Mutual Exclusivity in Cross-Situational Statistical Learning

摘要: Mutual Exclusivity in Cross-Situational Statistical Learning Daniel Yurovsky (dyurovsk@indiana.edu) Department of Psychological and Brain Science, Cognitive Science Program 1101 East 10 th Street Bloomington, IN 47408 USA Chen Yu (chenyu@indiana.edu) Street, ultimately discover correct word-referent associations. With respect to the above example, a learner may hear words toma bosa ambiguous context seeing both loofah gyroscope without any information as which word refers object. Although mappings cannot be determined from this single situation, could nonetheless solve problem by keeping track co- occurrences non-occurrences across situations. Using if viewed second scene with spatula while hearing diti toma, combine co-occurrence probabilities these two situations, correctly infer that maps gyroscope, “bosa” loofah, “diti” spatula. Recent empirical evidence has shown successful cross-trial learning situations adults (Yu & Smith, 2007) children (Smith Yu, 2008). Smith (2007) exposed series 27 trials, each containing 4 unknown possible referents. Their participants learned 9.5 out 18 less than 6 minutes. Thus, humans can use statistical multiple learn word-object storing, computing, continuously reducing set referents over time. The present study intends investigate ME constraint cross-situational paradigm. Can one-to-many pairings number trials? Since most previous experiments have focused on one-trial learning, language used one-to- one stimuli, it is unclear what expect. There are potential outcomes. One human learners rely mutual exclusivity much same way single-trial learning. As result, they still build one-to-one ignoring additional regularities training data, or fail at all due confusion. This would suggest an Alternatively, break demonstrating mapping test. result provide support for fundamental mechanism supporting general, particular. Abstract (ME) – preference object been powerful aid new words. We ask whether meanings through computation co-occurrences highly subject constraint. Our results show one-to-two when separated time interleaved. demonstrates robustness then participants’ ratings their knowledge after individual trials shed light underlying mechanism. applied points along within trial, test explain its residual effects found traditional literature. Keywords: exclusivity; learning; acquisition Introduction building plays critical role In Markman Wachtel’s (1988) classic experiment, child presented known (ball) (gyroscope) asked experimenter bring “toma”. Having never before heard “toma,” will select novel referent, but not ball, already name. original exlcusivity reliably demonstrated various studies. Subsequent research also proposed several explanations children’s behavior, such Principle Contrast (Clark, 1983), Novel-Name-Nameless-Category (Golinkoff, Hirsh-Pasek, Bailey, Wenger, 1992), Pragmatic Account (Diesendruck Markson, 2001). Further, provides reliable deal reference uncertainty (Quine, 1960) trial. Given referents, apply rule those names determine referent Recently, alternative solution learning: statistics 2007). A who unable trial accumulate individually Experiment 1 Following paradigm (2007), we