Two Distinct Sequence Learning Mechanisms for Syntax Acquisition and Word Learning

The ability to acquire spoken language depends in part on a sensitivity to the sequential regularities contained within linguistic input. In this chapter, the authors propose that language learning operates via two distinct sequence-learning processes: probabilistic sequence learning, which supports the acquisition of syntax and other structured linguistic patterns, and repetition sequence learning, which supports word learning. First, the authors review work from their lab and others illustrating that performance on tasks that require participants to learn non-linguistic sequential patterns is empirically associated with different measures of language processing. Second, they present recent work from their lab specifically highlighting the role played by probabilistic sequence learning for acquiring syntax in a sample of deaf and hard-of-hearing children. Finally, the authors demonstrate that the learning of repeating sequences is related to vocabulary development in these children. These findings suggest that there may be at least two relatively distinct domain-general sequential processing skills, with each supporting a different aspect of language acquisition.

Learning Words by Imitating

This chapter proposes a single imitation-learning algorithm capable of simultaneously learning linguistic as well as nonlinguistic tasks, without demonstrations being labeled. A human demonstrator responds to an environment that includes the behavior of another human, called the interactant, and the algorithm must learn to imitate this response without being told what the demonstrator was responding to (for example, the position of an object or a speech utterance of the interactant). Since there is no separate symbolic language system, the symbol grounding problem can be avoided/dissolved. The types of linguistic behavior explored are action responses, which includes verb learning but where actions are generalized to include such things as communicative behaviors or internal cognitive operations. Action responses to object positions are learnt in the same way as action responses to speech utterances of an interactant. Three experiments are used to validate the proposed algorithm.

A Dynamic Neural Field Model of Word Learning

Word learning is a complex phenomenon because it is tied to many different behaviors that are linked to multiple perceptual and cognitive systems. Further, recent research suggests that the course of word learning builds from effects at the level of individual referent selection or noun generalization decisions that accumulate on a moment-to-moment timescale and structure subsequent word learning behaviors. Thus, what is needed for any unified theory of word learning is 1) an account of how individual decisions are made across different contexts, including the details of how objects are encoded, represented, and selected in the course of a word learning behavior; and 2) a mechanism that builds on these individual, contextually specific decisions. Here, the authors present a Dynamic Neural Field (DNF) Model that captures processes at both the second-to-second and developmental timescales and provides a process-based account of how individual behaviors accumulate to create development. Simulations illustrate how the model captures multiple word learning behaviors such as comprehension, production, novel noun generalization (in yes/no or forced choice tasks), referent selection, and learning of hierarchical nominal categories. They also discuss how the model ties developments in these tasks to developments in object perception, working memory, and the representation and tracking of objects in space. Finally, the authors review empirical work testing novel predictions of the model regarding the roles of competition and selection in forced-choice and yes/no generalization tasks and the role of space in early name-object binding.

I Think I Have Heard That One Before

In this chapter, the authors present a model for learning Word-Like Units (WLUs) based on acoustic recurrence, as well as the results of an application of the model to simulated child-directed speech in human-robot interaction. It is a purely acoustic single-modality model: the learning does not invoke extralinguistic factors such as possible references of words or linguistic constructs including phonemes. The main target phenomenon is the learner’s perception that a WLU has been repeated. To simulate it, a Dynamic Time Warping (DTW)-based algorithm is introduced to search for recurrent utterances of similar acoustic features. The authors then extend this model to incorporate interaction, corrective feedback in particular, and assess the ameliorating effect of caregiver correction when a WLU, which is close to the real word, is uttered by the learner.

Making Use of Multi-Modal Synchrony

If they are to learn and interact with humans, robots need to understand actions and make use of language in social interactions. Hirsh-Pasek and Golinkoff (1996) have emphasized the use of language to learn actions when introducing the idea of acoustic packaging in human development. This idea suggests that acoustic information, typically in the form of narration, overlaps with action sequences, thereby providing infants with a bottom-up guide to attend to relevant parts and to find structure within them. The authors developed a computational model of the multimodal interplay of action and language in tutoring situations. This chapter presents the results of applying this model to multimodal parent-infant interaction data. Results are twofold and indicate that (a) infant-directed interaction is more structured than adult-directed interaction in that it contains more packages, and these packages have fewer motion segments; and (b) the synchronous structure within infant-directed packages contains redundant information making it possible to solve the reference problem when tying color adjectives to a moving object.

