An interlingua aiming at communication on the Web: How language-independent can it be?
What it is about
The paper reports on the Universal Networking Language project, an interlingua meant for many-to-many exchange on the Web. Each language has an encoder and a decoder, built by native speakers, with language-specific lexicons and grammars, plus a shared base of universal knowledge for gaps. The authors examine how language-independent the interlingua really is. Its concept labels are English words, so an English distinction can leak into it: spatial prepositions such as "on", "over" and "above" are one form in Portuguese, and one English sentence can have two readings (agent or object) that need different target constructions. The authors propose shared testing procedures across teams and a round-trip evaluation: encode a text, decode it back into the same language, and compare with the original. They report Brazilian Portuguese results qualitatively, with a lexicon of over 65,000 entries.
Ideas
- I1. Keep the round-trip score per source language and report it apart from the cross-author score. The round-trip is the cheapest gate for the meta-world reader, and it shows whether the reader loses information on its own.
- I2. Keep two readings where the reader cannot choose, with a flag, as the paper's default analysis does. Count the meta tokens that carry two readings in the Bible verses, such as an object or agent reading of the same verb, and see how many the encoder then gets wrong.
- I3. Measure the English leak in the concept names. For each concept, check whether its split is needed by a language other than English, using the Hebrew reader's output. Concepts that split only English words are the language-dependent ones to watch.
For Mova
The round-trip gate is the existing roundtrip tool in tools/roundtrip, and the leak check belongs in tools/translate (meta.rs), where the Hebrew reader in hebrew.rs gives the second language. The paper's ambiguity point ties to the attachment work in open problem 1. Its results are qualitative, so each idea starts with a count on the dev verses.
Cite
@inproceedings{martins2000interlingua,
title = {{An interlingua aiming at communication on the Web: How language-independent can it be?}},
author = {Martins, Ronaldo Teixeira and Rino, Lucia Helena Machado and Nunes, Maria das Gracas Volpe and Montilha, Gisele and Oliveira, Osvaldo Novais de},
booktitle = {NAACL-ANLP 2000 Workshop: Applied Interlinguas: Practical Applications of Interlingual Approaches to NLP},
year = {2000},
url = {https://aclanthology.org/W00-0204/}
}References
- Chomsky, N. (1957). Syntactic Structures. The Hague, Mouton. find
- Chomsky, N. (1965). Aspects of the Theory of Syntax. MIT Press, Cambridge, MA. find
- Fillmore, C. (1968). The case for case. In Bach, E. and Harms, R.T. (orgs.), Universals in linguistic theory, pp. 1-88. Rinehard and Winston, New York. find
- Frege, G. (1892). On Sinn and Bedeutung. In Beaney, M. (ed.), The Frege Reader. find
- Blackwell Publishers, Malden, MA, 1997. find
- Martins, R.T., Rino, L.H.M., Nunes, M.G.V. (1998a). As Regras Gramaticais para a Decodtfica~ao UNL-Portugu~s no Projeto UNL. Relat6rio T6cnico 67. Instituto de CiSncias Matem~iticas e da Computa~ao. Universidade de S~o Paulo, Sao Carlos. find
- Martins, R.T.; Rino, L.H.M.; Nunes, M.G.V.; Oliveira Jr., O.N. (1998b). Can the syntactic realization be detached from the syntactic analysis during generation of natural ldnguage sentences? III Encontro para o processamento computacional da lingua portuguesa escrita e falada (PROPOR'98). Porto Alegre - RS. Novembro. find
- Uchida, H. (1996). UNL: Universal Networking Language - An Electronic Language for Communication, Understanding, and Collaboration. UNU/IAS/UNL Center. Tokyo, Japan. find
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/martins-2000-interlingua-aiming-communication-web-how/