30 Syntactical rules forCD and CD’s shortcomings

Bhushan Trivedi

Introduction

In this module, we will extend our discussion about CD. We will discuss about how syntactic rules can be used to construct CD representations of the natural statement. We will also see how a complex statement can be constructed using multiple conceptualizations with relation to each other. However good conceptual dependencies are, they are not without flaws. We will soon see the flaws in the CD theory and provide a platform for discussing scripts; which extends the CD to knowledge representation which automatically adds required additional information implicitly assumed in the statement.

Syntax rules

Table 30.1 describes 17 different rules that one can use to represent natural language statements. The author has collected them from various papers of Schank and also from other sources. A few books that the author has referred to, use little different notations and also little different types of links. Though the difference is not big enough to introduce any serious problem, and it is also easy for a reader to conceptually get the idea even when things are represented in little different way (for example in one case a solid arrow is used while in the other case three line arrow is used to represent the same concept) we will not elaborate that further. Some authors use a single rule to provide multiple options and also some authors order the rules differently so that might also differ from one book to another book.

The rules described in 30.1 are structured in a way that the example is preceded by rules. Let us pick up each rule and describe how it works.

Rule 1: – this is most common rule which connects an action (ACT) with an actor (PP). This rule is also used in most other representations. The example showcases the ACT as PTRANS and PP as Milkhasingh. The running action is dependent on Milkhasingh and Milkhasingh is dependent on the action of running, as long as this statement is concerned. The two way dependency indicates that the neither the actor nor an action is primary (in terms of dependency, it is both ways)

Rule 2: – This is a rule which describes an attribute (PA) of an actor (PP). The example indicates that

PP  (Irfan) has height greater than average. Look at the state representation using state (value). If we have stated Irfan is 6 feet tall, the state indicator would be height (6). Many state indicators are represented as numeric scales as height is mentioned here. Some of them are mentioned like fuzzy set membership grade representation. For example happy might be indicated by MentalState (5) while ecstatic might be indicated by MentalState (10).

Rule 3: – this rule showcases another common relation present in many other examples that come later as well. An action might be acting on some object. When it is so, the relationship is indicated by that link described using ‘o’ (object case relation). In our example Dhony riding a bike is basically propelling it. Thus the object which is being propelled is bike. That relation is shown in the rule.

Rule 4: – This rule indicates that actions have some indicator for direction sometimes so this is the way to represent them in CD. In our case, Sachin has gone to Los Angeles. The PTRANS act is moving nothing else but Sachin himself to LA. The source is not mentioned in the statement so we have kept it as ‘?’.Such question marks represent important knowledge elements which can relate multiple statements. For example the earlier statement indicates that Sachin has played a match in Ahmedabad, India. It can be a good guess to assume the source to be India and replace that ‘?’ with India. Also see the D which indicates direction case relationship.

Rule 5: – This is an example we have seen in the previous module. This describes the relation between an ACT and source and recipient of object ownership being changed by this ACT. Sachin is the donor and Sarfaraz is the beneficiary of the ATRANS and thus are source and recipient of the bat. Interestingly, if we try to represent the statement “Sarfaraz took the bat from Sachin” does the representation remains the same? No. the actor is Sarfaraz and not Sachin now. It can be represented as follows.

Rule 6: – the object case relationship can be complex in the sense that the objects themselves can be conceptualizations. This rule provides that information. The example indicates that Mahesh hurt Jayesh by doing something (throwing stone), thus acting as in object case relationship of that ACT. Look at both conceptualizations. Throwing stone is represented by PROPELling the stone hurled at Jayesh. Also see simpler representation of verb hurt. In a true CD representation the verb hurt is represented by doing something which reduces mental and health state which we have ignored for a simpler representation. Such structures, though not allowed in pure CD theory, are often used for pedagogy purpose.

Rule 7: – act is using a conceptualization as an instrument of doing what it is doing. The example shown indicates Nobita eating rice is one conceptualization. The eating process is represented as Nobita INGESTing rice using another action as an instrument. The other ACT indicates that Nobita is doing something (indicated by a placeholder Do) with stick as an object. The second ACT and INGEST are connected by the link containing I which indicates instrument.

