Language and perception. Investigating linear and hierarchical implicit statistical learning across the visual, auditory, and tactile sensory domains.

This thesis explored how humans process and form recursive hierarchical structures arising from temporally ordered sequences of stimuli, across the visual, auditory, and tactile sensory domains. As we will explain throughout this thesis, we posit that the ability to form recursive hierarchical abstract representations from temporally ordered stimuli is a cognitive ability involved in human syntax processing and acquisition. Language unfolds in a linear fashion. Words follow one another, creating sentences that, on the surface, appear as linear sequences of sounds or symbols. However, a purely sequential arrangement of words alone falls short in encompassing the complexities of human language syntax. It is evident that the syntax of human languages has a fundamental hierarchical dimension, where constituents are organized in a way that is intricately linked to their linear order. Among the various syntactic phenomena that depend on this hierarchical organization, recursion is one of the most fascinating and controversial in the study of language. Recursion in human syntax, understood as the characteristic of embedding constituents within constituents of the same kind, has long been considered a fundamental and distinctive feature of human language. Therefore, the cognitive ability to deal with recursion has been viewed as crucial for language capacity, possibly representing a uniquely human faculty at the core of language ability. However, this topic is highly controversial. Despite the importance attributed to recursion in linguistics, several questions remain open. What is the role of recursion in human language? Is the ability to handle recursion specifically tied to the human language faculty? What is the mechanism underlying the cognitive ability to form recursive abstract representations in language, considering both the linear and hierarchical nature of syntax? To analyze this topic, this thesis will delve into three critical issues at the core of theoretical and experimental linguistic debates. The first issue addresses the debated role of recursion in human language syntax. The second issue examines the contributions of recursive hierarchical abstract representation and statistical learning to the acquisition and processing of human syntax. The third issue, intimately connected to the second, examines the existence of domain-specific representational and learning constraints, alongside the influence of domain-general learning abilities on this process. Our research had two main objectives: Firstly, we aimed to determine whether sequential statistical learning and the formation of recursive hierarchical abstract representation operate independently as distinct levels of language analysis or if they work together synergistically as complementary learning mechanisms. If they complement each other, we sought to understand the cognitive processes involved in transitioning from linear to recursive hierarchical dimensions. Secondly, we investigated whether the ability to form recursive hierarchical abstract structures from sequential stimuli is a language-specific ability or a domain-general ability, shared across different modalities and whether there are domain-specific differences in this ability between sensory domains. To address these inquiries, we employed the Artificial Grammar Learning paradigm, conducting three Serial Reaction Time tasks. Three distinct groups of adult participants were presented with a sequence of stimuli featuring the rules of a non-canonical binary grammar belonging to the Lindenmayer systems: The Fibonacci grammar (Fib). The choice to use this grammar was driven by its exceptional suitability for thoroughly investigating this research topic in all its various facets. On one hand, it allows for the investigation of the application of recursive algorithms for predicting points in the string, while simultaneously examining the relationship between sequential statistical learning and the creation of recursive hierarchical representations. On the other hand, this paradigm permits the examination and direct comparison of these cognitive abilities across different sensory modalities. In the three tasks, the symbols of Fib were encoded onto auditory tones, vibrotactile impulses, or colorful visual shapes. Through analysis of reaction times and accuracy data in response to perceived stimuli, we explored whether participants implicitly learned the regularities of Fib across all three sensory domains and potentially domain-specific learning differences. Our findings suggested a close linkage between the ability to form recursive hierarchical representations and the capacity to grasp low-level transitional regularities. With this regard, we introduced a cognitive parsing algorithm hypothesizing the cognitive mechanisms involved in transitioning from sequence to hierarchy. Furthermore, we observed that the cognitive ability to process and learn these structures, which underpin human language, is a domain-general ability present across diverse sensory domains. However, we also identified domain-specific differences, with auditory and tactile modalities exhibiting a distinct advantage over the visual domain. In summary, our results indicated that sequential statistical learning and recursive hierarchical abstract representation synergize as complementary modes of learning, rather than operating as distinct levels of language analysis. Moreover, our findings suggest that the capability to from recursive hierarchical abstract structures arising from temporally ordered stimuli is not a language-specific ability but rather a domain-general capacity present across different sensory modalities, potentially interacting with language in specific ways.