Sebastian Walter

Evaluation of a layered approach to question answering over linked data

We present a question answering system architecture which processes natural language questions in a pipeline consisting of five steps: i) question parsing and query template generation, ii) lookup in an inverted index, iii) string similarity computation, iv) lookup in a lexical database in order to find synonyms, and v) semantic similarity computation. These steps are ordered with respect to their computational effort, following the idea of layered processing: questions are passed on along the pipeline only if they cannot be answered on the basis of earlier processing steps, thereby invoking computationally expensive operations only for complex queries that require them. In this paper we present an evaluation of the system on the dataset provided by the 2nd Open Challenge on Question Answering over Linked Data (QALD-2). The main, novel contribution is a systematic empirical investigation of the impact of the single processing components on the overall performance of question answering over linked data.

A Corpus-Based Approach for the Induction of Ontology Lexica

While there are many large knowledge bases (e.g. Freebase, Yago, DBpedia) as well as linked data sets available on the web, they typically lack lexical information stating how the properties and classes are realized lexically. If at all, typically only one label is attached to these properties, thus lacking any deeper syntactic information, e.g. about syntactic arguments and how these map to the semantic arguments of the property as well as about possible lexical variants or paraphrases. While there are lexicon models such as lemon allowing to define a lexicon for a given ontology, the cost involved in creating and maintaining such lexica is substantial, requiring a high manual effort. Towards lowering this effort, in this paper we present a semi-automatic approach that exploits a corpus to find occurrences in which a given property is expressed, and generalizing over these occurrences by extracting dependency paths that can be used as a basis to create lemon lexicon entries. We evaluate the resulting automatically generated lexica with respect to DBpedia as dataset and Wikipedia as corresponding corpus, both in an automatic mode, by comparing to a manually created lexicon, and in a semi-automatic mode in which a lexicon engineer inspected the results of the corpus-based approach, adding them to the existing lexicon if appropriate.

Survey on challenges of Question Answering in the Semantic Web

Semantic Question Answering (SQA) removes two major access requirements to the Semantic Web: the mastery of a formal query language like SPARQL and knowledge of a specific vocabulary. Because of the complexity of natural language, SQA presents difficult challenges and many research opportunities. Instead of a shared effort, however, many essential components are redeveloped, which is an inefficient use of researcher’s time and resources. This survey analyzes 62 different SQA systems, which are systematically and manually selected using predefined inclusion and exclusion criteria, leading to 72 selected publications out of 1960 candidates. We identify common challenges, structure solutions, and provide recommendations for future systems. This work is based on publications from the end of 2010 to July 2015 and is also compared to older but similar surveys.

Applying Semantic Parsing to Question Answering Over Linked Data: Addressing the Lexical Gap

Question answering over linked data has emerged in the past years as an important topic of research in order to provide natural language access to a growing body of linked open data on the Web. In this paper we focus on analyzing the lexical gap that arises as a challenge for any such question answering system. The lexical gap refers to the mismatch between the vocabulary used in a user question and the vocabulary used in the relevant dataset. We implement a semantic parsing approach and evaluate it on the QALD-4 benchmark, showing that the performance of such an approach suffers from training data sparseness. Its performance can, however, be substantially improved if the right lexical knowledge is available. To show this, we model a set of lexical entries by hand to quantify the number of entries that would be needed. Further, we analyze if a state-of-the-art tool for inducing ontology lexica from corpora can derive these lexical entries automatically. We conclude that further research and investments are needed to derive such lexical knowledge automatically or semi-automatically.

M-ATOLL: A Framework for the Lexicalization of Ontologies in Multiple Languages

Many tasks in which a system needs to mediate between natural language expressions and elements of a vocabulary in an ontology or dataset require knowledge about how the elements of the vocabulary (i.e. classes, properties, and individuals) are expressed in natural language. In a multilingual setting, such knowledge is needed for each of the supported languages. In this paper we present M-ATOLL, a framework for automatically inducing ontology lexica in multiple languages on the basis of a multilingual corpus. The framework exploits a set of language-specific dependency patterns which are formalized as SPARQL queries and run over a parsed corpus. We have instantiated the system for two languages: German and English. We evaluate it in terms of precision, recall and F-measure for English and German by comparing an automatically induced lexicon to manually constructed ontology lexica for DBpedia. In particular, we investigate the contribution of each single dependency pattern and perform an analysis of the impact of different parameters.

ATOLL-A framework for the automatic induction of ontology lexica

There is a range of large knowledge bases, such as Freebase and DBpedia, as well as linked data sets available on the web, but they typically lack lexical information stating how the properties and classes they comprise are realized lexically. Often only one label is attached, if at all, thus lacking rich linguistic information, e.g. about morphological forms, syntactic arguments or possible lexical variants and paraphrases. While ontology lexicon models like lemon allow for defining such linguistic information with respect to a given ontology, the cost involved in creating and maintaining such lexica is substantial, requiring a high manual effort. Towards lowering this effort we present ATOLL, a framework for the automatic induction of ontology lexica, based both on existing labels and on dependency paths extracted from a text corpus. We instantiate ATOLL with respect to DBpedia as dataset and Wikipedia as corresponding corpus, and evaluate it by comparing the automatically generated lexicon with a manually constructed one. Our results clearly corroborate that our approach shows a high potential to be applied in a semi-automatic fashion in which a lexicon engineer can validate, reject or refine the automatically generated lexical entries, thus having a clear potential to contributing to the reduction of the overall cost of creating ontology lexica.

Automatic Acquisition of Adjective Lexicalizations of Restriction Classes: a Machine Learning Approach

There is an increasing interest in providing common Web users with access to structured knowledge bases such as DBpedia, for example by means of question answering systems. An essential task of such systems is transforming natural language questions into formal queries, e.g. expressed in SPARQL. To this end, such systems require knowledge about how the vocabulary elements used in the available ontologies and datasets are verbalized in natural language, covering different verbalization variants, possibly in multiple languages. An important part of such lexical knowledge is constituted by adjectives. In this paper, we present and evaluate a machine learning approach to extract adjective lexicalizations from DBpedia. This is a challenge that has so far not been addressed. Our approach achieves an accuracy of

Question Answering over Linked Data (QALD-4)

91.15 \%