Re: [bpfk] BPFK work

On Sun, Oct 10, 2010 at 6:01 PM, Robert LeChevalier <lojbab@lojban.org> wrote:

Jorge Llambías wrote:

On Sun, Oct 10, 2010 at 6:42 PM, Robert LeChevalier <lojbab@lojban.org> wrote:

Jorge Llambías wrote:

I see the disadvantages (you lose the information that a text has a
speaker which is referred to as "mi" in the text and an audience which
is referred to as "do" in the text), but I still can't see the
advantages. What are the advantages of treating the whole conversation
as one text?

Syntactically, only that it takes one pass through the parser rather than
many.

Except when the parse of the full thing fails, in which case you still
have to try each piece again to see if the individual pieces parse,
right? Or are you saying that when the individual pieces parse but the
concatenation fails, the conversation as a whole fails?

Back then, the concept was that if the parse fails, the text was bad Lojban, period. Stop there. There was no error analysis. If I fed the entire conversation to the parser as a text, and it did not have the ".i" with each new speaker, it failed, and therefore was ungrammatical.

At the same time we recognized that the parser was imperfect and the YACC grammar did not address all of the metalinguistic rules.

We thought we had a superior grammar merely by allowing a much larger set of grammatical pieces of less than bridi length to be valid Lojban, recognizing that human beings would likely tend to speak less than perfect Lojban and less than complete sentences.

Do you no longer agree with what you taught in your lessons?

The lessons didn't really address the issue. CLL didn't really address the issue.

The "machine language" and the "human language" are not necessarily the same. The question of "what is a text?" was always considered to be a "machine language" question and hence not something we "taught".

You can legitimately say that our examples did not generally include a leading ".i", and hence taught-by-example that a leading ".i" was not important.

If I were conversing with a computer, I would expect that the computer would need the separators.

Pragmatically, while "mi" and "do" change their meanings with time (as do
ri, ra, di'u etc), most referents hold their value regardless of time,
speaker, etc. Any semantic analysis has to treat them as a single text.
Otherwise, somewhat-ambiguous semantics become unintelligible semantics.

Of course any semantic analysis needs to take context into account,
and the preceding text is most definitely part of the context of a
given text.

I'm still not seeing the advantage of treating a conversation
syntactically as a single text. It doesn't always work, so is your
rule: "if you can blend everything into one text, do it, otherwise
treat each part as a separate text"? Or what?

My rule, to the extent that I had one, has always been to treat a conversation syntactically as a single text, and if it didn't work, insert .i. at the start of each turn until it did work. I didn't really "teach" this, because it wasn't something that was part of the machine grammar (which doesn't even recognize the possibility of multiple speakers or multiple texts).

I never really considered one person finishing another person's sentence to be "part of the language", but can accept that people might want a way to do so.

lojbab