The secret is in the sound: from unsegmented speech to lexical categories

When learning language, young children are faced with many seemingly formidable challenges, including discovering words embedded in a continuous stream of sounds and determining what role these words play in syntactic constructions. We suggest that knowledge of phoneme distributions may play a crucial part in helping children segment words and determine their lexical category, and we propose an integrated model of how children might go from unsegmented speech to lexical categories. We corroborated this theoretical model using a two‐stage computational analysis of a large corpus of English child‐directed speech. First, we used transition probabilities between phonemes to find words in unsegmented speech. Second, we used distributional information about word edges – the beginning and ending phonemes of words – to predict whether the segmented words from the first stage were nouns, verbs, or something else. The results indicate that discovering lexical units and their associated syntactic category in child‐directed speech is possible by attending to the statistics of single phoneme transitions and word‐initial and final phonemes. Thus, we suggest that a core computational principle in language acquisition is that the same source of information is used to learn about different aspects of linguistic structure.     

A critique of Mark D. Allen's "the preservation of verb subcategory knowledge in a spoken language comprehension deficit"

Allen [Allen, M. (2005). The preservation of verb subcategory knowledge in a spoken language comprehension deficit. Brain and Language, 95, 255-264.] reports a single patient, WBN, who, during spoken language comprehension, is still able to access some of the syntactic properties of verbs despite being unable to access some of their semantic properties. Allen claims that these findings challenge linguistic theories which assume that much of the syntactic behavior of verbs can be predicted from their meanings. I argue, however, that this conclusion is not supported by the data for two reasons: first, Allen focuses on aspects of verb syntax that are not claimed to be influenced by verb semantics; and second, he ignores aspects of verb syntax that are claimed to be influenced by verb semantics.     

On the semantic content of subcategorization frames

This paper investigates relations between the meanings of verbs and the syntactic structures in which they appear. This investigation is motivated by the enigmas as to how children discover verb meanings. Well-known problems with unconstrained induction of word meanings from observation of world circumstances suggest that additional constraints or sources of information are required. If there exist strong and reliable parallels between the structural and semantic properties of verbs, then an additional source of information about verb meanings is reliably present in each verb's linguistic context. Five experiments are presented which investigate the following hypothesis regarding the scope of these relations: The closer any two verbs in their semantic structure, the greater the overlap should be in their licensed syntactic structures. To investigate this hypothesis, data of two kinds were collected from different groups of subjects: (a) One group of subjects was asked to judge the semantic relatedness of verbs by selecting the semantic outlier in triads presented to them. (b) A second group of subjects was asked to judge the grammaticality of these same verbs in a large range of syntactic environments. These two types of data were then compared to assess the degree of correspondence in the two partitionings (syntactic and semantic) of the verb set. The findings, overall, support the view that the syntax of verbs is a quite regular, although complex, projection from their semantics. In conclusion, we discuss the kinds of features that are formally marked in syntactic structure and relate these to the problem of verb-vocabulary acquisition in young children.     

How toddlers begin to learn verbs

Any theory of language must account for how children learn verbs, the gateway to grammar. Yet verbs can be difficult to learn. Building on Gentner's 'natural partitions hypothesis' we suggest that, to learn a verb, infants must conceptualize components of events and map verbs in the ambient language onto those components. Although toddlers detect and categorize at least some of the conceptual underpinnings of verb categories, the mapping of verbs onto these representations is not transparent. Mapping is a difficult problem in its own right. The Emergentist Coalition Model that has been used to explain noun learning also begins to explain how children move from perceptual to social and then to linguistic information to link verbs to actions and events.     

Age of acquisition: its neural and computational mechanisms

The acquisition of new skills over a life span is a remarkable human ability. This ability, however, is constrained by age of acquisition (AoA); that is, the age at which learning occurs significantly affects the outcome. This is most clearly reflected in domains such as language, music, and athletics. This article provides a perspective on the neural and computational mechanisms underlying AoA in language acquisition. The authors show how AoA modulates both monolingual lexical processing and bilingual language acquisition. They consider the conditions under which syntactic processing and semantic processing may be differentially sensitive to AoA effects in second-language acquisition. The authors conclude that AoA effects are pervasive and that the neural and computational mechanisms underlying learning and sensorimotor integration provide a general account of these effects.     

