Application Semantics via Rules in Open Vocabulary English

1

Application Semantics via Rules in Open Vocabulary English

Adrian Walker

www.reengineeringllc.com

Presentation for theSci entific Discourse Meeting

July 11 2011

http://www.w3.org/wiki/HCLSIG/SWANSIOC/Actions/RhetoricalStructure/meetings/20110711

2

Abstract

There has been much progress assigning semantics to data. However the meaning that resides in an application (or in a SPARQL query) should be taken into account. Even if data identifiers and ontologies have really fine readable meanings, an application can change the semantics completely. And, unless there are explanations of what the app has done, no-one will be any the wiser unless the error is egregious (eg -- the Eiffel tower is a dog).

This talk describes a system on the Web that combines three kinds of semantics: (a) data -- as in SQL or RDF, (b) inference -- via a theory of declarative knowledge, and (c) open vocabulary English. The combination is used to answer questions over networked databases, and to explain the results in hypertexted English. The subject knowledge needed to do this can be acquired in social network style, by typing executable English into browsers.

3

Agenda• The World Wide Database vision

• Only experts have the skills to use the current tools

• An easier future for Semantic Technology -- combine:

– Semantics1 - Data Semantics = the current Technology

– Semantics2 - what a reasoner should do

– Semantics3 - Application Semantics = English meanings at the UI / AI

• A browser-based system for writing and running applications in English

• Examples : Semantics of ontology data, and of oil-industry SQL data

• Google finds applications that are written in executable English

• Summary

4

The World Wide Database vision

"If HTML and the Web made all the online documents look like one huge

book, the Semantic Web will make all the data in the world look like one

huge database”

-- Tim Berners-Lee

What is the Semantic Web?

“Data integration across application, organizational boundaries”

-- Tim Berners-Lee

5


• An advantage of RDF is that data from diverse sources can, in principle,

be freely merged and repurposed.

• Yet we cannot always expect meaningful results from simply merging

previously unseen RDF data under an existing application

• An application adds meaning to the data

6

negotiable semantic distance Manufacturer’s Englishmodel of the world


Retailer’s Englishmodel of the world

7


<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"><rdf:Alt rdf:about="http://retailer.org/node"/>

</rdf:RDF>

<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"><rdf:Alt rdf:about="http://manuf.org/node"/>

</rdf:RDF>

negotiable semantic distance



8



</rdf:RDF>


</rdf:RDF>



X semantic disconnects X


9



</rdf:RDF>


</rdf:RDF>





10






– Semantics3 - Application Semantics = English meanings at the UI



• Google indexes and searches applications that are written in English

• Summary

11

Only Experts have the Skills to Use the Current Tools

KnowledgeDiscovery

Data MiningText Mining

SemanticWeb

Sub Topic

12


Researcher

Adrian

Bob

ClaireKnowledgeDiscovery


SemanticWeb

Sub TopicInstance

13


Researcher

Adrian

Bob



SemanticWeb

Sub TopicInstance

Does research on

14


Researcher

Adrian

Bob



SemanticWeb

Sub TopicInstance

Does research on

New user asked: how can I use RDF and Owl to find out from the above that

“Bob does research into Semantic Web” ?

15


Researcher

Adrian

Bob



SemanticWeb

Sub TopicInstance

Does research on

New user asked: how an I use RDF and Owl to find out from the above that

“Bob does research into Semantic Web” ?Expert replied: “You can do it by declaring subtopic to be transitive and by using a rule such as

ObjectPropertyAtom( worksIn, ?x, ?y) IF ObjectPropertyAtom( worksIn, ?x, ?z) AND ObjectPropertyAtom( subtopic, ?z, ?y)

Such rules can be expressed in RuleML or in SWRL, but you would have to find aninference tool for them.”

16










• Summary

17

KnowledgeDiscovery


SemanticWeb

Sub Topic

An Easier Future for Semantic Technology

this-item is a sub topic of this-topic===================================Data Mining Knowledge DiscoveryText Mining Knowledge DiscoveryKnowledge Discovery Semantic Web

Facts:

18

KnowledgeDiscovery


SemanticWeb

Sub Topic

Researcher

Adrian

Bob

Claire

Instance



this-person is a researcher===================Adrian Bob Claire

Facts:

19

KnowledgeDiscovery


SemanticWeb

Sub Topic

Researcher

Adrian

Bob

Claire

Instance

Does research on


this-person does research into this-topic==============================Adrian Knowledge DiscoveryBob Data MiningClaire Text Mining


this-person is a researcher===================Adrian Bob Claire

Facts:

