---
title: "Classifying Amharic News Text Using Self-Organizing Maps (2005)"
description: "The paper tackles information access for Amharic, the Ethiopian lingua franca. 10.5% precision. 26 references, cited by 2 papers here."
authors: ["Samuel Eyassu", "Björn Gambäck"]
year: 2005
venue: "Proceedings of the ACL Workshop on Computational Approaches to Semitic Languages"
source: "https://aclanthology.org/W05-0710/"
licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)"
licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/eyassu-2005-classifying-amharic-news-text-self/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/eyassu-2005-classifying-amharic-news-text-self/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2005](https://md.research.elamur.ai/free/en/papers/2005/) / Classifying Amharic News Text Using Self-Organizing Maps

# Classifying Amharic News Text Using Self-Organizing Maps

**Authors:** Samuel Eyassu and [Björn Gambäck](https://md.research.elamur.ai/free/en/authors/bjorn-gamback/) · **Year:** 2005 · **Venue:** [Proceedings of the ACL Workshop on Computational Approaches to Semitic Languages](https://md.research.elamur.ai/free/en/venues/acl-workshop-on-computational-approaches-to-semitic-languages/) ([ACL Workshop on Computational Approaches to Semitic Languages 2005](https://md.research.elamur.ai/free/en/venues/acl-workshop-on-computational-approaches-to-semitic-languages/2005/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/)

**Links:** [Publisher's page](https://aclanthology.org/W05-0710/) · [PDF](https://aclanthology.org/W05-0710.pdf) · [OpenAlex](https://openalex.org/W1964428610)

**Citations:** 20 citations (OpenAlex, 2026-10-14) · 2 in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper tackles information access for Amharic, the Ethiopian lingua franca. Amharic is written in its own Ge'ez-derived syllabic script and had few language resources in 2005. The authors map Amharic queries onto a small collection of Amharic news items using Self-Organizing Maps (SOM), an unsupervised neural network built with the Matlab Neural Network Toolbox. They selected 101 training documents, 105 test documents and 25 queries. From these they built a term–document matrix, weighted it with log-entropy, and compressed it with SVD before training the map. Documents landing near a query on the map were ranked by cosine similarity.

Raw unnormalised vectors performed very poorly, at 10.5% precision. The SVD-reduced space reached 60.0% eleven-point average precision on unseen data. A further setup treated each query's content as a class label, which gave 72.8% on training data and 69.5% on test data. This edges past earlier vector-space and LSI results on Amharic. The work matters as an early baseline for retrieval in an under-resourced African Semitic language.

## Key points
- Data: 101 training news items, 105 test items and 25 queries, with 16 queries used in testing. Relevance was judged by experts.
- Pipeline: a term–document matrix with log-entropy weighting, then SVD reduction. Queries were shortened using the global weight vector taken from the training matrix.
- Retrieval precision was 10.5% with the plain matrix versus 60.0% with SVD-reduced vectors. Average classification accuracy peaked around 5,000 training epochs, out of a sweep from 100 to 20,000.
- Classification scored 72.8% on training and 69.5% on test data. Earlier Amharic work by Theodros (GebreMeskel, 2003) reported 69.1% with the vector space model and 71.6% with LSI on training data. LSI on Reuters-21578 reaches 88–90%, but that corpus is about 100 times larger.
- Script facts: there are 33 base symbols, each with seven vowel "orders", plus labialised forms, for 275 fidEls in total. Unicode support for the script arrived only in 2000.

