Method and system for automatically disambiguating the synonymic links in a dictionary for a natural language processing system

• US 5,237,503 A
• Filed: 10/28/1991
• Issued: 08/17/1993
• Est. Priority Date: 01/08/1991
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Parisot, Pierre, Bedecarrax, Chantal, Warnesson, Isabelle
• Primary Examiner(s): Envall, Jr., Roy N.
• Assistant Examiner(s): CHUNG TRANS, XUONG MY
• Application Number:
  US07/783,286
• Time in Patent Office: 659 Days
• Field of Search: 364/419
• US Class Current: 704/10
• CPC Class Codes: G06F 40/247 Thesauruses; Synonyms
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique

Abstract

A method and system for automatically disambiguating the synonymic links in a dictionary for a natural language processing system. Said dictionary is stored in the memory of a data processing system and includes a list of headwords with their respective synonym lists. The same headword can have different meanings, each of which has its own synonym list, each of which also can have different meanings. Disambiguation of the synonymic links is performed by reading from said dictionary a "meaning-entry", the words from which it is a synonym and its own list of synonyms, to build a synonymic environment table. A similarity index is computed for each pair of words of said environment and the words having the greatest similarity with the aggregate are incrementally clustered. The final cluster is then validated and the "meaning-entry" and its disambiguated synonyms are written back into the dictionary.

Claims

1. A method for automatically disambiguating the synonymic links in a dictionary for a natural language processing system, wherein said dictionary is stored in the memory of a data processing system and includes a list of "meaning-entries" with their respective synonym lists, characterized in that it comprises the following steps:

reading from said dictionary a "meaning-entry", the words from which it is a synonym and its own synonyms, to build a synonymic environment table;

digitizing said environment table to build an environment matrix;

building a similarity matrix by computation of similarity indexes between the elements of said environment matrix;

incrementally clustering the words having the greatest similarity index, first with said "meaning-entry" and thereafter with the previous clusters; and

writing the "meaning-entry" and the obtained cluster containing the disambiguated synonyms back into said dictionary.

2. A method according to claim 1, characterized in that said environment matrix is obtained by associating said "meaning-entry" and the entries in its synonymic environment with the set of words contained in the synonym lists of all of these words and assigning a "1" value, when a synonymic relationship exists and a "0" value when a synonymic relationship does not exist.

3. A method according to claim 2, characterized in that said similarity indexes are computed by comparison of word profiles obtained by replacement of the words in the synonym lists of each entry by "1" or "0", according to the presence or absence, respectively, of said word in said lists.

4. A method according to claim 3, characterized in that said similarity index is defined as follows:

##EQU8## for each pair of words belonging to the synonymic environment.

5. A method according to claim 4, characterized in that in said incremental clustering step, it is computed, for each candidate for integration in the cluster, a score corresponding to the mean of their similarity indexes on the cluster.

6. A method according to claim 5, characterized in that it further comprises the step of comprising the score of a candidate for integration in the cluster with the mean computed on the basis of the whole set of words provided by the synonymic environment, to determine whether or not it has to be iterated in the cluster.

7. A method according to claim 6, characterized in that it further comprises the step validating the cluster by comparison of the score or each word of the cluster with a validation threshold.

8. Data processing system comprising a first memory, a second memory, an electronic dictionary including a list of "meaning entries" with their respective synonym lists stored in said second memory:

means for reading from said dictionary stored in said second memory, a "meaning-entry", the words from which it is a synonym and its own synonyms, to build a synonymic environment table into said first memory;

means for digitizing said environment table to build an environment matrix;

means for computing similarity indexes between the elements of said environment matrix to build a similarity matrix;

means for incrementally clustering the words having the greatest similarity index; and

means for writing back into said dictionary stored in said second memory and "meaning-entry" and the words of the cluster, whereby the synonymic links in the dictionary are disambiguated.

9. System according to claim 8 characterized in that said dictionary is stored in a relational data base maintained on a Direct Access Storage Device (DASD).