Where there’s a will—there’s no intuition. The unintentional basis of semantic coherence judgments

Abstract

It is broadly agreed that the processing of a word triad with a common remote associate (coherent triad) leads to its partial activation, which is the process underlying intuitive coherence judgments. The present studies demonstrate that this process not only is independent of the intention to find the common associate (CA), but rather may be impaired by it. In Experiment 1, incidentally reading a triad did automatically activate the CA without participants being aware of the underlying semantic structure of triads. However, intentionally searching for the CA did not. Memorizing the triad even inhibited the activation of the CA. Also, it was found that coherent triads are memorized better than incoherent triads, irrespective of mindset. Experiment 2 ruled out task-switching costs and anxiety as alternative explanations. In Experiment 3, intentionally searching for the CA decreased the accuracy of intuitive coherence judgments compared to merely reading the triad.

Introduction

Imagine you have invited three friends for dinner and you ask them for their favorite food: the first is crazy about Asian dishes, the second likes hot marinades and the third loves poultry. Will you be able to make them all happy? Your gut feeling may say yes and after a minute of rumination, you find the solution: tandoori-chicken! The challenge here was to accommodate your friends’ predilections by finding their smallest common denominator. Although the explicit solution popped into your mind only after some time and effort, an intuitive hunch was there much earlier and without expending much energy. Did this hunch occur only because you faced the problem and were eager to find a solution? Or was it automatically generated independently from your intentions?

An intuitive hunch of a solution before explicitly retrieving the solution satisfies many of the criteria for intuition that are found in the literature: (1) Outside awareness. Intuition occurs with very little awareness about the underlying process (e.g., Deutsch and Strack, 2008, Hammond, 1996, Liebermann, 2000, Reber, 1989, Smith and DeCoster, 2000, Wilson et al., 2000) or conscious knowledge that is used (Dienes & Perner, 1996). (2) Efficiency and speed. Intuitive judgments are described as fast (e.g., Gigerenzer, Todd and the ABC Research Group, 1999, Hamm, 2008, Hogarth, 2001, Strack and Deutsch, 2004) and effortless (e.g., Kahneman and Frederick, 2002, Liebermann, 2000, Stanovich and West, 2000). (3) Feeling. The product of this underlying tacit, efficient, and fast mechanism is often qualified as having the qualities of a subtle affect or ‘vibe’ (Epstein, 1991) or being a ‘cognitive feeling’ (e.g., Clore et al., 2001, Hart, 1965, Kahneman and Frederick, 2002, Loewenstein et al., 2001). (4) Intentionality. It is ‘automatic’ (e.g., Betsch, 2008, Epstein, 1991, Epstein, 1994, Hogarth, 2001) in the sense of ‘unintentional’ or ‘uncontrollable’. These four features do not necessarily overlap (cf. Bargh, 1994, Moors and De Houwer, 2006) and can be disentangled in a thorough analysis.

While tacitness (e.g., Reber, 1967, see also below for semantic coherence judgments) and speed (e.g., Bolte & Goschke, 2005) are most frequently examined as features of intuitive judgments, intentionality has rarely been investigated (Shames, 1994). To remedy this shortcoming, the present study asks whether the mental mechanisms that underlie intuitive judgments depend on people’s mindset, i.e., their actual intentions to generate an intuitive judgment.

One example of experimentally testable intuitions are judgments of semantic coherence, which require spontaneous guesses whether or not three given words have a common associate (CA, Bowers, Regehr, Balthazard, & Parker, 1990). Similar to the earlier tandoori-chicken hunch, individuals are able to use their intuition that there exists a solution even though they are unable to retrieve the CA (Bolte and Goschke, 2005, Bolte et al., 2003). We assume this intuition to be a cognitive feeling emerging from associative processing when activation spreads efficiently and independently from intentions from active representations to related concepts. In the case of a coherent word triple, the confluent associative spread (cf. Anderson, 1983) may cause the partial activation of the CA and may additionally produce the cognitive feeling. When asked whether three words have a CA, participants may use this feeling in an explicit judgment about the semantic coherence of the word triad (see Topolinski & Strack, submitted for publication, for a detailed account and experimental testing of how the partial activation may trigger a cognitive feeling, see also Topolinski, Likowski, Weyers, & Strack, in press).

Given these considerations, the confluent semantic spreading is a necessary condition for judgments of semantic coherence (Bolte et al., 2003, Dorfman et al., 1994; see also, for different intuitions, Bowden and Beeman, 1998, Greenwald, 1992, Yaniv and Meyer, 1987). Therefore, we focus on this initial mechanism and analyze its dependency on intention.

We assume the partial activation of the solution concept to be a fully automatic mechanism that is independent from the participant’s intention to find the CA. This assumption is in line with earlier theoretical accounts (e.g., Anderson, 1983, Neely, 1977, Smith and DeCoster, 1998, Smith and DeCoster, 2000, Tversky and Kahneman, 1973, Yaniv and Meyer, 1987) and with most of the priming literature (see the next section).

