Luca L. Bonatti

Linguistic Constraints on Statistical Computations The Role of Consonants and Vowels in Continuous Speech Processing

Speech is produced mainly in continuous streams containing several words. Listeners can use the transitional probability (TP) between adjacent and non-adjacent syllables to segment “words” from a continuous stream of artificial speech, much as they use TPs to organize a variety of perceptual continua. It is thus possible that a general-purpose statistical device exploits any speech unit to achieve segmentation of speech streams. Alternatively, language may limit what representations are open to statistical investigation according to their specific linguistic role. In this article, we focus on vowels and consonants in continuous speech. We hypothesized that vowels and consonants in words carry different kinds of information, the latter being more tied to word identification and the former to grammar. We thus predicted that in a word identification task involving continuous speech, learners would track TPs among consonants, but not among vowels. Our results show a preferential role for consonants in word identification.

Pure Reasoning in 12-Month-Old Infants as Probabilistic Inference

Twelve-month-old infants employ Bayesian statistics. Many organisms can predict future events from the statistics of past experience, but humans also excel at making predictions by pure reasoning: integrating multiple sources of information, guided by abstract knowledge, to form rational expectations about novel situations, never directly experienced. Here, we show that this reasoning is surprisingly rich, powerful, and coherent even in preverbal infants. When 12-month-old infants view complex displays of multiple moving objects, they form time-varying expectations about future events that are a systematic and rational function of several stimulus variables. Infants’ looking times are consistent with a Bayesian ideal observer embodying abstract principles of object motion. The model explains infants’ statistical expectations and classic qualitative findings about object cognition in younger babies, not originally viewed as probabilistic inferences.

Finding Words and Rules in a Speech Stream Functional Differences Between Vowels and Consonants

We have proposed that consonants give cues primarily about the lexicon, whereas vowels carry cues about syntax. In a study supporting this hypothesis, we showed that when segmenting words from an artificial continuous stream, participants compute statistical relations over consonants, but not over vowels. In the study reported here, we tested the symmetrical hypothesis that when participants listen to words in a speech stream, they tend to exploit relations among vowels to extract generalizations, but tend to disregard the same relations among consonants. In our streams, participants could segment words on the basis of transitional probabilities in one tier and could extract a structural regularity in the other tier. Participants used consonants to extract words, but vowels to extract a structural generalization. They were unable to extract the same generalization using consonants, even when word segmentation was facilitated and the generalization made simpler. Our results suggest that different signal-driven computations prime lexical and grammatical processing.

On Consonants, Vowels, Chickens, and Eggs

On Consonants, Vowels, Chickens, and Eggs Luca L. Bonatti, Marcela Pena, Marina Nespor, and Jacques Mehler International School for Advanced Studies, Trieste, Italy; Universite de Nantes, Nantes Atlantique Universites, Labecd (EA 3259), Nantes, France; Laboratorio de Neurociencias, Escuela de Psicologia, Pontificia Universidad Catolica de Chile, Macul, Chile; University of Ferrara, Ferrara, Italy; and Laboratoire de Science Cognitive et Psycholinguistique, CNRS and EHESS, Paris, France

Large scale brain activations predict reasoning profiles

Deduction is the ability to draw necessary conclusions from previous knowledge. Here we propose a novel approach to understanding the neural basis of deduction, which exploits fine-grained inter-participant variability in such tasks. Participants solved deductive problems and were grouped by the behavioral strategies employed, i.e., whether they were sensitive to the logical form of syllogistic premises, whether the problems were solved correctly, and whether heuristic strategies were employed. Differential profiles of neural activity can predict membership of the first two of these groups. The predictive power of activity profiles is distributed non-uniformly across the brain areas activated by deduction. Activation in left ventro-lateral frontal (BA47) and lateral occipital (BA19) cortices predicts whether logically valid solutions are sought. Activation of left inferior lateral frontal (BA44/45) and superior medial frontal (BA6/8) cortices predicts sensitivity to the logical structure of problems. No specific pattern of activation was associated with the use of a non-logical heuristic strategy. Not only do these findings corroborate the hypothesis that left BA47, BA44/45 and BA6/8 are critical for making syllogistic deductions, but they also imply that they have different functional roles as components of a dedicated network. We propose that BA44/45 and BA6/8 are involved in the extraction and representation of the formal structure of a problem, while BA47 is involved in the selection and application of relevant inferential rules. Finally, our findings suggest that deductive reasoning can be best described as a cascade of cognitive processes requiring the concerted operation of several, functionally distinct, brain areas.

Finding Words and Word Structure in Artificial Speech: The Development of Infants' Sensitivity to Morphosyntactic Regularities.

