KLEO is a bootstrapping learning-by-reading system that builds a knowledge base in a fully automated way by reading texts for a domain. KLEO’s initial knowledge base is a small knowledge base that consists of domain independent knowledge and KLEO expands the knowledge base with the information extracted from texts. A key facility in KLEO is knowledge integration which combines new information gleaned from individual sentences of the texts, along with prior knowledge, to form a comprehensive and computationally useful knowledge base. This paper introduces the architecture of KLEO, especially the knowledge integration facility, and presents our evaluation plan. The knowledge acquisition bottleneck has been the major obstacle to building large-scale knowledge bases. Despite enormous past efforts, it is still costly and tedious to build knowledge bases manually. As a solution to this problem, a new approach has been gaining much attention due to the advance of natural language processing and the proliferation of texts on the Internet. The approach is to construct a knowledge base with knowledge extracted from texts. KLEO 1 is a such Learning-by-Reading system which operates in the following steps : 1. It reads a text to form a semantic representation. The knowledge base provides the information required to understand the text. 2. It adds the semantic representation to the knowledge base. Kleo repeats these steps with a corpus of texts. Note that these two steps constitute a bootstrapping cycle in which reading extends the knowledge base (step2) and the extended knowledge base in turn improves the reading performance (step1). A key to this approach is knowledge integration the task of (1) combining semantic representations for individual sentences to form a coherent representation for the text and (2) combining the new information with the prior knowledge base. Knowledge integration is an important facility in the Learning-by-Reading task because, withCopyright c © 2009, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. The web interface to KLEO is available at http://www.cs.utexas.edu/users/onue5 out it, the learned knowledge base is often fragmented, incoherent and hence computationally useless. This paper is not intended to provide technical details on the KLEO system. Rather, it will present an overview of the challenges in building a Learning-by-Reading system, along with a system architecture and evaluation plan for KLEO, focusing on the Knowledge Integration facility.