Mixed Effects of Training on Transfer

Mixed Effects of Training on Transfer Sebastien Helie (helie.sebastien@courrier.uqam.ca) Department of Computer Science, C.P. 8888 Succ. Centre-Ville Montreal, PQ H3C 3P8 (CANADA) Denis Cousineau (denis.cousineau@umontreal.ca) Department of Psychology, C.P. 6128 Succ. Centre-Ville Montreal, PQ H3C 3J7 (CANADA) search task and transferred to the visual and memory search task while the other half did the opposite. The results showed that training in visual and memory search fully transferred to memory search. On the other hand, training in memory search transferred only partially to the visual and memory search task. Hillstrom and Logan concluded that there is a global component, common to both tasks, and a private component, present only in visual and memory search. Moreover, results from their Experiment 3 suggested that the private component was closely related to the stimuli used. However, in Hillstrom and Logan’s experiments, the memory search condition was entirely embedded in the visual and memory search condition. Therefore, the observed transfer is not a surprise. Another task in which learning transfer was found is the string verification task (Haider & Frensch, 1996; 1999; 2002). In Haider and Frensch’s experiments, participants were asked to verify the validity of letter strings of the kind “A [4] F G H”. The task was to determine if the letters were an ordered segment from the alphabet. In the second position, a number always appeared in brackets. It indicated the number of letters skipped between the first and third positions. The string length varied between three and seven. The important points are that the number was always four, it was always in the second position and, if the string was not an ordered segment of the alphabet, the problem was always at the third position. Haider and Frensch (1996; 1999) postulated that participants would notice the consistency of the error position and ignore the remaining of the string (the reduction of information theory). Results supported their hypothesis: After extensive training, participants were presented with new strings and the ability to ignore useless information transferred to these new strings. Nevertheless, response times of the new strings were still slower then those of the original (training) strings. They concluded that these slower response times revealed the presence of another component, specific to the stimuli, which did not transfer. Haider and Frensch’s decomposition of skills (1996) was further studied in a perceptual learning setting (Goldstone, 1998, Doane et al., 1996). Doane and her colleagues independently tested stimulus-related and task-related knowledge in same-different tasks (Bamber, 1969) involving abstract polygons. Specifically, the discrimination difficulty in the learning phase was varied in order to measure its effect on transfer. Results showed that a harder learning phase lead to better performance on novel stimuli. Abstract The study of learning transfer yields conflicting patterns of results. While some research shows strong effects of previous learning, others show no such effects. This is a consequence of the absence of consensus on what parts of a skill is transferred. In the present paper, we suggest that learning can be divided into general task-related components and specific stimulus-related ones. In one condition, participants were transferred to a new set of stimuli while continuing to perform the same task. Results show an absence of benefit right after the transfer and the presence of long-lasting interference. In the opposite condition, the results show no effects of previous training. These diverging results are best explained by a model of higher-level skill acquisition: knowledge partitioning. Introduction The decomposition of performance has a long history in psychology. It was first proposed by Donder (1868), who proposed the subtractive method to measure the speed of “psychological acts”. This method was further developed by Sternberg in the late sixties (Sternberg, 1969) and more recently using the mean interaction contrast model (Thomas, 2000). In all these models, the decomposition occurs at the level of a single response and the goal is to identify the sequence of operations between the stimulation and the response (e.g. encoding, decision, response selection, etc.). More recently, other researchers have proposed another way of decomposing cognitive processes. In this approach, the performance is decomposed by whether improvement results from repeated exposure to the stimuli or repeated exposure to the task (Haider & Frensch, 1996, 1999). Therefore, it does not aim at identifying the stimulus- response chain of processing. However, this level of analysis is particularly interesting in the study of learning transfer because, during transfer, either the stimuli or the task are changed in part or in whole. Hillstrom and Logan (1998) have tested this approach using memory search vs. visual and memory search. In the two conditions of their Experiment 1, participants first had to memorize a set of letters prior to a block of trials. On each trial, participants saw letters in the test display and had to decide whether there was an element from the memorized set or not. The only difference between the two conditions was the test display. In the memory search condition, a single letter appeared in the display. In the visual and memory search condition, an array of letters appeared in the display. Half of the participants were trained in the memory

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.