Elamur Research⌕
☰
home / papers / 2012 / A Formalized Reference Grammar for UNL-Based Machine Translation between English and Arabic

A Formalized Reference Grammar for UNL-Based Machine Translation between English and Arabic

Paper · 2012 · Sameh Alansary

Summary

The paper tests whether the UNL interlingua can support English–Arabic translation in both directions, from English to Arabic and from Arabic to English. It also proposes a shared, formalized "reference grammar" that grammars for different languages could be built and compared against. The author took 500 sentences from English grammar books, chosen to cover common phenomena: unknown words, numerals, determiners, prepositions, conjunctions, noun phrases, time expressions, verb forms, pronouns and sentence types. These were hand-annotated into UNL and translated into Arabic. The resulting parallel resource, Corpus500, is published on UNLweb.

Rule-based grammars were written for both directions:

  • UNLization (analysis): modules for tokenization, numerals, attributes, X-bar syntax and syntax-semantics. Arabic adds a transliteration module.
  • NLization (generation): the reverse path.

Analysis runs in the IAN engine and generation in EUGENE. Precision and recall are high in all four directions, around 0.96–0.99. The author presents this as a basis for more robust interlingua MT and as a benchmark that other UNL languages could align with.

Key points

  • Results (precision / recall / F-measure):
    • English→UNL: 0.979 / 0.96 / 0.969
    • Arabic→UNL: 0.98 / 0.96 / 0.974
    • UNL→English: 0.97 / 0.96 / 0.964
    • UNL→Arabic: 0.989 / 0.96 / 0.974
  • Evaluation references: analysis output was checked against manually built semantic networks, and generation output against manual translations.
  • Engines: IAN is semi-automatic, with a specialist doing word-sense disambiguation and optional rules narrowing the candidates. EUGENE generates fully automatically.
  • Dictionaries: both engines use two kinds. Enumerative dictionaries list every inflected form, while generative dictionaries store lexemes with a morphological paradigm number.
  • Future work: the plan is to expand to 1000 structures with at least 5 sentences each, giving about 5000 sentences.

Interesting ideas

  • I1. A test set organised by grammatical structure, taken from grammar books, lets grammars for different languages be measured on the same phenomena rather than on arbitrary text.
  • I2. Tokenization uses longest-match against a dictionary and marks unmatched strings as TEMP; a rule then keeps a whole unknown span, such as a URL, as one node instead of splitting it into dictionary fragments.
  • I3. Spelled-out numbers are converted to digits by a small dictionary plus composition rules: tens are stored incomplete ("twenty" as 2) and completed by a zero or a following unit, while ordinals and "N times" become attributes on the number.
  • I4. Function words are replaced by abstract features: quantifiers become attributes such as @paucal or @all on the next word, and pronouns become empty concepts carrying person, number and gender.
  • I5. Generation runs in stages from list to tree, tree to tree, and back to list, with attributes realised as affixes (@def, @pl). The designers also choose to produce an overt pronoun so the output stays explicit.

For Mova

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

  • I1 (learn): We could keep a phenomenon-indexed regression suite (numerals, quantifiers, pronouns, time expressions) and report scores per structure, so a repair rule's gains and losses are visible by construction.
  • I2 (tokens): The tokenizer could flag out-of-lexicon spans explicitly and keep them whole up to a space or the sentence end, so the parser sees one opaque unit instead of garbage pieces.
  • I3 (tokens): Normalising number words into a digit value with ordinal or multiplicative flags before parsing would give maths word problems clean, traceable quantities.
  • I4 (world): Entities in the text world could store quantity and person/number/gender as explicit features instead of keeping the function word, which makes coreference and counting checks simpler and explainable.
  • I5 (back): Generating from the meaning graph through an explicit tree stage, with a documented policy on dropping or inserting pronouns, would keep text→graph→text round trips inspectable.

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

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

🧭 Topics 3Research subjects and languages the work is about.Machine translation · 1,105 worksEnglish · 2,379 worksSemitic languages · 3,066 works

Cite

Cite as Sameh Alansary. 2012. A Formalized Reference Grammar for UNL-Based Machine Translation between English and Arabic. Proceedings of COLING 2012: Posters.

