Language Learning and Development · 2009 · 49 citations · 32 references
Language StudiesComputer ScienceLinguisticsLanguage AcquisitionNatural Language ProcessingPsycholinguisticsDomain ModelingLanguage LearningLanguage ModelingLearning TheoriesSecond Language AcquisitionCognitive Neuroscience ModelingDomain GeneralizationLearner Modeling From Text
Abstract We identify three components of any learning theory: the representations, the learner's data intake, and the learning algorithm. With these in mind, we model the acquisition of the English anaphoric pronoun one in order to identify necessary constraints for successful acquisition, and the nature of those constraints. Whereas previous modeling efforts have succeeded by using a domain-general learning algorithm that implicitly restricts the data intake to be a subset of the input, we show that the same kind of domain-general learning algorithm fails when it does not restrict the data intake. We argue that the necessary data intake restrictions are domain-specific in nature. Thus, while a domain-general algorithm can be quite powerful, a successful learner must also rely on domain-specific learning mechanisms when learning anaphoric one. ACKNOWLEDGMENTS This paper has benefited from discussion with Amy Weinberg, Norbert Hornstein, Colin Phillips, Sandy Waxman, Bill Idsardi, Terry Regier, Charles Yang, Gerry Altmann, Susan Goldin-Meadow, the Psycholinguistics-Acquisition group at the University of Maryland, the Cognitive Neuroscience of Language lab at the University of Maryland, the Center for Language Sciences at the University of Rochester, and six anonymous reviewers. Special thanks to LouAnn Gerken whose comments improved the explication of these ideas immeasurably. Needless to say, any remaining errors are our own. This work was supported by an NSF graduate fellowship to LP, NSF grant BCS-0843896 to LP, NSF grant BCS-0418309 to JL, and NIH grant R03-DC006829 to JL. Notes 1Many thanks to an anonymous reviewer for pointing this out. 2It is important to note that this claim is based on realistic estimates of the variables involved. If we alter n and m such that the ratio is unrealistically high (e.g., above 0.85), then this model can in fact converge on the correct interpretation (e.g., for a ratio of 0.85, just over 50% of the time). However, we feel that it is reasonable to focus on realistic variables estimates as we are attempting to explain realistic child learning behavior. 3The sudden jumps in the trajectory result from the learner actually encountering an unambiguous data point in the input. The rest of the time, the learner does not learn anything from the input. 4Thanks to an anonymous reviewer for highlighting the relevance of local versus global uncertainty. 5This is where the Size Principle comes into play. The subset is the N0 category, and so it has a higher likelihood of having generated the observed data. The ratio of subset to superset determines how much the subset is favored.
32
A Mathematical Theory of Communication
Claude E. Shannon · Bell System Technical Journal · 1948 · 80.7K citations