Learning Common Sense Knowledge Through Human-Robot Dialogue
Current deep learning methods excel in pattern recognition and synthesis tasks, particularly in fields such as speech processing and computer vision. Artificial Neural Networks (ANNs) can readily train on high-dimensional, real-world data, which is often noisy. However, these methods have several limitations, including a limited capability for relational reasoning, a critical aspect for Natural Language Processing (NLP) and intelligent assistant applications.
The video presents a hybrid framework that merges ANNs with logic-based learning to facilitate common-sense reasoning. Inductive Logic Programming (ILP) involves the learning of a hypothesis that, in conjunction with background knowledge, explains a set of positive and negative examples. The learned hypothesis, when integrated with the background knowledge, can subsequently be used for inference.
This framework employs a deep learning model to convert audio input into a speech string and extract a dependency tree through natural language parsing. From this dependency tree, a structured representation is derived for reasoning. The system can then infer an answer to a user's question using its knowledge base. If the answer is incorrect, the system can learn a hypothesis using factual knowledge derived from experience and observation.