited Erika Rosenberg and Will Irwin, his student and lab assistant at that time, respectively, to help in the task of interpreting FACS scores. They had knowledge of the concepts about emotion interpretation used in Wallace V. Friesen's program, but also used additional information to interpret FACS scores. Their approach was "clinical" in that it examined each FACS score and assigned an interpretation. They stored FACS scores and their interpretations in a flat-file database, but the limitations of this approach soon became obvious. They needed a more flexible way to retrieve the information and a more efficient way to store it.

At this time, about 1994, Joseph C. Hager was working in Ekman's lab on a project using computers to recognize FACS Action Units. He had scored tens of thousands of facial events using the EMFACS approach during his tenure as Ekman's graduate student and lab assistant in the early 1980s, and later as part of his consulting business, and so understood the issues involved. He dumped the information from the flat-file into a Paradox ® relational database and wrote a user interface in REXX to enable easier access to the data. This system was named "FACS Affect Interpretation Dictionary" or "FACSAID." Eventually, about 6000 FACS scores were incorporated into this database. In 1998, after the platform for running these programs had been discarded in Ekman's lab, the data was moved to a DB2 ® system capable of network operation. A group of FACSAID users partially funded a new user interface written in Java that ran as an applet on a Web page until late in 2003. This program provided much improved services beyond the REXX program. Unfortunately, in these five years, the Web platform and its use of Java applets changed significantly, the programs became difficult to run in recent browsers, and the business that underwrote the maintanence costs of the system ceased operations. A new version of FACS with changed AUs and intensity scoring was introduced in 2002, requiring additional changes to the database. Many people have expressed interest in the interpretation of facial expressions, but are not versed in either the techniques of emotion interpretation or FACS. A new approach to providing the FACSAID service was needed, changes to the database and updated user interface programs were required. These changes are now being addressed.

Currently, access to the database is provided on two levels. On this site, individual FACS scores can be entered on Web pages and an interpretation obtained via a JSP-Java Servlet infrastructure. No authorization is required to use this facility; the user needs only to know FACS. Its limitation is that only a single FACS score can be interpreted at one time. A Java application will soon be available that runs on the user's Web-connected desktop microcomputer. This program has the features of the Java applet previously available online. Improvements to the database are also being implemented to make it more useful generally to a wider audience and bring it up to date with the FACS Manual. Eventually, a new user interface will incorporate these database changes.

This database approach used by FACSAID differs from a rule-based, expert system, in which a set of rules, such as that used by the Friesen and Ekman program, is used to "predict" the meaning of new behaviors. Each behavior in the database is examined by experts before an interpretation is assigned and then the two are stored for later retrieval. This approach permits future additions to the database of different behavioral measurements and theories of interpretation, allowing a more flexible approach to assigning meanings to behaviors. The database approach also allows for collecting and storing other information about the behaviors for later retrieval and analysis. One of the weaknesses in the database in its current form is that the rules for interpreting behaviors are not explicit, the authority of the experts who interpreted them being the only basis for confidence in their accuracy. The next version of the database will address this problem. Another weakness of the database approach is that many possible FACS scores have not been evaluated by experts, and are not available when using FACSAID. Each newly observed score must be considered by the experts and an interpretation assigned. We are working on increasing the number of scores that have interpretations and on improving the procedures for obtaining new interpretations. Other planned improvements include adding more didactic materials about the behaviors with links to example expressions.