1.6.3 For demonstration of machine translation (MT).
http://mind.sourceforge.net/m4thuser.html#1.6.3
Although for current versions of the original, un****ted MindForth
machine translation still lies in the future, the MT capability is
inherently there, embedded in the cognitive architecture of the AI.
Mind.Forth has three separate layers of mental activity involved in
the generation of sentences of thought in natural human language.
The top layer is the auditory memory array, which indiscriminately
holds quasi-phonetic memory engrams of words in all the languages
known by the AI -- all mixed in together and separated only by the
word-fetch access tags. The words all have in common that they are
composed of sounds, regardless of what human language they belong to.
In an AI Mind that speaks both English and German, it is the job of
the linguistic superstructure for English to keep track of all the
English auditory engrams, and it is the job of the German syntax
superstructure to keep track of all the German auditory engrams --
with some overlap of German terms like Kindergarten and Gesundheit
being used in English, and English words like manager and software
being used in German.
The middle layer of linguistic generation is the semantic memory
holding
the lexical array of word-tags as vocabulary items for each known
language.
Although the AI Forthmind might know both English and German, only
the
fetch-tags for the words as stored in audition, but not the actual
words,
are stored in the lexical array -- which easily functions as if it
were
two separate arrays, one for English and one for German, because
access
to English words and access to German words are strictly separated by
the
associative tags coming up from conceptual thinking and the fetch-
tags
going up to the actual words stored in the auditory memory.
The bottom layer of linguistic generation is the deep conceptual
array
of pre-verbal concepts interacting on spreading activation and not
yet
finding expression as components of thought in any particular
language.
Although English or German syntax may guide the formation of thought,
the same thought that occurs under the control of one language may
bubble up as a translated thought in any other available language,
insofar as the AI Mind contains adequate vocabulary and grammar to
express itself easily in more than one language.
When the AI Forthmind proceeds to add German to its repertoire of
known human languages, it will be possible to activate either English
or German in the AI Mind simply by speaking (entering input) to the
AI in either English or German. Since the AI will know its English
words and its German words only by means of associative tags, the
same
tags will, as it were, vote for which human language to activate
in the mind of the computer or robot. Simply saying one German word
like Gesundheit may not launch the German-speaking mechanism
in the AI Mind, but entering a whole sentence of nothing but German
words should get the ***ulative idea across that it is time to think
and speak in German.
It is not yet time for MindForth to claim a machine translation
ability
off-the-shelf, so to speak, because MindForth first needs to grow
larger in memory size, in availability of syntactic modules, and in
bootstrap vocabulary for both English and German and for any other
human language being requested for purposes of machine translation.
The AI may also need to grow larger in terms of the subtle
interactions
of subconscious concepts, so that nuances of thought in one language
may survive translation into another language. But the writing is on
the wall and on the Web -- Mind.Forth AI is your good bet for future
developments in intelligent machine translation.
As time goes by, MindForth or its derivatives may be expected to grow
into AI software that starts out as a blank slate where language is
concerned, with the ability to learn human languages in the same way
as a human baby learns language. In the meantime, there may be some
hybrid AI Minds which are pre-programmed to speak English or German
or whatever, but which also have syntactic concatenation routines
(easy to program in Forth on the basis of nodal activation levels)
that permit the AI to learn new syntactic sequences, either beyond
what the AI already knows in one language, or in the acquisition of
a human language previously unknown to the AI, or even a made-up
language like Esperanto or Lojban. For any known human language,
living or dead, there may eventually be at least one AI Mind
(if not a parliament of Minds) tasked with knowing that human
language superbly well so as to serve as an ultimate authority.
--
http://mentifex.virtualentity.com/m4thuser.html#1.6.3
|
