During the past two weeks we have focused on adults' and childrens' concept of intentionality and on the complex process of inferring intentions and goals (along with inferences of agency, action labels, and intentionality). During the following weeks we will explore topics for which the concept of intentionality and related inference processes play a central role: explanations of behavior, communication, and responsibility judgments.

Adults' explanations of behavior

Against the background we now have -- philosophy of action and theory of mind research -- it seems obvious to ask how people differently explain (a) behavior perceived as intentional and (b) behavior perceived as unintentional. Only the conception of intentional behavior presupposes that the agent has certain beliefs and desires that she somehow brings together in a reasoning process culminating in her intention to act. People's explanation of such behaviors will therefore make heavy use of the agent's subjective beliefs and desires and take into account the (more or less rational) reasoning process that channels them into intentions and actions. The folk perception of unintentional behaviors, by contrast, does not presuppose any of these subjective and rational processes; unintentional behaviors happen more like physical events in the world (billiard balls, complex machinery, etc.). Such events can be explained via "objective" causal chains and do not require reference to a rational and subjectively representing decision maker.

This model of how adults explain behavior within their folk psychology has long been ignored in social psychology. The dominant paradigm has been "attribution theory," which posits that people consider all behaviors pretty much alike and merely look for these behaviors' causes -- which are also very much alike except that some are located in the person and others are located in the situation. The explainer's inferences that lead to such "person attributions" and "situation attributions" should conform to normative constraints (similar to the logic of experimental design), and people are flawed in that they often disobey these constraints (they commit a variety of "attribution errors").

This is the setup for my "Intentionality in attribution" paper. In it I lay out what appears to be people's conceptual framework for perceiving behavior and derive a model of what their explanations must look like within this framework. I develop the model in some detail, both conceptually and linguistically, and then use it both for critiquing standard attribution theory and for deriving novel predictions and topics of research on behavior explanations.

In the seminar we may focus on the explainer's assumptions of subjectivity and rationality because they are most relevant to the controversial issue of simulation/empathy and to the question of when and how children explain behavior.

Simulation/empathy vs. theory

In developmental psychology circles, the "theory theory" position has gained remarkable dominance in the last 5-10 years. The idea is that people deal with others' minds (e.g., infer their mental states, explain their behavior) with the help of a theory of minds, which is literally assumed to be very similar to a scientific theory. It makes ontological assumptions (what exists and what doesn't), it changes as new, contradictory evidence comes in, and it provides a causal model of the world (see Gopnik's optional orgasmic article; for a critique see Goldman's article). It seems worth discussing in the seminar how much we would want to support a strong (i.e., literal) analogy between theory of mind and scientific theory.

A contrasting position (one that the dominant position likes to deem a mere curiosity) is "simulation theory." It posits, with some variation among authors, that people deal with other minds by running simulations of what that other person may feel or think or do, assuming facts about the context, about the other's values, goals, etc. Some authors claim that people always deal with other minds via the simulation route, others say that children learn to deal with other minds via that route (and may later develop theory-like structures). The articles by Goldman (required) and Gordon (optional) offer many good examples and some good arguments for taking simulation seriously, in one way or another. Personally, I find the simulation account very convincing as a description of what often goes on phenomenologically, at least in adults. But I also think that simulations cannot work efficiently without a "grammar," a conceptual framework that dictates what is being simulated, according to which rules. How this grammar and on-line simulations interact, specifically during explanations of behavior, may be a point of discussion in the seminar.

Children's explanations of behavior

The articles by Wellman, Hickling, & Schult and by Kalish have several things in common: They point to children's growing capacity to offer different types of explanations for (what they must consider to be) different types of events -- psychological vs. physical vs. biological explanations (Wellman et al), or reason vs. cause explanations (Kalish). Both articles show the familiar increase in performance among 4 and 5-year olds compared to 3-year olds. However, both articles also suggest that the basic "reasoning systems" are already well in place among 3-year olds, whereas their "discrimination accuracy" is yet to improve. Wellman et al. review studies that demonstrate the early onset (around 2 years) of verbal explanations of the "reason" (belief/desire) mode. Kalish shows that the complexity of reason explanations (their interplay between beliefs, intentions, and physical constraints) is only partially appreciated even among 3-4 year-olds. As so often, different methodologies paint rather different pictures of children's performance and invite different inferences about their competence.

A detail about both Kalish's and Wellman et al.'s methodology is that both research teams do not very clearly define what "reason explanations" or "psychological explanations" are. In fact, the lack of theoretically refined definitions creates slippage in coding decisions. For example, Wellman et al,. classify "I didn't cough in Mommy's face cause then she might get the germs" (see Table 3) as a biological mode, whereas I would strongly argue that it is a reason explanation (thus psychological). Moreover, Wellman et al. code all explanations referring to the agent's mental states and emotions as "psychological," which also includes cause explanations (in Kalish's system) in which, say, an emotion, directly brings about an unintentional behavior (e.g., "He frowned because he was confused"). Kalish is not sufficiently detailed about his coding rules either -- I suspect that the "neutral" codes would harbor many codeable units if one adopted a tighter theoretical framework.

I doubt that this slippage alters the major findings of these studies, but it seems important to take notice of such limitations. As in other topic domains (e.g., judgments of intentionality proper), lacking a coherent and empirically-based model of adults' performance and competence leaves developmental researcher resort to their intuitions and philosophers' speculative models.

A final issue that runs through some of the readings:

What's worth explaining

Wellman et al. point to the distinction between requests for explanation and offerings of explanation. Most current research focuses on offerings. The question when and under what conditions people request explanations (when they "wonder why"), is discussed in Gopnik's (optional) article and explored empirically in the Malle & Knobe (optional) article. The latter provides a modest "theory of wondering" from which a number of predictions follow about the kinds ofbehaviors actors and observers will tend to wonder about and explain. The corresponding findings also relate back to our seminar's discussions of action streams and intention inferences; the findings demonstrate how much "observers" are focused on other people's actions and how much "actors" are focused on their own experiences. This would imply that people need to learn how to parse and interpret other's actions, but they also need to learn how to parse and interpret their own experiences -- a direction of research that may ultimately raise the status of a (reformed) simulation position.