An Embodied Model of Young Children’s Categorization and Word Learning

Children learn words with remarkable speed and flexibility. However, the cognitive basis of young children’s word learning is disputed. Further, although research demonstrates that children’s categories and category labels are interdependent, how children learn category labels is also a matter of debate. Recently, biologically plausible, computational simulations of children’s behavior in experimental tasks have investigated the cognitive processes that underlie learning. The ecological validity of such models has been successfully tested by deploying them in robotic systems (Morse, Belpaeme, Cangelosi, & Smith, 2010). The authors present a simulation of children’s behavior in a word learning task (Twomey & Horst, 2011) via an embodied system (iCub; Metta, et al., 2010), which points to associative learning and dynamic systems accounts of children’s categorization. Finally, the authors discuss the benefits of integrating computational and robotic approaches with developmental science for a deeper understanding of cognition.

The Miracle Year

In recent years, a functional perspective on infant communication has emerged whereby infants’ production of vocal sounds is understood not only in terms of the acoustic properties of those sounds, but also in terms of the sounds that regulate and are regulated by social interactions with those hearing them. Here, the authors synthesize findings across several disciplines to characterize this holistic view of infant language learning. The goal is to interpret classic and more recent behavioral findings (e.g., on infants’ preferences) in light of data on pre- and postnatal neurophysiological responses to the environment (e.g., fetal heart rate, cortical blood flow). Language learning is a complex process that takes place at multiple levels across multiple systems; this review is an attempt to embrace this complexity and provide an integrated account of how these systems interact to support language learning.

Timing Matters

What is the nature of sensitive periods in lexical development? In this chapter, the authors propose a novel dynamic view of sensitive periods. They suggest that they are periods of heightened interaction and adaptation between organism and environment that are the emergent result of the changing developmental landscape. In support of this perspective, the authors first provide an extended model of word learning to show that language moves through a predictable sequence of sensitive periods, each serving as a building block for the prior. Next, they show how changes in the timing of sensitive periods can affect early word learning in the case of two populations—preterm infants and children with cochlear implants. Finally, the authors provide a theoretical overview of how typically developing infants move from basic perception to full-blown language across several domains of language, and how changes in the timing of the input and response can lead to changes in developmental outcomes.

A Sensory-Motor Solution to Early Word-Referent Learning

Many theories of word learning begin with the uncertainty inherent to learning a word from its co-occurrence with a visual scene. However, the relevant visual scene for infant word learning is neither from the adult theorist’s view nor the mature partner’s view, but is rather from the learner’s personal view. Here, the authors review recent studies on 18-month-old infants playing with their parents in a free-flowing interaction. Frame-by-frame analyses of the head camera images at and around naming moments were conducted to determine the visual properties at input that were associated with learning. The main contribution is that toddlers, through their own actions, often create a personal view that consists of one dominating object. Parents often (but not always) name objects during these optimal sensory moments, and when they do, toddlers learn the object name. The results are discussed with respect to early word learning, embodied attention, and robotics.

Learning Words from Experience

How does experience influence children’s acquisition of word meanings? In this chapter, the authors discuss the evidence from two bodies of literature that take different perspectives to answer this question. First, they review evidence from the “experience” literature, which has demonstrated that different experiential factors (e.g., differences in the quantity and quality of maternal speech) are related to individual differences in children’s early vocabularies. Although the results of the studies within this literature are interesting, the authors argue that they do not clarify how experience influences children’s vocabulary development. They posit that this question can best be answered by marrying the “experience” literature and the “cognitive” literature, which has identified the skills and knowledge that children possess that help them determine the meanings of words. The authors demonstrate how integrating both literatures will provide a valuable framework from which research can be designed and hypotheses tested. In doing so, their framework will provide a comprehensive understanding of how experience influences children’s lexical development.