Dissertation/Thesis Abstract

Lojbanic English, An Interlingua for Parallel Machine Translation
by Immes, Luke P., Ph.D., University of Massachusetts Lowell, 2019, 164; 13880804
Abstract (Summary)

We investigated machine translation using the interlingua Lojban, and our own extensions, i.e., Lojbanic English. Lojban avoids ambiguity because of its 1342 primitive predicates, and no polysemy. Lojbanic English was tested for a wide variety of sentence types—yielding 503 Lojban/Lojbanic English sentence tests. We developed a translator from English to Lojbanic English, using our 35 generic sentence patterns. Ambiguity was avoided, but unforeseen patterns were yet to be considered.

We also investigated other anomalies that Lojban would be (mostly) able to avoid—grammatical usage errors by Swan. We implemented 80 of most common 130 errors. The test suite was The Brown corpus consisting of 55889 sentences. Our system detected 35 true positives distributed among 15 of Swan’s rules. A low true positive rate, 35/55889, had been expected. No false positives were detected.

When writing in Lojban one adds new predicates incrementally; this is very time consuming, To address Lojban's insufficient vocabulary, we developed an interactive algorithm which will recast WordNet synonym sets’ definitions into existing Lojban primitive predicates. The output is in terms of our Lojbanic English. If a relevant subset, e.g., 1/10, of the unique synonym-set definitions—totaling 116718, are converted into Lojban predicates, then 1.945 man years would be required for this effort.

To avoid unforeseen syntactic patterns or implicit semantics, our Lojbanic English is such that the user writes only in terms of (Lojban) structure words, named entities and Lojbanic English predicates. Off-line, English phrases of the input sentence are mapped into Lojbanic English, e.g., linking clauses or phrases. Sufficient generality is employed to allow for reuse of the English phrases. At runtime, given an arbitrary English sentence, any errors in the final Lojbanic English generated are detected and corrected. Swan usage errors are avoided. User's skill is required.

This specification of Lojbanic English gives rise to a bijective function—English words can be automatically replaced with foreign language counter parts, in parallel. We are able to translate Lojbanic English into some Lojbanic foreign language, and back (via Lojban); the result is identical to the original text. Thus, we are completely confident in the translation.

Indexing (document details)
Advisor: Levkowitz, Haim
Commitee: Beke, Tibor, Mwaura, Jonathan
School: University of Massachusetts Lowell
Department: Computer Science
School Location: United States -- Massachusetts
Source: DAI-B 80/11(E), Dissertation Abstracts International
Subjects: Translation studies, Computer science
Keywords: Bijective function, Interlingua, Lojban, Lojbanic English, Machine translation, Parallel
Publication Number: 13880804
ISBN: 978-1-392-26531-4
Copyright © 2020 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy