EARSHOT: A Minimal Neural Network Model of Incremental Human Speech Recognition

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.     

The relative strengths of indirect and direct word learning

Indirect word learning lacks many of the overt social-pragmatic cues to reference available in direct word learning, yet the two result in equally robust mappings when comprehension is assessed immediately after learning. The 3 studies reported here investigated how 3-year-olds (N=96) respond to more challenging tests of the relative strengths of indirect and direct word learning. In Study 1, children's comprehension of indirectly and directly learned proper and common names was tested after a 2-day delay. Both types of learning resulted in proper name mappings that picked out an individual and in common name mappings that could be extended to another category member. In Studies 2 and 3, children's comprehension was tested after they had been provided with additional, and sometimes inconsistent, information about the scope of previously learned words. There was a hint of a difference between indirect and direct word learning. but results overall suggested that the two were equivalent.     

Differences in neuronal activity explain differences in memory forming abilities of different populations of Lymnaea stagnalis

The ability to learn and form long-term memory (LTM) can enhance an animal's fitness, for example, by allowing them to remember predators, food sources or conspecific interactions. Here we use the pond snail, Lymnaea stagnalis, to assess whether variability between natural populations (i.e., strains) in memory forming capabilities correlates with electrophysiological properties at the level of a single neuron, RPeD1. RPeD1 is a necessary site of LTM formation of aerial respiratory behaviour following operant conditioning. We used strains from two small, separate permanent ponds (TC1 and TC2). A comparison of the two populations showed that the TC1 strain had enhanced memory forming capabilities. Further, the behavioural phenotype of enhanced memory strain was explained, in part, by differences in the electrophysiology of RPeD1. Compared to RPeD1 from the naive TC2 strain, RPeD1 from the TC1 strain has both a decreased resistance and decreased excitability. Moreover, 24h after a single 0.5h training session, those membrane properties, as well as the firing and bursting rate, decrease further in the TC1 strain but not in the TC2 strain. The initial differences in RPeD1 properties in the TC1 strain coupled with their ability to further change these properties with a single training session suggests that RPeD1 neurons from the TC1 strain are "primed" to rapidly form memory.     

Sikh American millennials at work: Institution building,activism, and a renaissance of cultural expression

ABSTRACT

This paper explores the contributions of Sikh American millennials in institution building, activism, and cultural expression in the post 9/11 period. The discussion is based on interviews and fieldwork conducted by the author. I describe the motivations and concerns of Sikh American millennial activists. I also discuss the digital fluencies they apply in their fight against discrimination and violence. They value innovative leadership, civil rights, gender equality, social justice, and youth programs. I conclude that millennial institution building is shaping the community by expanding civil society engagement, promoting youth leadership, and highlighting social problems within the community like sexism and classism.     

Privacy of Moral Perspective

Journal of Indian Council of Philosophical Research - This paper attempts to delve into Wittgenstein’s unique notion of solipsism and its centrality in his proposal of transcendental ethics....     

Learning words: children disregard some pragmatic information that conflicts with mutual exclusivity

Children tend to infer that when a speaker uses a new label, the label refers to an unlabeled object rather than one they already know the label for. Does this inference reflect a default assumption that words are mutually exclusive? Or does it instead reflect the result of a pragmatic reasoning process about what the speaker intended? In two studies, we distinguish between these possibilities. Preschoolers watched as a speaker pointed toward (Study 1) or looked at (Study 2) a familiar object while requesting the referent for a new word (e.g. ‘Can you give me the blicket?’). In both studies, despite the speaker's unambiguous behavioral cue indicating an intent to refer to a familiar object, children inferred that the novel label referred to an unfamiliar object. These results suggest that children expect words to be mutually exclusive even when a speaker provides some kinds of pragmatic evidence to the contrary.     

The power of print: Children’s trust in unexpected printed suggestions

How do children evaluate the veracity of printed text? We examined children’s handling of unexpected suggestions conveyed via print versus orally. In Experiment 1 (N = 131), 3- to 6-year-olds witnessed a speaker either read aloud an unexpected but not completely implausible printed label (e.g., fish for a bird-like animal with some fish features) or speak the label without accompanying text. Pre-readers accepted labels in both conditions. Early readers often rejected spoken labels yet accepted them in the print condition, and in Experiment 2 (N = 55) 3- to 6-year-olds continued to apply them even after the print was obscured. Early readers accept printed testimony that they reject if only spoken, and the influence of text endures even when it is no longer visible.     

Turning Believers into Skeptics: 3-Year-Olds' Sensitivity to Cues to Speaker Credibility

Under most circumstances, children (and adults) can safely assume that the testimony they hear is true. In two studies, we investigated whether 3-year-olds (N = 100) would continue to hold this assumption even if the person who provided the testimony behaved in an uncertain, ignorant, and/or distracted manner. In Study 1, children were less likely to trust that, for example, a key-like object was a spoon if the speaker indicated uncertainty about her testimony (e.g., “I think this is a spoon”) than if she simply labeled the object ostensively (e.g., “This is a spoon”). In Study 2, 3-year-olds were also more skeptical about a speaker's testimony when she had earlier made an obvious naming error and seemed distracted, but not when she either made an error or seemed distracted. These results indicate that 3-year-olds can respond differently to the same testimony, depending on the speaker's behavior.     

Looks Aren't Everything: 24-Month-Olds' Willingness to Accept Unexpected Labels

A label can efficiently convey nonobvious information about category membership, but this information can sometimes conflict with one's own expectations. Two studies explored whether 24-month-olds (N = 56) would be willing to accept a category label indicating that an animal (Study 1) or artifact (Study 2) that looked like a member of one familiar category was actually a member of a different familiar category. Results showed that children were receptive to these unexpected labels and used them as the basis for inference. These findings indicate that linguistic information can lead even toddlers to “disbelieve their eyes.”     

The Plausibility and Significance of Underdetermination Arguments

Journal of Indian Council of Philosophical Research - Underdetermination of theory choice claims that empirical evidence fails to provide sufficient grounds for choosing a theory over its rivals....