Ambiguities To Resolve

Types of Ambiguity

In any language, there are two types of ambiguities.

  • General ambiguity: the kind that really can have two or more meanings or interpretations, are comparatively rare, which Discovery must ask the user to resolve, since we can't expect computers to resolve such ambiguities any better than a human could.
  • Computer ambiguity: extremely common, the kind which is entirely clear to human beings, but one from which a computer—despite all its processing power—would detect two or more meanings or interpretations, some of which may border on the ridiculous.

This is due entirely to a computer's total lack of human awareness not only of what the concepts the words in a sentence represent, but also of the context in which the sentence is expressed among others in a verbal exchange. Many strategies can be used to overcome this deficiency, such as syntactic constraints, frequency in context, selectional restrictions (semantic constraints), "recency" rules, parallel structure, world knowledge, textual coherence and speaker intent.

For ambiguities it would require the user to resolve, Discovery would simply limit all possible interpretations to an applicable few from which the user could choose. Nevertheless, it has been found that in actual practice that the more refined the programming has become, the more Discovery has overcome these deficiencies to more closely mimic a human's ability.

These two types of ambiguity fall under another set of categories, which are supplemented by examples, some of which have been resolved:

• Lexical, when more than one meaning can apply to one or more words

I saw (see in past/cut with a saw in the present) a bat (flying mammal/wooden club). (genuine)

The bat ate its dinner. (computer) The verb to eat has extended syntaxes which limit the subject that performs it to a living thing; hence, a flying mammal.

The sick bat lay on the ground. (computer) One definition of the adjective sick is associated with the lexical category adjective.organic, which by lexical category combination with the verb to be (copula) can only modify a noun associated with a lexical category noun.person or noun.animal; hence, a flying mammal.

• Syntactic, where sentences have more than one parse tree, i.e. can be diagrammed in more than one way.

º Phrase attachment, where prepositional phrases in succession may modify more than one noun, whether immediately or remotely preceding.

Mary ate a salad with spinach from California for lunch on Tuesday. (computer)

Discovery finds a combination of lexical categories which conforms to the categories assigned to the verb, object noun and prepositional phrases. On the basis of this combination, in context, one or more definitions for each word assigned a lexical category apply, while those remaining do not:

eat: verb.consumption
salad: noun.food
with spinach: preposition.accomp + noun.food
from California: preposition.src + noun.location
for lunch: preposition.goal + noun.food
on Tuesday: preposition.rel + noun.time

The combination also specifies which object nouns in the prepositional phrases are modified by others:

eat [salad { (with spinach) (from California) } { (for lunch (on Tuesday) ) } ]

º Conjunction, where a noun can be joined by a conjunction with more than one noun, whether immediately or remotely preceding.

Mary ate a salad with spinach from California for lunch on Tuesday and (on) Wednesday. (computer)

Resolved with the same strategy as above, but when identical lexical categories are moreover found for two consecutive nouns or prepositional phrases:

on Tuesday: preposition.rel + noun.time
(on) Wednesday: (preposition.rel + ) noun.time

º Noun group structure, where nouns in succession may modify more than one noun, whether immediately or remotely preceding.

New York University Martin Luther King Jr. scholarship program projects coordinator Susan Reid. (computer)

Resolved by lexical category combination match of successive nouns:

New York: noun.location
University: noun.artifact/noun.group
Martin Luther King, Jr.: noun.person
scholarship: noun.possession
program: noun.cognition
projects: noun.cognition
coordinator: noun.person
Susan Reid: noun.person

• Semantic, where either the class to which a noun belongs or the relationship between two or more nouns in an action or state expressed in the same sentence is unclear.

Lucy owns a parrot (existentially quantified) that is larger than a cat (either universally quantified or means "typical cats"). (computer)

Resolved by imposing article class constraints to nouns evaluated within particular definition rules.

• Anaphoric, where a pronoun may substitute more than one noun.

Margaret invited Susan for a visit, and she (Margaret) gave her (Susan) a good lunch. (computer)

Margaret invited Susan for a visit, but she (Susan) told her (Margaret) she (Susan/Margaret) had to go to work. (computer)

On the train to Boston, George chatted with another passenger. The man (another passenger) turned out to be a professional hockey player. (computer)

Bill told Amy that he had decided to spend a year in Italy to study art. (computer)
That (art) would be his life's work. (computer)
After he had done that (spending a year in Italy), he would come back and marry her. (computer)
That (deciding) was the upshot of his thinking the previous night. (computer)
That (telling Amy) started a four-hour fight. (computer)

As yet to be resolved in later development.

© 2014 Webworld-v2. All rights reserved | Template by  W3Layouts