Evidence of acoustic coding in long-term memory

Contrary to previous indications, retroactive interference in long-term paired associate learning was found to be a function of acoustic similarity. Experimental groups were exposed to the A-B, A'-C paradigm where corresponding stimuli were homophones. Their retention scores were substantially and significantly lower than control groups run with an A-B, C-D paradigm. The failure of previous studies to reveal effects of acoustic similarity in this way is attributed to the use of an insufficiently high degree of similarity.     

Modelling intransitive preferences: A random-effects approach

Current models for individual-level paired comparison data are based on the three assumptions that (1) pairwise judgments are independent, (2) the utility of an item remains invariant across trials, and (3) pair-specific variability can account for intransitive choice behaviour. All three assumptions seem strong and likely to be violated in empirical applications. This paper introduces a new framework for the analysis of paired comparison data which relaxes these three assumptions and considers the utilities associated with the same item across trials to be neither independent nor identical, but related. The proposed approach provides new insights about the reliability and consistency of paired comparison judgments and can account for systematic violations of transitivity. An application, based on a replication of Tversky's [(1969). Intransitivity of preference. Psychological Review, 76, 31-48] gamble study, illustrates the usefulness of the new approach in modelling both transitive and intransitive preferences.     

Psychophysical discontinuity and pseudosequence effects

A psychophysical model is advanced which explicitly allows for a major discontinuity in the response continuum, and for stochastic transition between response states, where each state is uniquely associated with one segment of the psychophysical relation. It is shown that the form of mean psychophysical gain function thus generated is consistent with second-order departures from simpler models, and that pseudosequential effects, up to a lag of two trials, can be generated if the gain function is assumed to be continuous when in fact it contains a major discontinuity. Previous evidence on apparent sequential effects must be reevaluated in the light of this result.     

Intention to respond in a special way offers some protection against forgetting associations

In a continuous memory-updating paradigm, subjects studied name-color associations and were tested later for the color associate given the name. The default color response was made in one location, but on predesignated trials the response was required to be made in a special location. Memory for the color associated with a given name was assessed after short and long retention intervals when both default and special responses were required. Separate measures were examined of memory for the intention to respond in a particular way (default or special) and of memory for the color associations paired with the names. Memory for color associates was better overall with short than with long retention intervals and was better when special (rather than default) responses were required, especially at the long retention interval. These results imply that the requirement to respond in a special way protects associations from loss due to forgetting.     

A note on mathematical models for learning

It is shown that a rational learning curve developed by Estes for paired associate learning is a special case of a more general function. The latter is the product of two functions and assumes that the discovery and fixation aspects of learning are independent. The indications are that the form of one of the functions has not been tested sharply by Estes in the paired associate learning setting.     

A general framework for the analysis of concept identification tasks

A general framework is presented for concept identification based on hypothesis-testing theory. It is a modification of the duoprocess theory presented by Chumbley (1972). It is shown how Markov models for various complex concept identification tasks may be derived from this framework and how such models may be analyzed by making use of probability generating functions. Two experiments are described. In experiment 1 three tasks were used: two simple tasks, where the subject either only had to select the relevant dimension in order to solve the problem or only had to learn a short list of paired-associates, and a more complex task, where both processes were needed to reach the solution. The results were in general favorable to the theory. Experiment 2 was designed to test the application of the theory to the four-choice concept problem. The predictions of the theory are compared to those of the subproblem learning theory of trabasso and Bower (1964), modified to include a ‘learning-on-errors’ assumption. The fit of the duoprocess theory was reasonably good and superior to that of the subproblem learning theory.     

Short-term retention of response outcome as a determinant of serial reversal learning

The correlation between short-term retention of the outcome of the preceding response and overall learning proficiency was investigated for serial reversal learning. Pigeons were trained to asymptote on a serial reversal problem and then were presented a percentage reinforcement schedule where only some correct trials were rewarded. Nonrewarded correct trials were treated exactly as incorrect trials. The difference in error probability following the two types of correct trials was then used as a measure of short-term retention. When intertrial intervals (ITI) were short (6 sec), substantial differences occurred. When the ITI's were increased, the difference in accuracy declined regularly to no difference at an ITI of 60 sec. This demonstration of a short-term retention gradient, coupled with the finding that overall reversal learning was much better with the shorter ITI's, suggests that a primary mechanism of improvement in serial reversal learning is the acquisition of a conditional discrimination based on the outcome of the preceding response.     