## Interesting ideas
- The authors manually inspected misses and found a pattern: rejected documents held only a line or two relevant to the query, while false hits merely shared co-occurring words.
- Documents relevant to two topics landed on map nodes where the two topic regions meet, so position on the grid itself exposed the ambiguity.
- New queries are projected into the existing space using weights learned on the training collection, so nothing is recomputed.
- The same SOM was reused in two modes: an unsupervised retrieval map, and a supervised version where the query content acts as the class label. Both were compared against prior vector-space and LSI numbers on the same language.
- Instead of fixing a training length in advance, the authors swept epochs widely and kept the best-scoring network, observing a clear peak.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (learn): Evaluation reports could sort each failure into "relevant signal too sparse" or "spurious lexical overlap". This would separate content-coverage errors from shallow-similarity errors.
- **I2** (world): When placing entities or events in the ontology, an item sitting between two class regions could be flagged as genuinely dual. It would then be kept with both readings rather than forced into one class.
- **I3** (world): Compiled knowledge such as verb-class or WordNet-derived similarity spaces could absorb new lexical items in the same way. Each new word would be projected through fixed global weights instead of triggering a full rebuild.
- **I4** (learn): A component trained without labels could be re-run with silver labels from witness parsers. Comparing the two modes against a simple baseline would show how much the labels actually add.
- **I5** (tree): Training of the perceptron parser or gparse could record such accuracy-versus-epoch curves on held-out data. This would document where performance saturates and justify the chosen checkpoint.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Swedish Institute of Computer Science](https://md.research.elamur.ai/free/en/ngos/swedish-institute-of-computer-science/) — Nonprofit, Kista, Sweden · 6 works · 649 citations
- [Addis Ababa University](https://md.research.elamur.ai/free/en/universities/addis-ababa-university/) — University, Addis Ababa, Ethiopia · 15 works · 77 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Samuel Eyassu and Björn Gambäck. 2005. Classifying Amharic News Text Using Self-Organizing Maps. Proceedings of the ACL Workshop on Computational Approaches to Semitic Languages.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{eyassu2005classifying,
  title = {{Classifying Amharic News Text Using Self-Organizing Maps}},
  author = {Samuel Eyassu and Björn Gambäck},
  booktitle = {Proceedings of the ACL Workshop on Computational Approaches to Semitic Languages},
  year = {2005},
  doi = {10.3115/1621787.1621801},
  url = {https://doi.org/10.3115/1621787.1621801}
}
```

- [← previous A Probabilistic Setting and Lexical Coocurrence Model for Textual Entailment](https://md.research.elamur.ai/free/en/papers/glickman-2005-probabilistic-setting-lexical-coocurrence-model/)
- [next → Modifying a Natural Language Processing System for European Languages to Treat Arabic in Information Processing and Information Retrieval Applications](https://md.research.elamur.ai/free/en/papers/grefenstette-2005-modifying-natural-language-processing-system/)
- [↑ back to papers of 2005](https://md.research.elamur.ai/free/en/papers/2005/)

## References

26 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Nega Alemayehu and Peter Willett. 2002. Stemming of Amharic words for information retrieval. Literary and Linguistic Computing, 17(1):1–17. [↗](https://doi.org/10.1093/llc/17.1.1)
2. Atelach Alemu, Lars Asker, and Gunnar Eriksson. 2003a. An empirical approach to building an Amharic treebank. In Proc. 2nd Workshop on Treebanks and Linguistic Theories, Växjö University, Sweden. [find](https://scholar.google.com/scholar?q=Atelach+Alemu%2C+Lars+Asker%2C+and+Gunnar+Eriksson%2E+2003a%2E+An+empirical+approach+to+building+an+Amharic+treebank%2E+In+Proc%2E+2nd+Workshop+on+Treebanks+and+Linguistic+Theories%2C+Va%CC%88xjo%CC%88+University%2C+Sweden%2E)
