Over the course of the first year of life, infants develop from being generalized listeners, capable of discriminating both native and non-native speech contrasts, into specialized listeners whose discrimination patterns closely reflect the phonetic system of the native language(s). Recent work by Maye, Werker and Gerken (2002) has proposed a statistical account for this phenomenon, showing that infants may lose the ability to discriminate some foreign language contrasts on the basis of their sensitivity to the statistical distribution of sounds in the input language. In this paper we examine the process of enhancement in infant speech perception, whereby initially difficult phonetic contrasts become better discriminated when they define two categories that serve a functional role in the native language. In particular, we demonstrate that exposure to a bimodal statistical distribution in 8-month-old infants ’ phonetic input can lead to increased discrimination of difficult contrasts. In addition, this exposure also facilitates discrimination of an unfamiliar contrast sharing the same phonetic feature as the contrast presented during familiarization, suggesting that infants extract acoustic/phonetic information that is invariant across an abstract featural representation.

A majority of English words are stressed on their first syllable. Infants use stress as a cue to word segmentation, but it is unclear how infants discover the correlation between stress and word boundaries. We exposed English-learning infants to a list of words stressed on their second syllable to discover whether infants can learn a new relation between stress and word boundaries. English-learning infants treat stressed syllables as word onsets, which is incorrect in words where stressed syllables occur second (iambic words). A brief exposure allowed infants to subsequently segment iambic words correctly, whether the exposure consisted of 100 % or 80 % iambic words. We also trained 7-month-olds—who typically rely on transitional probabilities—to use stress as a cue to word segmentation. The results suggest that infants are sensitive to the distribution of stress across word position and that altering this distribution affects their segmentation strategies. Languages are complex systems characterized by multiple interdependent levels of organization, such as sound, meaning, and syntactic structure. And yet, infants, who are less cognitively advanced than adults, must learn them. At first glance, it seems surprising that languages are so complex given that cognitively immature infants must learn them. However, it may be the case that the complexity of language is not an obstacle for infants to overcome. The complexity of language may

Infants'phonological acquisition during the first 18 months of life has been studied within experimental psychology for some 30 years. Current research themes include statistical learning mechanisms, early lexical development, and models of phonetic category perception. So far, linguistic theories have hardly been taken into account. These theories are based upon the assumption that there is a common core of innate phonologicalknowledge across speakers of all human languages, and they contain detailed proposals concerning phonological representationsand the derivations by which abstract underlying forms are mapped onto concrete surface forms. It remains to be investigated experimentally if there is innate phonological knowledge and how the language-specific phonological grammar is acquired.

Phonological constraints can, in principle, be classified according to whether they are natural (founded in principles of Universal Grammar (UG)) or unnatural (arbitrary, learned inductively from the language data). Recent work has used this distinction as the basis for arguments about the role of UG in learning. Some languages have phonological patterns that arguably reflect unnatural constraints. With experimental testing, one can assess whether such patterns are actually learned by native speakers. Becker, Ketrez, and Nevins (2007), testing speakers of Turkish, suggest that they do indeed go unlearned. They interpret this result with a strong UG position: humans are unable to learn data patterns not backed by UG principles. This article pursues the same research line, locating similarly unnatural data patterns in the vowel harmony system of Hungarian, such as the tendency (among certain stem types) for a final bilabial stop to favor front harmony. Our own test leads to the opposite conclusion to Becker et al.: Hungarians evidently do learn the unnatural patterns. To conclude we consider a bias account—that speakers are able to learn unnatural environments, but devalue them relative to natural ones. We outline a method for testing the strength of constraints as learned by speakers against the strength of the corresponding patterns in the lexicon, and show that it offers tentative support for the hypothesis that unnatural constraints are disfavored by language learners.

Learning classes of sounds in infancy

During language acquisition, children learn to segment speech into phonemes, syllables, morphemes, and words. We examine word segmentation specifically, and explore the possibility that children might have generalpurpose chunking mechanisms to perform word segmentation. The Voting Experts (VE) and Bootstrapped Voting Experts (BVE) algorithms serve as computational models of this chunking ability. VE finds chunks by searching for a particular information-theoretic signature: low internal entropy and high boundary entropy. BVE adds to VE the ability to incorporate information about word boundaries previously found by the algorithm into future segmentations. We evaluate the general chunking model on phonemicallyencoded corpora of child-directed speech, and show that it is consistent with empirical results in the developmental literature. We argue that it offers a parsimonious alternative to specialpurpose linguistic models. 1

Two factors have been proposed as the main determinants of phonological typology: channel bias, phonetically systematic errors in transmission, and analytic bias, cognitive predispositions making learners more receptive to some patterns than others. Much of typology can be explained equally well by either factor, making them hard to distinguish empirically. This study presents evidence that analytic bias is strong enough to create typological asymmetries in a case where channel bias is controlled. I show that (i) phonological dependencies between the height of two vowels are typologically more common than dependencies between vowel height and consonant voicing, (ii) the phonetic precursors of the heightheight and height-voice patterns are equally robust and (iii) in two experiments, English speakers learned a height-height pattern and a voice-voice pattern better than a height-voice pattern. I conclude that both factors contribute to typology, and discuss hypotheses about their interaction. 1

We present a computational model of grammar learning that combines domain-general learning mechanisms with rich representations of linguistic knowledge, world knowledge and situational and discourse context. These representations support processes of language understanding and inference (including both constructional analysis and reference resolution) that help the learner make sense of utterances in context. The learner then draws on generalization and statistical induction techniques to form new constructions that better capture correlations between linguistically identified and contextually inferred information. Our work is part of the larger Neural Theory of Language (NTL) project, whose goal is to build models of cognition and language that satisfy convergent constraints from biology, psychology, linguistics and computation (Chang,