Forgetting curves: implications for connectionist models

Forgetting in long-term memory, as measured in a recall or a recognition test, is faster for items encoded more recently than for items encoded earlier. Data on forgetting curves fit a power function well. In contrast, many connectionist models predict either exponential decay or completely flat forgetting curves. This paper suggests a connectionist model to account for power-function forgetting curves by using bounded weights and by generating the learning rates from a monotonically decreasing function. The bounded weights introduce exponential forgetting in each weight and a power-function forgetting results when weights with different learning rates are averaged. It is argued that these assumptions are biologically reasonable. Therefore power-function forgetting curves are a property that may be expected from biological networks. The model has an analytic solution, which is a good approximation of a power function displaced one lag in time. This function fits better than any of the 105 suggested two-parameter forgetting-curve functions when tested on the most precise recognition memory data set collected by. Unlike the power-function normally used, the suggested function is defined at lag zero. Several functions for generating learning rates with a finite integral yield power-function forgetting curves; however, the type of function influences the rate of forgetting. It is shown that power-function forgetting curves cannot be accounted for by variability in performance between subjects because it requires a distribution of performance that is not found in empirical data. An extension of the model accounts for intersecting forgetting curves found in massed and spaced repetitions. The model can also be extended to account for a faster forgetting rate in item recognition (IR) compared to associative recognition in short but not long retention intervals.     

Dynamics of activation in semantic and episodic memory

Spreading-activation models for the structure of semantic and episodic memory postulate a network of interconnected nodes in which activation spreads from a source node to recipient nodes. These models account for a broad range of memory-related processes, including word recognition, sentence verification, prose comprehension, and sentence production. A fundamental question regarding this account concerns the nature of activation growth at each node in the network. Two mutually exclusive possibilities are (a) that activation grows in a discrete fashion, making abrupt transitions between two or more distinct states and (b) that activation grows continuously from a resting level to an asymptotic level. In the present article, we characterize this dichotomy with examples from the literature, and we apply an adaptive priming procedure for testing discrete versus continuous activation models. Our procedure involves the presentation of prime stimuli at various moments before a test stimulus; subjects are required to make a lexical (word/nonword) decision about the test stimulus. The duration of the interval between the prime and test stimuli is varied adaptively on the basis of subjects' performance. Reaction times are recorded as a function of this duration. According to discrete activation models, there is a unique reaction-time distribution associated with each possible state of node activation. The distribution of reaction times observed when the test stimulus appears near the moment of transition between discrete states should therefore constitute a finite mixture of the underlying basis distributions associated with the individual discrete activation states. The mixture proportion will depend on the relation between the priming interval and the distribution of state-transition times. Continuous activation models assert instead that activation grows continuously over time and that there is a unique reaction-time distribution associated with any given degree of intermediate priming. Such models predict that no finite mixture distribution will emerge when the priming interval has a fixed intermediate duration. Two experiments with the adaptive priming procedure are reported to test these alternative predictions. In Experiment 1, the prime and test stimuli were semantically associated words (e.g., bread-butter). In Experiment 2, episodic associations between the prime and test stimuli were established through paired associate learning. For both cases, the mixture prediction failed, and two-state discrete activation models were rejected.(ABSTRACT TRUNCATED AT 400 WORDS)     

Associative isolation: Unifying associative and list memory

Rather than treating paired associate and serial learning as involving the acquisition of distinct types of information [e.g. Murdock (1974). Human memory: Theory and data. New York: Wiley] I propose an Isolation Principle which treats the two as ends of a continuum. According to this principle, consecutive pairs of items are relatively isolated from other studied items in paired associates learning, but not isolated in serial list learning. The consequence is that variability that dominates forward and backward probed recall is highly correlated in pairs but less so, due to differential interference, in lists. This can explain an important dissociation: whereas forward and backward probes of pairs are nearly perfectly correlated, the correlation is only moderate for serial lists. I demonstrate this in a chaining model by varying item-to-item associative strengths and in a positional coding model by varying the similarity structure of item positions. This enables a range of models to account for data on pairs and lists, as well as potential intermediate or hybrid paradigms, within a single theoretical framework.     

Does test-enhanced learning transfer for triple associates?

Test-enhanced learning and transfer for triple-associate word stimuli was assessed in three experiments. In each experiment, training and final-test trials involved the presentation of two words per triple associate (triplet), with the third word having to be retrieved. In agreement with the prior literature on different stimuli, training through testing with feedback yielded markedly better final-test performance than did restudy. However, in contrast to the positive transfer reported for paired associate stimuli, minimal or no positive transfer was observed, relative to a restudy control, from a trained cue combination (e.g., A, B, ?) to other cue combinations from the same triplet that required a different response (e.g., B, C, ?). That result also held when two unique cue combinations per triplet were tested during training, and for triplets with low and high average associative strengths. Supplementary analyses provided insight into the overall transfer effect: An incorrect response during training appears to yield positive transfer relative to restudy, whereas a correct response appears to yield no, or even negative, transfer. Cross-experiment analyses indicated that test-enhanced learning is not diminished when two or three cue combinations are presented during training. Thus, even though learning through testing is highly specific, testing on all possible stimulus–response combinations remains the most efficient strategy for the learning of triple associates.     

