Data‐driven approaches to information access

This paper summarizes three lines of research that are motivated by the practical problem of helping users find information from external data sources, most notably computers. The application areas include information retrieval, text categorization, and question answering. A common theme in these applications is that practical information access problems can be solved by analyzing the statistical properties of words in large volumes of real world texts. The same statistical properties constrain human performance, thus we believe that solutions to practical information access problems can shed light on human knowledge representation and reasoning.     

Oxytocin attenuates racial categorization in 14-month-old infants

Intergroup bias – the preferential attitudes one holds towards one's social group – is a ubiquitous socio-cognitive phenomenon. In fact, studies show that already in the first months of life, infants manifest a preference for members of their own social group. This points to the possibility of inborn mechanisms involved in social group cognition. Here we assess the effect of a biological activation of infants’ affiliative motivation on their social categorization capacity. In a first visit to the lab, mothers self-administered either Oxytocin (OT) or placebo (PL) via a nasal spray and then engaged in a face-to-face interaction with their 14-month-old infants, a procedure previously shown to increase OT levels in infants. Infants then performed a racial categorization task presented on an eye-tracker. Mothers and infants returned a week later and repeated the procedure while self-administering the complementary substance (i.e., PL or OT, respectively). In total, 24 infants completed the two visits. We found that whereas infants in the PL condition on the first visit exhibited racial categorization, infants in the OT condition in their first visit did not. Moreover, these patterns remained a week later despite the change in substance. Thus, OT inhibited racial categorization when infants first encountered the to-be-categorized faces. These findings highlight the role of affiliative motivation in social categorization, and suggest that the neurobiology of affiliation may provide insights on mechanisms that may be involved in the downstream prejudicial consequences of intergroup bias.     

Immediate and long-term retention of basic-level categories in 24-month-olds

This study assessed the effect of presenting single versus multiple exemplars of basic-level categories on 24-month-olds' long-term memory for categorical information. Sixty-four infants were tested in a paired-comparison recognition memory paradigm immediately and one week after familiarization with basic-level categories. Infants were randomly assigned to one of two familiarization conditions and one of two test conditions. In one familiarization condition (Varied Exemplar), they briefly viewed four different exemplars of each of 16 basic-level categories. In the other familiarization condition (Single Exemplar), they viewed only one instance of each category on four different trials. Test trials consisted of either an exemplar seen during familiarization (Familiar-Exemplar Test) or novel intracategory exemplar (Unfamiliar-Exemplar Test) paired with a stimulus from a novel category. Both immediately following familiarization and one week later, infants looked longer at test stimuli from the novel category irrespective of familiarization or test condition.     

Linguistic Distributional Knowledge and Sensorimotor Grounding both Contribute to Semantic Category Production

The human conceptual system comprises simulated information of sensorimotor experience and linguistic distributional information of how words are used in language. Moreover, the linguistic shortcut hypothesis predicts that people will use computationally cheaper linguistic distributional information where it is sufficient to inform a task response. In a pre-registered category production study, we asked participants to verbally name members of concrete and abstract categories and tested whether performance could be predicted by a novel measure of sensorimotor similarity (based on an 11-dimensional representation of sensorimotor strength) and linguistic proximity (based on word co-occurrence derived from a large corpus). As predicted, both measures predicted the order and frequency of category production but, critically, linguistic proximity had an effect above and beyond sensorimotor similarity. A follow-up study using typicality ratings as an additional predictor found that typicality was often the strongest predictor of category production variables, but it did not subsume sensorimotor and linguistic effects. Finally, we created a novel, fully grounded computational model of conceptual activation during category production, which best approximated typical human performance when conceptual activation was allowed to spread indirectly between concepts, and when candidate category members came from both sensorimotor and linguistic distributional representations. Critically, model performance was indistinguishable from typical human performance. Results support the linguistic shortcut hypothesis in semantic processing and provide strong evidence that both linguistic and grounded representations are inherent to the functioning of the conceptual system. All materials, data, and code are available at https://osf.io/vaq56/ .     

