CogniSieve

Contents

What Is It
How does it work
What Can It Be Used For
Technical Specifications
Limitations
Download Instructions

What Is It?

CogniSieve is an example of a Cognitron - a robotic component that has an ability to recognize. Cognitrons are not something totally new. The Newton's Smart Assist feature which recognizes and "understands"
simple commands such as "dinner with John" is an early example. The latest example is Apple's recently announced Data Detectors. Bill Gates believes they will be prevalent within 10 years or less.

In our case CogniSieve recognizes semantic entities in unformatted text.

How does it work?

Recognition is achieved by a three step process:
 
• Words are tokenized.
• The tokens are scanned to see if there are any familiar patterns. If found, the entire pattern is replaced by their holonym (Holonym/Meronym ). Sometimes this is just a single token, but it can also be a pattern with the same number of tokens except the unknown tokens have been "recognized" and assigned based upon their context.
• These tokens are scanned. If a generic class token is found it is replaced with the generic class token. In other words they are promoted into their Hypernyms (more generic class Hypernym/Hyponym).
 

The program repeatedly cycles between steps 2 and 3 until no further changes occur.

What Can It Be Used For?

The shareware example recognizes only telephone numbers, email addresses and web addresses. Users can use CogniSieve to extract these entities from text or web pages.

Custom Recognition Targets can be recognized by editing the rules that control CogniSieve. Some examples:
 

• Extract a particular column from a table of stock prices
• All items that are preceded by a keyword such as "Item Name : "
• Only sentences that use a keyword in a particular sense. E.g.. sentences about "party"

(a fun occasion) as opposed to involvement, (to be a party to).
 
Tasks where CogniSieve can provide productivity gains.
 
• Building up mailing lists by extracting names and addresses
• Verifying that items in a mailing list are valid entities
• Extracting sub sets of data from a larger database
• Automated extraction of email surveys or questionnaires
• Automated grading of multiple-choice exams or surveys

Technical Specifications

Size :                    1.2 Megs
Memory Required :         2 Megs
OS :                     7.x or 8
Hardware :                PowerPC Macs.
                                                Win95 version in a couple of months (hopefully).

Limitations

• Maximum clipboard size is 32k.
• Player version does not allow editing of rules

Download Instructions

Check the modification dates to see whether you have latest versions.
 

File Name	Size	Modification date
CogniSieve Player 0.8 B	448 k	30 Sept 1997	Download