A research team at Carnegie Mellon University (CMU), is running the Never-Ending Language Learning (NELL) system based on a self-learning crawler and an attached machine learning system. It categorises content (or extracts entities) into a couple of hundred predefined categories and tries to attach basic relationships (again predefined) between them. While it is able to revise assumptions, some mistakes are still corrected by humans, which means the system is supervised in two important aspects:
- predefined semantic categories / entity sets and relations.
- manually corrected relationships
Retrieval is implemented via a minimalistic Question Answering system that currently acts more like a search engine.
In contrast to IBM Watson, the CMU NELL is not focusing on one topic but is driven by a type of curiosity, i.e. it decides what to learn and look for on its own. Also, it seems there is no economic force behind the system.
From the Cognitive Workbench perspective, CMU NELL could be build using the following components: