Ingo Glöckner

An application of automated reasoning in natural language question answering

The LogAnswer system is an application of automated reasoning to the field of open domain question answering. In order to find answers to natural language questions regarding arbitrary topics, the system integrates an automated theorem prover in a framework of natural language processing tools. The latter serve to construct an extensive knowledge base automatically from given textual sources, while the automated theorem prover makes it possible to derive answers by deductive reasoning. In the paper, we discuss the requirements to the prover that arise in this application, especially concerning efficiency and robustness. The proposed solution rests on incremental reasoning, relaxation of the query (if no proof of the full query is found), and other techniques. In order to improve the robustness of the approach to gaps of the background knowledge, the results of deductive processing are combined with shallow linguistic features by machine learning.

Evaluation of quantified propositions in generalized models of fuzzy quantification

Abstract The quantifiers found in natural language (NL) are not restricted to the absolute and proportional types usually considered in fuzzy set theory. In order to handle the wealth of NL quantifiers including quantifiers of exception (“all except about ten”), cardinal comparatives (“many more than”) and others, it is necessary to consider generalized models of fuzzy quantification. Starting from an analysis in terms of semi-fuzzy quantifiers (specifications) and fuzzification mechanisms (prototypical models), the sequel develops a precise notion of generalized models which rests on a formalization of linguistic adequacy criteria. It also presents concrete examples of such models which generalize the FG-count and OWA approaches to fuzzy quantification. In order to let applications profit from the improved coverage and coherence of interpretations, the sequel is especially concerned with the issue of practical implementation. It presents efficient methods for implementing the main types of quantifying propositions which demonstrate the computational feasibility of the proposed models.

An Integrated Approach to Semantic Evaluation and Content-Based Retrieval of Multimedia Documents

We present an overview of a large combined querying and retrieval system that performs content-based on-line searches in a large database of multimedia documents (currently text, tables and colour images). Queries are submitted as sentences in natural language and are transformed into the language of the target database. The documents are analyzed semantically for their information content; in a data fusion step the individual pieces of information extracted from these documents are aggregated into cognitively adequate result documents. There is no pre-indexing necessary when new documents are stored into the system. This retains a high degree of flexibility with respect to the questions that may be asked. It implies, however, that both huge amounts of data must be evaluated rapidly and that intelligent caching strategies must be employed. It is therefore mandatory that the system be equipped with dedicated high-speed hardware processors. The complete system is currently available as a prototype; the paper outlines its architecture and gives examples of some real sample queries in the knowledge domain of weather data documents.

A natural language question answering system as a participant in human Q&A portals

LogAnswer is a question answering (QA) system for the German language, aimed at providing concise and correct answers to arbitrary questions. For this purpose LogAnswer is designed as an embedded artificial intelligence system which integrates methods from several fields of AI, namely natural language processing, machine learning, knowledge representation and automated theorem proving. We intend to employ LogAnswer as a virtual user within Internet-based QA forums, where it must be able to identify the questions that it cannot answer correctly, a task that normally receives little attention in QA research compared to the actual answer derivation. The paper presents a machine learning solution to the wrong answer avoidance (WAA) problem, applying a meta classifier to the output of simple term-based classifiers and a rich set of other WAA features. Experiments with a large set of real-world questions from a QA forum show that the proposed method significantly improves the WAA characteristics of our system.

Efficient question answering with question decomposition and multiple answer streams

The German question answering (QA) system IRSAW (formerly: InSicht) participated in QA@CLEF for the fifth time. IRSAW was introduced in 2007 by integrating the deep answer producer InSicht, several shallow answer producers, and a logical validator. InSicht builds on a deep QA approach: it transforms documents to semantic representations using a parser, draws inferences on semantic representations with rules, and matches semantic representations derived from questions and documents. InSicht was improved for QA@CLEF 2008 mainly in the following two areas. The coreference resolver was trained on question series instead of newspaper texts in order to be better applicable for follow-up questions. Questions are decomposed by several methods on the level of semantic representations. On the shallow processing side, the number of answer producers was increased from two to four by adding FACT, a fact index, and SHASE, a shallow semantic network matcher. The answer validator introduced in 2007 was replaced by the faster RAVE validator designed for logic-based answer validation under time constraints. Using RAVE for merging the results of the answer producers, monolingual German runs and bilingual runs with source language English and Spanish were produced by applying the machine translation web service Promt. An error analysis shows the main problems for the precision-oriented deep answer producer InSicht and the potential offered by the recalloriented shallow answer producers.

Logic-Based Question Answering

Question answering systems aim to provide concise and correct responses to arbitrary questions, communicating with the user in a natural language. This way they help making the knowledge of large textual sources accessible in an intuitive manner which goes beyond the capabilities of conventional search engines. In the LogAnswer project the universities of Hagen and Koblenz cooperate to build a German language question answering system which combines computational linguistics and automated reasoning to deduce answers from a knowledge base derived from Wikipedia.

Combining logic and machine learning for answering questions

LogAnswer is a logic-oriented question answering system developed by the AI research group at the University of Koblenz-Landau and by the IICS at the University of Hagen. The system addresses two notorious problems of the logic-based approach: Achieving robustness and acceptable response times. Its main innovation is the use of logic for simultaneously extracting answer bindings and validating the corresponding answers. In this way the inefficiency of the classical answer extraction/answer validation pipeline is avoided. The prototype of the system, which can be tested on the web, demonstrates response times suitable for real-time querying. Robustness to gaps in the background knowledge and errors of linguistic analysis is achieved by combining the optimized deductive subsystem with shallow techniques by machine learning.

Natural language navigation in multimedia archives: an integrated approach

The paper presents the design and prototypical implementation of an integrated retrieval system (HPQS) which provides natural language access to multimedia documents in restricted topic areas. It supports new flexible ways of querying by combining a semantically rich retrieval model based on fuzzy set theory with domain-specific methods for document analysis which can be applied online (i.e. the search criteria are not restricted to combinations of anticipated descriptors). Emphasis is put on the retrieval methodology and on the interplay of the system components: because of its provision of computationally demanding direct search methods, it is crucial to the system that all components cooperate to ensure acceptable response times.

LogAnswer - A Deduction-Based Question Answering System (System Description)

LogAnswer is an open domain question answering system which employs an automated theorem prover to infer correct replies to natural language questions. For this purpose LogAnswer operates on a large axiom set in first-order logic, representing a formalized semantic network acquired from extensive textual knowledge bases. The logicbased approach allows the formalization of semantics and background knowledge, which play a vital role in deriving answers. We present the functional LogAnswer prototype, which consists of automated theorem provers for logical answer derivation as well as an environment for deep linguistic processing.

Optimal Selection of Proportional Bounding Quantifiers in Linguistic Data Summarization

Summary. Proportional bounding quantifiers like “Between p1 and p2 percent” are potentially useful for expressing linguistic summaries of data. Given p1, p2, existing methods for data summarization based on fuzzy quantifiers can be used to assign a quality score to the summary. However, the problem remains how the optimal choice of p1, p2 in the range 0 ≤ p1 ≤ p2 ≤ 100% can be established. Moreover, the proposed quality indicators are rather heuristic in nature. The paper presents a method for computing the optimal bounding quantifier which best summarizes the given data. Specifically, the most specific quantifier will be chosen which results in the highest validity score of the summary given a constraint on the the percentage range p2 − p1. The method not only assigns validity scores to the quantifiers of interest but also determines the best choice of quantifier in O(N log m) time, where N is the size of the base set and m the number of different membership grades in the fuzzy arguments.