As things become complex and interdependent, it becomes necessary to experiment to try things and observe what happens - rather than have predefined methods. A simple example is the eight queens problem how to put eight queens on a chessboard so that they do not attack each other. No method is known that does not require experimentation. Orion has structural backtrack, so we can build structure and backtrack out of the building of the sentence structure if we encounter failure.
In language, it is easy to find examples of ambiguity. We start with
The computer generated images.
generated is an active verb phrase.
The computer generated images of the aircraft.
The computer generated images of the aircraft lost in battle.
The computer generated images of the aircraft lost in battle were blurry.
generated becomes a verb trapped in a noun phrase. We cant decide on the role of the word generated until we can see the entire sentence. The way we see the whole sentence is by trying alternatives for the things we are not sure about.
If we assume generated is an active verb phrase, we will get an error we will have were blurry left over. If we assume generated is a participial, we will get an error because of the object phrase, so we are left with only one alternative.
Not all hypothesising is black and white we will also need to differentiate among shades of grey.
The man practised running marathons.
The man can run marathons, so this is what he practised.
The man practised training scenarios.
You cant train a scenario, so this must mean
The man practised the scenarios used for training.
An ambiguous example
The man practised learning scenarios.
This could mean
The man practised the learning scenarios.
The man practised learning the scenarios.
Another example
The system enables the display of special features and the behaviour of the vehicle.
This could be
The system enables (the display of special features) and (the behaviour of the vehicle).
Or
The system enables (the display of special features and the behaviour of the vehicle).
Both are grammatically acceptable without error, only one of them is more in keeping with the meaning of the rest of the document. That is, we need to use meaning to determine which alternative was meant by the writer.
When hypothesising, we will need to know whether we are dealing with a black and white or a grey case, where we have to consider fitness of purpose. If it is only a black and white case, as soon as we have a successful result, we can stop.
The types of objects in a sentence whose purpose cannot be known without a larger view of the structure of the sentence:
Splitting Nouns
When we have a string of nouns, we cant be sure whether they are joined in one long noun phrase, or should be split in two. Sometimes the context will provide clues.
The resource company records show is nearly exhausted.
Is this
The resource that company records show is nearly exhausted.
The resource company that records show is nearly exhausted.
NounVerbPhrase
Could be a noun or a verb if a noun, could be part of a prior noun phrase
The airplane costs more than doubled
The airplane costs more than double
InterimPresentParticipial
If the interimparticipial has compatible noun phrases on either side of it, and the following phrase is not a time phrase, it is a participial.
If there are incompatible noun phrases on either side, and there is not a prior transparticipial (he kept the engine running), it must be a noun, and is merged with the preceding noun phrase.
If it is preceded by a RelationControl relation (he practised running marathons) and has an acceptable object, it becomes part of the verb phrase.
InterimPastParticipial
Could be a Past Participial or a verb phrase, or a verb captured in a noun phrase (the computer generated image was blurred). If the preceding noun phrases are not compatible as an object, it must be a verb phrase.
SubordinatePrepositional
Could be a subordinate conjunction or a preposition. If there is no following verb phrase, or no leading verb phrase, it is taken as a preposition.
CoordinateConjunction
Could be grouping objects, or joining clauses. In some cases, we need to resolve ambiguity.
He talked to Jack and Jill and Fred went home.
The person training staff required for the store opening needs to exercise care.
person training staff looks like a participial a person training the staff fits the modelling of to train, although the new recruit training staff is possible it has person on either side of training. If new recruit is modelled as untrained, then the person who does training can be modelled as a trainer someone who has been trained, and thus reject new recruit as a subject.
required this can be a participial or a verb phrase. The object of a participial cannot be the subject of a verb phrase, so the only possible subject is person. If a participial, staff can be required. Two alternatives.
for can be a subordinate conjunction or a preposition. We cannot be sure whether there is a following verb phrase. Two alternatives.
opening stores open every day. We cant be sure about opening needs (it could be a variant on initial needs), as the following needs is a nounverb phrase. Two alternatives.
needs this can be a noun, or it can be a verb phrase. The possible subjects are person and store opening (if for is a subordinate conjunction). Two alternatives.
care can be a noun or a verb. The odds on it being a noun here are great, but it is possible to say if those people who wish to exercise care to come this way. Unknown number of possible subjects. Two alternatives.
We now have six objects with alternatives, and need to choose the most appropriate combination. This is about the limit for humans.
We try required as a verb phrase. There seems no good object for required. We make it a participial. Its object can be either staff or person. Each is possible we choose the nearer one. This forces for to be a preposition, as there is no verb phrase to its left.
We do not yet have a verb phrase for person. It could be needs or care. Person care does not work a singular noun and a plural verb, so it must be needs.
We have no spare subject for care, so it must be a noun.
In this example, we had to make choices and see where they lead. We combined grammatical knowledge with relation modelling knowledge.
Some questions:
How do we represent the alternatives so the state of hypothesising survives undoing?
Do we use the existing building mechanism to build the structures, undoing where necessary? This would seem expensive.
We could build the relations with unknowns, and then use IS operators to switch the various hypothetical connections. This is the easiest conceptually.
If we switch on one hypothesis, we need to switch off others that would conflict with it. We can do this with XOR operators if we get an inconsistency in doing so, we can already rule out the alternative.
That is, if we make person the subject of required, it cant be the subject of needs or care, and staff cant be the object of required. If required is not a verb phrase, for cannot be a subordinate conjunction, as there is no other possible verb phrase on its left.
The connection between the alternative of required being a verb phrase and for being a subordinate conjunction requires special logical modelling
True verb phrase allows true or false subordinate conjunction
False verb phrase requires false subordinate conjunction, which forces a true preposition
If for is a preposition, then the store is at least part of a prepositional noun phrase. to exercise requires a person as the subject, and we only have person or staff.
If a verb phrase, we have
The person required
For (as conjunction)
The store needs to exercise care
Required has no object. We conclude that required is unlikely to be a verb phrase. If required is a past participial, then for is a preposition which swallows the store opening and we have
The person needs to exercise care
We turn required into a past participial, for into a preposition, needs into a verb phrase, and care into a noun, remove the hypothesising structure and complete the building of the sentence structure.
Hypothesising is time consuming, but offers a more compact solution than an enormous set of patterns. It is the only approach that allows two alternatives to be considered "side by side" when there are subtle differences in meaning between them.