Lexical semantics examples

Background

Books and other text documents keep much of the human knowledge. For that reason, it is important to develop computational tools that extract and synthesize information from natural language text with the aim of building large-scale knowledge bases. The task of machine understanding of textual documents mainly parses and transforms unstructured text into a structured representation. This representation should be unambiguous - making it suitable for machine reading and machine interpretation .

With the advent of the Semantic Web, the need for methods that perform automatic extraction of semantic data from texts becomes even more relevant . Angeli and Manning , for example, propose to use the relations extracted from texts to enlarge databases of known facts and to predict facts, introducing the notion of fact similarity. Novel smartphone interfaces using mining techniques such RevMiner that navigates and analyzes reviews are also based on relations extracted from texts.

Information extraction (IE) systems aim to identify structured relations, like tuples, from unstructured sources such as documents or web pages. IE methods can be used to help building knowledge representation models that report relations between words, like ontologies, semantic networks, and thesauri, among others. According to Fader et al. ‘typically, IE systems learn an extractor for each target relation from labeled training examples’. They are usually domain dependent, and their adaptation to a new domain requires manual labor comprising specification and implementation of new patterns of relationships or corpora annotation . Moreover, this approach is not scalable to corpora with a large number of target relationships or where the target relationships cannot be specified in advance .