|
In the previous pages on the interpretation of language we considered what is referred to as syntactic knowledge. Roughly speaking, syntactic knowledge includes knowledge about:
More abstractly, syntactic knowledge may be defined as linguistic knowledge that can be stated without any reference to whatever the words may refer. Semantic knowledge is linguistic knowledge that does depend on properties of a word's referent. For example,
is a syntactically correct sequence of words, but it is semantically anomalous. Stones can't sing beautifully; in fact, they can't sing at all! This is semantic knowledge because it states a regularity that depends on the referent of the word stones. If the referent of stones was the Welsh people, then many would not find the sentence semantically anomalous. |
||
|
Charles Fillmore was one of the first linguists to introduce a representation of linguistic knowledge that blurred this strong distinction between syntactic and semantic knowledge of a language. He introduced what was termed case structure grammar and this representation subsequently had considerable influence on psychologists as well as computational linguists. The figure to the right presents the basic ideas that define a case structure grammar. Notice that linguistic knowledge is organized around verbs or more precisely, a verb sense. (A verb may have more than one sense or meaning and these are represented separately. For example, to run for office is a different sense of run than to run to first base, and these would be two different representations in a case grammar.) Associated with each verb sense is a set of cases. Some of the cases are obligatory and others are optional. A case is obligatory if the sentence would be ungrammatical if it were omitted. For example, John gave the book is ungrammatical. There are two notable features that are illustrated in the example representation. First, the cases associated with a verb seem to be associated with questions that we one would naturally ask about an event. Who did what to whom when? The representation seems well adapted to the retrieval of the information provided in a sentence. This feature was particularly appealing to psychologists and computational linguists A second interesting feature is that the same representation is provided to both the active and passive forms of the sentence. In the figure the active form is shown above the representation and the passive form below. This feature would be consistent with a finding that we rarely recall the exact syntactic form of the sentence but do recall the basic information provided by the sentence. |
 |
|
|
In the next figure to the right, the green portion illustrates the basic case structure representation of the sentence information. But, when we hear this sentence we also know the set of inferences shown; namely, that John had a book at some time b and Jim has the book at some time a and time b is before time a. The blue box illustrates the way in which these inferences can be linked to the case structure representation. The case structure representation served to inspire the development of what was termed a frame-based representation in AI research. Within a frame-base architecture it is quite natural to have these type of inferences triggered by the representation of the sentence. (For those familiar with certain types of Object Oriented programming language; the frame-based architecture in AI was a somewhat more complicated and elaborated programming environment.) One of the consistent findings in human sentence understanding is that we seem to draw these inferences automatically. And, we rarely remember whether or not such information was explicitly stated in the sentence. This observation is consistent with some of the features of a frame-based representation as suggested by case structure grammar Another aspect of the case grammar representation is that it can be effectively used to parse incomplete or noisy sentences. For example, while John gave book is not grammatical, it is still possible to create an appropriate case grammar parse of this string of words. However, case grammar is not a particularly good representation for use in parsing sentences that involve complex syntactic constructions. The web page on representing textual information will give you some appreciation of this difficulty. |
 |
|
Understanding, Interpreting and Remembering Events |
||
| © Charles F. Schmidt | ||