(800) 681-3114

Implicit Learning: Robustness in the Face of Psychiatric Disorders

The performance of a group of psychiatric inpatients on two different cognitive tasks was compared with that of a control group of college undergraduates. The task in the first experiment was implicit learning of a complex, synthetic grammar; the task in the second experiment was explicit learning of relatively simple letter-to-number matching rules. In the first experiment, differences between the normals and the psychiatrically impaired were found on the preliminary memorization task but not on the implicit grammar learning task; in the second experiment, differences were observed on all phases of the experiment, with the inpatients performing no better than chance. These findings provide support for the hypothesis that, under appropriate conditions, individuals suffering from serious disorders may show no deficits when working with complex and abstract stimulus domains while showing serious performance problems when working with relatively simple, concrete stimuli. The key factor is that the former were presented as tasks that tap nonreflective, implicit processes, whereas the latter were put forward as ones that recruit conscious, explicit processes.

For some 20 years now the term implicit learning has been employed to characterize the manner in which people develop intuitive knowledge of the underlying structure of a complex stimulus environment.

The experiments are based in part on studies reported in Abrams (1987). Preparation of this article was supported in part by a grant from the City University of New York PSC-CUNY Research Award Program, Brooklyn College of CUNY.

Brooklyn College and the Graduate Center of CUNY, Brooklyn, New York 11210. Address all correspondence to Michael Abrams, Psychological Medicine Corporation, Jersey City, New Jersey 07306, or to Arthur S. Reber, Brooklyn College of CUNY, Brooklyn, New York 11210.

1967, 1988; Schacter, 1987). In the standard representation, an implicit process is marked by two essential features: The acquisition operation is unconscious and automatic, and the resulting knowledge base is abstract and rule-governed.

The classic paradigm used to study implicit learning has evolved the use of complex stimulus displays whose properties are defined by a rich rule system that generates the displays. The earliest work here used artificial grammars based on finite-state (Markovian) systems (Reber, 1967), and most of the subsequent research on these implicit, nonreflective processes has used such synthetic, semantic-free systems (Brooks, 1978; Dulany, Carlson, & Dewey, 1984; Fried & Holyoak, 1984; Howard & Ballas, 1980; Morgan & Newport, 1981; Reber, 1976; Reber & Allen, 1978; Reber, Kassin, Lewis, & Cantor, 1980). The generality of the findings, however, is well known, and a number of other stimulus environments has also been used by various researchers with congruent results. Broadbent and his colleagues showed that people learn to control complex, simulated economic and production systems in an implicit manner (Berry & Broadbent, 1984; Broadbent, FitzGerald, & Broadbent, 1986), Lewicki has reported similar automatized processes in tasks as diverse as those involving stochastic predictions of the location of visual stimuli and the acquisition of person perception dispositions (Lewicki, 1985, 1986a, 1986b), and Reber and Millward have found implicit processes operating in the classic probability learning task (Millward & Reber, 1968, 1972; Reber, 1988; Reber & Millward, 1968, 1971).

A perusal of this literature makes it quite clear that these implicit, automatic, nonreflective cognitive processes differ in significant and important ways from the more typically studied explicit, conscious, reflective ones of problem solving, concept formation, associative learning, and the like, where the learning is conscious and explicit and the resulting knowledge base can be reasonably well communicated to others (Reber, 1988; Schacter, 1987).

Recently, Hasher and Zacks and their co-workers (Hasher & Zacks, 1979, 1984; Zacks, Hasher, & Sanft, 1982) have pursued ~/ line of research that can be seen as conceptually parallel to that on implicit learning. They have argued that information about particular classes of stimuli is also encoded without any cognitive or attentional effort. The kinds of stimulus information that they have worked with, however, are considerably less complex than those used in the implicit learning literature. Specifically, they have proposed that information about the frequency and temporal and spatial location of events is acquired without conscious effort or reflection. They also, interestingly, proposed that this kind of information acquisition is extremely robust and can be seen in undiminished form in severely impaired individuals.

