Browse Prior Art Database

Classification of Vague Natural Language and Representation in a Relational Data Base

IP.com Disclosure Number: IPCOM000118459D
Original Publication Date: 1997-Feb-01
Included in the Prior Art Database: 2005-Apr-01
Document File: 8 page(s) / 186K

Publishing Venue

IBM

Related People

Friede, K: AUTHOR [+3]

Abstract

Disclosed is a method for the mapping of vague context-dependent natural language expressions into two-valued logic which enables precise answers to queries against a database to cope with the vagueness of natural language attributes in queries against databases. The proposed concept particularly utilizes a decision tree with local decision procedures at each node.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 47% of the total text.

Classification of Vague Natural Language and Representation in a
Relational Data Base

      Disclosed is a method for the mapping of vague
context-dependent natural language expressions into two-valued logic
which enables precise answers to queries against a database to cope
with the vagueness of natural language attributes in queries against
databases.  The proposed concept particularly utilizes a decision
tree with local decision procedures at each node.

      Vagueness is characteristic for human language and the human
conceptual system.  The meaning of the adjective "big" depends on the
contextual environment where it is used.  For instance, an animal
which is "big" compared with a termite is relatively small compared
with an elephant.  That phenomenon cannot only be observed with
adjectives but also with substantives (nouns), since as compared with
an ocean, a lake is just a pool.

      It is now described how the above semantics (1, 2, 3) are
implemented in a relational database.  The exemplary substantive
"tree" is a vague term insofar as a plant 'A' can be more likely a
tree than a plant 'B'.  Compared with an oak, an elder is not a tree.
But compared with a violet, an elder is a kind of tree.  Thus, "tree"
can be considered a relative term.  This example is illustrated with
reference to Fig. 1, where an according graphical tree is depicted.

      This tree is generated by applying the word "tree" to an
aggregate D wherein, at each node, tree and non-tree objects are
separated.  It is emphasized that objects which are assigned to
non-trees are only regarded to be less likely a tree than the objects
which are assigned to trees.  But this does not mean that non-trees
are not trees at all!  Therefore, an aggregate corresponding to
non-trees can be subdivided into tree and non-tree objects at the
next level node of the tree.

      Now, referring to Figs. 2 and 3, it is shown how the above
information can be stored in a relational database as a binary
number.  Fig. 2 accords to the value table shown in Fig. 3, which is
a representation of a relation in a relational database.

      The 'arrows' shown in Fig. 2 represent selections of the noun
"tree" and the little boxes the results of a respective selection.
In the first selection step, "tree" selects the objects {S1,...,S11}
as trees od {S1,...,S12}.  S12 is singled out as non-tree.  In the
second stage, the objects {S1,...,S10} are selected as trees in the
aggregate {S1,...,S11}.  S11 is singled out as a non-tree object.

This procedure is continued until all objects are separated from the
original aggregate in linear order.  The advantage of the proposed
methodology is that the value table can be set up as a nxm matrix
and, thus, for each word, such a matrix can be provided.

      The tree structure shown in Fig. 2 is only exemplary; a further
embodiment is depicted in Fig. 4, and the respective value table is
depicted in Fig. 5.  The em...