This paper presents an implemented computational model of word acquisition which learns directly from raw multimodal sensory input. Set in an information theoretic framework, the model acquires a lexicon by finding and statistically modeling consistent cross‐modal structure. The model has been implemented in a system using novel speech processing, computer vision, and machine learning algorithms. In evaluations the model successfully performed speech segmentation, word discovery and visual categorization from spontaneous infant‐directed speech paired with video images of single objects. These results (...) demonstrate the possibility of using state‐of‐the‐art techniques from sensory pattern recognition and machine learning to implement cognitive models which can process raw sensor data without the need for human transcription or labeling. (shrink)

Are similar, or even identical, mechanisms used in the computational modeling of speech segmentation, serial image processing, and music processing? We address this question by exploring how TRACX2, a recognition‐based, recursive connectionist autoencoder model of chunking and sequence segmentation, which has successfully simulated speech and serial‐image processing, might be applied to elementary melody perception. The model, a three‐layer autoencoder that recognizes “chunks” of short sequences of intervals that have been frequently encountered on input, is trained on the tone intervals of (...) melodically simple French children's songs. It dynamically incorporates the internal representations of these chunks into new input. Its internal representations cluster in a manner that is consistent with “human‐recognizable” melodic categories. TRACX2 is sensitive to both contour and proximity information in the musical chunks that it encounters in its input. It shows the “end‐of‐word” superiority effect demonstrated by Saffran et al. (1999) for short musical phrases. The overall findings suggest that the recursive autoassociative chunking mechanism, as implemented in TRACX2, may be a general segmentation and chunking mechanism, underlying not only word‐ and image‐chunking, but also elementary melody processing. (shrink)

Is formal simplicity a guide to learning in humans, as simplicity is said to be a guide to the acceptability of theories in science? Does simplicity determine the difficulty of various learning tasks? I argue that, similarly to how scientists sometimes preferred complex theories when this facilitated calculations, results from perception, learning and reasoning suggest that formal complexity is generally unrelated to what is easy to learn and process by humans, and depends on assumptions about available representational and processing primitives. (...) “Simpler” hypotheses are preferred only when they are also easier to process. Historically, “simpler”, easier‐to‐process, scientific theories might also be preferred if they are transmitted preferentially. Empirically viable complexity measures should build on the representational and processing primitives of actual learners, even if explanations of their behaviour become formally more complex. (shrink)

The remarkable successes of the physical sciences have been built on highly general quantitative laws, which serve as the basis for understanding an enormous variety of specific physical systems. How far is it possible to construct universal principles in the cognitive sciences, in terms of which specific aspects of perception, memory, or decision making might be modelled? Following Shepard (e.g.,1987), it is argued that some universal principles may be attainable in cognitive science. Here, 2 examples are proposed: the simplicity principle (...) (which states that the cognitive system prefers patterns that provide simpler explanations of available data); and the scale‐invariance principle, which states that many cognitive phenomena are independent of the scale of relevant underlying physical variables, such as time, space, luminance, or sound pressure. This article illustrates how principles may be combined to explain specific cognitive processes by using these principles to derive SIMPLE, a formal model of memory for serial order (Brown, Neath, & Chater, 2007), and briefly mentions some extensions to models of identification and categorization. This article also considers the scope and limitations of universal laws in cognitive science. (shrink)

Prelinguistic infants must find a way to isolate meaningful chunks from the continuous streams of speech that they hear. BootLex, a new model which uses distributional cues to build a lexicon, demonstrates how much can be accomplished using this single source of information. This conceptually simple probabilistic algorithm achieves significant segmentation results on various kinds of language corpora - English, Japanese, and Spanish; child- and adult-directed speech, and written texts; and several variations in coding structure - and reveals which statistical (...) characteristics of the input have an influence on segmentation performance. BootLex is then compared, quantitatively and qualitatively, with three other groups of computational models of the same infant segmentation process, paying particular attention to functional characteristics of the models and their similarity to human cognition. Commonalities and contrasts among the models are discussed, as well as their implications both for theories of the cognitive problem of segmentation itself, and for the general enterprise of computational cognitive modeling. (shrink)

Children learn their native language by exposure to their linguistic and communicative environment, but apparently without requiring that their mistakes be corrected. Such learning from “positive evidence” has been viewed as raising “logical” problems for language acquisition. In particular, without correction, how is the child to recover from conjecturing an over-general grammar, which will be consistent with any sentence that the child hears? There have been many proposals concerning how this “logical problem” can be dissolved. In this study, we review (...) recent formal results showing that the learner has sufficient data to learn successfully from positive evidence, if it favors the simplest encoding of the linguistic input. Results include the learnability of linguistic prediction, grammaticality judgments, language production, and form-meaning mappings. The simplicity approach can also be “scaled down” to analyze the learnability of specific linguistic constructions, and it is amenable to empirical testing as a framework for describing human language acquisition. (shrink)

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. (shrink)

