OT: THEORY Fusion Grammar

Is this a new theory, or have I just re-discovered
something old?

Hypothesis: For any language, when two elements of a
sentence have an important relationship with each
other, those two elements will be immediately adjacent
in the sentence.

Definitions:
     Tag: A string appended to a word or element to
identify its role in the sentence.
     Sentence Element: a single word, or any number of
words that have been fused together into a single
element.
     Fusion: The process of joining two tagged
elements to form a new element which is then given its
own tag.
     Fusion Rule: A rule which specifies the two tags
types that may be fused and the resulting tag type of
the fused element.
     A Fusion Grammar: A collection of fusion rules
which is complete in the sense that any sentence of
the language may, by repeated application of the
rules, eventually be fused into a single element which
represents the complete fusion parse of that sentence.

Example: In this example I won't bother to tag the
words and elements. Their roles should be clear for
such a simple example.

Sentence: The big ugly dog that Mary brought home
yesterday has been barking all night.

The sentence is scanned from left to right until a
fusion rule is found that can be applied.

Result: The big (ugly dog) that Mary brought home
yesterday has been barking all night.

Here "ugly" and "dog" have been fused into a single
element. We could, if we wanted to, coin a new single
word for that element and paraphrase the sentence
using that coined word. E.g. (blending "motly" and
"mutt" we coin "mott", meaning "ugly dog". Then the
paraphrase is: The big mott that Mary brought home
yesterday has been barking all night.

Next, we return to the beginning of the sentence and
scan again until a rule can be applied. This time
"big" is fused with "(ugly dog)" giving us: The (big
(ugly dog)) that Mary brought home yesterday has been
barking all night. Again, we could, if we so desired,
coin a new word for the element (big (ugly dog)), for
example, "ubermott", giving us the paraphrase: The
ubermott that Mary brought home yesterday has been
barking all night.

Next we fuse "the" with (big (ugly dog)) (or
"ubermott", giving "d'ubermott"?)

(The (big (ugly dog))) that Mary brought home
yesterday has been barking all night.

Without going into excruciated details of each step,
we can see that eventually "all" will fuse with
"night" to create an element (all night)
representative of a span of time. "has" and "been"
fuse together to create an element with a tense
modifier tag (has been), and this element fuses with
the adjacent ING-tagged verb "barking" to form the
element ((has been) barking). The fused element (Mary
brought) fuses with a location to give us ((Mary
brought) home), an action qualified by location. This
element is then fused with the adjacent temporal
modifier "yesterday" to give us an action located in
both time and space: (((Mary brought home) yesterday),
and then marked as equivalent to a post-positioned
adjective: (that (((Mary brought home) yesterday)).

This element is then fused with the thing that it is
adjacent to, giving us the single element:
((The (big (ugly dog))) (that (((Mary brought) home)
yesterday))). This entire element is tagged as being
equivalent to a noun, and again we might be able to
coin a single word to paraphrase that entire element,
say "Fido" for example. The sentence paraphrase is
then: (Fido) ((has been) barking) (all night)).

Next we join the noun-like element named "Fido" to the
adjacent verb-like element ((has been) barking) to
create an element tagged SV which represents an event.
Finally, the temporal frame of reference (all night)
is fused to the adjacent event giving us a single
element; an event with a temporal frame of reference:
(((The (big (ugly dog))) (that (((Mary brought) home)
yesterday)))(((has been) barking) (all night))))

This is the complete fusion parse of the sentence.

If there are multiple ways to fuse pairs or multiple
ways to tag elements then each alternative is
developed in parallel.

If the sentence is inherently unambiguous only one
parse will be capable of completion at the end. If the
sentence is inherently ambiguous then multiple
complete parses will be produced.

Hypothesis: For any natural language, related elements
are always immediately adjacent and there exists a
complete fusion grammar for that language.

Comments? Counterexamples?

--gary

Replies