Common use of Approaches Clause in Contracts

Approaches. For PANACEA it was decided that the majority of SCF acquisition would focus on “inductive classifiers” as exemplified by (Messiant, 2008). Such classifiers are relatively domain- independent because they take parsed data as input and learn SCFs based on observed verbal argument patterns, without any preconceived inventory of SCFs. They are also lightweight and suitable for large scale web service provision. The goal here was not necessarily to improve on state-of-the-art accuracy, but to make the tools available in a lexical acquisition platform. Two different inductive classifiers have been developed, using slightly different methods for deciding on the features to be included in the SCFs. The initial versions of these inductive classifiers were developed and tested on two different languages, Italian and English, respectively, but the newest versions implemented generalized, language- and tagset- independent SCF classifiers that can use the native tagset of the parser output to learn SCFs. These classifiers adopt the same acquisition methodology as the language-specific ones, but rely on very general extraction rules, with a customizable interface. These classifiers have been deployed as web services and integrated in the PANACEA platform. Using these classifiers, SCF lexica for three languages (Italian, English, and Spanish) have been acquired. In addition to inductive classifiers, several other approaches have been pursued within PANACEA. The remainder of this section details different approaches to SCF acquisition which have been investigated in PANACEA. When web services have been deployed, they are described under each approach. A full summary table of web services is collected in Section 3. The major papers associated with each approach are summarized here. A full list of papers is collected in Section 2.7, and the papers themselves are available in Annex A.