Intra-sentential word processing

The sentence processing supertask is responsible for low-level (mainly intrasentential) understanding, and includes a variety of tasks aimed at this first level of comprehension. sentence processing is given an English sentence, although this need not be a well-formed sentence. The output of the supertask is a concept set representing the meaning of the sentence. In most cases, if the sentence denotes an action, by means of a subject and a verb, the main concept produced will fall on the action column of the knowledge   ontology. Another likely occurrence is for the main output to be a state; e.g., John is tall is a state description. It is also possible that the output will be an object, especially if the input was not a complete sentence. For example, when parsing the title of the earlier Hartman story, Lycanthrope, the best concept which can be produced to capture the meaning is simply the internal concept lycanthrope.

Consider the first sentence of the example story: In Space Quadrant 964, eight days away from Colony Beta Six, the Enterprise is trapped in orbit around an uncharted planet. This sentence consists of five clauses--the core of the sentence is the Enterprise is trapped; the remaining four clauses are all prepositional and serve to act as descriptors. In order to translate the sentence into a useful internal representation, a number of tasks need to successfully occur.

First, lexical retrieval is necessary. Each word in the sentence maps onto a concept within the mind of the reasoner. For the purposes of my research, I have restricted this phase of the sentence processing to previously known words--in other words, no lexical string entering the reader is unknown. This eliminates the need for novel word identification which other theories have addressed. It also eliminates the need to determine the part of the speech a word belongs to; all necessary ones are encoded prior to read time. This does not restrict the level of novelty which can be handled; it merely limits the novelty to conceptual novelty rather than lexical novelty. Because of these restrictions, lexical retrieval is accomplished in my computer model simply through each concept being tagged with the English words which can invoke it. Consider the first clause: In Space Quadrant 964. Lexically, this is a prepositional phrase consisting of the preposition in and the noun group Space Quadrant 964.

The next task of sentence processing is disambiguation. Most words actually have multiple conceptual meanings, and it is necessary to determine which is intended. Some of this is possible through simply parsing rules; other forms of disambiguation requires conceptual information. In the first clause, there are actually a number of in concepts which the word could be mapped to (in a few minutes, for example, versus in the house). Since Space Quadrant 964 is a concept denoting location, the reader disambiguates the preposition to be a location designate.

The next phrase, eight days away from Colony Beta Six, is determined to be a distance designate. Another possibility considered is a time reference; however, this is a more difficult understanding to make, so the distance one is preferred (see Chapter 6 for an explanation of how possible understandings are selected from.). The core phrase is next: the Enterprise is trapped. This is a passive sentence--this fact will be used by other supertasks in order to provide   anticipations as to what information may be needed to arrive at a complete understanding. Finally, the remaining two clauses are location designates, describing where the Enterprise is being trapped. Notice that we have actually been filling in information about the Enterprise as we have been reading this first sentence--for example, we could answer the question, Where is the Enterprise?