nils reiter

Abstract: This paper presents a supervised approach for identifying generic noun phrases in context. Generic statements express rule-like knowledge about kinds or events. Therefore, their identification is important for the automatic construction of knowledge bases. In particular, the distinction between generic and non-generic statements is crucial for the correct encoding of generic and instance-level information.
Generic expressions have been studied extensively in formal semantics.
Building on this work, we explore a corpus-based learning approach for identifying generic NPs, using selections of linguistically motivated features. Our results perform well above the baseline and existing prior work.

Abstract: This paper gives an overview of an interdisciplinary research project that is concerned with the application of computational linguistics methods to the analysis of the structure and variance of rituals, as investigated in ritual science. We present motivation and prospects of a computational approach to ritual research, and explain the choice of specific analysis techniques. We discuss design decisions for data collection and processing, present the general NLP architecture, and give first results on the automatic exploitation of the domain.

Abstract: This chapter is concerned with lexical enrichment of ontologies, i.e. how to enrich a given ontology with lexical information derived from a semantic lexicon such as WordNet or other lexical resources. We present an approach towards the integration of both types of resources, in particular for the human anatomy domain as represented by the Foundational Model of Anatomy and for the molecular biology domain as represented by an ontology of biochemical substances. The chapter describes our approach on enriching these biomedical ontologies with information derived from WordNet and Wikipedia by matching ontology class labels to entries in WordNet and Wikipedia. In the first case we acquire WordNet synonyms for the ontology class label, whereas in the second case we acquire multilingual translations as provided by Wikipedia. A particular point of emphasis here is on selecting the appropriate interpretation of ambiguous ontology class labels through sense disambiguation, which we address by use of a simple algorithm that selects the most likely sense for an ambiguous term by statistical significance of co-occurring words in a domain corpus. Acquired synonyms and translations are added to the ontology by use of the LingInfo model, which provides an ontology-based lexicon model for the annotation of ontology classes with (multilingual) terms and their linguistic properties.

Abstract: The applicability of ontologies for natural language processing depends on the ability to link ontological concepts and relations to their realisations in texts. We present a general, resource-poor account to create such a linking automatically by extracting Wikipedia articles corresponding to ontology classes. We evaluate our approach in an experiment with the Music Ontology.

Abstract: This paper is concerned with lexical enrichment of ontologies, i.e. how to enrich a given ontology with lexical entries derived from a semantic lexicon. We present an approach towards the integration of both types of resources, in particular for the human anatomy domain as represented by the Foundational Model of Anatomy (FMA). The paper describes our approach on combining the FMA with WordNet by use of a simple algorithm for domain-specific word sense disambiguation, which selects the most likely sense for an FMA term by computing statistical significance of synsets on a corpus of Wikipedia pages on human anatomy. The approach is evaluated on a benchmark of 50 ambiguous FMA terms with manually assigned WordNet synsets (i.e. senses).

PDF BibTeX

Abstract: Most end-user applications of natural language processing such as question answering or information retrieval – and especially a research-oriented task like textual entailment – need to process entailment and contradiction in one way or another. In order to detect entailment and contradiction, a system needs to access very different kinds of resources, such as lexicons, semantic word nets or knowledge ontologies.

One such knowledge ontology is SUMO. SUMO represents upper-model knowledge in an abstract way, using a lisp-like format whose expressivity equals that of first-order logic. A direct connection between the syntactic level in the form of free or syntactically preprocessed text and the semantics in the form of SUMO does not exist. There is no obvious way to represent the information contained in a text or sentence using SUMO.

This thesis is concerned with this syntax-semantics interface. We will propose an algorithm that combines FrameNet and SUMO. Methods and tools to achieve FrameNet annotations of a given sentence do exist already. Our combination algorithm is based on this FrameNet annotation and links frames to SUMO concepts and frame elements to SUMO relations.

PDF BibTex

Abstract: This paper discusses our contribution to the third RTE Challenge – the SALSA RTE system. It builds on an earlier system based on a relatively deep linguistic analysis, which we complemented with a shallow overlap component based on word overlap. We evaluate their (combined) performance on various data sets. However, earlier observations that the combination of features improves the overall accuracy could be replicated only partly.

PDF Bibtex