20

KnowledgeDiscovery


SemanticWeb

Sub Topic

Researcher

Adrian

Bob

Claire

Instance

Does research on


some-subject is a sub topic of some-subject1that-subject1 is a sub topic of some-topic-----------------------------------------------------that-subject is a sub topic of that-topic

A rule:

21


some-person does research into some-subjectthat-subject is a sub topic of some-topic------------------------------------------------------that-person does research into that-topic

some-subject is a sub topic of some-subject1that-subject1 is a sub topic of some-topic-----------------------------------------------------that-subject is a sub topic of that-topic

Another rule:

-- To run or change this example, please point Firefox or IE to the demo OwlResearchOnt at www.reengineeringllc.com

KnowledgeDiscovery


SemanticWeb

Sub Topic

Researcher

Adrian

Bob

Claire

Instance

Does research on

22


Question: Bob does research into some-topic?

KnowledgeDiscovery


SemanticWeb

Sub Topic

Researcher

Adrian

Bob

Claire

Instance

Does research on

23


Question: Bob does research into some-topic?

Answer: Bob does research into this-topic===========================

Data MiningKnowledge DiscoverySemantic Web

-- To run or change this example, please point Firefox or IE to the demo OwlResearchOnt at www.reengineeringllc.com

KnowledgeDiscovery


SemanticWeb

Sub Topic

Researcher

Adrian

Bob

Claire

Instance

Does research on

24


Bob does research into Data Mining Data Mining is a sub topic of Semantic Web --------------------------------------------------------Bob does research into Semantic Web

Explanation:

KnowledgeDiscovery


SemanticWeb

Sub Topic

Researcher

Adrian

Bob

Claire

Instance

Does research on

25


Bob does research into Data Mining Data Mining is a sub topic of Semantic Web --------------------------------------------------------Bob does research into Semantic Web

Data Mining is a sub topic of Knowledge Discovery Knowledge Discovery is a sub topic of Semantic Web ------------------------------------------------------------------Data Mining is a sub topic of Semantic Web

Explanation:

KnowledgeDiscovery


SemanticWeb

Sub Topic

Researcher

Adrian

Bob

Claire

Instance

Does research on

26

• Combine, in one system for non-expert authors and users


27


• Semantics1 - Data Semantics

• the current technology


28




• Semantics2 -Mathematical Theory of Declarative Knowledge

• specifies what a reasoner should do


29




• Semantics2 -Mathematical Theory of Declarative Knowledge

• specifies what a reasoner should do

• Semantics3 – Natural Language Application Semantics

• English meanings at the Author/User Interface


30










• Summary

31

A browser-based system for writing and running applications in English

Business Policy Agents

End User / Author

Writes Business Rules in open vocabulary

English Directly into a browser

Runs the Rules Using the browser

Sees English explanations of the Results

Who does researchInto the Semantic Web?

Semantics3

32


Programmer

Theory ofDeclarativeKnowledge

End User / Author

Writes Business Rules in open vocabulary English Directly into a browser


Sees Englishexplanations of the Results

Who does researchInto the Semantic Web?

Semantics3

Semantics2

33


Programmer



InternetBusiness Logic

Application Independent

End User / Author




Who does research.Into the Semantic Web?

Semantics3

Semantics2

34


Programmer





End User / Author





SQL

RDF

Semantics3

Semantics2

Semantics1

35










• Summary

36

A retailer orders computers from a manufacturer

In the retailer's terminology, a computer is called a PC for Gamers, while in the manufacturer's terminology, it is called a Prof Desktop.

The retailer and the manufacturer agree that both belong to the class Worksts/Desktops

Use semantic resolution to find out to what extent a Prof Desktop has the required memory, CPU and so forth for a PC for Gamers

-- Example based on “Semantic Resolution for E-Commerce”, by Yun Peng, Youyong Zou, Xiaocheng Luan ( UMBC ) and Nenad Ivezic, Michael Gruninger and Albert Jones ( NIST )

Ex 1: English semantics of ontology data

37



</rdf:RDF>


</rdf:RDF>





38

A retailer orders computers from a manufacturer -- facts

for the retailer the term PC for Gamers has super-class this-class in the this-ns namespace==================================================================

Computers to order retailerWorksts/Desktops sharedComputers shared


39

A retailer orders computers from a manufacturer -- facts



for the manufacturer the term Prof Desktop has super-class this-class in the this-ns namespace=====================================================================

Desktop manufacturerWorksts/Desktops sharedComputer Systems manufacturerComputers shared


40

A retailer orders computers from a manufacturer -- facts and a rule



for the manufacturer the term Prof Desktop has super-class this-class in the this-ns namespace=====================================================================

