Automatic Amharic text News Classification: A Neural Networks Approach
No Thumbnail Available
Date
2005-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Parsing is important in Natural Language Processing. A parser allows a
collection of Sentences to be analyzed in terms of its well- formlessness
according to a per-defined grammar. A parser can be applied in Machine
translation, spelling correction, grammar checking, summary, speech
processing, question-answering systems. This paper presents a rule based
parser for Amharic sentence disambiguation.
The parser analyzes structurally ambiguous sentences with a rule base
method. Using an active chart, the parser first generates all possible
parses of a sentence according to given grammar rules. It employees
depth-first search strategy in invoking rules and bottom up parsing
strategy in constructing parses. In order to resolve structurally
ambiguous sentences, the parser incorporates a dictionary that
containing the headwords of noun phrases and verb phrases and their
categories that is used to uniquely identify the groups of headwords.
Structural rules are also provided for each category of the headwords in
the dictionary.
The performance of the parser is measured with accuracy, which is
evaluated by comparing the automatically parsed sentences against the
hand-parsed sentences in the test set. The result shows that 86% of the
sentences are parsed correctly . The result achieved is very encouraging.
Given the small sample size available for rules induction, the parser can
be improved by evaluating it on a large collection of Amharic sentences in
future .
Description
Keywords
Information Science