Elamur Research⌕
☰
home / papers / 2014 / Automatic Construction of Amharic Semantic Networks from Unstructured Text Using Amharic WordNet

Automatic Construction of Amharic Semantic Networks from Unstructured Text Using Amharic WordNet

Paper · 2014 · Alelgn Tefera and Yaregal Assabie

Summary

The paper addresses the lack of any automatic way to build semantic networks for Amharic, a morphologically rich Semitic language. The authors start from a small, hand-built Amharic WordNet of noun synsets. They use it as seed knowledge to discover new "type-of" (hyponymy) and "part-of" (meronymy) links in raw news text.

The pipeline has five steps:

  1. Clean and stem the text with an affix-stripping stemmer.
  2. Index it with TF-IDF.
  3. Build a WordSpace model, compressing a term–document matrix through random projection.
  4. For each WordNet concept, retrieve its top-k neighbours by cosine similarity as candidate related concepts.
  5. Check candidate pairs against lexical patterns. These patterns come from the words found between known WordNet pairs in the corpus, with the concepts replaced by placeholders. A pair is linked when its filled-in pattern occurs often enough.

The resulting network, AMSNet, stores facts as predicates over concept pairs. Linguists judged accuracy at roughly 70%, since no gold standard exists. The work matters as a first step toward lexical-semantic resources for an under-resourced language.

Key points

  • Seed lexicon: the manually built Amharic WordNet holds 890 single-word nouns in 296 synsets, linked by synonymy, "type-of", "part-of" and antonym relations.
  • Corpus: 4325 news documents in total.
    • 1064 POS-tagged items from Walta Information Center were used for concept extraction.
    • 3261 items from the Ethiopian National News Agency were used for frequency counting.
  • Tools: Apache Lucene for indexing, Semantic Vectors for the WordSpace, Java for the extraction steps, and a modified Alemayehu and Willett (2002) stemmer.
  • Patterns: n-grams with 2 ≤ n ≤ 7 matching "C1 X* C2", where X is at most five tokens. The concepts are then generalized to var1/var2 slots.
  • Results: manual evaluation by linguists gave 68.5% average accuracy for "type-of" and 71.7% for "part-of".

Interesting ideas

  • I1. Known lexicon pairs (e.g. a country and its continent) serve as seeds whose connecting phrases become reusable templates for finding new pairs of the same relation. This bootstraps a relation extractor without labelled training data.
  • I2. Candidate pairs are first narrowed by distributional similarity and only then checked with patterns. A cheap filter thus limits how many expensive pattern verifications are needed.
  • I3. The corpus is split so that one part proposes concepts and a separate part supplies the occurrence counts used to accept a relation, which works like held-out confirmation.
  • I4. Relations are stored as predicates R(C1,C2), so properties can be inherited down "type-of" links (if every country has a government, Ethiopia does too).

For Mova

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

  • I1 (world): WordNet hypernym/meronym pairs could seed templates for extending Mova's ontology, using UD dependency paths between the two nodes instead of surface n-grams to stay structural.
  • I2 (learn): when inducing new knowledge-graph edges, a fast similarity pre-filter could select pairs before costlier witness-parser or LLM-teacher checks.
  • I3 (learn): silver facts mined from one text pool could be accepted only if re-attested above a threshold in a disjoint pool, which reduces self-confirming errors.
  • I4 (answer): readers could answer questions about facts never stated in a text by inheriting properties along explainable ontology chains.

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

🎓Addis Ababa UniversityAddis Ababa · Ethiopia17 works · 77 citations

Also on the byline: Jigjiga University.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 19,503 worksArtificial Intelligence · 18,805 worksArts and Humanities · 17,959 worksLanguage and Linguistics · 17,789 worksBiochemistry, Genetics and Molecular Biology · 273 worksMolecular Biology · 270 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 3,068 works

Cite

Cite as Alelgn Tefera and Yaregal Assabie. 2014. Automatic Construction of Amharic Semantic Networks from Unstructured Text Using Amharic WordNet. Proceedings of the Seventh Global Wordnet Conference.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/w14-0123

@inproceedings{tefera2014automatic,
  title = {{Automatic Construction of Amharic Semantic Networks from Unstructured Text Using Amharic WordNet}},
  author = {Alelgn Tefera and Yaregal Assabie},
  booktitle = {Proceedings of the Seventh Global Wordnet Conference},
  year = {2014},
  doi = {10.18653/v1/w14-0123},
  url = {https://doi.org/10.18653/v1/w14-0123}
}

References

14 references, 0 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 Willet. 2002. Stemming of Amharic Words for Information Retrieval, In Literary and Linguistic Computing, Vol 17, Issue 1, pp. 1-17. find
  2. Getahun Amare. 2010. ዘመናዊ የአማርኛ ሰዋስው በቀላል አቀራረብ (Modern Amharic Grammar in a Simple Approach). Addis Ababa, Ethiopia. find
  3. Tim Berners-Lee. 2001. The Semantic Web, Scientific American , Vol 284, Issue 5, pp. 34-43. find
  4. Widdows Dominic and Cohen Trevor. 2010. The Semantic Vectors Package: New Algorithms and Pub-ሃዲያ Hadiya ወረዳ district አቢሲንያ Abyssinia Key Synonym Part-of Type-of lic Tools for Distributional Semantics, In Fourth IEEE International Conference on Semantic Computing (IEEE ICSC2010). Carnegie Mellon University, Pittsburgh, PA, USA. find
  5. Christiane Fellbaum. 1998. WordNet: An Electronic Lexical Database. Cambridge, MIT Press. find
  6. Tsatsaronis George, Iraklis Varlamis and Michalis Vazirgiannis. 2010. Text Relatedness Based on a Word Thesaurus, Journal of Artificial Intelligence Research, vol. 37, pp. 1-39. find
  7. Brian Harrington and Stephen Clark. 2007. ASKNet: Automated Semantic Knowledge Network, In Proc. 22nd National Conf. on Artificial Intelligence, Vancouver, Canada. pp. 889-884. find
  8. Paul Lewis, Gary Simons and Charles Fennig 2013; Ethnologue: Languages of the World, Seventeenth edition. Dallas, Texas: SIL International. find
  9. Stephen Richardson, William Dolan and Lucy Vanderwende. 1998. MindNet: Acquiring and structuring semantic information from text, In Proceedings of the 17th COLING, Montreal, Canada. pp. 1098-1102. find
  10. Magnus Sahlgren. 2006. The Word-Space Model: Using distributional analysis to represent syntagmatic and paradigmatic relations between words in high-dimensional vector space. PhD Thesis, Stockholm University, Sweden. find
  11. Andrew Smith. 2003. Automatic Extraction of Semantic Networks from Text using Leximancer, In Proceedings of HLT-NAACL, Edmonton. find
  12. Mark Steyvers and Joshua Tenenbaum. 2005. The Large-Scale Structure of Semantic Networks: Statistical Analyses and a Model of Semantic Growth, Cognitive Science, Vol 29, Issue 1, pp. 41–78. find
  13. Xiaoli Fern and Carla Brodley. 2003. Random Projection for High Dimensional Data Clustering: A Cluster Ensemble Approach, In Proc. of the 20th Int. Conf. on Machine Learning (ICML-2003), Washington, DC. find
  14. Baye Yimam. 2000. የአማርኛ ሰዋስው (Amharic Grammar). Addis Ababa, Ethiopia. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tefera-2014-automatic-construction-amharic-semantic-networks/