Return-Path: Message-Id: <9112090720.AA27146@relay1.UU.NET> Date: Mon Dec 9 06:35:06 1991 Reply-To: Logical Language Group Sender: Lojban list From: Logical Language Group Subject: Ooops! and re Chris Handley 12/9/91 X-To: lojban@cuvmb.cc.columbia.edu To: John Cowan , Ken Taylor Status: RO X-From-Space-Date: Mon Dec 9 06:35:06 1991 X-From-Space-Address: cbmvax!uunet!cuvma.bitnet!LOJBAN Dave (misquoting Nick, who didn't leave it out) >Nick (in story): le maxri lei manti se sudri'a cu -- lojbab (.oiro'a), who should know better (.ianaise'i), doubles, and redoubles the error in two successive sentences: >lenu le solri dirce glare le terdi sudri'a lei maxri cu cu -- -- >The Sun radiatingly-warming the Earth dried the wheat. >lei manti sudgau lei maxri cu -- >The ant(s) (agentively doing something) dried the wheat. Maybe I should make sure I'm awake when posting. Apologies. .e'o fraxu lojbab Side note in response to Chris Handley: Computers will have to learn Lojban at least partially the way they learn any natural languages - initially with lots of help from humans. At this point in Lojban's evolution it is ridiculous to claim that the language has reached the goals of stability and 'logicalness', because no one yet really thinks in the language. To teach computers, we will need at least the following steps: 1. Teaching the computer a large number of Lojban words, including their place structures, probably in addition with a substantial knowledge base showing how these concepts are related (AI frames or whatever). 2. Teaching the computer to make intelligent guesses about place structures of words it doesn't know. This is more than merely coming up with guidelines like dikyjvo - it means having a capability to analyze the stuff that is found in a given place, determining its category among concepts (not sure yet what I mean by that but it may suggest something to others). 3. Most important - teaching the computer to recognize human errors becasue the human beings won;t be perfect at remembering place strutures, or in applying any rules or guidelines that are developed. The computer must therefore recognize nonsense for what it is, be able to make an attempt using the processes of 2. to ascertain what place strcture the human WAS using, and, because Lojban must allow a speaker to speak nonsense intentionally, the computer must be prepared to ask the human whether indeed the analysis is correct; i.e. whether to choose the apparently erroneous/nonsensical interpretation, or the 'corrected' version. For a long time to come, speaking nonsense in Lojban will be difficult because listeners will generally assume that the speaker has made an error and will tend to modify the speaker's statement into one that is not nonsense. This is a hypothesis, of course, because that is what we do in English and other languages, including conlangs and some computer languages even. At some point there will hopefully be this level of competence. (A Lojban speaker can metalinguistically assert correctness using the discursive affirmation "jo'a"; pragmatically, this should tell a computer to presume that the place structures are being filled according to the rules even if this seems nonsensical.) lojbab