Abstract not found

How do infants learn the sound patterns of their native language? By the end of the 1st year, infants have acquired detailed aspects of the phonology and phonotactics of their input language. However, the structure of the learning mechanisms underlying this process is largely unknown. In this study, 9-month-old infants were given the opportunity to induce specific phonological patterns in 3 experiments in which syllable structure, consonant voicing position, and segmental position were manipulated. Infants were then familiarized with fluent speech containing words that either fit or violated these patterns. Subsequent testing revealed that infants rapidly extracted new phonological regularities and that this process was constrained such that some regularities were easier to acquire than others. Months before infants speak their first words, they have acquired extensive and detailed knowledge about the sound patterns of their native language. Indeed, sounds are the infant’s entrance point into spoken language acquisition, beginning with the rhythmic patterns of the infant’s language—knowledge acquired before birth (e.g., Mehler et al., 1988)—and extending during the 1st year to include language-typical phonological patterns such as lexical

It is widely assumed that human learning and the structure of human languages are intimately related. This relationship is frequently suggested to be rooted in a language-specific biological endowment, which encodes universal, but arbitrary, principles of language structure (a universal grammar or UG). How might such a UG have evolved? We argue that UG could not have arisen either by biological adaptation or non-adaptationist genetic processes. The resulting puzzle concerning the origin of UG we call the logical problem of language evolution. Because the processes of language change are much more rapid than processes of genetic change, language constitutes a “moving target ” both over time and across different human populations, and hence cannot provide a stable environment to which UG genes could have adapted. We conclude that a biologically determined UG is not evolutionarily viable. Instead, the original motivation for UG—the mesh between learners and languages—arises because language has been shaped to fit the human brain, rather than vice versa. Following Darwin, we view language itself as a complex and interdependent “organism, ” which evolves under selectional pressures from human learning and processing mechanisms. That is, languages are themselves undergoing severe selectional pressure from each generation of language users and learners. This suggests that apparently arbitrary aspects of linguistic structure may result from general learning and processing biases, independent of language. We illustrate how this framework can integrate evidence from different literatures and methodologies to explain core linguistic phenomena, including binding constraints, word order universals, and diachronic language change. 1.

The time course of spoken word recognition depends largely on the frequencies of a word and its competitors, or neighbors (similar-sounding words). However, variability in natural lexicons makes systematic analysis of frequency and neighbor similarity difficult. Artificial lexicons were used to achieve precise control over word frequency and phonological similarity. Eye tracking provided time course measures of lexical activation and competition (during spoken instructions to perform visually guided tasks) both during and after word learning, as a function of word frequency, neighbor type, and neighbor frequency. Apparent shifts from holistic to incremental competitor effects were observed in adults and neural network simulations, suggesting such shifts reflect general properties of learning rather than changes in the nature of lexical representations. Current models of spoken word recognition share a set of core assumptions that correspond to what Marslen-Wilson (1993) called the macrostructure of spoken word recognition: As speech is heard, multiple lexical candidates are activated and compete for recognition with strengths proportional to their similarity with the input and their prior probabilities (frequencies of occurrence).

We compare our model of unsupervised learning of linguistic structures, ADIOS [1, 2, 3], to some recent work in computational linguistics and in grammar theory. Our approach resembles the Construction Grammar in its general philosophy (e.g., in its reliance on structural generalizations rather than on syntax projected by the lexicon, as in the current generative theories) , and the Tree Adjoining Grammar in its computational characteristics (e.g., in its apparent affinity with Mildly Context Sensitive Languages). The representations learned by our algorithm are truly emergent from the (unannotated) corpus data, whereas those found in published works on cognitive and construction grammars and on TAGs are hand-tailored. Thus, our results complement and extend both the computational and the more linguistically oriented research into language acquisition.

A crucial component of language acquisition involves organizing words into grammatical categories and discovering relations between them. Many studies have argued that phonological or semantic cues or multiple correlated cues are required for learning. Here we examine how distributional variables will shift learners from forming a category of lexical items to maintaining lexical specificity. In a series of artificial language learning experiments, we vary a number of distributional variables to category structure and test how adult learners use this information to inform their hypotheses about categorization. Our results show that learners are sensitive to the contexts in which each word occurs, the overlap in contexts across words, the non-overlap of contexts (or systematic gaps), and the size of the data set. These variables taken together determine whether learners fully generalize or preserve lexical specificity.

l), and, eventually, the neurobiological, reality of the theoretical constructs. Many examples of this process can be found in the study of human vision, where, as in language, direct observation of the underlying mechanisms is difficult; for instance, the concept of multiple parallel spatial frequency channels, introduced in the late 1960s, was completely vindicated by purely behavioral means over the following decade; see, e.g., [2]. In linguistics, the nature of the requisite evidence is well described by Townsend and Bever: "What do we test today if we want to explore the behavioral implications of syntax? . . . the psychological basis for the two primary and ever-present operations, merge and move." [3], p.82. Unfortunately, to our knowledge, no experimental evidence has been offered to date that suggests that Merge and Move are real (in the same 1 sense as the spatial frequency channels in human vision are). Generative linguists typically respond to calls for evidence for the r

Abstract not found

In the real world causal variables do not come pre-identified or occur in isolation, but instead are imbedded within a continuous temporal stream of events. A challenge faced by both human learners and machine learning algorithms is identifying subsequences that correspond to the appropriate variables for causal inference. A specific instance of this problem is action segmentation: dividing a sequence of observed behavior into meaningful actions, and determining which of those actions lead to effects in the world. Here we present two experiments investigating human action segmentation and causal inference, as well as a Bayesian analysis of how statistical and causal cues to segmentation should optimally be combined. We find that both adults and our model are sensitive to statistical regularities and causal structure in continuous action, and are able to combine these sources of information in order to correctly infer both causal relationships and segmentation boundaries.

The poverty of stimulus argument is one of the most controversial arguments in the study of language acquisition. Here we follow previous approaches challenging the assumption of impoverished primary linguistic data, focusing on the specific problem of auxiliary fronting in polar interrogatives. We develop a series of child-directed corpus analyses showing that there is indirect statistical information useful for correct auxiliary fronting in polar interrogatives, and that such information is sufficient for producing grammatical generalizations even in the absence of direct evidence. We further show that there are simple learning devices, such as neural networks, capable of exploiting such statistical cues, producing a bias to correct aux-questions when compared to their ungrammatical counterparts. The results suggest that the basic assumptions of the poverty of stimulus argument need to be reappraised.