Despite the lack of invariance problem (the many-to-many mapping between acoustics and percepts), human listeners experience phonetic constancy and typically perceive what a speaker intends. Most models of human speech recognition (HSR) have side-stepped this problem, working with abstract, idealized inputs and deferring the challenge of working with real speech. In contrast, carefully engineered deep learning networks allow robust, real-world automatic speech recognition (ASR). However, the complexities of deep learning architectures and training regimens make it difficult to use them to provide direct insights into mechanisms that may support HSR. In this brief article, we report preliminary results from a two-layer network that borrows one element from ASR, long short-term memory nodes, which provide dynamic memory for a range of temporal spans. This allows the model to learn to map real speech from multiple talkers to semantic targets with high accuracy, with human-like timecourse of lexical access and phonological competition. Internal representations emerge that resemble phonetically organized responses in human superior temporal gyrus, suggesting that the model develops a distributed phonological code despite no explicit training on phonetic or phonemic targets. The ability to work with real speech is a major advance for cognitive models of HSR. 相似文献
The current experiments examined the creation of nonbelieved true and false memories after imagining bizarre and familiar actions using the imagination inflation procedure (Goff & Roediger, 1998). In both experiments, participants took part in three sessions. In Session 1, participants had to perform or imagine simple familiar actions (e.g., “stir the water with the spoon”) and bizarre actions (e.g., “balance the spoon on your nose”). A day later, participants needed to imagine simple actions of which some were new actions, and some were old actions that appeared in the first session. After a week, the participants completed a recognition task. For those actions that were correctly or incorrectly remembered as having been performed, the participant was challenged that the action was not performed in order to evoke nonbelieved true and false memories. In general, we found that the imagination inflation procedure can successfully induce participants to produce nonbelieved memories. In Study 1, we successfully induced nonbelieved memories for bizarre actions, although in general nonbelieved memory rates were low. In Study 2, more participants formed nonbelieved memories for bizarre actions than for familiar actions. Also, we found that especially belief was more susceptible to revision when memories were challenged than recollection. In two experiments, we showed that nonbelieved memories can successfully be induced for both familiar and bizarre actions. 相似文献
Over the past two decades, researchers consistently demonstrated the importance of science teaching approaches and student self-efficacy in influencing their science achievement. These findings have become the foundation of science education reform. However, empirical supports of these relationships are limited to direct relationships and small-scale studies. Therefore, little is known about the mechanism of how teaching approaches and student self-efficacy affect student achievement. In order to fill these gaps, this study used a multilevel structural equation modeling approach to analyze the direct and indirect relationships between teaching approaches, student self-efficacy, and science achievement by using the data of US eighth grade students in the 2011 TIMSS assessment. The results indicated that none of the teaching approaches identified in this study were directly associated with student science achievement, but significant mediation effect was found between generic teaching and student science achievement through student self-efficacy. Implications of these results for US educational system and reform were discussed.