This last suggestion has received considerable attention recently and has been supported by a substantial body of research indicating that individuals suffering from a wide range of mental disabilities can automatically encode information to essentially the same extent as people without impairments. Students with learning handicaps (Ceci, 1982, 1983), adults with depressive disorders (Remien, 1986; Roy, 1982), and even those with the severe memory impairments of Alzheimer's disease and Korsakoff's syndrome (Knopman & Nissen, 1987; Strauss, Weingartner, & Thompson, 1985; Hirst & Volpe, 1985) show markedly less impairment in automatic encoding of event frequency than in effortful memory tasks. Newman, Weingartner, Smallberg, and Calne (1984) even showed that artificially induced impairments effected by the administration of 1-dopa had an impact on effortful cognitive processes but not on automatic frequency encoding.

The fact that the automatic induction of fundamental information is resistant to impairments raises the question as to whether the processes of implicit learning are similarly immune. The inference here derives from the argument put forward in several places that implicit learning should properly be viewed as a complex, covariational form of frequency counting.

For example, in an earlier paper (Reber & Lewis, 1977) it was noted that the subjects' underlying representation of an implicitly learned system like a Markovian grammar was captured by bi- and trigram covariations of a relational form. A recent report by Mathews, Buss, and Stanley (1987) confirmed this characterization; subjects typically code the complex letter sequences into two- and three-letter clusters on the basis of the degree of covariation that these letters display in the rule system and marked for location. Using a somewhat different context, Lewicki came to a similar conclusion about the importance of the encoding of covariations of events. In his recent review of nonconscious cognitive processes in social settings (Lewicki, 1986a), he asserted that the most basic operation involved in processing information about episodes is the encoding of cooccurrence and covariation of events. He has put forward the even stronger argument that there is no other way to acquire a concept than to discover cooccurrences between (some of) its features, and there is no other way to evaluate novel instances for its category membership than to check whether the exemplars' features occur properly.

In light of considerations such as these, it seemed not unreasonable to hypothesize that implicit learning of a system as complex as a synthetic grammar might, like the Hasher and Zacks type of encoding of fundamental information, be robust in the presence of various forms of psychological disorders. To explore this hypothesis, two experiments were carried out in which performance of psychiatric patients was compared with that of normal college students on a standard implicit learning task as well as on tasks that required explicit, reflective cognitive processing. Following the line of reasoning above, we anticipated large differences between our populations on the explicit tasks but more muted differences on the implicit.


This study is a direct comparison between a mixed group of psychiatric patients and a group of college undergraduates on the standard grammar learning task that has been used on numerous other occasions by a variety of researchers.


Psychiatric Patients. This group consisted of 25 male inpatients at the Brooklyn Veteran's Administration Hospital. All either had been hospitalized for major disorders, primarily schizophrenia and depression, or were inpatients in treatment for chronic alcoholism. All of the alcoholics exhibited some signs of organic brain impairment. In all cases, the diagnosis made by the hospital staff was accepted as accurate.

Some comments on this rather unusual subject population are called for here. First, we were not particularly concerned with distinguishing between various diagnostic categories in this study. The primary interest here was to provide, if we could, a kind of "existence demonstration." That is, we wanted to explore whether the line of reasoning that led us to this experiment had a basic validity to it. To that end, any subtleties that may exist between the cognitive capacities of these various psychiatric categories were bypassed in this study.

Control Subjects. The comparison group consisted of 36 undergraduate students at Brooklyn College who served as part of the requirements of a course in introductory psychology.

Generally speaking, the demographic characteristics of the groups differed considerably. The average age of the psychiatric patients was 39 years; of the undergraduate students, 21 years. In addition, the student population, not surprisingly, was better educated and healthier, and, of course, did not spend their time in an institutional setting. These differences are important in that they make the comparison in performance between the two groups that much more stringent a test of the overall hypothesis.


Fig. 1. Schematic diagram of the finite-state grammar used to generate the stimuli for Experiment 1. Stimuli are generated by following any path of arrows leading from the initial state to any of the possible terminal states.


Learning Stimuli. The stimuli for the learning phase of the study consisted of letter strings of lengths 2 through 6 using combinations and permutations of the letters J, T, V, X. These strings were produced by the simple, finite-state grammar shown in schematic form in Figure 1. Each of these "grammatical" strings is generated by entering the system at State 1 and following the arrows that define permissible transitions, arriving at States 3, 4, or 5, all of which can terminate the string. Twenty-five strings were selected at random from those the grammar can generate and served as the learning stimuli.

