Understanding paired associate transfer effects based on shared stimulus components
|PIs||Pavlik, MacWhinney, Wu, Koedinger|
|Faculty||MacWhinney, Wu, Koedinger|
|Others with > 160 hours||Dozzi|
|Number of students||300|
|Total Participant Hours||>700|
|Datashop?||Expected date 4/15|
Over the course of approximately 448 practice trials of Chinese vocabulary practice, certain schedules of practice were delivered to test several transfer hypotheses implied (but not implemented) by the model of practice used for in-vivo experiment optimal training schedules.
These schedules made use of the fact that there are 6 ways to test (Pinyin->English, Sound->English, Hanzi->English, English->Pinyin, Sound->Pinyin, Hanzi->Pinyin) vocabulary knowledge given current tutor capability. There were 21 total within-subjects conditions each with a particular sequence of the 6 types of trials. For instance, to look for unit learning we compared a condition where subjects practiced a Hanzi-Pinyin pair with a study (a presentation of the entire pair) followed by a drill 4 trials later with a condition where the preparation was identical expect that prior to Hanzi->Pinyin practice subjects had 2 trials of English->Pinyin practice for the same word. In this case, if the Hanzi->Pinyin practice benefits from prior English->Pinyin practice it seems to imply that the Pinyin response is being learned as a unit somewhat independently of any particular association.
There were 3 main hypotheses described below, which we call unit knowledge component learning, resonant learning, and stimulus mapping.
In all three cases the hypothesis was confirmed.
- Optimal Spacing Interval
- Expanding Spacing interval
- Wide spacing Interval
- Narrow Spacing Interval
What sorts of complexity needs to be accounted for to make the ACT-R declarative memory model adequately represent the various common types of transfer in vocabulary learning?
Background and significance
There is a venerable literature (e.g. Soloman & Asch, 1962) on effects like those described in this work, but there appears to be no unified modeling of such phenomena in the context of a practice scheduling algorithm. The assumption is that increased accuracy of the model will increase the accuracy of the predictions it makes when calculating schedules during in-vivo learning.
For example, based on the idea that there is desirable difficulty for practice, one might assume that it is better to start by giving Foreign->English drill and then reverse to English->Foreign drill after an initial proficiency is reached. Unfortunately, determining this initial proficiency necessary before transitioning to a more difficult task must be done on a case by case basis unless there is a detailed model that encompasses both tasks. This study helps to provide such a model.
Naive subjects were screened for no prior Chinese experience. Effects of conditions were assessed approximately one minute after treatments.
Transfer was measured in the unit knowledge component learning conditions and in the resonant learning conditions.
Alternative structures of instructional schedule for knowledge component training based on the predictions of an ACT-R based cognitive model. In this study the manipulations all involved the effects of having related (but not identical practice) prior to some comparison practice.
- Unit knowledge component learning - This hypothesis proposes that the stimulus items (sound file, Hanzi character, pinyin, or English) are learned as individual components somewhat independent of the pairings they occur in. Supports the notion of knowledge decomposition.
- Resonant learning - This hypothesis proposes that people spontaneously recall related knowledge components (spreading activation) when prompted to recall a specific pair. Further, this covert practice results in measurable learning.
- Stimulus mapping - This is the straightforward notion that learning of the connection between an orthography and a sound is advantaged because there are mapping rules (knowledge components) that allow this conversion.
All three hypotheses were confirmed. A paper is in preparation describing these important results.
- Unit knowledge component learning - The results supporting knowledge component learning appear to suggest that unitary knowledge (a pinyin response for instance) learned in one context applies directly to other contexts that require the same knowledge components.
- Resonant learning - These results show that learning is an active process that is not constrained by the specific stimulus demand, but rather extends to other recently learned information. The result also suggests that building more complex relational knowledge structures may have advantages due to improved reflective learning.
- Stimulus mapping - This result implies that rules(procedures) are learned which result in a general abilities to translate between certain vocabulary representations (primarily orthography-phonology).