Abstract. Artificial agents trying to achieve communicative goals in situated interactions in the real-world need powerful computational systems for conceptualizing their environment. In order to provide embodied artificial systems with rich semantics reminiscent of human language complexity, agents need ways of both conceptualizing complex compositional semantic structure and actively reconstructing semantic structure, due to uncertainty and ambiguity in transmission. Furthermore, the systems must be open-ended and adaptive and allow agents to adjust their semantic inventories in order to reach their goals. This paper presents recent progress in modeling open-ended, grounded semantics through a unified software system that addresses these problems. 1

A neural cognitive architecture would be an architecture based on simulated neurons, that provided a set of mechanisms for all cognitive behaviour. Moreover, this would be compatible with biological neural behaviour. As a result, such architectures can both form the basis of a fully-fledged AI and help to explain how cognition emerges from a collection of neurons in the human brain. The development of such a neural cognitive architecture is in its infancy, but a protoarchitecture in the form of behaving agents entirely based on simulated neurons is described. These agents take natural language commands, view the environment, plan and act. The development of these agents has led to a series of questions that need to be addressed to advance the development of neural cognitive architectures. These questions include long posed ones where progress has been made, such as the binding and symbol grounding problems; issues about biological architectures including neural models and brain topology; issues of emergent behaviour such as short and long-term Cell Assembly dynamics; and issues of learning such as the stability-plasticity dilemma. These questions can act as a road map for the development of neural cognitive architectures and AIs based on them.

Abstract Naming is a powerful cognitive tool that facilitates categorization by forming an association between words and their referents. There is evidence in child development literature that strong links exist between early word-learning and conceptual development. A growing view is also emerging that language is a cultural product created and acquired through social interactions. Inspired by these studies, this paper presents a novel learning architecture for category formation and vocabulary acquisition in robots through active interaction with humans. This architecture is open-ended and is capable of acquiring new categories and category names incrementally. The process can be compared to language grounding in children at single-word stage. The robot is embodied with visual and auditory sensors for world perception. A human instructor uses speech to teach the robot the names of the objects present in a visually shared environment. The robot uses its perceptual input to ground these spoken words and dynamically form/organize category descriptions in order to achieve better categorization. To evaluate the learning system at word learning and category formation tasks, two experiments were conducted using a simple language game involving naming and corrective feedback actions from the human user. The obtained results are presented and discussed in detail.

Aiding categorization by grounding spoken words- an infant inspired approach to concept formation and language acquisition

Although many researchers have suggested that compositional concepts should be sensorimotor grounded, the method to accomplish this remains unclear. This paper introduces a second-order neural network with parametric biases (sNNPB) that learns compositional structures based on sensorimotor time series data. The data was produced by a simulated robot that executed distinct object interactions (moveto and orient-toward). We show that various sNNPB setups can learn to compositionally imitate object-interactions beyond the interactions that were specifically trained, which was not possible with previous neural network (NN) architectures, including recurrent neural networks (RNNs). We also show that these imitation capabilities are accomplished by developing a self-organized, geometrically-arranged compositional concept structure in the PB values and task-oriented, Braitenberglike sensory encodings in hidden sensory layers. Since second-order connections were necessary to accomplish the task, we hypothesize that such connections may be essential to drive the learning of both sensorimotor-grounded compositional

Recently cultural theories of language evolution have gained significant momentum in explaining natural language. This paper reviews agent-based modeling, one of the key methodologies which is in part responsible for these developments. We discuss the most important challenges for a theory of cultural language evolution and the resulting dominant experimental paradigm. The discussion is framed along examples of experiments conducted within the methodology. We focus, in particular, on spatial language as an example of a complex and cognitively central domain treated in a series of robotic experiments. 1. Cultural Theories of Language Evolution Cultural theories of language evolution trace, explain and model the cultural development of the languages of the world. Explaining both the past and present of language is a daunting goal. The languages of the world have developed into ingenious systems for communicating enormous subtleties about the inner and outer world. An example of a part of language in which this creativity is very tangible is spatial language (Levinson, 2003; Svorou, 1994; Levinson & Wilkins, 2006). We know now from different studies in spatial language that human languages vary tremendously in how people, for instance, talk about the spatial configuration of objects. The following phrase from Tzeltal which is a Mayan language shows an example of a geocentric spatial language strategy (Brown & Levinson, 1993). (1) ay

Grounding the meanings in sensorimotor behavior using reinforcement learning