Testing Stimuli. The stimuli for the testing phase were 50 letter strings made up using the same four letters. Of these, 25 were grammatical strings and 25 were not. These "nongrammatical" items were created by introducing a violation into a grammatical string. Of the 25 grammatical strings, four were "old" strings, which were used as learning stimuli.


Learning. Letter strings were presented one at a time on a computer screen with a 3-second exposure time, after which the screen went blank and the subject was prompted to reproduce the string on the keyboard. Each string was presented up to three times if a subject failed to reproduce it correctly. The full list of 25 strings was run through twice.

Testing. Letter strings were presented on the computer screen for 7 seconds each. Following each presentation, the subject was prompted to categorize that string as either a correct (or acceptable) string or one that contained an error and was unacceptable. Responses were made by pressing either the Y or the N key. The notion of correctness was described by telling subjects to regard the first 50 strings that they had reproduced as examples of acceptable or correct strings.

The full set of 50 strings was presented twice, resulting in a total of 100 of these "well-formedness" trials for each subject. Subjects were informed about the 50-50 split in grammatical and nongrammatical strings but were not told about the repetition of the items, nor were they given any feedback about the correctness of their responses. After the full testing session was over, subjects were queried as to the basis of their well-formedness decisions.

Instructions. At the beginning of the learning phase, half the subjects in each group were given neutral instructions in which the task was described as a simple memory task and half were given instructions informing them about the rule-governed nature of the stimuli and were encouraged to try to figure out what the rules were. These instructions appeared on the computer screen and were simultaneously read aloud by the experimenter. After the learning phase was complete, all subjects were told about the rule-governed nature of the letter strings prior to beginning the testing phase. For the neutrally instructed subjects, this was the first time any explicit information about the letter strings was provided.


Learning. An analysis of variance revealed a significant difference between the two subject populations in the amount of time it took to complete the learning task, with the normal student group taking much less time (11.4 vs. 15.6 seconds) than the inpatient population (F(1, 59) = 7.7, p < .01, MSe = 34.5).


Owing to a computer failure, we were unable to retrieve complete data on the number of trials to criterion and the number and type of errors made by the subjects during this task. The only data available were total times to learn. While not entirely satisfactory, this measure does provide a rough estimate of the difficulty that individual subjects had with this task. The learning task is one that taps overt, conscious processes, the kinds of cognitive functions usually under the control of conscious, reflective operations. As such, we take this difference to be an indication that on such tasks the impaired subject population performs relatively poorly. There were no differences between the instruction groups and no interaction between instructions and population. There were, in fact, no effects of instructional set anywhere in the experiment, and in all analyses these subgroups have been combined.

Testing. Both groups were significantly above chance on the well-formedness task. The student group had a mean proportion correct of 66.4 (t(35) = 17.1, p < .001) and the inpatient group a mean of 65.3 (t(24) = 10.4, p < .00i). There was no difference between the groups in this ability to discriminate well-formed from ill-formed strings. There was a slight but nonsignificant tendency for both groups of subjects to accept the "old" strings from the learning phase as well formed more frequently than the novel, grammatical strings, but the groups did not differ from each other. There was no bias toward accepting and/or rejecting grammatical and nongrammatical strings in either group. As is displayed in Table I, even when the various subject populations are subdivided into those with primary diagnosis of psychosis and those on the special ward for alcoholics, no differences are found either between the impaired subgroups or the normals in their ability to discriminate between strings that conformed to the grammar and those that did not. In fact, as is shown in Table I, there were no significant differences in subjects' performance on any of the measures of string discrimination.

The post experimental debriefing session, during which we attempted to get as much detail as possible from our subjects concerning the basis upon which they were making their decisions, yielded few data of value. Neither group of subjects was particularly introspective, and few accurate characterizations of the actual rule system were given. Not surprisingly, the normal control subjects were considerably more coherent and expressive than the inpatient population, but, surprisingly, they were no more accurate in presenting justifications for their well-formedness judgments. This general finding using this interrogation technique is in keeping with results reported elsewhere (Reber, 1967; Reber, Allen, & Regan, 1985), although we recognize that compared with procedures used by Dulany et al. (1984) and Mathews et al. (1987), it may yield an underestimate of subjects' conscious knowledge.


The key finding is the significant difference in performance between the two subject populations when confronted with a task that calls on explicit and conscious cognitive systems combined with the lack of a difference when they are presented with one that recruits implicit, nonreflective processes. These results are in keeping with the characterization of implicit learning sketched above. They also represent a very interesting finding: Individuals suffering from serious psychological disorders that disrupt "ordinary," overt, cognitive processes, such as short-term memory, perform with virtually no deficits when dealing with stimulus environments of the complexity of the synthetic grammar used here. As outlined above, a number of recent studies (for reviews, see Hasher & Zacks, 1984; Schacter, 1987) lend support to the proposition that relatively simple, nonreflective processes, such as frequency encoding of single events, are robust in the face of serious psychiatric disorders. These results imply that an extension of this robustness hypothesis can be made to complex processes, such as the implicit learning of an artificial grammar.

While we find these results quite exciting, we want to emphasize that we regard this experiment as a sophisticated pilot study--a compelling one to be sure, but still one whose findings should be regarded as preliminary in nature. There is a number of reasons for our caution. First, we feel that any unprecedented result needs confirmation and replication before it is taken unquestioningly. Second, we recognize that this study is flawed, first, by the failure to have a specific overt and conscious task where differences between our subject populations might be observed and, second, by the loss of the fine-grain data from the learning phase. To pursue these issues in a more concrete fashion, a second experiment was run using a relatively simple problem-solving task, one where the application of overt, conscious processes should yield success. Here, we anticipated finding differences between our two populations.


