Young infants show unexplained asymmetries in the exclusivity of categories formed on the basis of visually presented stimuli. We describe a connectionist model that shows similar exclusivity asymmetries when categorizing the same stimuli presented to the infants. The asymmetries can be explained in terms of an associative learning mechanism, distributed internal representations, and the statistics of the feature distributions in the stimuli. We use the model to explore the robustness of this asymmetry. The model predicts that the asymmetry will persist when a category is acquired in the presence of mixed category exemplars. A study with 3- to 4-month-olds show that asymmetric exclusivity continues to persist in the presence of a mixed familiarization, thereby corroborating the model's predictions. We suggest that by interpreting asymmetric exclusivity effects as manifestations of interference in an associative memory system, the model can also be extended to account for interference e...

The term “aphasia ” refers to acute or chronic impairment of language, an acquired condition that is most often associated with damage to the left side of the brain, usually due to trauma or stroke. We have known about the link between left-hemisphere damage and language loss for more than a century (Goodglass, 1993). For almost as long, we have also known that the lesion/symptom correlations observed in adults do not appear to hold for very young children (Basser, 1962; Lenneberg, 1967). In fact, in the absence of other complications, infants with congenital damage to one side of the brain (left or right) usually go on to acquire language abilities that are well within the normal range (Eisele & Aram, 1995; Feldman, Holland, & Janosky, 1992; Vargha-Khadem, Isaacs, & Muter,

We present morphological analyses of verb inflections produced by 15 Spanish-speaking children (age range: 1 ;7 to 4;7) taken from longitudinal and cross-sectional samples of spontaneous speech and narratives. Our main observation is the existence of a dissociation between regular and irregular processes in the distribution of errors: regular suffixes and unmarked (non-alternating) stems are over-extended to irregulars in children's inflection errors, but not vice versa. We also found that overregularization errors at all ages are only a small minority of the children's irregular verbs, that the period of overregularization is preceded by a stage without errors, and that the onset of overregularizations is connected to the emergence of obligatory finiteness markings. These findings are explained in terms of the dual-mechanism model of inflection.

& The ERP repetition priming paradigm has been shown to be sensitive to the processing differences between regular and irregular verb forms in English and German. The purpose of the present study is to extend this research to a language with a different inflectional system, Spanish. The design (delayed visual repetition priming) was adopted from our previous study on English, and the specific linguistic phenomena we examined are priming relations between different kinds of stem (or root) forms. There were two experimental conditions: In the first condition, the prime and the target shared the same stem form, e.g., ‘‘ando–andar’ ’ [I walk–to walk], whereas in the second condition, the prime contained a marked (alternated) stem, e.g., ‘‘duermo– dormir’ ’ [I sleep–to sleep]. A reduced N400 was found for unmarked (nonalternated) stems in the primed condition, whereas marked stems showed no such effect. Moreover, control conditions demonstrated that the surface form properties (i.e., the different degree of phonetic and orthographic overlap between primes and targets) do not explain the observed priming difference. The ERP priming effect for verb forms with unmarked stems in Spanish is parallel to that found for regularly inflected verb forms in English and German. We argue that effective priming is possible because prime target pairs such as ‘‘ando–andar’ ’ access the same lexical entry for their stems. By contrast, verb forms with alternated stems (e.g., ‘‘duermo’’) constitute separate lexical entries, and are therefore less powerful primes for their corresponding base forms. &

Children face an enormously difficult task in learning their native language. It is widely believed that they do not receive or make little use of negative evidence (Marcus, 1993), and yet it has been proven that many classes of languages less powerful than natural languages cannot be learned in the absence of negative evidence (Gold, 1964). In this paper we present an approach to learning good approximations to members of one such class of languages, the regular languages, based on positive evidence alone. 1. Introduction The ability to communicate through spoken language is widely regarded as the hallmark of human intelligence. Children acquire their native tongue with remarkable ease, mastering the vast majority of that language before they enter school. However, the facility with which children acquire language belies the complexity of the task. For example, children clearly receive positive evidence (examples of sentences in the language), but it is widely believed that children ...