3. Atelach Alemu, Lars Asker, and Mesfin Getachew. 2003b. Natural language processing for Amharic: Overview and suggestions for a way forward. In Proc. 10th Conf. Traitement Automatique des Langues Naturelles, Batz-sur-Mer, France, pp. 173–182. [find](https://scholar.google.com/scholar?q=Atelach+Alemu%2C+Lars+Asker%2C+and+Mesfin+Getachew%2E+2003b%2E+Natural+language+processing+for+Amharic%3A+Overview+and+suggestions+for+a+way+forward%2E+In+Proc%2E+10th+Conf%2E+Traitement+Automatique+des+Langues+Natur)
4. Atelach Alemu, Lars Asker, Rickard Cöster, and Jussi Karlgren. 2004. Dictionary-based Amharic–English information retrieval. In 5th Workshop of the Cross Language Evaluation Forum, Bath, England. [find](https://scholar.google.com/scholar?q=Atelach+Alemu%2C+Lars+Asker%2C+Rickard+Co%CC%88ster%2C+and+Jussi+Karlgren%2E+2004%2E+Dictionary%2Dbased+Amharic%E2%80%93English+information+retrieval%2E+In+5th+Workshop+of+the+Cross+Language+Evaluation+Forum%2C+Bath%2C+England%2E)
5. Saba Amsalu. 2001. The application of information retrieval techniques to Amharic. MSc Thesis, School of Information Studies for Africa, Addis Ababa University, Ethiopia. [find](https://scholar.google.com/scholar?q=Saba+Amsalu%2E+2001%2E+The+application+of+information+retrieval+techniques+to+Amharic%2E+MSc+Thesis%2C+School+of+Information+Studies+for+Africa%2C+Addis+Ababa+University%2C+Ethiopia%2E)
6. Marvin Bender, Sydney Head, and Roger Cowley. 1976. The Ethiopian writing system. In Bender et al., eds, Language in Ethiopia. Oxford University Press. [find](https://scholar.google.com/scholar?q=Marvin+Bender%2C+Sydney+Head%2C+and+Roger+Cowley%2E+1976%2E+The+Ethiopian+writing+system%2E+In+Bender+et+al%2E%2C+eds%2C+Language+in+Ethiopia%2E+Oxford+University+Press%2E)
7. Michael Berry, Susan Dumais, and Gawin O’Brien. 1995. Using linear algebra for intelligent information retrieval. SIAM Review, 37(4):573–595. [↗](https://doi.org/10.1137/1037127)
8. Thomas Bloor. 1995. The Ethiopic writing system: a profile. Journal of the Simplified Spelling Society, 19:30–36. [find](https://scholar.google.com/scholar?q=Thomas+Bloor%2E+1995%2E+The+Ethiopic+writing+system%3A+a+profile%2E+Journal+of+the+Simplified+Spelling+Society%2C+19%3A30%E2%80%9336%2E)
9. Lijuan Cai and Thomas Hofmann. 2003. Text categorization by boosting automatically extracted concepts. In Proc. 26th Int. Conf. Research and Development in Information Retrieval, pp. 182–189, Toronto, Canada. [find](https://scholar.google.com/scholar?q=Lijuan+Cai+and+Thomas+Hofmann%2E+2003%2E+Text+categorization+by+boosting+automatically+extracted+concepts%2E+In+Proc%2E+26th+Int%2E+Conf%2E+Research+and+Development+in+Information+Retrieval%2C+pp%2E+182%E2%80%93189%2C+Toronto%2C)
10. John Cowell and Fiaz Hussain. 2003. Amharic character recognition using a fast signature based algorithm. In Proc. 7th Int. Conf. Image Visualization, pp. 384–389, London, England. [find](https://scholar.google.com/scholar?q=John+Cowell+and+Fiaz+Hussain%2E+2003%2E+Amharic+character+recognition+using+a+fast+signature+based+algorithm%2E+In+Proc%2E+7th+Int%2E+Conf%2E+Image+Visualization%2C+pp%2E+384%E2%80%93389%2C+London%2C+England%2E)
11. Scott Deerwester, Susan Dumais, George Furnas, Thomas Landauer, and Richard Harshman. 1990. Indexing by latent semantic analysis. Journal of the American Society for Information Science, 41(6):391–407. [find](https://scholar.google.com/scholar?q=Scott+Deerwester%2C+Susan+Dumais%2C+George+Furnas%2C+Thomas+Landauer%2C+and+Richard+Harshman%2E+1990%2E+Indexing+by+latent+semantic+analysis%2E+Journal+of+the+American+Society+for+Information+Science%2C+41%286%29%3A391%E2%80%93407)
12. Susan Dumais. 1991. Improving the retrieval of information from external sources. Behavior Research Methods, Instruments and Computers, 23(2):229–236. [↗](https://doi.org/10.3758/bf03203370)
13. Sisay Fissaha and Johann Haller. 2003. [Application of corpus-based techniques to Amharic texts](https://md.research.elamur.ai/free/en/papers/fissaha-2003-application-corpus-based-techniques-amharic/). In Proc. MT Summit IX Workshop on Machine Translation for Semitic Languages, New Orleans, Louisana.
