[lojban] Re: Loglish: A Modest Proposal

["Ben Goertzel" <lojban-out@lojban.org>]

> I assume that -- as is usual with
> AI prospectuses -- as these problems are pointed
> out, devices will be found (or at least claimed)
> to deal with them.

Yes -- I explicitly stated that my posting on Loglish was just a preliminary
idea-sketch and not a fully-fleshed-out, rigorous proposal.  I posted my
preliminary ideas in order to get feedback from others, which I am getting,
and I much appreciate it...

If Loglish actually takes off (e.g., if I implement or convince someone else
to implement a parser for it) then I will take discussion of it off this
mailing list and onto another list, since I realize this is a list for
discussing Lojban not Loglan, Loglish or other Lojban-related languages.
However, this list seemed the best place for tossing around ideas about
Loglish.  At least, the people on this list seem to know what the hell I'm
talking about, whereas when I floated the idea on an AI-related list I got
no reply but general dumbfoundedness... ;-)

> Fow now (as usual, agqin) we
> are left with a good display and unremarked
> amounts of pre-editing, post-editing and just
> plain cheating in the middle.  Of the latter, I
> note that neither "murder" nor "kill," its
> supposed surrogate, is three-placed in English

"Kill" is not three-placed in English *syntax*, but it has more than one
argument in English *semantics*.

For instance, FrameNet

http://framenet.icsi.berkeley.edu/index.php

gives the following case-roles associated with the English verb "kill":

use		Core
---------------------------
Degree	Peripheral
Depictive	Extra-Thematic
Instrument	Peripheral
Killer	Core
Manner	Peripheral
Means		Peripheral
Place		Peripheral
Purpose	Extra-Thematic
Reason	Extra-Thematic
Result	Extra-Thematic
Time		Peripheral
Victim	Core


Just as I'd suggest WordNet as the standard reference for Loglish
word-senses, I'd suggest FrameNet as the standard reference for Loglish
argument-structures.

In this case "instrument" maps into the FrameNet argument slot called
"Instrument"

> ("He killed her a knife" is not English -- or,
> rather, means that he bagged a wild knife for
> her) and neither is connected with conflict or
> violence to make a connection with "weapon," so
> the analysis does not go off as it appears to in
> the text.  That does not mean it does not go off,
> of course, but it must be more complex (and
> possibly more ad hoc) than suggested.  Lacking a
> quick connection between "murder" and "plier quu
> weapon" one wonders how to keep this from being
> "Ben murdered the chicken who had a plier."

Because the parser (be it a computer parser or the parser in the mind of the
human Loglish-listener) sees

"plier quu weapon"

and then has the job of matching "weapon" with one of the argument-slots of
kill according to FrameNet.  Out of all these argument-slots, it has to
figure out that "weapon" matches "instrument" better than any of the others.

Of course, the parser's job is even easier if the speaker uses

"pliers quu instrument"

i.e. if the speaker is a bit familiar with FrameNet lingo, but I gave
"pliers quu weapon" as an example just to show that there's some room for
improvisation/sloppiness on the speaker's part.

> Presumably Loglish must have some way of marking
> off modification of arguments from modification
> of the main predicate.  Given the paucity of
> reference to Lojban, it is not clear what it
> might be (it probably can't be the Lojban device
> with {be}, given the conflict with English "be."

The English word "be" wouldn't exist in Loglish, nor would any other English
"helping" or "being" verbs.  This was not explicitly stated in my brief
initial Loglish document, but it should have been.

As I did say in my document, I assume that *all Lojban cmavo* (including
"be") are supposed to be carried over into Loglish.

I screwed up by mistakenly leaving out "la", "cu" and "lo" in the Loglish
example I emailed out last night, but Jorge fortunately corrected this silly
error on my part.

> I think the practical issues motivating this
> suggestion have real merit; I just don't think it
> concerns Lojban (nor -- yet -- predicate logic).

No insights into predicate logic are likely to come from Lojban or Loglish;
both languages are intended to serve as front ends to predicate logic...

As to the irrelevance to Lojban -- well, if this discussion is generally
annoying people, I can take it off this list.  But for the moment, I at
least am enjoying and benefiting from the responses ;-)

-- Ben G




To unsubscribe from this list, send mail to lojban-list-request@lojban.org
with the subject unsubscribe, or go to http://www.lojban.org/lsg2/, or if
you're really stuck, send mail to secretary@lojban.org for help.