Last week we discussed the conceptual framework that specifies what goes into the concept of intentionality. The adult concept appears to be as refined as any published philosophical conception (e.g., by Searle, Thalberg, or others) and specifies five conditions that need to be met for people to ascribe intentionality. We agreed that there may be several noteworthy deviations from this complex set of conditions-e.g., when adults ascribe intentionality to children or animals, or when children ascribe intentionality to (e.g., the desire-intention distinction appears to emerge around age 4).
The question for this week is how people make judgments about intentionality for ongoing behaviors. This large question breaks down in a series of smaller questions:
It appears as if these tasks follow a temporal sequence: parsing --> judging intentionality --> inferring specific intentions/goals. However, this assumption quickly runs into difficulties. For example, to parse an action of "phoning somebody" the perceiver needs to have the concept of "phoning" and needs to ascribe the intention of phoning to a particular sequences of movements. So an infant (or a Martian) who does not have the concept of phoning will not be able to parse that action. But can the infant (or Martian) at least parse that the agent does something that is distinguished from other things without knowing what the agent is doing?
The answer to this question depends on how we interpret Newtson's work (exemplified in Newtson & Enquist, 1976). He shows quite convincingly that adults can parse a stream of behavior into "meaningful units," that the breakpoints between these units are processed attentively, and that they are critical to the interpretation of the entire stream. However, what constitutes those breakpoints? Change of movement patterns? Start of a new "action"? Betrayal of the next "intention"? Newtson's work does not answer what the units are.
Will an infant be able to parse behavior streams? Several scholars suggest that certain features of movement (e.g., self-propelled, directed, speedy) indicate basic intentions that even infants can "read off." Premack & Premack, for example, claim that infants can detect persistence and vigor of a movement, and that indicates "trying" to them. Moreover, infants recognize certain basic goals in movement patterns, namely, escape, responsiveness, and overcoming gravity. It may be true that infants are responsive to these features of movement; however, that they hence perceive intentions or intentionality is a very strong claim, which we will need to examine in discussion. Because many behaviors or movements can be conceptualized in many different ways, finding the "appropriate" conceptualization ("he was making a phone call" vs. "he was cleaning the receiver") may require an inference that this is what the agent intended to do. So application of certain action concepts already requires inferences of intentions. True, some basic action concepts (e.g., flight, attack, response) may be abstractions from characteristic movement patterns and are easily recognized by their (reliable) links to certain results, objects, and contexts. But many other action concepts require consideration of a variety of beliefs and desires in the agent to verify or falsify them (e.g., deceiving).
Heider's proposal (for how adults infer intentions) shares similarities with Premack & Premack's (they must have read Heider but do not reference his work), but his includes some of the higher-level inferences that Premack & Premack ignore: Heider says that judgments of intentionality rely on judgments of ability and of trying. Trying has an exertion aspect (see vigor and persistence in Premack & Premack) and an intention aspect (content of trying-Heider often calls it goal). Some actions, Heider says, clearly suggest a particular goal (in modern terms, they activate a "script"); for others one needs to infer the goal from knowledge about the person, the context, and the object towards which the behavior is directed. Because Heider's examples and proposals require high-level inferences (with a large amount of world knowledge already assumed), they are likely to be limited to adults. So one interesting question is how we can bridge the gap between the Premack & Premack model and the Heider model (or some variants of it).
Even farther into the high-level cognition domain travels the short section from my dissertation (Malle, 1994). I presented verbal descriptions of ambiguous behaviors (thus providing verbally parsed behavior streams) and asked people to tell us how they would find out whether someone acted intentionally. This question may precede the question about what the agent's intention is; I suggest, however, that the two are often answered simultaneously. I try to back this proposal by saying that if we take the five conditions of intentionality and ask which ones people will "explore" when making judgments of behavior, people will very often reduce the question of intentionality to the question of intention. That is because, for social behaviors, skill and awareness are usually assumed. To find out whether someone had an intention to act one might ask the agent directly; but if that is ruled out, people seem to use a variety of strategies to derive the answer from belief, desire, or mixed belief/desire information. That is, under the assumption of rationality, people may try to piece together the agent's network of beliefs and desires and then see whether they can derive an intention to act in the specified way. (This procedure sounds a bit like simulation, and I have a hunch that people use simulation a lot to make those kinds of judgments; we will take up this issue in our session on explanations of behavior.) (Note that there is major limitation to this study: The experimental question pulled for people's strategies, such as observing, challenging, etc., not for the kind of information that they hoped to gain (beliefs, goals). In our coding we tried to infer that information, but further studies need to target this information more directly.)
What we need to address in our discussion is how people are able to parse a continuous stream of movements into a sequence of interpretable units, and how those units get specifically interpreted within an intentional framework. Are judgments of action, intentionality, and content of intention and goals made simultaneously? Are they similar to a parallel constraints satisfaction process? When do scripts, assumptions, inferences, strategies come into play? As a taste for next weeks discussion we may also discuss how little is enough (or how much is needed) for an infant to genuinely infer intentions.