14. Jane Furzey. 1996. Enpowering socio-economic development in Africa utilizing information technology. A country study for the United Nations Economic Commission for Africa, University of Pennsylvania. [find](https://scholar.google.com/scholar?q=Jane+Furzey%2E+1996%2E+Enpowering+socio%2Deconomic+development+in+Africa+utilizing+information+technology%2E+A+country+study+for+the+United+Nations+Economic+Commission+for+Africa%2C+University+of+Pennsylvania%2E)
15. Theodros GebreMeskel. 2003. Amharic text retrieval: An experiment using latent semantic indexing (LSI) with singular value decomposition (SVD). MSc Thesis, School of Information Studies for Africa, Addis Ababa University, Ethiopia. [find](https://scholar.google.com/scholar?q=Theodros+GebreMeskel%2E+2003%2E+Amharic+text+retrieval%3A+An+experiment+using+latent+semantic+indexing+%28LSI%29+with+singular+value+decomposition+%28SVD%29%2E+MSc+Thesis%2C+School+of+Information+Studies+for+Africa%2C+Ad)
16. Timo Honkela, Samuel Kaski, Krista Lagus, and Teuvo Kohonen. 1997. WEBSOM — Self-Organizing Maps of document collections. In Proc. Workshop on Self-Organizing Maps, pp. 310–315, Espoo, Finland. [↗](https://doi.org/10.1016/s0925-2312(98)00039-3)
17. Teuvo Kohonen, Samuel Kaski, Krista Lagus, Jarkko Salojärvi, Jukka Honkela, Vesa Paatero, and Antti Saarela. 2000. Self organization of a massive document collection. IEEE Transactions on Neural Networks, 11(3):574–585. [↗](https://doi.org/10.1109/72.846729)
18. Teuvo Kohonen. 1999. Self-Organization and Associative Memory. Springer, 3 edition. [find](https://scholar.google.com/scholar?q=Teuvo+Kohonen%2E+1999%2E+Self%2DOrganization+and+Associative+Memory%2E+Springer%2C+3+edition%2E)
19. Xia Lin, Dagobert Soergel, and Gary Marchionini. 1991. A self-organizing semantic map for information retrieval. In Proc. 14th Int. Conf. Research and Development in Information Retrieval, pp. 262–269, Chicago, Illinois. [find](https://scholar.google.com/scholar?q=Xia+Lin%2C+Dagobert+Soergel%2C+and+Gary+Marchionini%2E+1991%2E+A+self%2Dorganizing+semantic+map+for+information+retrieval%2E+In+Proc%2E+14th+Int%2E+Conf%2E+Research+and+Development+in+Information+Retrieval%2C+pp%2E+262%E2%80%93269)
20. Hwee Tou Ng, Wei Boon Goh, and Kok Leong Low. 1997. Feature selection, perceptron learning, and a usability case study for text categorization. In Proc. 20th Int. Conf. Research and Development in Information Retrieval, pp. 67–73, Philadelphia, Pennsylvania. [↗](https://doi.org/10.1145/278459.258537)
21. Miguel Ruiz and Padmini Srinivasan. 1999. Hierarchical neural networks for text categorization. In Proc. 22nd Int. Conf. Research and Development in Information Retrieval, pp. 281–282, Berkeley, California. [find](https://scholar.google.com/scholar?q=Miguel+Ruiz+and+Padmini+Srinivasan%2E+1999%2E+Hierarchical+neural+networks+for+text+categorization%2E+In+Proc%2E+22nd+Int%2E+Conf%2E+Research+and+Development+in+Information+Retrieval%2C+pp%2E+281%E2%80%93282%2C+Berkeley%2C+Calif)
22. David Rumelhart, Geoffrey Hinton, and Ronald Williams. 1986. Learning internal representations by error propagation. In Rumelhart and McClelland, eds, Parallel Distributed Processing, vol 1. MIT Press. [↗](https://doi.org/10.21236/ada164453)
23. Gerard Salton and Michael McGill. 1983. Introduction to Modern Information Retrieval. McGraw-Hill. [find](https://scholar.google.com/scholar?q=Gerard+Salton+and+Michael+McGill%2E+1983%2E+Introduction+to+Modern+Information+Retrieval%2E+McGraw%2DHill%2E)
24. Fabrizio Sebastiani. 2002. [Machine learning in automated text categorization](https://md.research.elamur.ai/free/en/papers/sebastiani-2002-machine-learning-automated-text-categorization/). ACM Computing Surveys, 34(1):1–47. [↗](https://doi.org/10.1145/505282.505283)
25. George Tambouratzis, N. Hairetakis, S. Markantonatou, and G. Carayannis. 2003. Applying the SOM model to text classification according to register and stylistic content. Int. Journal of Neural Systems, 13(1):1–11. [↗](https://doi.org/10.1142/s0129065703001352)
26. Daniel Yacob. 2005. Developments towards an electronic Amharic corpus. In Proc. TALN 12 Workshop on NLP for Under-Resourced Languages, Dourdan, France, June (to appear). [find](https://scholar.google.com/scholar?q=Daniel+Yacob%2E+2005%2E+Developments+towards+an+electronic+Amharic+corpus%2E+In+Proc%2E+TALN+12+Workshop+on+NLP+for+Under%2DResourced+Languages%2C+Dourdan%2C+France%2C+June+%28to+appear%29%2E)

*Updated 1 November 2026*