Desktop manufacturerWorksts/Desktops sharedComputer Systems manufacturerComputers shared

for the retailer the term some-item1 has super-class some-class in the some-ns namespacefor the manufacturer the term some-item2 has super-class that-class in the that-ns namespace----------------------------------------------------------------------------------------------------------------------the retailer term that-item1 and the manufacturer term that-item2 agree - they are of type that-class

-- To run or change this example, please point Firefox or IE to the demo SemanticResolution1 at www.reengineeringllc.com


41

A retailer orders computers from a manufacturer -- answer table

this-result : retailer this-item1 is matched by manufacturer this-item2 on the property this-prop for part this-comp====================================================================================NEED PC for Gamers *missing-item* Size Graphics CardOK PC for Gamers Prof Desktop Size CPUOK PC for Gamers Prof Desktop Size MemoryOK PC for Gamers Prof Desktop Size Sound Card



42

A retailer orders computers from a manufacturer -- explanation/proof of an answer

retailer PC for Gamers is matched by manufacturer Prof Desktop on the property Size for part Memory ------------------------------------------------------------------------------------------------------------------------------------OK : retailer PC for Gamers is matched by manufacturer Prof Desktop on the property Size for part Memory



43



the retailer term PC for Gamers and the manufacturer term Prof Desktop agree - they are of type Worksts/Desktops for the retailer the term PC for Gamers has part Memory with property Size >= 256 in the shared namespace for the manufacturer the term Prof Desktop has part Memory with property Size = 512 in the shared namespace = 512 meets the requirement >= 256 ----------------------------------------------------------------------------------------------------------------------------------------------retailer PC for Gamers is matched by manufacturer Prof Desktop on the property Size for part Memory



44



the retailer term PC for Gamers and the manufacturer term Prof Desktop agree - they are of type Worksts/Desktops for the retailer the term PC for Gamers has part Memory with property Size >= 256 in the shared namespace for the manufacturer the term Prof Desktop has part Memory with property Size = 512 in the shared namespace = 512 meets the requirement >= 256 ----------------------------------------------------------------------------------------------------------------------------------------------retailer PC for Gamers is matched by manufacturer Prof Desktop on the property Size for part Memory

for the retailer the term PC for Gamers has super-class Worksts/Desktops in the shared namespace for the manufacturer the term Prof Desktop has super-class Worksts/Desktops in the shared namespace --------------------------------------------------------------------------------------------------------------------------------------------the retailer term PC for Gamers and the manufacturer term Prof Desktop agree - they are of type Worksts/Desktops



45



</rdf:RDF>


</rdf:RDF>





46



</rdf:RDF>


</rdf:RDF>


the retailer term PC for Gamers and the manufacturer term Prof Desktop agree -they are of type Worksts/Desktops

for the manufacturer the term Prof Desktop has part Memory with property Size = 512 in the shared namespace

English explanations bridge the semantic gap between people

and machines



47










• Summary

48

Ex 2: English semantics of oil-industry SQL data

• A customer needs 1000 gallons of product y in October

• Products x and z can be substituted for product y, but only in the Fall

• Combine products x, y and z to fill the order

• Combination depends on:

• How much of each product is available from each refinery

• Available transportation from each refinery to the customer area

-- Example based on “Oil Industry Supply Chain Management Using English Business Rules Over SQL” by Ted Kowalski and Adrian Walker,www.reengineeringllc.com/Oil_Industry_Supply_Chain_by_Kowalski_and_Walker.pdf

49


estimated demand this-id in this-region is for this-quantity gallons of this-finished-product in this-month of this-year===================================================================================

523 NJ 1000 product-y October 2005

Facts:

50




Facts:

in this-season an order for this-product1 can be filled with the alternative this-product2==============================================================

Fall product-y product-xFall product-y product-z

51




Facts:



in this-month the refinery this-name has committed to schedule this-amount gallons of this-product=======================================================================

October Shell Canada One 500 product-yOctober Shell Canada One 300 product-xOctober Shell Canada One 800 product-zOctober Shell Canada One 10000 product-w

52




Facts:



in this-month the refinery this-name has committed to schedule this-amount gallons of this-product=======================================================================

October Shell Canada One 500 product-yOctober Shell Canada One 300 product-xOctober Shell Canada One 800 product-zOctober Shell Canada One 10000 product-w

we have this-method transportation from refinery this-name to region this-region==========================================================

truck Shell Canada One NJrail Shell Canada One NJ

53

Ex 2: English semantics of oil-industry SQL data Rules:

estimated demand some-id in some-region is for some-quantity gallons of some-finished-productin some-month of some-year

for estimated demand that-id some-fraction of the order will be some-product from some-refinerythat-quantity * that-fraction = some-amount------------------------------------------------------------------------------------------------------------------------------------------------for demand that-id that-region for that-quantity that-finished-product we use that-amount that-product from that-refinery

