Interlingua for Multilingual Machine Translation

What it is about

The CICC interlingua represents a sentence as a set of head concepts joined by labelled binary relations, plus unary attributes that restrict a concept. The relation set is flat: about two dozen labels such as agent, object, instrument, manner, time-from, time-to, place, condition, sequence and possessor, with no hierarchy, so that the structure stays simple. Tense, aspect, mood and sentence style are moved out of the words into separate frames: the speaker's perspective, the speaker's intention, and the sentence structure. This is meant to make the interlingua language-universal, so that the analysis of each language can share the same concept dictionary. The system was under development for Japanese, Chinese, Indonesian, Malay and Thai. The paper describes the design and does not report scores.

Ideas

• I1. Move tense and mood out of the concept. In the meta-world, the tense and mood of a clause come from the UD features and sit in a separate frame, not in the concept. Test it on verse pairs where the same verb appears in different tenses, and count how many merges stay stable.
• I2. Compare the relation labels with the UD relations in the frames. Count how many UD relation types map onto a single label and how many need splitting. Use the result to decide where a label is too coarse, for example where two relations share one label. The code is tools/roundtrip (frames) and the roles in prag.
• I3. Test a flat relation set of about two dozen labels as a deterministic enum, in the style of the rule that keys are enums. Measure whether the set covers the relation labels that appear in the residual of the meta-world.

For Mova

The separation of tense and mood from the concept is the same question as open problem 4 and the FEATS work in open problem 5. The relation inventory matters for tools/roundtrip and prag. The paper gives no coverage or accuracy figures, so each idea starts with a count on the verse pairs, run twice to check for identical output.

Topics: Machine translation

Cite

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

@inproceedings{uchida1993interlingua,
  title = {{Interlingua for Multilingual Machine Translation}},
  author = {Uchida, Hiroshi and Zhu, Meiying},
  booktitle = {Proceedings of Machine Translation Summit IV},
  year = {1993},
  url = {https://aclanthology.org/1993.mtsummit-1.14/}
}

References

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

1. [ l] Hiroshi UCHIDA, Meiying ZHU: An Interlingua for Multilingual machine Translation. 89-NL-72-9. Information Processing Society of Japan, 1989. find our card
2. [2] Hiroshi UCHIDA: ATLAS II : A Machine Translation System Using Conceptual Structures as an Interlingua. Machine Translation Summit, pp. 85-92 1989. find
3. [3] Itsuki HOSOE : An inquiry into the moaning of tense in the English verb, Shinozaki-Shorin. 1931. find
4. [4] Nagao. Tanaka. Makino. 1989. find
5. Uchida. Ishizaki : [5] Japan Electronic Dictionary Research Institute. Ltd. : Concept Dictionary, TR- 27, 1990. find
6. [6] Donna Gates, Dawn HABERLACH, etc. : Vol.4. No.l . 1989. find
7. [7] Sergei NIRENBURG, Lori LEVIN. : Knowledge Representation Support. MACHINE TRANSLATION, Vol.4. No.l 1989. find
8. [8] Sergei NIRENBURG : Knowledge-Based Machine Translation. MACHINE TRANS-LATION. Vol. 4, No.l 1989. find
9. [9] Tanaka, Tokunaga, etc.: A Study on Automatic Extraction of Interlingual Concepts from Bilingual Dictionaries. 89-NL-72-3, Information Processing Society of Japan, 1989 find
10. [10] Uchida, Zhu: Interlingua (Final Edition). CICC. 1993.3 find
11. [11] Masuoka. Takubo: Fundamental Japanese Grammar, Kurosio Shuppan, 1989.9 find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/uchida-1993-interlingua-multilingual-machine-translation/