Small-world Structure in Children’s Featured Semantic Networks

Background: Knowing the development pattern of children’s language is applicable in developmental psychology. Network models of language are helpful for the identification of these patterns. Objectives: We examined the small-world properties of featured semantic networks of developing children. Materials & Methods: In this longitudinal study, the featured semantic networks of children aged 18-30 months were obtained using R software version 3.5.2 and the igraph software package. The data of 2000 English (British)-speaking children, half boy and half girls, were gathered from existing databases of MCDI (between 2000 and 2007) and McRae feature norms. The growth pattern of these networks was illustrated by graph measures. Comparing these measures with those of the reference random networks, the small-world structure can be examined. Results: To have a comparison between path length and clustering coefficient of featured semantic networks with those of random networks, we computed the Q quotient. The results showed that the values of the Q quotient at 18, 22, 26, and 30 months of age were all more than 1, which confirms the small-world characteristic of the networks. Conclusion: Featured semantic networks of children exhibited a small-world structure, in which there was a local structure in the form of clusters of words. For global access, some words act as bridges connecting semantically distant clusters. These networks possess small-world property from the early months of age. The small-world structure cannot be seen in the less dense networks built with a higher cut-off threshold.

Transformer Networks of Human Concept Knowledge

We present a computational model capable of simulating aspects of human knowledge for thousands of real-world concepts. Our approach involves fine-tuning a transformer network for natural language understanding on participant-generated feature norms. We show that such a model can successfully extrapolate from its training dataset, and predict human knowledge for novel concepts and features. We also apply our model to stimuli from twenty-three previous experiments in semantic cognition research, and show that it reproduces fifteen classic findings involving semantic verification, concept typicality, feature distribution, and semantic similarity. We interpret these findings using established properties of classic connectionist networks. The success of our approach shows how the combination of natural language data and psychological data can be used to build cognitive models with rich world knowledge. Such models can be used in the service of new psychological applications, such as the cognitive process modeling of naturalistic semantic verification and knowledge retrieval, as well as the modeling of real-world categorization, decision making, and reasoning.

A cross-linguistic comparison of Spanish and English semantic norms: Looking at core features

It has been suggested that human communities that share their basic cultural foundations evince no remarkable differences concerning the characterization of core concepts. However, the small but existing differences among them reflect their sociocultural diversity. This study compares 219 concrete concepts common to both Spanish and English semantic feature norms in order to assess whether core features of concepts follow a universal or cultural language-specific pattern. Concepts were compared through a geometric technique of vector comparison in the Euclidean n-dimensional space alongside the calculation of the network’s degree of centrality. The role of cognate status was also explored by repeating the former analysis separating cognate from noncognate words. Taken together, our data show that languages are structurally similar independent of the cognate status of words, further suggesting that there are some sort of core features common to both languages.

Learning Neural Audio Embeddings for Grounding Semantics in Auditory Perception

Multi-modal semantics, which aims to ground semantic representations in perception, has relied on feature norms or raw image data for perceptual input. In this paper we examine grounding semantic representations in raw auditory data, using standard evaluations for multi-modal semantics. After having shown the quality of such auditorily grounded representations, we show how they can be applied to tasks where auditory perception is relevant, including two unsupervised categorization experiments, and provide further analysis. We find that features transfered from deep neural networks outperform bag of audio words approaches. To our knowledge, this is the first work to construct multi-modal models from a combination of textual information and auditory information extracted from deep neural networks, and the first work to evaluate the performance of tri-modal (textual, visual and auditory) semantic models.

Using semantic feature norms to investigate how the visual and verbal modes afford metaphor construction and expression

abstractIn this study, two modalities of expression (verbal and visual) are compared and contrasted, in relation to their ability and their limitations to construct and express metaphors. A representative set of visual metaphors and a representative set of linguistic metaphors are here compared, and the semantic similarity between metaphor terms is modeled within the two sets. Such similarity is operationalized in terms of semantic features produced by informants in a property generation task (e.g., McRae et al., 2005). Semantic features provide insights into conceptual content, and play a role in deep conceptual processing, as opposed to shallow linguistic processing. Thus, semantic features appear to be useful for modeling metaphor comprehension, assuming that metaphors are matters of thought rather than simple figures of speech (Lakoff & Johnson, 1980). The question tackled in this paper is whether semantic features can account for the similarity between metaphor terms of both visual and verbal metaphors. For this purpose, a database of semantic features was collected and then used to analyze fifty visual metaphors and fifty verbal metaphors. It was found that the number of semantic features shared between metaphor terms is predicted by the modality of expression of the metaphor: the terms compared in visual metaphors share semantic features, while the terms compared in verbal metaphors do not. This suggests that the two modalities of expression afford different ways to construct and express metaphors.