Extending an Indonesian Semantic Analysis-based Question Answering System with Linguistic and World Knowledge Axioms

Extending an Indonesian Semantic Analysis-based Question Answering System with Linguistic and World Knowledge Axioms* Rahmad Mahendra a, Septina Dian Larasati a, and Ruli Manurung a aFaculty of Computer Science, University of Indonesia, Depok 16424, Indonesia rama42@ui.edu, septina.larasati@gmail.com, maruli@cs.ui.ac.id Abstract. We adopt a previously developed model of deep syntactic and semantic processing to support question answering for Bahasa Indonesia, and extend it by adding a number of axioms designed to encode useful knowledge for answering questions, thus increasing the inferential power of the QA system. We believe this approach can increase the robustness of semantic analysis-based QA systems, whilst simultaneously lightening the burden of complexity in designing semantic attachment rules that transduce logical forms from syntactic structures. We show how these added axioms enable the system to answer questions which previously could not have been answered. Keywords: bahasa Indonesia, knowledge representation, QA systems, semantic analysis. 1. Introduction A question anwering (QA) system seeks to provide answers to questions expressed in natural language, where the answers are to be found in a given collection of documents. QA systems typically require more sophisticated linguistic analysis than conventional information retrieval, as they need to reason about various other factors, among others the types of questions, predicate argument structure, and result aggregation. In the work presented in this paper, we start from a unification-based grammar augmented with lambda-calculus rules that constructs semantic representations of Indonesian declarative sentences. To these representations we subsequently combine a suite of axioms designed to encode linguistic and world knowledge, and assert them into a knowledge base. A separate QA module answers queries by unifying the question semantic representation with the augmented set knowledge base. In Sections 2 and 3 we first discuss some relevant past work. Section 4 presents the overall framework of our system, and Section 5 discusses the semantic representation underlying our approach, arguing for some form of axiomatic post-processing. Finally, Sections 6 and 7 present the axioms themselves, along with some examples of how they contribute to the QA process. 2. Lightweight semantic approaches to Question Answering In general, there are two approaches to QA: the bottom-up approach employs “shallow” statistical methods such as keyword-based retrieval, which benefit from the sheer size of large electronic collections of documents nowadays available (e.g. the web), and are very robust. Unfortunately, these probabilistic methods are sometimes unable to perform the required inference for answering complex questions. On the other hand, the top-down approach uses “deeper” linguistic methods to obtain semantic representations of both the question and (a subset of) documents. The resulting logical forms enable precise identification of answers, sometimes in cases where they are not explicitly stated in the source documents. However, these * Copyright 2008 by Rahmad Mahendra, Septina Dian Larasati, and Ruli Manurung 262 22nd Pacific Asia Conference on Language, Information and Computation, pages 262–271 deep methods typically require carefully-engineered, language-specific resources that are very costly to produce and not very robust. More recently, work has been done in developing QA systems that try to combine the …

Content truncated to 500 words