2 APPLYING QDC STRAINTS GRAPHICALLY. IN SUCH A GRAPH THE ENTITIES AN...

3.2 Applying QDC

straints graphically. In such a graph the entities and

values are represented as nodes, and the constraints

In order to automatically apply QDC during ques-

and questions as edges.

tion answering, several problems need to be ad-

It is not clear how possible, or desirable, it is to

dressed. First, criteria must be developed to

automatically develop such constraint networks

determine when this process should be invoked.

(other than the simple one for reciprocal questions),

Second, we must identify the set of question types

since so much real-world knowledge seems to be

that would potentially benefit from such an ap-

required. To illustrate, let us look at the constraints

Geographic (“Where is X”). Neighboring entities

required for the earlier example. A more complex

are in the same part of the world.

constraint system is used in our experiments de-

Kinship (“Who is married to X”). Most kinship

scribed later. For our Leonardo da Vinci example,

relationships have named reciprocals e.g. husband-

wife, parent-child, and cousin-cousin. Even though

the set of constraints applied can be expressed as

these are not in practice one-one relationships, we

follows

¹

:

can take advantage of sufficiency even if necessity is

Date(Died) <= Date(Born) + 100

not entailed.

Date(Painting) >= Date(Born) + 7

Definitional (“What is X?”, “What does XYZ stand

Date(Painting) <= Date(Died)

for?”) For good definitions, a term and its defini-

tion are interchangeable.

The corresponding graphical representation is in

Part-whole. Sizes of parts are no bigger than sizes

Figure 1. Although the numerical constants in these

of wholes. This fact can be used for populations,

constraints betray a certain arbitrariness, we found it

areas, etc.

a useful practice to find a middle ground between