To achieve language proficiency, infants must find the building blocks of speech and master the rules governing their legal combinations. However, these problems are linked: words are also built according to rules. Here, we explored early morphosyntactic sensitivity by testing when and how infants could find either words or within-word structure in artificial speech snippets embodying properties of morphological constructions. We show that 12-month-olds use statistical relationships between syllables to extract words from continuous streams, but find word-internal regularities only if the streams are segmented. Seven-month-olds fail both tasks. Thus, 12-month-olds infants possess the resources to analyze the internal composition of words if the speech contains segmentation information. However, 7-month-old infants may not possess them, although they can track several statistical relations. This developmental difference suggests that morphosyntactic sensitivity may require computational resources extending beyond the detection of simple statistics.

Infants anticipate probabilistic but not deterministic outcomes.

Infants look at physically impossible events longer than at physically possible events, and at improbable events longer than at probable events. Such behaviors are generally interpreted as showing that infants have expectations about future events and are surprised to see them violated. It is unknown, however, whether and under what conditions infants form proactive expectations about the future, as opposed to realizing post hoc that outcomes do not comply with their previous knowledge or experience. Here we investigate the relation between expectation and surprise at probabilistic or deterministic events in preverbal infants. When a situation is uncertain, 12-month-olds anticipate probable outcomes and are surprised at improbable continuations of the scene. However, they do not anticipate the only possible outcome of a physically deterministic situation, although they are surprised when it does not occur. The results suggest that infants are sensitive to the tradeoff between information gain and programming efforts, showing higher propensity to anticipate those future events that carry novel knowledge.

Words, rules, and mechanisms of language acquisition

We review recent artificial language learning studies, especially those following Endress and Bonatti (Endress AD, Bonatti LL. Rapid learning of syllable classes from a perceptually continuous speech stream. Cognition 2007, 105:247-299), suggesting that humans can deploy a variety of learning mechanisms to acquire artificial languages. Several experiments provide evidence for multiple learning mechanisms that can be deployed in fluent speech: one mechanism encodes the positions of syllables within words and can be used to extract generalization, while the other registers co-occurrence statistics of syllables and can be used to break a continuum into its components. We review dissociations between these mechanisms and their potential role in language acquisition. We then turn to recent criticisms of the multiple mechanisms hypothesis and show that they are inconsistent with the available data. Our results suggest that artificial and natural language learning is best understood by dissecting the underlying specialized learning abilities, and that these data provide a rare opportunity to link important language phenomena to basic psychological mechanisms. For further resources related to this article, please visit the WIREs website.

A Dissociation Between Judged Causality and Imagined Locations in Simple Dynamic Scenes

To mentally extrapolate the trajectory of a moving object that disappears from sight, different sources of information can be exploited: memory of its last visible position, its inferred movement through time, and general understanding of the causal structure of the scene. It is often assumed that these cues are integrated into unified analog mental representations. In our experiment, participants predicted the position of an object that disappeared behind an occluder and estimated the degree to which the movement was caused by another object. They made considerable errors in predicting imagined displacements. Moreover, their predictions were misaligned with their judgments of causality. They predicted the positions of the invisible moving objects better in events that they judged less causally correct than in events that they judged more causally correct. These results suggest that physical and cognitive parameters of imagined dynamic events do not merge into unitary mental models simulating actual states of the world.

Precursors of logical reasoning in preverbal human infants

The infant as philosopher Visual behaviors, such as a shift in ones gaze or a prolonged stare, can be diagnostic of internal thoughts. Cesana-Arlotti et al. used these measures to demonstrate that preverbal infants can formulate a logical structure called a disjunctive syllogism (see the Perspective by Halberda). That is, if A or B is true, and A is false, then B must be true. Presenting infants with scenes where the outcome revealed B to be false evoked looks of surprise. Science, this issue p. 1263; see also p. 1214 The development of logical reasoning rekindles debate about the contributions of language to thinking. Infants are able to entertain hypotheses about complex events and to modify them rationally when faced with inconsistent evidence. These capacities suggest that infants can use elementary logical representations to frame and prune hypotheses. By presenting scenes containing ambiguities about the identity of an object, here we show that 12- and 19-month-old infants look longer at outcomes that are inconsistent with a logical inference necessary to resolve such ambiguities. At the moment of a potential deduction, infants’ pupils dilated, and their eyes moved toward the ambiguous object when inferences could be computed, in contrast to transparent scenes not requiring inferences to identify the object. These oculomotor markers resembled those of adults inspecting similar scenes, suggesting that intuitive and stable logical structures involved in the interpretation of dynamic scenes may be part of the fabric of the human mind.