Contracting,equal, and expanding learning schedules: The optimal distribution of learning sessions depends on retention interval

In laboratory and applied learning experiments, researchers have extensively investigated the optimal distribution of two learning sessions (i.e., initial learning and one relearning session) for the learning of verbatim materials. However, research has not yet provided a satisfying and conclusive answer to the optimal scheduling of three learning sessions (i.e., initial learning and two relearning sessions) across educationally relevant time intervals. Should the to-be-learned material be repeated at decreasing intervals (contracting schedule), constant intervals (equal schedule), or increasing intervals (expanding schedule) between learning sessions? Different theories and memory models (e.g., study-phase retrieval theory, contextual variability theory, ACT-R, and the Multiscale Context Model) make distinct predictions about the optimal learning schedule. We discuss the extant theories and derive clear predictions from each of them. To test these predictions empirically, we conducted an experiment in which participants studied and restudied paired associates with a contracting, equal, or expanding learning schedule. Memory performance was assessed immediately, 1 day, 7 days, or 35 days later with free- and cued-recall tests. Our results revealed that the optimal learning schedule is conditional on the length of the retention interval: A contracting learning schedule was beneficial for retention intervals up to 7 days, but both equal and expanding learning schedules were better for a long retention interval of 35 days. Our findings can be accommodated best by the contextual variability theory and indicate that revisions are needed to existing memory models. Our results are practically relevant, and their implications for real-world learning are discussed.     

An analysis of some conditions for representingN state Markov processes as general all or none models

Recently Markov learning models with two unidentifiable presolution success states, an error state, and an absorbing learned state, have been suggested to handle certain aspects of data better than the three state Markov models of the General All or None model type. In attempting to interpret psychologically, and evaluate statistically the adequacy of various classes of Markov models, a knowledge of the relationship between the classes of models would be helpful. This paper considers some aspects of the relationship between the class of General All or None models and the class of Stationary Absorbing Markov models withN error states, andM presolution success states.     

The Effect of Erroneous Knowledge of Results on Skill Acquisition when Augmented Information is Redundant

Even though it can be shown that verbal knowledge of results (KR) is redundant with sensory feedback for learning certain motor skills, such findings do not eliminate the possibility that when KR is available it influences underlying learning processes. In order to examine the function of KR more closely, two experiments were designed in which the subjects received conflicting information about their own sensory feedback and the KR presented by the experimenter. In Experiment 1, two erroneous-KR groups, a correct-KR group, and a no-KR group performed 150 practice trials on a simple anticipation timing task and then performed three no-KR retention tests of 30 trials each following intervals of 10 minutes, 1 week, and 1 month. The results supported previous findings that providing correct KR is redundant in anticipation tasks. However, learning was influenced by KR as subjects performed according to the erroneous KR information, thereby ignoring their sensory feedback even after a 1-month interval. In Experiment 2, subjects practised a more complex striking response for the anticipation task for 75 trials and then performed no-KR retention trials either immediately, or 1 day or 1 week later. One of the groups received erroneous KR after 50 practice trials with correct KR. The results confirmed and extended those from Experiment 1, as erroneous KR, even after initial practice with correct KR, influenced retention performance. These results indicate that although KR provides information that is not needed to learn anticipation timing skills, this augmented verbal information is a dominant source of information that influences underlying cognitive processes involved in learning motor skills.     

Word-learning performance in beginning readers.

This investigation examined word-learning performance in beginning readers. The children learned to read words with regular spelling-sound mappings (e.g., snake) more easily than words with irregular spelling-sound mappings (e.g., sword). In addition, there was an effect of semantics: Children learned to read concrete words (e.g., elbow) more successfully than abstract words (e.g., temper). Trial-by-trial learning indicated that children made greater use of the regularity and semantic properties at later trials as compared with early trials. The influence of cognitive skills (paired associate learning and phonological awareness) on word-learning performance was also examined. Regression analyses revealed that whereas paired associate learning skills accounted for unique variance in the children's learning of both regular and irregular words, phonological awareness accounted for unique variance only in the acquisition of regular words.     

Discovering the Sequential Structure of Thought

Multi‐voxel pattern recognition techniques combined with Hidden Markov models can be used to discover the mental states that people go through in performing a task. The combined method identifies both the mental states and how their durations vary with experimental conditions. We apply this method to a task where participants solve novel mathematical problems. We identify four states in the solution of these problems: Encoding, Planning, Solving, and Respond. The method allows us to interpret what participants are doing on individual problem‐solving trials. The duration of the planning state varies on a trial‐to‐trial basis with novelty of the problem. The duration of solution stage similarly varies with the amount of computation needed to produce a solution once a plan is devised. The response stage similarly varies with the complexity of the answer produced. In addition, we identified a number of effects that ran counter to a prior model of the task. Thus, we were able to decompose the overall problem‐solving time into estimates of its components and in way that serves to guide theory.     

