The neural basis of conditional reasoning: An event-related potential study

Abstract

The spatiotemporal analysis of brain activation during the execution of conditional reasoning tasks (the four inference forms: Modus Ponens (MP), Modus Tollens (MT), affirming the consequent (AC), and denying the antecedent (DA)) and one baseline task (BS) was performed in 12 normal young adult participants using high-density event-related brain potentials (ERPs). Results showed that the early components elicited by the five task types were not significantly different. Reasoning tasks elicited a more negative EPR deflection (N600) than did the BS task in the time window of 500–700 ms after onset of the minor premise. Dipole source analysis of the difference wave (MP − BS) suggested that a generator localized in the left anterior cingulate cortex (BA 24) was involved in the activation and the application of the inference rules. ERP components of the five tasks were similar in the subsequent time period between 700 and 1700 ms. Following that period, a greater negativity in the reasoning tasks, in comparison to the BS task, developed between 1700 and 2000 ms poststimulus over the left fronto-central scalp regions. A generator of this effect was located in the right anterior cingulate cortex (BA 24) and was possibly related to cognitive control. The results indicate that the cingulate cortex was activated by conditional reasoning tasks with purely abstract materials and support the view that human reasoning is not a unified phenomenon but is content-sensitive.

Introduction

Conditional statements such as If A then C are the most studied of the connectives. A typical experiment contains an ordinary conditional as a major premise and then its antecedent or consequent (or a negation of its antecedent or consequent) as the minor premise (Evans & Simon, 2003). Participants are asked to assume these premises to be true and are then asked whether they endorse a conclusion that may validly follow. For example, participants can be given conditionals of the form If A then C as the major premise and A as the minor premise (Modus Ponens, MP). The frequency of endorsement of the conclusion C is then recorded. Other participants can be given the same major premise but not-C as the minor premise (Modus Tollens, MT), and the frequency of endorsement of the conclusion not-A in the valid inference is recorded. Simple MP arguments are correctly endorsed at rates of at least 90%, whereas correct responses for MT inferences are at rates of approximately 60% (Johnson-Laird & Byrne, 2002). Thus, different logical inferences yield different behavioral results.

There are two non-valid inference forms related to the conditional rule. One is called the “Denial of the Antecedent” (DA). As the name suggests, this arises when the antecedent of a conditional is negated, as in If A then C along with not-A, which often leads one to conclude not-C. In the event that a participant is presented with these two kinds of premises, the logically appropriate response is to say Inconclusive (Cannot Tell). The other non-valid inference form, the Affirmation of the Consequent (AC), where one is confronted with if A then C and the information C, is also Inconclusive (Johnson-Laird & Byrne, 2002).

Three major psychology theories of the ordinary indicative conditional in natural language were developed to try to explain the results of experiments with these conditionals. Mental logic theories propose that the participant has an underlying knowledge of the inferential role of the closed-form, or logical terms, of the language (e.g. ‘all’, ‘some’, ‘if–then’) and uses this knowledge to infer the conclusion (Braine & O’Brien, 1998; Rips, 1994). Such theories explain the differing results with MP and MT by pointing out that MP is the basic rule of if-elimination, whereas MT is a derived rule requiring more computational steps and thus more mental resources (Braine & O’Brien, 1998; Rips, 1994). Mental model theories explain human inference, not by hypothesizing that there is a mental natural deduction system, but rather by suggesting that people manipulate ‘mental models’ (Johnson-Laird, 1983, Johnson-Laird, 1994; Johnson-Laird & Byrne, 2002). Only one explicit model construction is required for MP, whereas one or two implicit models based on the explicit model are required for MT, AC, and DA. According to this theory, participants draw incorrect conclusions for MT, AC, and DA because implicit models cannot be constructed contemporarily within the limits of working memory capability. The theory of dual-process reasoning suggests that there are two sets of different reasoning process systems with relevant neurobiological foundation, in which System 1 is a rapid, parallel, and automatic process mainly in the frontal–temporal pathway, and System 2 is a slow, serial process together with working memory in the parietal–occipital pathway (Evans, 2003, Goel, 2003).

Neuroimaging techniques provide new methods of examining these issues in ways that transcend the differences among the models (Goel, Buchel, Frith, & Dolan, 2000; Goel & Dolan, 2003; Goel, Gold, Kapur, & Houle, 1998; Knauff, Mulack, & Greenlee, 2002; Noveck, Goel, & Smith, 2004; Osherson et al., 1998; Ruff, Knauff, & Spreer, 2003). Goel (2003) suggest that, if the mental model theory is correct, then the reasoning trials should result in right hemisphere and parietal activation (visuo-spatial areas), but if the mental logical theory is correct, the left frontal and temporal lobe regions (language areas) should be activated. In fact, Goel and Dolan's experiments have found that a parietal–frontal pathway is activated when participants reason with arbitrary materials, whereas a temporal–frontal system, linked to language areas, is activated when participants reason with syllogisms using realistic statements (Goel & Dolan, 2003). Thus, it appears that the reasoning activity is exclusively linked neither to language areas nor to visual–spatial areas. Noveck et al. (2004) used brain imaging to study conditional reasoning with arbitrary materials, such as “If there is a black triangle, then there is a red square”. They found that the left superior parietal lobule, the left temporal lobe, and language areas were activated with MP in comparison to BS and that the left superior parietal lobe and the left frontal and prefrontal gyrus were activated with MT in comparison to BS. In the analysis of MT minus MP, the left dorsal lateral prefrontal cortex, the left inferior parietal lobe temporal lobe, and the inferior prefrontal cortex were activated, findings which are not consistent with rule theory or model theory.

Overall, it is unknown whether people solve reasoning problems by means of a set of inference rules or by means of visual–spatial models. In addition, although PET and fMRI studies provide important results, the time course of brain activity using such techniques lacks good temporal resolution. Many of these studies examined the brain activation during the whole reasoning process in a blocked fashion and thus could not distinguish reasoning-related processes during different stages of problem processing. For example, Fangmeier, Knauff, Ruff, and Sloutsky (2006) assumed that the reasoning process proceeds in three temporally separable phases: (1) the premise processing phase, (2) the premise integration phase, and (3) the validation phase in which reasoners decide whether a conclusion logically follows from the premises. Event-related potentials (ERPs) may provide a means to evaluate timing of cognitive processes prior to a response. In the ERP technique, recordings are made of the electrical activity of the brain that is time-locked to the presentation of an external stimulus. Thus, ERP data allow for more precise examinations of the time course of activation for different stages of reasoning and provide more valuable results for determining whether people solve reasoning problems by means of a set of inference rules or by means of visual–spatial models.

The purpose of the present study was to investigate the spatiotemporal pattern of brain activation in the performance of four different formats of conditional reasoning tasks and one BS task (memory retrieval) using high-density (64 channels) ERP recording and dipole source analysis (BESA software). Although a previous study found that different brain areas are activated by comparing MP with MT (Noveck et al., 2004), it is not known which ERP components are involved in different conditional reasoning tasks (MP, MT, AC, and DA) and whether there are differences among these components. Additionally, an fMRI study found activation of different brain areas to be involved in cognitive processes of reasoning and, thus, supported the theory of dual-process reasoning (Goel et al., 2000, Noveck et al., 2004). The methods of high-density (64 channel) ERP recordings and the dipole source analysis provide critical spatiotemporal information for analyzing the functional neuroanatomy of cognitive process of different inference forms, thereby enabling the testing of the different hypotheses.