3-Year-olds’ comprehension,production, and generalization of Sesotho passives

Researchers have long been puzzled by the challenge English passive constructions present for language learners, with adult-like comprehension and production emerging only around the age of 5. It has therefore been of significant interest that researchers of other languages, including the Bantu language Sesotho, have reported acquisition of the passive by the age of 3 (Demuth, 1989). Such reports have typically been based on spontaneous production data, calling for further investigation. This study carried out a series of experiments with Sesotho-speaking 3-year-olds, testing their ability to comprehend the passive, produce the passive, and generalize novel verbs to passive frames. The results showed that passive comprehension was good, with no effect of actional/non-actional verb type. Elicited production of the passive was also good, with no difference between adversive and non-adversive verbs. Finally, all participants made both active and passive generalizations to novel verbs. These findings provide strong evidence that Sesotho-speaking 3-year-olds have robust, abstract knowledge of passive syntax. The paper concludes with a discussion of the factors that contribute to the early learning of the Sesotho passive, why acquisition of the passive may be delayed in English, and the implications for understanding grammatical development more generally.     

A Computational Model for the Item‐Based Induction of Construction Networks

According to usage‐based approaches to language acquisition, linguistic knowledge is represented in the form of constructions—form‐meaning pairings—at multiple levels of abstraction and complexity. The emergence of syntactic knowledge is assumed to be a result of the gradual abstraction of lexically specific and item‐based linguistic knowledge. In this article, we explore how the gradual emergence of a network consisting of constructions at varying degrees of complexity can be modeled computationally. Linguistic knowledge is learned by observing natural language utterances in an ambiguous context. To determine meanings of constructions starting from ambiguous contexts, we rely on the principle of cross‐situational learning. While this mechanism has been implemented in several computational models, these models typically focus on learning mappings between words and referents. In contrast, in our model, we show how cross‐situational learning can be applied consistently to learn correspondences between form and meaning beyond such simple correspondences.     

Motion-Verb Generalizations in English and Spanish: Influences of Language and Syntax

English and Spanish speakers differ in the ways they talk about motion events, but how have these different modes of expression become instantiated as differing generalizations—as syntactic rules, lexical patterns, or both? In two studies, we asked English- and Spanish-speaking adults to interpret novel motion verbs presented in three types of sentence frames. Overall, English speakers expected novel verbs to encode the manner of motion, whereas Spanish speakers expected the verbs to encode the path of motion. The sentence frames also significantly affected how the speakers interpreted the novel verbs. We conclude that speakers of different languages represent their different generalizations about the composition of motion verbs both lexically and syntactically, and discuss how these generalizations might be important for issues of language acquisition and linguistic relativity.     

The learnability of abstract syntactic principles

Children acquiring language infer the correct form of syntactic constructions for which they appear to have little or no direct evidence, avoiding simple but incorrect generalizations that would be consistent with the data they receive. These generalizations must be guided by some inductive bias – some abstract knowledge – that leads them to prefer the correct hypotheses even in the absence of directly supporting evidence. What form do these inductive constraints take? It is often argued or assumed that they reflect innately specified knowledge of language. A classic example of such an argument moves from the phenomenon of auxiliary fronting in English interrogatives to the conclusion that children must innately know that syntactic rules are defined over hierarchical phrase structures rather than linear sequences of words (e.g., [Chomsky, 1965] , [Chomsky, 1971] , [Chomsky, 1980] and [Crain and Nakayama, 1987] ). Here we use a Bayesian framework for grammar induction to address a version of this argument and show that, given typical child-directed speech and certain innate domain-general capacities, an ideal learner could recognize the hierarchical phrase structure of language without having this knowledge innately specified as part of the language faculty. We discuss the implications of this analysis for accounts of human language acquisition.     

