Introducing S-rules

The T-expressions in the START system are built using the pattern <subject relation object> at every level of embedding and thus mimic the hierarchical organization of English sentences and parallel the representational characteristics of natural language. A language-based knowledge representation system has many advantages: it is very expressive and easy to use; it provides a uniform symbolic representation for parsing and generation; and it makes it possible to automatically create large knowledge bases from natural language texts.

However, a representation mimicking the hierarchical organization of natural language syntax has one undesirable consequence: sentences differing in their surface syntax but close in meaning are not considered similar by the system. Thus, given sentence (10) as input, START will create T-expressions (11), whereas a near paraphrase, sentence (12), will generate T-expressions (13):

(10) Bill surprised Hillary with his answer.

(11) <<Bill surprise Hillary> with answer>

      <answer related-to Bill>

(12) Bill's answer surprised Hillary.

(13) <answer surprise Hillary>

      <answer related-to Bill>

Speakers of English know (at least implicitly) that in sentence (10), the subject (Bill) brings about the emotional reaction (surprise) by means of some property expressed in the with phrase. Sentence (12) describes the same emotional reaction as in (10) despite different syntactic realizations of some of the arguments; namely, in (12), the property and its possessor are collapsed into a single noun phrase. It seems natural that this kind of knowledge be available to a natural language system. However, START, as described so far, does not consider T-expressions (11) and (13), which are associated with these sentences, to be similar.

The difference in the T-expressions becomes particularly problematic when START is asked a question. Suppose the input text includes the surprise sentence (10) that is stored in the knowledge base using T-expressions (11). Now suppose the user asked the following question:

(14) Whose answer surprised Hillary?

Although a speaker of English could easily answer this question after being told sentence (10), START would not be able to answer it because T-expressions (15) produced by question (14) will not match T-expressions (11) in the knowledge base.

(15) <answer surprise Hillary>

      <answer related-to whom>

To be able to handle such questions, the START system should be made aware of the interactions between the syntactic and semantic properties of verbs. Interactions similar to the one just described pervade the English language and, therefore, cannot be ignored in the construction of a natural language system.

The surprise example illustrates that START needs information that allows it to deduce the relationship between alternate realizations of the arguments of verbs. In this instance, we want START to know that whenever A surprised B with C, then it is also true that A's C surprised B. We do this by introducing rules that make explicit the relationship between alternate realizations of the arguments of verbs. We call such rules S-rules. Here is the S-rule that solves the problem caused by the verb surprise:

(16) Surprise S-rule

      If <<subject surprise object1> with object2>

      Then <object2 surprise object1>

S-rules are implemented as a rule-based system. Each S-rule is made up of two parts, an antecedent (the If-clause) and a consequent (the Then-clause). Each clause consists of a set of templates for T-expressions, where the template elements are filled by variables or constants. The Surprise S-rule will apply only to T-expressions which involve the verb surprise and which meet the additional structural constraints.

S-rules operate in two modes: forward and backward. When triggered by certain conditions, S-rules in the forward mode allow the system to intercept T-expressions produced by the understanding module, transform or augment them in a way specified by the rule, and then incorporate the result into the knowledge base. For instance, if the Surprise S-rule is used in the forward mode, as soon as its antecedent matches T-expressions (17) produced by the understanding module, it creates a new T-expression in (18) and then adds it to the knowledge base:

(17) <<Bill surprise Hillary> with answer>

      <answer related-to Bill>

(18) <answer surprise Hillary>

      <answer related-to Bill>

Now question (14) can be answered since T-expressions (15) associated with this question match against T-expressions (18). The generating module of START responds:

(19) Bill's answer surprised Hillary.

All additional facts produced by the forward S-rules are instantly entered in the knowledge base. The forward mode is especially useful when the information processed by START is put into action by another computer system because in such a situation START ought to provide the interfacing system with as much data as possible.

In contrast, the backward mode is employed when the user queries the knowledge base. Often for reasons of computational efficiency, it is advantageous not to incorporate all inferred knowledge into the knowledge base immediately. S-rules in the backward mode trigger when a request comes in which cannot be answered directly, initiating a search in the knowledge base to determine if the answer can be deduced from the available information. For example, the Surprise S-rule used in the backward mode does not trigger when sentence (10) is read and T-expressions (11) are produced by START; it triggers only when question (14) is asked.