PSYCHOLOGICAL SCIENCE Research Article A SYSTEM FOR RELATIONAL REASONING IN HUMAN PREFRONTAL CORTEX James A. Waltz, Barbara J. Knowlton, Keith J. Holyoak, Kyle B. Boone, Fred S. Mishkin, Marcia de Menezes Santos, Carmen R. Thomas, and Bruce L. Miller University of California, Los Angeles Abstract—The integration of multiple relations between mental rep- resentations is critical for higher level cognition. For both deductive- and inductive-reasoning tasks, patients with prefrontal damage exhib- ited a selective and catastrophic deficit in the integration of relations, whereas patients with anterior temporal lobe damage, matched for overall IQ but with intact prefrontal cortex, exhibited normal relation- al integration. In contrast, prefrontal patients performed more accu- rately than temporal patients on tests of both episodic memory and semantic knowledge. These double dissociations suggest that integra- tion of relations is a specific source of cognitive complexity for which intact prefrontal cortex is essential. The integration of relations may be the fundamental common factor linking the diverse abilities that depend on prefrontal function, such as planning, problem solving, and fluid intelligence. Reasoning depends on the ability to form and manipulate mental representations of relations between objects and events. For example, transitive inference (a type of deductive reasoning, in which the truth of the premises ensures the truth of the conclusion) requires the abili- ty to integrate two relations that share an element (e.g., given that Bill is taller than Charles and Abe is taller than Bill, it follows that Abe is taller than Charles). Similarly, in drawing analogies (a type of induc- tive reasoning, in which the initial premises determine the plausibility of the conclusion), relational reasoning is also essential (e.g., in the problem “person is to house as bear is to what?” the shared roles, dweller and dwelling, constrain the inferred answer, “cave”). Problem solving and planning also necessarily depend on relational integration. For example, using the elementary problem-solving strategy of differ- ence reduction (Newell & Simon, 1972) requires integration of a dif- ference between the present state and the goal state (a relation, such as “this wall lacks a coat of paint”) with the expected change that would be produced by an operator applied to the present state (a second rela- tion, such as “painting the wall will add a coat of paint”). Forming subgoals by means-ends analysis also requires integration of multiple relations (e.g., “a paintbrush can be used to apply paint”; “I lack a brush”) to derive relevant subgoals (“I must find a brush”). There is a critical gap between the capacity to comprehend single relations and the capacity to integrate multiple relations (Halford & Wilson, 1980; Halford, Wilson, & Phillips, 1998; Maybery, Bain, & Halford, 1986). For example, understanding the single relation “Bill is taller than Charles” can be accomplished using perception (given a visual scene) or language (given a sentence); in contrast, integrating two relations, such as “Bill is taller than Charles” and “Abe is taller than Bill,” to draw the transitive inference requires more than percep- tual or linguistic processing alone. Although there is evidence that nonhuman primates have some capacity to process relations (Premack & Woodruff, 1978; Thompson, Oden, & Boysen, 1997; for a review, see Tomasello & Call, 1997), humans display far greater sophistica- tion in relational reasoning across a wide range of content domains. 1 Given the large increases in the size of prefrontal cortex in human evo- lution (Benson, 1993), prefrontal cortex may be the locus of a system for relational reasoning in humans (Holyoak & Kroger, 1995; Robin & Holyoak, 1995). This hypothesis is broadly consistent with evidence that prefrontal cortical dysfunction leads to selective decrements in performance on tasks involving hypothesis testing, categorization, planning, and prob- lem solving, all of which involve relational reasoning (e.g., Delis, Squire, Bihrle, & Massman, 1992; Milner & Petrides, 1984; Owens, Downes, Sahakian, Polkey, & Robbins, 1990; Shallice & Burgess, 1991). However, these complex tasks could be failed for many reasons and were not designed to test the specific hypothesis that the prefrontal cortex is critical for relational integration. In the present study, we examined performance on simple tasks requiring deductive or inductive reasoning, using closely matched variants that differed specifically in whether or not success required integration of multiple relations. We hypothesized that patients with prefrontal cortical dysfunction would exhibit impaired performance when asked to integrate multiple relations, yet would perform normal- ly when only one relation needed to be considered. In order to rule out the possibility that any performance deficits in relational reasoning observed in prefrontal patients could be attributable to some general cognitive impairment, we compared their performance on several tasks with the performance of patients with anterior temporal lobe damage, as well as that of normal control subjects. It has long been established that patients with lesions of medial temporal lobe struc- tures show impairments in acquisition of new declarative memories, despite the preservation of other cognitive abilities (Scoville & Milner, 1957; Squire, Knowlton, & Musen, 1993). Recent results indicate that patients with lateral anterior temporal lobe lesions exhibit deficits in semantic knowledge (Graham & Hodges, 1997; Hodges, Patterson, Oxbury, & Funnell, 1992). We predicted that prefrontal patients would be impaired on tasks requiring relational integration, but would per- form more accurately than patients with anterior temporal lobe lesions on tests of semantic memory and episodic recognition. VOL. 10, NO. 2, MARCH 1999 Copyright © 1999 American Psychological Society 119 Address correspondence to James Waltz or Barbara Knowlton, Department of Psychology, UCLA, Los Angeles, CA 90095-1563; e-mail: waltz@lifesci. ucla.edu or knowlton@lifesci.ucla.edu. 1. Chimpanzees (Gillan, 1981), monkeys ( Harris & McGonigle, 1994), rats (W.A. Roberts & Phelps, 1994), and pigeons (Terrace, 1991) are capable of learning serial orderings when the items are introduced in sequential order and multiple trials are provided so that adjacent pairs are overlearned. However, these various types of associative transitivity observed in nonhuman animals can be distinguished from inferential transitivity (D’Amato, 1991). Only humans over 5 years of age appear to be able to make transitive inferences reli- ably in one trial by integrating multiple premises (Halford, 1984).