On grounding language with neural networks

The author attempts to answer the following questions: Is there evidence for symbols in cognition-if yes, where call we find them? Can neural networks help us in building models of grounded cognition? How can robots understand language? These questions deal with three issues that usually receive quite independent treatment in artificial intelligence (AI) and cognitive science: robotics, language, and symbols. The aim of the article is to show that these issues have much in common, by showing that: symbols-if at all-can be found in language and hardly anywhere else; instead of symbol grounding in the sense of S. Harnad (1990), cognitive science should focus on language grounding (via the prior problem of grounding concepts); robots-or at least situated beings, simulated or physically embodied-should be the main test bed for it