Lexical or syntactic control of sentence formulation? Structural generalizations from idiom production

To compare abstract structural and lexicalist accounts of syntactic processes in sentence formulation, we examined the effectiveness of nonidiomatic and idiomatic phrasal verbs in inducing structural generalizations. Three experiments made use of a syntactic priming paradigm in which participants recalled sentences they had read in rapid serial visual presentation. Prime and target sentences contained phrasal verbs with particles directly following the verb (pull off a sweatshirt) or following the direct object (pull a sweatshirt off). Idiomatic primes used verbs whose figurative meaning cannot be straightforwardly derived from the literal meaning of the main verb (e.g., pull off a robbery) and are commonly treated as stored lexical units. Particle placement in sentences was primed by both nonidiomatic and idiomatic verbs. Experiment 1 showed that the syntax of idiomatic and nonidiomatic phrasal verbs is amenable to priming, and Experiments 2 and 3 compared the priming patterns created by idiomatic and nonidiomatic primes. Despite differences in idiomaticity and structural flexibility, both types of phrasal verbs induced structural generalizations and differed little in their ability to do so. The findings are interpreted in terms of the role of abstract structural processes in language production.     

Acquiring and processing verb argument structure: distributional learning in a miniature language

Adult knowledge of a language involves correctly balancing lexically-based and more language-general patterns. For example, verb argument structures may sometimes readily generalize to new verbs, yet with particular verbs may resist generalization. From the perspective of acquisition, this creates significant learnability problems, with some researchers claiming a crucial role for verb semantics in the determination of when generalization may and may not occur. Similarly, there has been debate regarding how verb-specific and more generalized constraints interact in sentence processing and on the role of semantics in this process. The current work explores these issues using artificial language learning. In three experiments using languages without semantic cues to verb distribution, we demonstrate that learners can acquire both verb-specific and verb-general patterns, based on distributional information in the linguistic input regarding each of the verbs as well as across the language as a whole. As with natural languages, these factors are shown to affect production, judgments and real-time processing. We demonstrate that learners apply a rational procedure in determining their usage of these different input statistics and conclude by suggesting that a Bayesian perspective on statistical learning may be an appropriate framework for capturing our findings.     

The role of frequency in the acquisition of English word order

Akhtar [Akhtar, N. (1999). Acquiring basic word order: Evidence for data-driven learning of syntactic structure. Journal of Child Language, 26, 339–356] taught children novel verbs in ungrammatical word orders. Her results suggested that the acquisition of canonical word order is a gradual, data-driven process. The current study adapted this methodology, using English verbs of different frequencies, to test whether children's use of word order as a grammatical marker depends upon the frequency of the lexical items being ordered. Ninety-six children in two age groups (2;9 and 3;9) heard either high frequency, medium frequency or low frequency verbs that were modeled in SOV order. Children aged 2;9 who heard low frequency verbs were significantly more likely to adopt the weird word order than those who heard higher frequency verbs. Children aged 3;9 preferred to use SVO order regardless of verb frequency. Furthermore, the younger children reverted to English word order using more arguments as verb frequency increased and used more pronouns than their older counterparts. This suggests that the ability to use English word order develops from lexically specific schemas formed around frequent, distributionally regular items (e.g. verbs, pronouns) into more abstract, productive schemas as experience of the language is accrued.     

The input to verb learning in Mandarin Chinese: a role for syntactic bootstrapping

The authors investigated the role of syntax in verb learning in Mandarin Chinese, which allows pervasive ellipsis of noun arguments. Two questions were investigated using the Beijing corpus on CHILDES: (a) Does the input to young children manifest syntactic-semantic correspondences as needed for acquiring verb meanings? (b) Are verbs presented in multiple frames? Over 6,000 child-directed utterances were parsed. Analyses revealed that transitive verbs, motion verbs, and internal/communication verbs were distinguished syntactically; moreover, the 60 target verbs were used in multiple sentence frames. These findings support a role for syntactic bootstrapping in Mandarin verb learning.     

成人言语输入对儿童早期单词获得影响的个案追踪

追踪观察一名婴儿(6~20个月), 分析其中看护者的言语输入特征及婴儿早期词汇获得的发展变化。主要对成人言语输入中动、名词比例、单词在句中的位置、具体环境等因素及婴儿早期动、名词理解和产生等方面进行探讨。结果显示, 成人言语输入中动词比例显著高于名词, 更多动词位于句首或句尾使得主语和宾语省略; 同时, 这种动词优势的输入特征促进儿童早期动词获得, 使得婴儿早期语言样本中动、名词理解相对比例与成人言语输入一致。这一结果表明, 婴儿词汇发展初期就已经利用言语或社会线索, 同时这种早期词汇组成中较高的动词比例进一步支持“名词优势”理论存在跨语言差异     

Argument structure and the child's contribution to language learning

One of the oldest questions in cognitive science asks whether children are able to learn language (or anything) because they are equipped with a very powerful general-purpose learning mechanism or because they are equipped with a domain-specific constrained language acquisition device. Recent advances in statistical approaches to language learning seem to boost the plausibility of general-purpose learning. However, in this article we propose that in the domain of verb learning, children rely more on their internally generated preconceptions about linguistic structure than on robust cues in the input, suggesting that at least in this aspect of language learning, domain-specific grammatical knowledge guides linguistic development.     

