Desktop version

Home arrow Communication

  • Increase font
  • Decrease font

<<   CONTENTS   >>


SOAR (Laird, 2012) is the oldest cognitive architecture developed (see Figure 4.1 for an overview). Originally the name coined by Newell and his colleagues was an acronym for State Operate And Result, which synthesized its main theme: the fact that all cognitive tasks are represented by symbolic problem spaces containing a series of states. Such spaces are searched by production rules grouped into operators. This heuristic search driven behaviour was inherited directly from the GPS system. Exactly as in the GPS, SOAR indeed accomplishes problem solving by selecting, given a certain goal state to reach, appropriate operators able to reduce the “symbolic distance” between the goal state and current state. In addition, SOAR was directly inspired by the cognitive psychology research for its memory system, which distinguishes between a Long-Term Memory - nowadays composed of a Semantic Memory and an Episodic Memory (storing information about facts) and a Procedural Memory (storing knowledge as production rules; i.e., rules of the type “if x then Y”) - and a Working Memory, also known as Short-Term Memory and used as a sort of buffer for the temporary and “short term” storage of the knowledge to handle while performing a given task. The production rules are read in parallel to producing reasoning cycles. From a representational perspective, SOAR exploits symbolic representations of knowledge (called “chunks”) and uses pattern matching to select relevant knowledge elements. SOAR activity is based on a decision cycle: given a goal to reach, when a production matches the contents of the working memory, the rule fires and the knowledge stored in one

The SOAR cognitive architecture, from Laird (2012), with permission from MIT Press

FIGURE 4.1 The SOAR cognitive architecture, from Laird (2012), with permission from MIT Press.

of the two declarative memories (Semantic or Episodic) is retrieved. In the cases where SOAR cannot proceed with the selection of the appropriate operator to solve the goal, it reaches a so-called impasse: a core notion in the SOAR architecture representing a trigger for learning. In particular, when an impasse arises in SOAR, the system recalibrates itself by assuming a new goal: the resolution of the impasse. In this way, the new goal (i.e., the solution of the impasse) becomes a subgoal of the original one (the whole process is known as universal subgoaling) and the original goal is returned to only once the subgoal is achieved and the impasse resolved. Learning in SOAR is strongly dependent on the subgoaling process. Indeed, whenever a subgoal has been achieved, the resolution procedure that has led to that achievement is added to the knowledge base to prevent the impasse that produced the subgoal from occurring again (this learning process is known as chunking). If an impasse occurs because the consequences of an operator are unknown, and in the subgoal these consequences are subsequently found, knowledge is added to SOAR’s memory about the consequences of that operator. An additional feature of the architecture concerns the possibility of using external inputs to extend its Semantic Memory. Such knowledge can be used to solve the impasse and can be incorporated into the learned rules. A crucial connection with cognitive constraints in SOAR, apart from the assumptions about the memory system and the heuristic search hypothesis, is represented by the strict connection between its internal information processing mechanisms and the so-called “Newell’s time scales of human action” (see Table 4.1).

Examples and application of the MCG 63

TABLE 4.1 Newell’s timescale of human actions


Scale (sec)

Time twits














10 minutes







10 seconds

Unit task


1 second



100 ms

Deliberate act



10 ms

Neural circuit


1 ms



100 (.is


Newell (1990).

This time scale was proposed by Newell (1990) as an important element for understanding cognitive behaviour. In particular, the underlying assumption of Newell’s time scale is that there are regularities at these different time scales (each corresponding to different “bands” of behaviour) that can be studied somewhat independently of the time scales above and below them. SOAR processes have a high mapping with the timing individuated in Newell’s time scale (Table 4.1).

<<   CONTENTS   >>

Related topics