<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0710">
  <Title>Classifying Amharic News Text Using Self-Organizing Maps</Title>
  <Section position="2" start_page="0" end_page="71" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Even though the last years have seen an increasing trend in investigating applying language processing methods to other languages than English, most of the work is still done on very few and mainly European and East-Asian languages; for the vast number of languages of the African continent there still remains plenty of work to be done. The main obstacles to progress in language processing for these are two-fold. Firstly, the peculiarities of the languages themselves might force new strategies to be developed. Secondly, the lack of already available resources and tools makes the creation and testing of new ones more difficult and time-consuming.</Paragraph>
    <Paragraph position="1"> [?]Author for correspondence.</Paragraph>
    <Paragraph position="2"> Many of the languages of Africa have few speakers, and some lack a standardised written form, both creating problems for building language processing systems and reducing the need for such systems. However, this is not true for the major African languages and as example of one of those this paper takes Amharic, the Semitic language used for countrywide communication in Ethiopia. With more than 20 million speakers, Amharic is today probably one of the five largest on the continent (albeit difficult to determine, given the dramatic population size changes in many African countries in recent years).</Paragraph>
    <Paragraph position="3"> The Ethiopian culture is ancient, and so are the written languages of the area, with Amharic using its own script. Several computer fonts for the script have been developed, but for many years it had no standardised computer representation1 which was a deterrent to electronic publication. An exponentially increasing amount of digital information is now being produced in Ethiopia, but no deep-rooted culture of information exchange and dissemination has been established. Different factors are attributed to this, including lack of digital library facilities and central resource sites, inadequate resources for electronic publication of journals and books, and poor documentation and archive collections. The difficulties to access information have led to low expectations and under-utilization of existing information resources, even though the need for accurate and fast information access is acknowledged as a major factor affecting the success and quality of research and development, trade and industry (Furzey, 1996).</Paragraph>
    <Paragraph position="4">  In recent years this has lead to an increasing awareness that Amharic language processing resources and digital information access and storage facilities must be created. To this end, some work has now been carried out, mainly by Ethiopian Telecom, the Ethiopian Science and Technology Commission, Addis Ababa University, the Ge'ez Frontier Foundation, and Ethiopian students abroad. So have, for example, Sisay and Haller (2003) looked at Amharic word formation and lexicon building; Nega and Willett (2002) at stemming; Atelach et al. (2003a) at treebank building; Daniel (Yacob, 2005) at the collection of an (untagged) corpus, tentatively to be hosted by Oxford University's Open Archives Initiative; and Cowell and Hussain (2003) at character recognition.2 See Atelach et al. (2003b) for an overview of the efforts that have been made so far to develop language processing tools for Amharic.</Paragraph>
    <Paragraph position="5"> The need for investigating Amharic information access has been acknowledged by the European Cross-Language Evaluation Forum, which added an Amharic-English track in 2004. However, the task addressed was for accessing an English database in English, with only the original questions being posed in Amharic (and then translated into English).</Paragraph>
    <Paragraph position="6"> Three groups participated in this track, with Atelach et al. (2004) reporting the best results.</Paragraph>
    <Paragraph position="7"> In the present paper we look at the problem of mapping questions posed in Amharic onto a collection of Amharic news items. We use the Self-Organizing Map (SOM) model of artificial neural networks for the task of retrieving the documents matching a specific query. The SOMs were implemented using the Matlab Neural Network Toolbox.</Paragraph>
    <Paragraph position="8"> The rest of the paper is laid out as follows. Section 2 discusses artificial neural networks and in particular the SOM model and its application to information access. In Section 3 we describe the Amharic language and its writing system in more detail together with the news items corpora used for training and testing of the networks, while Sections 4 and 5 detail the actual experiments, on text retrieval and text classification, respectively. Finally, Section 6 sums up the main contents of the paper.</Paragraph>
    <Paragraph position="9"> 2In the text we follow the Ethiopian practice of referring to Ethiopians by their given names. However, the reference list follows Western standard and is ordered according to surnames (i.e., the father's name for an Ethiopian).</Paragraph>
  </Section>
class="xml-element"></Paper>