Assigning Inflectional Paradigms to Named Entities by Linear Successive Abstraction

This paper describes how a supervised learning method is used for assigning inflectional paradigms to organizational named entities as the main prerequisite for generating a morphological lexicon of these entities. An inflectional paradigm consists of a set of rules for generating all forms of a lexicon entry. A morphological lexicon consists of lexicon entries and their corresponding forms. This type of language resource is crucial in tasks such as natural language generation (generating natural language business news from database data and news templates) and named entity identification (necessary step in data mining and business intelligence). The basic resource used in this research is a list of 106.530 named entities of organizations given in basic form (nominative) and ranked by relevance. On the first 5.000 manually tagged named entities 59 inflectional paradigm classes are defined. Using linear successive abstraction, a suffix model is trained, validated and tested on this tagged dataset. Morphological lexica of general language, personal names and settlements are used as additional resources in the decision process. The achieved accuracy on the test set is 98,70%.