[lojban] Re: Loglish: A Modest Proposal

["Ben Goertzel" <ben@goertzel.org>]

> > That is sort of correct; however Loglish as I
> > defined it will have *close to
> > unique* parsing IF quu is used correctly, so
> > that it's possible to resolve
> > the intended argument position from the quu
> > specifier using simple,
> > automated semantic inference.
>
> Which is only a gleam in some few technicians'
> eyes.

No, we (the Novamente AI team) have working code that carries out this sort
of semantic inference based on WordNet.

It's not very difficult; we built it as part of a larger language processing
system.

> > In nearly all cases it will be possible to
> > achieve successful results via
> > simple algorithms such as
> >
> > "Resolve 'X qui Y' to the sense of X whose
> > WordNet definition has the
> > smallest semantic distance to Y."
>
> Does wordNet really have quantized semantic
> distances? Is it likely to soon?

Yes.  For some links into the literature, see

www.d.umn.edu/~tpederse/tools.htm

http://www.citeulike.org/user/schaal/article/266098

But we wound up brewing our own code for this, since none of these
approaches were ideal.


> If X is not in the Loglish dictionary, how do we
> get a measure of semantic propinquity to X?  I am
> not quite sure how well this will work even with
> knowledgeable human speakers, let alone with
> machines.

If "murder" is not in the Loglish dictionary, but "kill" is, it's very easy
to determine that "murder" has a small WordNet semantic distance to "kill",
and then hypothetically port over the Loglish-dictionary-entry from "kill"
to "murder."

> > What I like is that this messiness of compound
> > formation -- like the
> > messiness of tanru -- occurs within a context
> > of predicate-logic-based
> > syntax/semantics (defined by the Loglish cmavo
> > and sentence structure)
>
> Yes, I too like (for professional reasons if no
> other) the underlying FOPL.  But it seems to me
> that, having started down that route, we should
> follow it as far as feasible.  Lojban seems to me
> to be pretty close to that goal (some further
> steps have turned up in various proposals over
> theyears that have not gotten incorporated into
> the development of Lojban, but they are mainly
> rather minor -- trimming, not the core.

The question is what does "feasible" mean...

So far as I can gather, Lojban is not yet quite feasible to use as an
all-purpose everyday conversational language, because of the lack of
vocabulary.  This problem may well be remedied with another 5 years of work
at vocabulary-building, however.

In terms of human-AI communication, Lojban would be feasible to use right
now IF

a) one never cared about teaching one's AI English, or

b) one's plan was to bring the AI to near-human-level-intelligence using
Lojban-only teaching, and then teach it English

However, if one wants to teach an AI using a combination of

-- instruction in some conversational language, with

-- knowledge gained from English-language documents using information
extraction

then Lojban is not yet feasible to use because of the lack of formal
Lojban-English translation tools and the general lack of high-quality Lojban
linguistic resources (such as a systematic ontology of the case-roles
implicit in Lojban predicates' argument-structure).  Clearly Lojban *could*
be made feasible for this purpose via the judicious application of a few
hundred thousand dollars to the creation of Lojban linguistic resources.
But I don't see that this is likely to happen any time soon.

The cost of making Loglish feasible for human-AI communication would be a
lot less, I feel.  On the other hand, Loglish demands more on the AI's part
than Lojban does.  However, my suspicion is that what Loglish demands on the
AI's part is not true general intelligence, but merely some
statistical-linguistics facility in manipulating WordNet (combined with a
straightforward parser a la Lojban).  I.e., it may be that given the current
state of computational linguistics technology, Lojban goes further than is
necessary in making human communication computer-like.

Of course, all this is ONLY about the goal of using logic-based human
languages to make near-term human-AI communication easier.  I'm not making
any statements about Loglish's viability for meeting Lojban's other goals.


-- Ben




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.