基于语义网络的创造性认知机制研究现状与展望

语义网络是主要的知识表征方式之一,对创造性研究具有重要意义。近年来不断发展的网络科学方法为相关研究提供了强大的方法论框架及量化工具,研究者得以分别从语义网络结构及语义相关认知操作的角度探讨语义网络与创造性的内在联系。已有证据表明语义网络中所储存的概念越丰富、概念间的连接通路越多元化或越灵活,则个体创造性水平越高;而个体创造性水平亦受制于个体对概念的检索、抑制、选择、整合及扩展等认知能力。在此方向上,未来研究工作应当更多地思考如何在一个相对整合的认知框架中深入探讨语义网络结构及相关认知过程对创造性的交互作用,并结合神经影像等技术手段,将语义网络之于创造性的认知机制嵌套于神经科学体系中进行解释。     

Age-Specific Effects of Lexical–Semantic Networks on Word Production

The lexical–semantic organization of the mental lexicon is bound to change across the lifespan. Nevertheless, the effects of lexical–semantic factors on word processing are usually based on studies enrolling young adult cohorts. The current study aims to investigate to what extent age-specific semantic organization predicts performance in referential word production over the lifespan, from school-age children to older adults. In Study 1, we conducted a free semantic association task with participants from six age-groups (ranging from 10 to 80 years old) to compute measures that capture age-specific properties of the mental lexicon across the lifespan. These measures relate to lifespan changes in the Available Richness of the mental lexicon and in the lexical–semantic Network Prototypicality of concrete words. In Study 2, we used the collected data to predict performance in a picture-naming task on a new group of participants within the same age-groups as for Study 1. The results show that age-specific semantic Available Richness and Network Prototypicality affect word production speed while the semantic variables collected only in young adults do not. A richer and more prototypical semantic network across subjects from a given age-group is associated with faster word production speed. The current results indicate that age-specific semantic organization is crucial to predict lexical–semantic behaviors across the lifespan. Similarly, these results also provide cues to the understanding of the lexical–semantic properties of the mental lexicon and to lexical selection in referential tasks.     

Socially emotional brain-inspired cognitive architecture framework for artificial intelligence

Artificial Intelligence of the next generation needs to interact with users socially, convincing them in its ability to understand human minds, including emotions. For this to happen, an artificial emotional intelligence is needed, capable of adequate, believable behavior in social emotional interactions. Building on previous developments, the present work extends the general framework of emotional Biologically Inspired Cognitive Architecture (eBICA: Samsonovich, 2013, 2018), endowing it with fluents describing, in addition to appraisals, somatic markers, feelings, emotions, moods, emotional reactions and biases. Key building blocks that integrate them are moral schemas and semantic maps. The model describes interaction of three factors: plans and commitments, moral and ethical values, and somatic comfort. Learning in this framework includes self-organization of semantic maps that in turn may provide guidance for active humanlike learning. Implications for empirical studies and practical applications are discussed together with the expected impact.     

The formation of a generalized categorization repertoire: effect of training with multiple domains,samples, and comparisons

The present experiment explored the effects of three variables on the spontaneous categorization of stimuli in perceptually distinct and novel domains. Each of six stimulus domains was created by morphing two images that were the domain endpoints. The endpoints of the domains were male and female faces, two abstract drawings, a car and a truck, two banded-elevation satellite land images, a tree and a cat, and two false-color satellite images. The stimulus variants at each end of a domain defined two potential perceptual classes. Training was conducted in a matching-to-sample format and used stimuli from one or two domains, one or three variants per class as samples, and one or three variants per class as comparisons. The spontaneous categorization of stimuli in the untrained stimulus domains showed the emergence of a generalized categorization repertoire. The proportion of spontaneously categorized stimuli in the new domains was positively related to the number of domains and samples used in training, and was inversely related to the number of comparisons used in training. Differential reaction times demonstrated the discriminability of the stimuli in the emergent classes. This study is among the first to provide an empirical basis for a behavior-analytic model of the development of generalized categorization repertoires in natural settings.