Theories of early language acquisition

What features of brain processing and neural development support linguistic development in young children? To what extent is the profile and timing of linguistic development in young children determined by a pre-ordained genetic programme? Does the environment play a crucial role in determining the patterns of change observed in children growing up? Recent experimental, neuroimaging and computational studies of developmental change in children promise to contribute to a deeper understanding of how the brain becomes wired up for language. In this review, the muttidisciplinary perspectives of cognitive neuroscience, experimental psycholinguistics and neural network modelling are brought to bear on four distinct areas in the study of language acquisition: early speech perception, word recognition, word learning and the acquisition of grammatical inflections. It is suggested that each area demonstrates how linguistic development can be driven by the interaction of general learning mechanisms, highly sensitive to particular statistical regularities in the input, with a richly structured environment which provides the necessary ingredients for the emergence of linguistic representations that support mature language processing. Similar epigenetic principles, guiding the emergence of linguistic structure, apply to all these domains, offering insights into phenomena ranging from the precocity of young infant's sensitivity to speech contrasts to the complexities of the problem facing the young child learning the arabic plural.     

Human simulations of vocabulary learning

The work reported here experimentally investigates a striking generalization about vocabulary acquisition: Noun learning is superior to verb learning in the earliest moments of child language development. The dominant explanation of this phenomenon in the literature invokes differing conceptual requirements for items in these lexical categories: Verbs are cognitively more complex than nouns and so their acquisition must await certain mental developments in the infant. In the present work, we investigate an alternative hypothesis; namely, that it is the information requirements of verb learning, not the conceptual requirements, that crucially determine the acquisition order. Efficient verb learning requires access to structural features of the exposure language and thus cannot take place until a scaffolding of noun knowledge enables the acquisition of clause-level syntax. More generally, we experimentally investigate the hypothesis that vocabulary acquisition takes place via an incremental constraint-satisfaction procedure that bootstraps itself into successively more sophisticated linguistic representations which, in turn, enable new kinds of vocabulary learning. If the experimental subjects were young children, it would be difficult to distinguish between this information-centered hypothesis and the conceptual change hypothesis. Therefore the experimental "learners" are adults. The items to be "acquired" in the experiments were the 24 most frequent nouns and 24 most frequent verbs from a sample of maternal speech to 18-24-month-old infants. The various experiments ask about the kinds of information that will support identification of these words as they occur in mother-to-child discourse. Both the proportion correctly identified and the type of word that is identifiable changes significantly as a function of information type. We discuss these results as consistent with the incremental construction of a highly lexicalized grammar by cognitively and pragmatically sophisticated human infants, but inconsistent with a procedure in which lexical acquisition is independent of and antecedent to syntax acquisition.     

A probabilistic computational model of cross-situational word learning

Words are the essence of communication: They are the building blocks of any language. Learning the meaning of words is thus one of the most important aspects of language acquisition: Children must first learn words before they can combine them into complex utterances. Many theories have been developed to explain the impressive efficiency of young children in acquiring the vocabulary of their language, as well as the developmental patterns observed in the course of lexical acquisition. A major source of disagreement among the different theories is whether children are equipped with special mechanisms and biases for word learning, or their general cognitive abilities are adequate for the task. We present a novel computational model of early word learning to shed light on the mechanisms that might be at work in this process. The model learns word meanings as probabilistic associations between words and semantic elements, using an incremental and probabilistic learning mechanism, and drawing only on general cognitive abilities. The results presented here demonstrate that much about word meanings can be learned from naturally occurring child-directed utterances (paired with meaning representations), without using any special biases or constraints, and without any explicit developmental changes in the underlying learning mechanism. Furthermore, our model provides explanations for the occasionally contradictory child experimental data, and offers predictions for the behavior of young word learners in novel situations.