Rule 8: – This rule describes how a PP (other than the actor) and a conceptualization can be related to each other. The example also indicates how a verb ‘watch’ is indicated by mental transfer of information from eyes to CP.

Rule 9: – Conceptualization can also be associated with time. The example is self-explanatory. The statement that we encountered in rule 1 is extended further by providing time related information which is provided in the representation.

Rule 10: – Conceptualization are also related to locations where they occur. The example indicates “Nagpur” being a place where the conceptualization occurred and thus indicated by the link shown in the example. This rule, otherwise, is quite similar to 9th.

Rule 11: – This is a case where ACT inflicts some change in some state value. The example indicates that when Ram killed Ravan, he did something which changed the health state of Ravan from other than -10 to -10. Again note the placeholder Do. Also look at the lower case of r, it indicates reason and not recipient case relationship like R.

Rule 12: – This rule indicates that ACT happens as a reason of state change. Note the simplification of kill verb (which can elaborated as depicted in previous rule). The r link is now in reverse direction.

Also look at the state change described in the form other than numerical. This fuzzy like style is also used in CD provided somewhere the conversion to numerical state is provided. (For example sad indicates MentalState as -5 and happy indicates it to be 5).

Rule 13: – A conceptualization might act as a reason for some other conceptualization. The example indicates that Sachin going to restaurant is prohibited by another conceptualization where terrorists are actors and the ACT is attack. Here also, we have simplified the representation by not elaborating attack in form of primitive conceptualization. There are a few more things to note here. First is use of ‘/’ to indicate negation. Another is the dependency between conceptualization and not ACTs. Earlier examples were about dependencies between other things and not conceptualizations themselves. Similarly, the source is again unknown so represented as ‘?’ yet again.

Rule 14: – This rule describes an equivalence relationship. The example says that Sachin and Cricketer are having that relationship.

Rule 15: – This rule describes attributes related to actions. The example is the same as the one we have seen in rule 1, but with an additional word fast which is an attribute of an action and shown accordingly.

Rule 16: – It is possible to connect two PPs with some relation. The example talks about the relation poss-by (possessed by).

Rule 17: – our final rule indicates an attribute association with an actor. Clever is an attribute of Sachin and thus is shown using this example.

Advantages and Shortcomings of CD

One of the important advantage of representing the statements in CD is to have a representation which is independent of the words used in the statement. For example Sachin took a taxi to reach Boriwali or Sachin reached Boriwali by taxi or Sachin picked up a cab to Boriwali all will have same representation. The other advantage is that CD can act as Interlingua and thus can facilitate translation from one language to another based on the idea of the statement and not just translation of the words used in the statement. Another advantage is that we can fill missing pieces as the fixed structures contain them. If the information is missing, we can fetch it from the context. For example Sachin went to LA is preceded by Sachin purchased diamond rings from Dubai, clearly indicate the source being Dubai, which is missing in the description of the first statement. It is not apparent from our discussion so far but it is easier to infer from this structure as well.

However, there are some shortcomings of CD as well. First, there is no ontology describing contextual information about any domain that we are working with which enable us to infer correctly. There is no ISA relationship or similar relationships which can help infer from those links. The set of primitive acts chosen by Schank is also criticised by many researchers to be either incomplete, inappropriate or altogether unacceptable. Another problem is that there are no higher level concepts, representing any reasonably complex statement requires large structures. There is no mechanism of abstraction which allows a complex statement to be composed of high level abstractions which in turn are defined as low level abstractions. Many concepts are not recognized because of no ontology. For example Sachin bought a bat indicates a human mind an idea of a store from where he purchased it, but the primitive based representation would miss them. Another disadvantage is what is touted as the advantage of the CD. When we provide a generalized representation, the finer meaning is lost. For example there is a difference between giving and gifting something which is not captured by CD.

One more interesting outcome of our discussion about CD is that many things are included which are not mentioned in the statement itself. For example the verb ‘shot’ is represented using bullets and gun and the recipient being dead. This is really nice. In fact this property can be enhanced further.

For example when we say that “I went to stadium to watch a match”, there are many things which would have happened. A human listener will conclude that I have purchased a match ticket, travelled to the venue before the scheduled time of the match, took my seat, and spend time watching the match and returned home after completion of the same. Schank has at least recognized this problem and improved the CD by providing another construct called Script which we will discuss in the next module.