Ontology-based Natural Language Understanding

Research Areas: 

The OntoNLU project works on improving domain-specific natural language understanding by combining research on a linguistically sophisticated interpretation process with the development of tools for automatic, easy portability across domains.


Methods and Research Questions: 

A basic aspect of natural human-machine interaction is the use of natural language. But in order to conversationally interact with humans, machines need an adequate and flexible way of understanding natural language input and linking it to the knowledge they possess.

OntoNLU focuses on the design and implementation of an interpretation process that employs linguistically rich grammar formalisms in order to align natural language input with a given ontology. This alignment is achieved by means of a lexicon-ontology interface that specifies possible verbalizations of ontology concepts. This ontological lexicon serves as basis for the generation of domain-specific grammars, which can then be used for parsing and interpretation, resulting in meaning representations that are tailored towards a given domain. OntoNLU develops tools for the automatic creation of such ontology-specific grammars, in order to ensure easy portability to new domains, including large and heterogenous datasets.

Interpreting natural language expressions with respect to a specific ontology helps to tackle one of the major challenges for the automatic interpretation of natural language: the prevalence of vague and ambiguous expressions. Their interpretation often heavily relies on linguistic context and background knowledge. Exploiting domain ontologies as context can help to specify vague expressions and resolve ambiguities, and also opens up a range of promising possibilities, such as the incorporation of reasoning capabilities in the interpretation process.


One application of the techniques employed by OntoNLU is ontology-based question answering, i.e. the development of query interfaces to databases that are based on natural language rather than keywords, thereby allowing the user to express complex information needs in an intuitive fashion. A particular outcome of OntoNLU is the question answering system Pythia. One of the major goals is to scale Pythia to the constantly growing amount of heterogenuous linked data on the Semantic Web.