From robin@xxxxxxx.xxx.xxx Tue Apr 27 08:58:49 1999 X-Digest-Num: 125 Message-ID: <44114.125.724.959273824@eGroups.com> Date: Tue, 27 Apr 1999 18:58:49 +0300 From: Robin Turner On Tue, 27 Apr 1999 13:51:02 +0300, Robin Turner > skripted: > > >Marcos - > > > >The unambiguous, machine-readable "superauxlang" you propose already > >exists. Lojban's grammar is unambiguously machine-parsable, and in fact a > >parser has been available on the Lojban FTP site for a few years. > > Yes, I know of Lojban unambiguity, and its parsability, yet for what I > know about it, it seems not the most appropiate possible conlang to be > machine translated to european languages, because of the relatively > big grammatical and semanthical differences there are between these > and lojban. However, an euroclone loglang would certainly not be the > best to translate to oriental languages. > To be easily and unambiguously machine-readable, a Eoruclone auxlang would have to depart from Indo-European grammar so radically that it wouldn't be a Euroclone any more. I don't see major problems with Lojban semantics, since they tend to be more, rather than less specified than European natlangs. > > There is > >also a web-based glosser, which can deconstruct compound words with almost > >total accuracy (in all the time I've used it, I've only known it make one > >mistake, and I have suspicions about the compound in question anyway). > >Since there is no homonymy in Lojban, and unmarked metaphor and idiom are > >strongly discouraged, a Lojban gloss should be pretty unambiguous in any > >language (you may get some problems with different semantic fields, but > >these are inevitable in any kind of translation, not just machine > >translation). Nora Chevalier has just finished the "beta version" of a > >parser-glosser, which will take Lojban sentences and spit them out in what > >could be called "Loglish" - understandable but not grammatical (and > >certainly not idiomatic) English. > > I'm not very well versed on parsers, but AFAIK they analize a sentence > and identify its components and the correspondant part-of-speech to > every word (please correct me if I'm wrong). What's what you get as > output from a parser? > Depends on the parser, but basically the structure of a sentence. Not a lot of use in itself, but an essential first stage in machine translation. The advantage Lojban has is that a grammatically correct sentence can only be parsed in one way, compared with natlangs (and probably most conlangs) which allow different parsings of the same sentence. For example Time flies like an arrow NP(N) VP{V, AdP(Ad, Art, N)} NP(N, N) VP(V, Art, N) V NP{N AdP(Ad, Art, N)} > > >So basically we have the "front end" of a machine translator. The tricky > >bit is at the natlang end - translating the parsed and glossed Lojban into > >acceptable English, Spanish, Chinese etc. I can understand the stuff that > >Nora's parser-glosser outputs, but maybe that's only because I'm familiar > >with Lojban (just as I can follow the weird English that Turkish students > >sometimes come out with, because I know what they're translating from). > > I wonder couldn't Lojban parser be implemented to an already existant > multilingual translator so you can get proper translations Lojban -> > NL? I've no idea, but it doesn't sound too likely. I'm CCing this to the Lojban list, since this is where the computer wizards hang out. co'o mi'e robin.