  • Abrams, M. (1987). Implicit learning in the psychiatrically disturbed. Unpublished doctoral dissertation, City University of New York.
  • Allen, R., & Reber, A.S. (1980). Very long term memory for tacit knowledge. Cognition, 8, 175-t85.
  • Berry, D.C., & Broadbent, D.E. (1984). On the relationship between task performance and associated verbalizable knowledge. Quarterly Journal of Experimental Psychology, 36, 209-231.
  • Broadbent, D.E., Fitzgerald, P., & Broadbent, M.H.P. (1986). Implicit and explicit knowledge in the control systems. British Journal of Psychology, 77, 33-50.
  • Brooks, L. (1978). Nonanalytic concept formation and memory for instances~ In E. Rosch & B.B. Lloyd (Eds.), Cognition and categorization. Hillsdale, NJ: Erlbaum.
  • Ceci, S.J. (1982). Extracting language meaning from stimuli: Automatic arid purposive processing of the language-based learning disabled. Topics in Learning and Learning Disabilities, 2, 46-53.
  • Ceci, &J. (1983). Automatic and purposive semantic processing characteristics of normal and language/learning-disabled children. Developmental Psychology, !9, 42%439.
  • Danks, J.H., & Gans, D.L. (1975). Acquisition and utilization of a rule structure. Journal of Experimental Psychology, 104, 201-208.
  • Dulany, D.E., Carlson, R.A., & Dewey, G.I. (1984). A case of syntactical learning and judgment: How conscious and how abstract? Journal of Experimental Psychology: General, 113, 541-555.
  • Fried, L.S., & Holyoak, K.S. (1984). Induction of category distributions: A framework for classification learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 234-257.
  • Hasher, L., & Zacks, R.T. (1979). Automatic and effortfut processes in memory. Journal of Experimental Psychology: General, 108, 356-388.
  • Hasher, L., & Zacks, R.T. (1984). Automatic processing of fundamental information. American Psychologist, 39, 1272-1388.
  • Hirst, W., & Volpe, B.T. (1985). Automatic and effortful encoding in amnesia. In M.
  • Gazzaniga (Ed.), Handbook of cognitive neuroscience. New York: Plenum Press.
  • Howard, J.H., & Ballas, J.A. (1980). Syntactic and semantic factors in the classification of nonspeech transient patterns. Perception and Psychophysics, 28, 431-4.39.
  • Knopman, D.S., & Nissen, M.J. (1987). Implicit learning in patients with probable Alzheimer's disease. Neurology, 37, 784-788. Lewicki, P. (1985). Nonconscious biasing effects of single instances on subsequent judgments. Journal of Personality and Social Psychology, 48, 563-574.
  • Lewicki, P. (1986a). Nonconscious social information processing. New York: Academic Press.
  • Lewicki, P. (1986b). Processing information about covariations that cannot be articulated. Journal of Experimental Psychology: Learning, Memory, and Cognition, 12, 135-146.
  • Mathews, R.C., Buss, R.R., & Stanley, W.B. (1987). Accessibility of implicitly acquired knowledge of artificial grammars (Technical report). Baton Rouge: Louisiana State University.
  • Millward, R.B., & Reber, A.S. (1968). Event-recall in probability learning. Journal of Verbal Learning and Verbal Behavior, 7, 980-989.
  • Millward, R.B., & Reber, A.S. (1972). Probability learning: Contingent-event sequences with lags. American Journal of Psychology, 85, 81-89.
  • Morgan, J.L., & Newport, E.L. (1981). The role of constituent structure in the induction of an artificial language. Journal of Verbal Learning and Verbal Behavior, 20, 67-85.
  • Newman, R.P., Weingartner, H., Smallberg, S.S., & Calne, D.B. (1984). Effortful and automatic memory: Effects of dopamine. Neurology, 34, 805-807.
  • Reber, A.S. (1967). Implicit learning of artificial grammars. Journal of Verbal Learning and Verbal Behavior, 6, 855-863.
  • Reber, A.S. (1976). Implicit learning of synthetic languages: The role of instructional set. Journal of Experimental Psychology: Human Learning and Memory, 2, 88-94.
  • Reber, A.S. (1988). Implicit learning and tacit knowledge. Journal of Experimental Psychology: General.
  • Reber, A.S., & Allen, R. (1978). Analogy and abstraction strategies in synthetic grammar learning: A functionalist interpretation. Cognition 6, 189-221.
  • Reber, A.S., Allen, R., & Regan, S. (1985). Syntactical learning and judgment, still unconscious and still abstract: Comment on Dulany, Carlson, and Dewey. Journal of Experimental Psychology: General, 114, 17-24.
  • Reber, A.S., Kassin, S.M., Lewis, S., & Cantor, G. (1980). On the relationship between implicit and explicit modes in the learning of a complex rule structure. Journal of Experimental Psychology: Human Learning and Memory, 6, 492-502.
  • Reber, A.S., & Lewis, S. (1977). Toward a theory of implicit learning: The analysis of the form and structure of a body of tacit knowledge. Cognition, 5, 333-36l.
  • Reber, A.S., & Millward, R.B. (1968). Event observation in probability learning. Journal of Experimental Psychology, 77, 317-327.
  • Reber, A.S., & Millward, R.B. (1971). Event tracking in probability learning. Journal of Experimental Psychology, 77, 317-327.
  • Remien, R.H. (1986). Depression and cognitive impairment: Some limits on a general theory of effort. Dissertation Abstracts International, 47, 3123.
  • Roy, F.A. (1982). Action and performance. In A.W. Ellis (Ed.), Normality and pathology in cognitive functions. London: Academic Press.
  • Schacter, D.L. (1987). Implicit memory: History and current status. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 501-518.
  • Strauss, M.E., Weingartner, H., & Thompson, K. (1985). Remembering words and how often they occurred in memory impaired patients. Memory and Cognition, 13, 1507-1510.
  • Weingartner, H., Burns, S., Diebel, R., & Lewitt, P.A. (1984). Cognitive impairments in Parkinson's disease: Distinguishing between effort-demanding and automatic cognitive processes. Psychiatry Research, 11, 223-235.
  • Weingartner, H., Grafman, J., Boutelle, W., Kaye, W., & Martin, P.R. (1983). Forms of memory failure. Science, 221, 381-383.
  • Zacks, R.T., Hasher, L., & Sanft, H. (1982). Automatic encoding of frequency: Further findings. Journal of Experimental Psychology: Learning, Memory, and Cognition, 8, I06-116.