Natural Language Processing with Few Computational Linguistic Resources: An Experiment with Automatic Sentence Parsing for Amharic Texts

摘要

The amount of work required to start from scratch in developing all aspects of natural language processing for a new language is huge. At the same time there is an urgent need for a variety of applications including local language spell-checkers, word processors, machine translation systems, search engines, etc. For these applications to be developed, the existence of computerized language resources and a well developed framework for research in this area is essential. Tree-banks, Part-of-speech taggers, computerized grammars, lexica, and parsers are all necessary parts of this framework. The study reported in this article describes an attempt to design and implement a prototype of an automatic sentence parser for Amharic text. Amharic is the official government language of Ethiopia and a language for which very few computational linguistic resources exist. To automatically parse sentences, the study used the Inside Outside algorithm with a bottom up chart parsing strategy. The probabilistic context free grammar was used as a grammatical formalism to represent the phrase structure rules of the language. A small sample corpus of 100 four-word sentences was selected from sentences in the language, and has been used to serve as a training and test set In spite of the limited amount of data and other resources available, the experiments show some promising results.