Sequential search processes in long-term memory

In an experiment employing the symbol-element recognition task (Mohs, Wescourt, & Atkinson, 1975), subjects first learned six lists consisting of four words (elements) each. Each list was associated with a unique consonant (symbol). Subsequently, on each of a series of test trials, subjects were presented with one, two, or four symbol-element pairs. A positive response was required if all test words were correctly paired with their associated consonants and a negative response if any one test word was incorrectly paired with a consonant. Of primary concern was the way reaction time (RT) varied with number of pairs presented, the type of response required, and, on negative trials, the position of the mismatched pair in the test display. RT increased with the number of pairs presented on a trial and the increase was greater for positive than for negative trials. For negative pair trials, RT increased with the distance of the mismatched pair from the top of the test display. On negative trials in which the top pair in the test display was the mismatched pair, RT increased with the total number of pairs presented on the trial. A serial, probabilistic order of processing model is proposed to account for these results, and applications of the model to other paradigms are discussed.     

Paired associate learning: Data on pair difficulty and variables that influence difficulty

Two hundred pairs of five-letter words were produced randomly from all five-letter words in Thorndike and Lorge (1944). The difficulty of each pair was established for paired associate learning by having 50 subjects learn lists of 20 pairs. The difficulty of a pair was found to be highly reliable and was not influenced by the particular list in which it was learned. Frequency of response terms was positively related to learning, but the frequency of the stimulus terms was not. Two-syllable five-letter words were learned more rapidly than one-syllable five-letter words. Two 20-pair lists were constructed, one consisting of homogeneous pairs of average difficulty and the other of 10 very easy pairs and 10 very difficult pairs. As anticipated, performance on the latter list was initially better than that on the former, with the performance on the two lists converging over trials. The difficulty of the pairs as determined by paired associate learning was unrelated to misses on a recognition test, but the false alarms decreased as difficulty decreased.     

Worms under cover: relationships between performance in learning tasks and personality in great tits (Parus major)

In animals, individual differences in learning ability are common and are in part explained by genetic differences, developmental conditions and by general experience. Yet, not all variations in learning are well understood. Individual differences in learning may be associated with elementary individual characteristics that are consistent across situations and over time, commonly referred to as personality or temperament. Here, we tested whether or not male great tits (Parus major) from two selection lines for fast or slow exploratory behaviour, an operational measure for avian personality, vary in their learning performance in two related consecutive tasks. In the first task, birds had to associate a colour with a reward whereas in the second task, they had to associate a new colour with a reward ignoring the previously rewarded colour. Slow explorers had shorter latencies to approach the experimental device compared with fast explorers in both tasks, but birds from the two selection lines did not differ in accomplishing the first task, that is, to associate a colour with a reward. However, in the second task, fast explorers had longer latencies to solve the trials than slow explorers. Moreover, relative to the number of trials needed to reach the learning criteria in the first task, birds from the slow selection line took more trials to associate a new colour with a reward while ignoring the previously learned association compared with birds from the fast selection line. Overall, the experiments suggest that personality in great tits is not strongly related to learning per se in such an association task, but that birds from different selection lines might express different learning strategies as birds from the different selection lines were differently affected by their previous learning performance.     

Time and order effects on causal learning

Five experiments were conducted to explore trial order and retention interval effects upon causal predictive judgments. Experiment 1 found that participants show a strong effect of trial order when a stimulus was sequentially paired with two different outcomes compared to a condition where both outcomes were presented intermixed. Experiment 2 found that a 48-h retention interval eliminates the trial order effect, so that participants gave a global judgment about the relationship between the stimulus and the two outcomes equivalent to the one given by participants that received the two phases intermixed. This result was replicated in Experiment 3 in a situation in which the probability of the outcome in the presence of the cue was changed from .5 for both outcomes (Experiments 1, 2, 4, and 5) to .75 and .25 for outcomes 1 and 2, respectively. Experiment 4 found that retention intervals ranging from 45 min to 48 h eliminated the trial order effect similarly. Experiment 5 found that a 10-min retention interval replicated the effect of time upon sequential training found in precedent experiments, regardless of whether participants remained within the laboratory during the retention interval or spent it outside. The combined results of this experimental series suggest that retention intervals reduce retroactive interference in causal learning by allowing participants to use all the information presented across phases, rather than differentially increasing or decreasing retrieval of information about each of them.