Contents Volume

Volume Contents

Contents Volume     

Knowledge level learning and the power law: A soar model of skill acquisition in scheduling

Zusammenfassung. Das hier vorgestellte Modell lernt graduell, Planungsaufgaben aus der Klasse der Maschinenbelegungsprobleme (job-shop-scheduling problems) zu lösen. Mit Hilfe des Chunking-Mechanismus von Soar wird episodisches Wissen über die Belegungsreihenfolge von Aufträgen auf Maschinen memoriert. Bei der Entwicklung des Modells wurden zahlreiche qualitative (z. B. Transfereffekte) und quantitative Befunde (z. B. Bearbeitungszeiten) aus einer früheren empirischen Untersuchung berücksichtigt. In einer Validierungsstudie wurden dieselben Aufgaben von 14 Probanden und dem Modell bearbeitet. Die Passung von Simulationsdaten und empirischen Ergebnissen fiel insgesamt gut aus. Allerdings löst das Modell die Aufgaben schneller und zeigt auch einen etwas besseren Lernverlauf als die Probanden. Das Modell liefert eine Erklärung für das Rauschen, das typischerweise bei Bearbeitungszeiten zu beobachten ist: es handelt sich um erworbenes Wissen, das mehr oder weniger gut und auch unterschiedlich häufig auf neue Situationen übertragen wird. Der Lernverlauf der Probanden entspricht nur für aggregierte Daten einer Potenzfunktion (power law). Der vorgestellte Mechanismus zeigt, wie ein symbolisches Modell der Informationsverarbeitung graduelle Verhaltensänderungen generiert und wie der offensichtliche Erwerb allgemeiner Prozeduren ohne explizites Lernen von deklarativen Regeln erfolgen kann. Es wird nahegelegt, daß es sich hier um die Modellierung einer Form impliziten Lernens handelt. Summary. The model presented here gradually learns how to perform a job-shop scheduling task. It uses Soar's chunking mechanism to acquire episodic memories about the order to schedule jobs. The model was based on many qualitative (e.g., transfer effects) and quantitative (e.g., solution time) regularities found in previously collected data. The model was tested with new data where scheduling tasks were given to the model and to 14 subjects. The model generally fit these data with the restrictions that the model performs the task (in simulated time) faster than the subjects, and its performance improves somewhat more quickly than the subjects' performance. The model provides an explanation of the noise typically found in problem solving times - it is the result of learning actual pieces of knowledge that transfer more or less to new situations but rarely by an average amount. Only when the data are averaged (i.e., over subjects) does the smooth power law appear. This mechanism demonstrates how symbolic models can exhibit a gradual change in behavior and how the apparent acquisition of general procedures can be performed without resorting to explicit declarative rule generation. We suggest that this may represent a type of implicit learning.     

The puzzle of language evolution

Linguistics must again concentrate on the evolutionary nature of language, so that language models are more realistic with respect to human natural languages and have a greater explanatory force. Multi-agent systems are proposed as a possible route to develop such evolutionary models and an example is given of a concrete experiment in the origins and evolution of word-meaning based on a multi-agent approach.     

Large-scale databases of proper names

Few tools for research in proper names have been available—specifically, there is no large-scale corpus of proper names. Two corpora of proper names were constructed, one based on U.S. phone book listings, the other derived from a database of Usenet text. Name frequencies from both corpora were compared with human subjects’ reaction times (RTs) to the proper names in a naming task. Regression analysis showed that the Usenet frequencies contributed to predictions of human RT, whereas phone book frequencies did not. In addition, semantic neighborhood density measures derived from the HAL corpus were compared with the subjects’ RTs and found to be a better predictor of RT than was frequency in either corpus. These new corpora are freely available on line for download. Potentials for these corpora range from using the names as stimuli in experiments to using the corpus data in software applications.     

Effects of response type on coordinated responses during arm movement

This paper reports some experimental results on the coordination of finger and vocal responses with passing through a target position in multjjoint arm movement. In Experiment 1, we found that the difference in the timing of finger and vocal responses cannot be attributed entirely to efferent or representational effects. Instead, it appears to reflect the extent to which information about the internal stimuli generated by the arm movement are available to the centers controlling these different responses. That is, it is a compatibility effect. In Experiment 2, the case in which a finger response is made on the same side of the body as the moving arm was compared with the case in which it is made with the contralateral hand, which remains static. The interaction effect observed suggests that the path-ways subserving coordinated responses are informationally encapsulated, so that information about arm movement is not shared between the neural centers controlling different coordinated responses.     

What affects strategy selection in arithmetic? The example of parity and five effects on product verification

The parity effect in arithmetic problem verification tasks refers to faster and more accurate judgments for false equations when the odd/even status of the proposed answer mismatches that of the correct answer. In two experiments, we examined whether the proportion of incorrect answers that violated parity or the number of even operands in the problem affected the magnitude of these effects. Experiment 1 showed larger parity effects for problems with two even operands and larger parity effects during the second half of the experiment. Experiment 2 replicated the results of Experiment 1 and varied the proportion of problems violating parity. Larger parity effects were obtained when more of the false problems violated parity. Moreover, all three effects combined to show the greatest parity effects in conditions with a high proportion of parity violations in problems containing two even operands that were solved during the second half of the experiment. Experiment 3 generalized the findings to the case of five rule (i.e., checking whether a false product ends in 5 or 0), another procedure for solving and verifying multiplication problems quickly. These results (1) delineate further constraints for inclusion in models of arithmetic processing when thinking about how people select among verification strategies, (2) show combined effects of variables that traditionally have been shown to have separate effects on people's strategy selection, and (3) are consistent with a view of strategy selection that suggests a bias either in the allocation of cognitive resources in the execution of strategies or in the order of execution of these strategies; they argue against a simple, unbiased competition among strategies.