Retrieval Optimization of Pertinent Answers for NL Questions with the E-Librarian Service Serge Linckels,Harald Sack, Christoph Meinel Hasso Plattner Institute for Software Systems Engineering (HPI) University of Potsdam, Postfach 900460, D-14440 Potsdam, Germany {linckels,meinel,sack}@hpi.uni-potsdam.de Allthough educational content in the WWW is increasing dramatically, its usage in an educational environment is poor, mainly due to the fact that there is too much of (unreliable) redundant and not relevant information. Finding appropriate answers is a rather difficult task being reliant on the user filtering the pertinent information from the noise [2,4]. Turning the WWW into useful educational resources requires to identify correct, reliable, and machine under- standable information, as well as to develop simple but efficient search tools with the ability to perform logical inferences over this information. We present the web-based e-Librarian Service CHESt 1 that is able to under- stand a user’s questions given in natural language (NL) and to retrieve seman- tically pertinent Learning Objects (LOs). By LO we refer to an entity about a precise subject that may be used for learning, education or training [5] such as a video clip including machine processable metadata. The basic building block of CHESt is a domain ontology, which is used for the translation of the NL user questions into Description Logics (DLs) and to provide semantic metadata for the LOs. CHESt implements a retrieval algorithm, which is based on the concept covering problem. Among all the LOs that have some common information with the user query, CHESt identifies the most pertinent match(es), keeping in mind that the user expects an exhaustive answer while preferring a concise answer with only little or no information overhead. In difference to Question Answering [3], our approach is not targeted to com- pute a coherent answer in NL. CHESt simply provides a set of interrelated LOs that contain the information necessary to answer the user’s question. The trans- lation of a NL user question into a DLs expression (including two non-standard DLs inferences such as the least common subsumer and the difference operation) is described in [1]. By LO retrieval we refer to answering a user’s question by identifying only the semantically most pertinent LOs. In addition, our system quantifies the quality of the yielded results by measuring the semantic distance between the user’s query and the identified LOs. Our retrieval algorithm is based on the concept covering problem and on the quantification of the semantic dif- ference. The novelty of our approach is that it always proposes a solution to the user, even if the system concludes that there is no exhaustive answer. By quantifying the missing and supplementary information, the system is able to compute and visualize the quality and pertinence of the yielded LO(s). 1 http://www.linckels.lu/chest