(you
may click
the number of the subfile to be viewed, or
scroll down)
This
file contains the following subfiles:
20 - tonal
inflection
21 - word prediction example
21.5 - search efficiency
(subfile 20: tonal inflection
Clearly, four different replies are required for the four versions of
the following sentence:
1) I want to go
home. (George doesn't want to, though.)
2) I WANT to go home.
(It is [only]my wish. It is not an urgent need.)
3) I want to GO home.
(I wish to travel to
my dwelling. The travel itself
is what's important.)
4) I want to go HOME.
(As opposed to going to work.)
Tonal inflection of this type can become visible to the algorithm by
defining a two-value axis for "inflection" and allowing the input stage
to include the assignment of accented words. The context-specific
(temporary) parts of the words' definitions include values on this axis
– values that are never saved to the “dictionary” database. Storing
inflections from input will of course result in the output of
inflection as well.
(A discussion of tonal emphasis also requires "bombs" and "transforms",
which see.) Suppose some input has arrived which begins “Do you
like....” and in which tonal emphasis is placed on “you”. Many
subsequent replies would have begun “I (do) like....”. Thus an element
of one type of transform would look just like a bomb from “you” to “I”.
Another fuzzy explosion of possibilities is available without any
computational strain, with different variants being appropriate to
different tonal stresses in the input:
I
George
You
like
dislikes
used to hate
apples
pears
bananas
period
question
mark
comma
(subfile 21: Predicting the existence of
words not yet seen by
the program)
Suppose training conversations have been limited to food and eating.
"Listening to conversations" requires that a number of relevant words
be defined; after defining words for the subject "eating", there would
be a
number of them that might differ along only one axis (these words
represent
the types of apples). This situation would allow the creation a
particular type of template: most spans are rather limited,
because nearly identical and specific values are present in all the
words' definitions (all fruits are 1. material objects 2. of moderate
size 3. subject to being eaten etc.), but one span, on one axis, would
be very wide, because the various words in the collection have very
different values for that parameter (apples come in several colors).
The
program is then in a position to search for a class-defining word in
meaning space, and to ask about it, if there is nothing at the location
specified. In this way our deaf and dumb computer could come to
realize, on its own, that a class-word for this set of objects might
exist.
(subfile 21.5: Recording and
advance-preparation of searches)
Suppose the program has been sent to a point in MS and finds no object
there. A painfully slow and inefficient search must be undertaken to
find the closest object. As the search progresses, a list of points
"checked" is kept, and when an object is found, pointers to that object
are inserted at each MS point traversed during the search. Thus one
need never repeat the same search.
Clearly collisions will arise when new objects are inserted; that is,
pointers that were inserted before the new object might no longer point
to the closest object, which could now be the new object. Fortunately
the program can always tell which of two possible pointers should be
kept, since distances can easily be calculated and the shorter pointer
discovered. In fact, both may be kept just as easily, at the cost of
more memory, if their dates are recorded.