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

@inproceedings{alansary2012formalized,
  title = {{A Formalized Reference Grammar for UNL-Based Machine Translation between English and Arabic}},
  author = {Sameh Alansary},
  booktitle = {Proceedings of COLING 2012: Posters},
  year = {2012},
  url = {https://aclanthology.org/C12-2004/}
}

References

19 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Alansary, S. )2010(. A Practical Application of the UNL+ 3 Program on the Arabic Language . In Proceedings of the 10th International Conference on Language Engineering, Cairo, Egypt. find
  2. Alansary, S. )2011(. Interlingua-based Machine Translation Systems: UNL versus Other Interlinguas. In Proceedings of the 11th International Conference on Language Engineering, Cairo, Egypt. find
  3. Alansary, S., Nagi, M., Adly, N. (2011). Understanding Natural Language through the UNL Grammar Work-bench , Conference on Human Language Technology for Development (HLTD 2011), Bibliotheca Alexandrina, Alexandria, Egypt. find
  4. Alansary, S., Nagi, M., Adly, N. (2010). UNL+ 3: The Gateway to a Fully Operational UNL System , 10th International Conference on Language Engineering, Ain Shams University, Cairo, Egypt. find
  5. AlAnsary, S. (2011), Interlingua-based Machine Translation Systems: UNL versus Other Interlinguas. 11th International Conference on Language Engineering , Ain Shams University, Cairo, Egypt. find
  6. Alansary, S. (2012). A UNL-based approach for building an Arabic computational lexicon , the 8th international conference on informatics and system (infos2012), Cairo, Egypt. find
  7. Boitet C. (2002). A rationale for using UNL as an interlingua and more in various domains . Proc. LREC-02 First International Workshop on UNL, other Interlinguas, and their Applications, Las Palmas, 26-31/5/2002, ELRA/ELDA, J. Cardeñosa ed., pp. 23—26. find
  8. Dorr, Bonnie J. (1987). UNITRAN: An Interlingua Approach to Machine Translation . Proceedings of the 6th Conference of the American Association of Artificial Intelligence, Seattle, Washington. find
  9. Dorr, Bonnie J. (1990). “ A cross-linguistic approach to translation ”. Proceedings of 3rd International Conference on Theoretical and Methodological Issues in Machine Translation of Natural Language, Linguistics Research Center, University of Texas, Texas. find
  10. Dorr, Bonnie J., Hovy, E., Levin, L. (2004). Machine Translation: Interlingual Methods, Encyclopedia of Language and Linguistics. 2nd ed., Brown, Keith (ed.). find
  11. Nyberg E.H., Mitamura T. (1992). The KANT System: Fast, Accurate, High-Quality Translation in Practical Domains, in Proceedings of the International Conference on Computation Linguistics, (COLING 1992), Nantes, France. find
  12. Nyberg E. H., Mitamura T., Carbonell J. (1997). The KANT Machine Translation System: From R&D to Initial Deployment, in Proceedings of LISA (The Library and Information Services in Astronomy) Workshop on Integrating Advanced Translation Technology, Hyatt Regency Crystal City, Washington D.C. find
  13. Martins, R., Avetisyan, V. (2009). Generative and Enumerative Lexicons in the UNL Framework, the seventh international conference on computer science and information technologies (CSIT 2009), 28 September – 2 October, 2009, Yerevan, Armenia. find
  14. Martins, R. (2012). Le Petit Prince in UNL, the 8th international conference on language resources and evaluation (LREC'12), 23-25 May 2012, Istanbul, Turkey. find
  15. Uchida .H , Zhu .M. (2005). UNL2005 for Providing Knowledge Infrastructure , in Proceedings of the Semantic Computing Workshop (SeC2005), Chiba, Japan, 2005 find
  16. Uchida H., Zhu M., (1993). Interlingua for Multilingual Machine Translation , in Proceedings of the Machine Translation Summit IV, Kobe, Japan.
  17. Uchida H. and M. Zhu (2005). UNL2005 for Providing Knowledge Infrastructure , in Proceedings of the Semantic Computing Workshop (SeC2005), Chiba, Japan. find
  18. Uchida H. (1996). UNL: Universal Networking Language – An Electronic Language for Communication, Understanding, and Collaboration , UNU/IAS/UNL Center, Tokyo, Japan. find
  19. Witkam T. (2006). History and Heritage of the DLT (Distributed Language Translation) project, Utrecht, The Netherlands: private publication. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/alansary-2012-formalized-reference-grammar-unl-based/