Children at about age 18 months experience acceleration in word learning. This vocabulary explosion is a robust phenomenon, although the exact shape and timing vary from child to child. One class of explanations, which we term collectively as leveraged learning, posits that knowledge of some words helps with the learning of others. In this framework, the child initially knows no words and so learning is slow. As more words are acquired, new words become easier and thus it is the acquisition (...) of early words that fuels the explosion in learning. In this paper we examine the role of leveraged learning in the vocabulary spurt by proposing a simple model of leveraged learning. Our results show that leverage can change both the shape and timing of the acceleration, but that it cannot create acceleration if it did not exist in the corresponding model without leveraging. This model is then applied to the Zipfian distribution of word frequencies, which confirm that leveraging does not create acceleration, but that the relationship between frequency and the difficulty of learning a word may be complex. (shrink)

This paper presents a computational model of the way humans inductively identify and aggregate concepts from the low-level stimuli they are exposed to. Based on the idea that humans tend to select the simplest structures, it implements a dynamic hierarchical chunking mechanism in which the decision whether to create a new chunk is based on an information-theoretic criterion, the Minimum Description Length (MDL) principle. We present theoretical justifications for this approach together with results of an experiment in which participants, exposed (...) to meaningless symbols, have been implicitly encouraged to create high-level concepts by grouping them. Results show that the designed model, called hereafter MDLChunker, makes precise quantitative predictions both on the kind of chunks created by the participants and also on the moment at which these creations occur. They suggest that the simplicity principle used to design MDLChunker is particularly efficient to model chunking mechanisms. The main interest of this model over existing ones is that it does not require any adjustable parameter. (shrink)

This paper reconsiders the diphone-based word segmentation model of Cairns, Shillcock, Chater, and Levy (1997) and Hockema (2006), previously thought to be unlearnable. A statistically principled learning model is developed using Bayes’ theorem and reasonable assumptions about infants’ implicit knowledge. The ability to recover phrase-medial word boundaries is tested using phonetic corpora derived from spontaneous interactions with children and adults. The (unsupervised and semi-supervised) learning models are shown to exhibit several crucial properties. First, only a small amount of language exposure (...) is required to achieve the model’s ceiling performance, equivalent to between 1 day and 1 month of caregiver input. Second, the models are robust to variation, both in the free parameter and the input representation. Finally, both the learning and baseline models exhibit undersegmentation, argued to have significant ramifications for speech processing as a whole. (shrink)

Poverty of the Stimulus Arguments have convinced many linguists and philosophers of language that a domain specific language acquisition device (LAD) is necessary to account for language learning. Here we review empirical evidence that casts doubt on the necessity of this domain specific device. We suggest that more attention needs to be paid to the early stages of language acquisition. Many seemingly innate language-related abilities have to be learned over the course of several months. Further, the language input contains rich (...) stochastic information that can be accessed by domain-general learning mechanisms. Computer simulation has shown how mechanisms that are not domain specific can exploit the information contained in language. We conclude that (i) Poverty of the Stimulus Arguments need to be conceptually clarified and (ii) more empirical research needs to be carried out before we can rule out that data driven general purpose mechanisms can account for language learning. (shrink)

Infants are sensitive to statistical regularities (i.e., transitional probabilities, or TPs) relevant to segmenting words in fluent speech. However, there is debate about whether tracking TPs results in representations of possible words. Infants show preferential learning of sequences with high TPs (HTPs) as object labels relative to those with low TPs (LTPs). Such findings could mean that only the HTP sequences have a word‐like status, and they are more readily mapped to a referent for that reason. But these findings could (...) also suggest that HTP sequences are easier to encode, just like any other predictable sequence. Here we aimed to distinguish between these explanations. To do so, we built on findings that infants become resistant to learning labels that are not typical of their native language as they approach 2 years of age and add words to their lexicons. If tracking TPs in speech results in identifying candidate words, at this age TPs may have reduced power to confer lexical status when they yield a unit that is very dissimilar to word forms that are typical of infants’ native language. Indeed, we found that at 20 months, English‐learning infants with relatively small vocabularies learned HTP Italian words (but not LTP words) as object labels, while infants with larger vocabularies resisted learning HTP Italian words. These findings suggest that the HTP sequences may be represented as candidate words, and more broadly, that TP statistics are relevant to word learning. (shrink)

Debates concerning the types of representations that aid reading acquisition have often been influenced by the relationship between measures of early phonological awareness (the ability to process speech sounds) and later reading ability. Here, a complementary approach is explored, analyzing how the functional utility of different representational units, such as whole words, bodies (letters representing the vowel and final consonants of a syllable), and graphemes (letters representing a phoneme) may change as the number of words that can be read gradually (...) increases. Utility is measured by applying a Simplicity Principle to the problem of mapping from print to sound; that is, assuming that the “best” representational units for reading are those which allow the mapping from print to sounds to be encoded as efficiently as possible. Results indicate that when only a small number of words are read whole-word representations are most useful, whereas when many words can be read graphemic representations have the highest utility. (shrink)