The authors asked whether there is evidence to support the existence of the vocabulary spurt, an increase in the rate of word learning that is thought to occur during the 2nd year of life. Using longitudinal data from 38 children, they modeled the rate of word learning with two functions, one with an inflection point (logistic), which would indicate a spurt, and one without an inflection point (quadratic). Comparing the fits of these two functions using likelihood ratios, they found that just 5 children had a better logistic fit, which indicated that these children underwent a spurt. The implications for theories of cognitive and language development are considered. Typically developing children utter their first words between 8 and 14 months of age. At this time, they add words to their repertoire at a slow rate. As they get older and their vocabulary increases, their rate of learning new words also increases—it has to if they are to reach an average vocabulary level of 300 words by 24 months (Fenson, Dale, Reznick, Bates, & Thal, 1994) and 60,000 words by 18 years (Aitchinson, 1994). In addition, it is widely held that children’s rate of vocabulary acquisition does not simply increase but undergoes a discrete transition at approximately 50 words. At this time, children putatively switch from an initial stage of slow vocabulary growth to a subsequent stage of faster growth. This transition has been referred to as the vocabulary spurt, the vocabulary burst, or the naming explosion (L.

This paper is predominantly theoretic, and addresses the question of naturalness in dialogue systems. It sketches requirements for ambitious 21-century dialogue systems, and argues that the biggest problem in making dialogue systems flexible, robust and natural is the knowledge acquisition problem. The more complex tasks the dialogue system is expected to cope with, the more complex knowledge is needed, and it is impossible simply to itemize the facts necessary for such complexity. Instead, the system must be seen as a learning system, equipped with capabilities to learn through interaction. The argumentation goes as follows: (1) naturalness of a dialogue system is not only dependent on the available interaction patterns but on the knowledge that the system is able to exploit, (2) the closed-world approach to build such knowledge into the systems is limited when the task becomes complex, due to the increased requirements for managing interactions between the system and its environment, (3) to equip dialogue systems with the appropriate knowledge, it is necessary to include world knowledge and a dynamic update procedure in the system architecture, (4) the view of the dialogue system as a service provider should be extended to include systems as interactive partners, and thus dialogue systems should be investigated as learning systems rather than static models.

Language is a unique hallmark of the human species. Although many species can communicate in limited ways about things that are physically present, only humans can construct a full narrative characterization of events occurring outside of the here and now. Humans are also unique in their ability to fashion tools such as arrow points, axes, traps, and clothing. By using language to control the social coordination of tool making, humans have produced a material society that has achieved domination over all the creatures of our world and often over Nature herself. The religions of the world have interpreted our unique linguistic endowment as a Special Gift bestowed directly by the Creator. Scientists have also been influenced by this view of language, often attributing the emergence of this remarkable species-specific ability to some single, pivotal salutatory event in human evolution. I will refer to this sudden evolutionary jump into true human language as linguistic saltation. Linguistic saltationists (Bickerton, 1990; Chomsky, 1975; Hauser, Chomsky, & Fitch, 2002) tend to see language as a very recent evolutionary event. They can note that the divergence of our hominid ancestors from the Great Apes occurred over 6 million years

We introduce symbolic dynamics to cognitive scientists with the aim of furthering constructive debate on representation. Symbolic dynamics is a mathematical framework in which both continuous and discrete states of a system can be considered jointly. We discuss a number of theoretical implications this framework has for cognitive science, and offer some consideration of the way in which it might be employed for comparing or conciliating discrete and continuous representational theories. Symbolic dynamics may thus serve as a common, level playing field for debate in theories of cognitive representation.

In the past, recurrent networks have been used mainly in neurocognitive or psycholinguistically oriented approaches of language processing. Here we examine recurrent neural networks for their potential in a dicult spoken language classi cation task. This paper describes an approach to learning classi cation of recorded operator assistance telephone utterances. We explore simple recurrent networks using a large, unique telecommunication corpus of spontaneous spoken language. Performance of the network indicates that a semantic SRN network is quite useful for learning classi cation of spontaneous spoken language in a robust manner, which may lead to their use in helpdesk call routing.