Description: This paper discusses creating annotated data collections. Abstract: Open Mind Word Expert is an implemented active learning system that aims to create large annotated corpora by tapping into the world's vast pool of knowledge. It does this by relying on the vast number of Web users who contribute their knowledge to data annotation. Open Mind Word Expert focuses on building semantically annotated corpora, by collecting word sense tagging from the general public over the Web. During the first nine months of activity, the system yielded 90,000 high quality tagged items at a much lower cost than the traditional method of hiring lexicographers.

Description: This paper discusses word sense disambiguation. Abstract: It is generally agreed that the success of a Word Sense Disambiguation (WSD) system depends, in large, on having enough sense annotated data available at hand, and a well-motivated sense inventory into which the disambiguations are made. The authors report a Web-based approach to (1) constructing large sense tagged corpora by exploiting agreement of Web users who contribute word sense annotation, and (2) deriving a coarse-grained sense inventory from a fine-grained inventory by exploiting disagreements of independent contributors about word senses. The authors investigate the quantity and quality of the sense tagged data collected with this approach over the past year. The authors present and evaluate an automatic clustering algorithm able to derive sense clusters that compare well with manually constructed clusters.

Description: This paper describes an unsupervised approach for natural language disambiguation, applicable to ambiguity problems where classes of equivalence can be defined over the set of words in a lexicon. Lexical knowledge is induced from non-ambiguous words via classes of equivalence, and enables the automatic generation of annotated corpora. The only requirements are a lexicon and a raw textual corpus. The method was tested on two natural language ambiguity tasks in several languages: part of speech tagging (English, Swedish, Chinese), and word sense disambiguation (English, Romanian). Classifiers trained on automatically constructed corpora were found to have a performance comparable with classifiers that learn from expensive manually annotated data.