In contrast, previous work on judgments of semantic coherence suggests that to find a CA, an intention is necessary for the solution concept to be activated (Dorfman et al., 1994, Kihlstrom et al., 1996, Shames, 1994). In the next two sections, we will report evidence for each of these contradictory positions, starting with the findings from semantic priming.

It is reasonable that the activation of the CA works via semantic priming, which is the phenomenon that the processing of a word is facilitated if it is preceded by another word that is semantically related (for an overview, see Hutchison, 2003, McNamara, 2005). This finding can be explained by mechanisms of spreading activation (Anderson, 1983), in which the processing of a word causes its concept to be activated, which spreads activation to associated concepts and therefore preactivates those concepts to a certain degree. If one of these related concepts is denoted by a word in a subsequent naming or lexical decision task, this word will be processed more quickly because it has already been activated by the prime. This mechanism is assumed to operate automatically and without intention (Anderson, 1983, Balota and Lorch, 1986, Neely, 1977).

There is substantial evidence in the priming literature that remote associates may jointly activate their CA in an automatic manner. Findings regarding indirect semantic priming (Kiefer, Marzinzik, Weisbrod, Scherg, & Spitzer, 1998) show that concepts do not only activate their closest, but also more remote associates. For example, Balota and Lorch (1986) found that pronouncing the word “fast” was facilitated after having been exposed to the word “turtle”—because of the mediating concept “slow” (see also Bargh, 2006, Kiefer et al., 2005, McNamara and Altarriba, 1988, Spitzer et al., 1993). Furthermore, findings concerning summative semantic priming (Balota & Paul, 1996) show that concepts can jointly activate another concept. For example, Balota and Paul (1996, Experiment 2) found that two primes that converge on a single semantic concept (e.g., “lion” and “stripes”) facilitated the processing of that implied concept (“tiger”) in an additive fashion. This was the case even if the two primes were semantically unrelated to each other. Finally, Beeman et al. (1994) used triads of remotely associated words to prime their own CAs. In their study, a target word was named faster if it was preceded by three prime words each weakly related to the target than if it was preceded by three words that were unrelated to the target word. Because the prime words in this study were presented for only 300 ms and the participants did not have the goal to find the CAs of the primes, this facilitation effect provides substantial support for an automatic semantic activation convergence.

In sum, these evidences imply that the partial activation of the CA in semantic coherence tasks runs automatically in the sense that it does not require intention. This stands in contrast to earlier findings of semantic coherence judgments, which will be reviewed in the next section.

Most of the research on judgments of semantic coherence is based on the Remote Associates Test (RAT, Mednick & Mednick, 1967) in which participants have to generate the CA of three given words that are remotely related to this solution concept (e.g., PIPE WOOD HOT, CA: STOVE). Bowers et al. (1990) used a modified version of RAT items and instructed participants to simply guess whether a presented word triad had a CA (coherent triad) or not (incoherent triad) if they could not come up with the solution word after up to 12 s. This coherence judgment can be called ‘intuitive’ because it meets the criterion of tacitness (i.e., participants had to use a hunch instead of an explicit representation of the solution concept). However, asking participants to find the correct solution was likely to instigate an explicit search for the CA.

The critical feature of speed in semantic coherence judgments was precisely examined by Bolte and Goschke (2005) who demonstrated that subjects could reliably differ above chance between coherent and incoherent word triads even after only 1500 ms of exposure to the triad and additional 500 ms for the judgment.

It is broadly agreed that this tacit and fast intuitive judgment is based on the partial activation of the solution concept (e.g., Bolte et al., 2003, Dorfman et al., 1994). This was demonstrated by Shames (1994) and recently replicated by Harkins (2006) who showed that processing of a coherent RAT triad leads to an increased activation of the triad’s CA. However, two important aspects of the priming procedure used in these studies are critical for the feature of intentionality in semantic coherence judgments: First, participants knew the existence of an underlying semantic structure and were instructed to find the correct solution word. Second, they were given sufficient time (3–7 s) to retrieve the CA. Taken together, this suggests that participants were likely to search for the CA. In determining whether this knowledge (and the resulting problem-solving mindset) is necessary for the activation of the CA, Shames (1994, Experiment 2) found that the priming effect disappears when participants were not informed about the semantic structure of the triads and were not instructed to search for the solution, but to merely memorize the triads.

These results indicate that the activation of the solution concept may be caused by more than pure associative spreading. Consequently, it has been claimed that for remote associate triads “activation does not spread automatically (…) to a node representing their common associate” and that instead, “activation of solutions only occurs when the subject is operating under a particular problem-solving set” (Dorfman et al., 1994, p. 295). These conclusions drawn from Shames’ result contradict the implications from semantic priming literature and suggest that the underlying semantic process in semantic coherence judgments depends on intentional processes.

The present studies explore the role of intention for the activation of the CA in processing RAT triads. In Experiment 1, an intentional search for the CA was contrasted to an incidental processing of word triads. Additionally, an alternative explanation for Shames’ (1994) findings was tested by implementing a third condition in which triads had to be memorized. In Experiment 2, two possible confounding factors, namely task-switching costs and anxiety, were ruled out. Finally, in Experiment 3, the impact of an intentional search on intuitive semantic coherence judgments (Bolte & Goschke, 2005) was analyzed.