Giovedì 21 Dicembre 2017, ore 12.00, aula F1 il Dr. Massimo Stella, Complex Multilayer Networks (CoMuNe), Fondazione Bruno Kessler, Trento terrà un seminario dal titolo:
Modelling Words in the Human Mind via Percolation, Markov Chains and Multiplex Networks
Over the last forty years, overwhelming empirical evidence has shown that word similarities influence
learning, storing and retrieving words in the human mind, hence the importance of a network representation of this mental lexicon of word relationships. This talk revolves around the development of theoretical single- and multi-layer network models for: (i) detecting constraints over sound similarities in words, and (ii) quantifying word learning strategies in young children and adults.
In cognitive network science, sound similarities among words are expressed in terms of phonological networks, where nodes represent words and links represent phonological similarities (i.e. two phonetic
transcriptions having edit distance one). From a graph theoretical perspective, phonological networks are labelled graphs, where nodes have the additional property of being labelled and connections depend on the labelling. Our null models based on percolation and Markov processes indicate that some features such as the heavy-tailed degree distribution or the assortative mixing by degree are induced by the spatial embedding of phonological networks in combinatorial structures we call layers. These induced features cannot be considered as real patterns of the mental lexicon structure. Instead, our growing network models suggest the presence of constraints in the assembly of phonemes in real words, so that (i) large degrees and (ii) triadic closure are avoided. These constraints are matched by experimental evidence about confusability of similar sounding words.
The second part of the talk introduces the framework of multiplex lexical networks for quantifying the influence that multiple semantic and phonological layers of word similarities combined can have on
cognitive development in toddlers and adults. The investigation of the multiplex structure highlights phenomena that are not observable in single-layer networks. In toddlers, word learning strategies based on the closeness of words on the whole multiplex structure match empirical word learning significantly better than strategies based on individual layers. This indicates that multiplexity is important for understanding and predicting early word acquisition. At later stages the multiplex structure evolves by displaying an explosive phase transition, i.e. a hybrid phase transition with large critical exponents, in the emergence of a multiplex network core of words. This phase transition happens in children around 7-8 years old, a well documented age corresponding to a sudden increase in linguistic proficiency and analytical reasoning. Through robustness experiments, we also show that this lexical core is of particular interest from a clinical perspective as it boosts the resilience of the mental lexicon against progressive cognitive impairments such as aphasia.