This study investigates the joint influences of three factors on the discovery of new word‐like units in a continuous artificial speech stream: the statistical structure of the ongoing input, the initial word‐likeness of parts of the speech flow, and the contextual information provided by the earlier emergence of other word‐like units. Results of an experiment conducted with adult participants show that these sources of information have strong and interactive influences on word discovery. The authors then examine the ability of different (...) models of word segmentation to account for these results. PARSER (Perruchet & Vinter, 1998) is compared to the view that word segmentation relies on the exploitation of transitional probabilities between successive syllables, and with the models based on the Minimum Description Length principle, such as INCDROP. The authors submit arguments suggesting that PARSER has the advantage of accounting for the whole pattern of data without ad‐hoc modifications, while relying exclusively on general‐purpose learning principles. This study strengthens the growing notion that nonspecific cognitive processes, mainly based on associative learning and memory principles, are able to account for a larger part of early language acquisition than previously assumed. (shrink)

In 2006, Perruchet and Pacton (2006) asked whether implicit learning and statistical learning represent two approaches to the same phenomenon. This article represents an important follow‐up to their seminal review article. As in the previous paper, the focus is on the formation of elementary cognitive units. Both approaches favor different explanations on what these units consist of and how they are formed. Perruchet weighs up the evidence for different explanations and concludes with a helpful agenda for future research.

Statistical learning is often taken to lie at the heart of many cognitive tasks, including the acquisition of language. One particular task in which probabilistic models have achieved considerable success is the segmentation of speech into words. However, these models have mostly been tested against English data, and as a result little is known about how a statistical learning mechanism copes with input regularities that arise from the structural properties of different languages. This study focuses on statistical word segmentation in (...) Arabic, a Semitic language in which words are built around consonantal roots. We hypothesize that segmentation in such languages is facilitated by tracking consonant distributions independently from intervening vowels. Previous studies have shown that human learners can track consonant probabilities across intervening vowels in artificial languages, but it is unknown to what extent this ability would be beneficial in the segmentation of natural language. We assessed the performance of a Bayesian segmentation model on English and Arabic, comparing consonant-only representations with full representations. In addition, we examined to what extent structurally different proto-lexicons reflect adult language. The results suggest that for a child learning a Semitic language, separating consonants from vowels is beneficial for segmentation. These findings indicate that probabilistic models require appropriate linguistic representations in order to effectively meet the challenges of language acquisition. (shrink)

The informativity of a computational model of language acquisition is directly related to how closely it approximates the actual acquisition task, sometimes referred to as the model's cognitive plausibility. We suggest that though every computational model necessarily idealizes the modeled task, an informative language acquisition model can aim to be cognitively plausible in multiple ways. We discuss these cognitive plausibility checkpoints generally and then apply them to a case study in word segmentation, investigating a promising Bayesian segmentation strategy. We incorporate (...) cognitive plausibility by using an age-appropriate unit of perceptual representation, evaluating the model output in terms of its utility, and incorporating cognitive constraints into the inference process. Our more cognitively plausible model shows a beneficial effect of cognitive constraints on segmentation performance. One interpretation of this effect is as a synergy between the naive theories of language structure that infants may have and the cognitive constraints that limit the fidelity of their inference processes, where less accurate inference approximations are better when the underlying assumptions about how words are generated are less accurate. More generally, these results highlight the utility of incorporating cognitive plausibility more fully into computational models of language acquisition. (shrink)

We introduce a set of biologically and computationally motivated design choices for modeling the learning of language, or of other types of sequential, hierarchically structured experience and behavior, and describe an implemented system that conforms to these choices and is capable of unsupervised learning from raw natural‐language corpora. Given a stream of linguistic input, our model incrementally learns a grammar that captures its statistical patterns, which can then be used to parse or generate new data. The grammar constructed in this (...) manner takes the form of a directed weighted graph, whose nodes are recursively (hierarchically) defined patterns over the elements of the input stream. We evaluated the model in seventeen experiments, grouped into five studies, which examined, respectively, (a) the generative ability of grammar learned from a corpus of natural language, (b) the characteristics of the learned representation, (c) sequence segmentation and chunking, (d) artificial grammar learning, and (e) certain types of structure dependence. The model's performance largely vindicates our design choices, suggesting that progress in modeling language acquisition can be made on a broad front—ranging from issues of generativity to the replication of human experimental findings—by bringing biological and computational considerations, as well as lessons from prior efforts, to bear on the modeling approach. (shrink)

We address the problem of predicting how people will spontaneously divide into groups a set of novel items. This is a process akin to perceptual organization. We therefore employ the simplicity principle from perceptual organization to propose a simplicity model of unconstrained spontaneous grouping. The simplicity model predicts that people would prefer the categories for a set of novel items that provide the simplest encoding of these items. Classification predictions are derived from the model without information either about the number (...) of categories sought or information about the distributional properties of the objects to be classified. These features of the simplicity model distinguish it from other models in unsupervised categorization (where, for example, the number of categories sought is determined via a free parameter), and we discuss how these computational differences are related to differences in modeling objectives. The predictions of the simplicity model are validated in four experiments. We also discuss the significance of simplicity in cognitive modeling more generally. (shrink)