54




estimated demand some-id in some-region is for some-quantity gallons of some-finished-product in some-month of some-year

for demand that-id for that-finished-product refinery some-refinery can supply some-amount gallons of some-productfor demand that-id the refineries have altogether some-total gallons of acceptable base productsthat-amount / that-total = some-long-fractionthat-long-fraction rounded to 2 places after the decimal point is some-fraction----------------------------------------------------------------------------------------------------------------for estimated demand that-id that-fraction of the order will be that-product from that-refinery

55




estimated demand some-id in some-region is for some-quantity gallons of some-finished-product in some-month of some-year

for demand that-id for that-finished-product refinery some-refinery can supply some-amount gallons of some-productfor demand that-id the refineries have altogether some-total gallons of acceptable base productsthat-amount / that-total = some-long-fractionthat-long-fraction rounded to 2 places after the decimal point is some-fraction----------------------------------------------------------------------------------------------------------------for estimated demand that-id that-fraction of the order will be that-product from that-refinery

estimated demand some-id in some-region is for some-amount gallons of some-product in some-month of some-yearsum a-num :

for demand that-id for that-product refinery some-name can supply some-num gallons of some-product1 = a-total-------------------------------------------------------------------------------------------------------------------------for demand that-id the refineries have altogether that-total gallons of acceptable base products

56


An answer table:

for demand this-id this-region for this-quantity this-finished-product we use this-amount this-product from this-refinery======================================================================================

523 NJ 1000 product-y 190.0 product-x Shell Canada One523 NJ 1000 product-y 310.0 product-y Shell Canada One523 NJ 1000 product-y 500.0 product-z Shell Canada One

To run or change this example, please point Firefox or IE to the demo Oil-IndustrySupplyChain1 at www.reengineeringllc.com

57

Ex 2: English semantics of oil-industry SQL data An explanation:

estimated demand 523 in NJ is for 1000 gallons of product-y in October of 2005 for estimated demand 523 0.19 of the order will be product-x from Shell Canada One 1000 * 0.19 = 190 ------------------------------------------------------------------------------------------------------for demand 523 NJ for 1000 product-y we use 190 product-x from Shell Canada One

58



estimated demand 523 in NJ is for 1000 gallons of product-y in October of 2005 for demand 523 for product-y refinery Shell Canada One can supply 300 gallons of product-x for demand 523 the refineries have altogether 1600 gallons of acceptable base products 300 / 1600 = 0.1875 0.1875 rounded to 2 places after the decimal point is 0.19 ------------------------------------------------------------------------------------------------------------------for estimated demand 523 0.19 of the order will be product-x from Shell Canada One

59



estimated demand 523 in NJ is for 1000 gallons of product-y in October of 2005 for demand 523 for product-y refinery Shell Canada One can supply 300 gallons of product-x for demand 523 the refineries have altogether 1600 gallons of acceptable base products 300 / 1600 = 0.1875 0.1875 rounded to 2 places after the decimal point is 0.19 ------------------------------------------------------------------------------------------------------------------for estimated demand 523 0.19 of the order will be product-x from Shell Canada One

estimated demand 523 in NJ is for 1000 gallons of product-y in October of 2005 sum eg-amount :

for demand 523 for product-y refinery eg-refinery can supply eg-amount gallons of eg-product1 = 1600 ---------------------------------------------------------------------------------------------------------------------------------for demand 523 the refineries have altogether 1600 gallons of acceptable base products


60


Rules for finding SQL data on the Internet:

we have this-method transportation from refinery this-name to region this-region==========================================================

truck Shell Canada One NJrail Shell Canada One NJ

A data table

url:www.example.com dbms:9i dbname:ibldb tablename:T1 port:1521 id:anonymous password:oracle-----------------------------------------------------------------------------------------------------------------------------------we have this-method transportation from refinery this-name to region this-region

A rule that says how to find the table on the internet


61

Programmer


End User / Author





Semantics3

Semantics2




SQL

RDF

Semantics1

62

Ex 2: English semantics of oil-industry SQL data A SQL query generated automatically from the business rules:

select distinct x6,T2.PRODUCT,T1.NAME,T2.AMOUNT,x5 fromT6 tt1,T6 tt2,T5,T4,T3,T2,T1,T6,(select x3 x6,T6.FINISHED_PRODUCT x7,T6.ID x8,tt1.ID x9,tt2.ID x10,sum(x4) x5 fromT6,T6 tt1,T6 tt2,((select T6.ID x3,T3.PRODUCT1,T1.NAME,T2.AMOUNT x4,T2.PRODUCT fromT1,T2,T3,T4,T5,T6,T6 tt1,T6 tt2 whereT1.NAME=T2.NAME and T1.REGION=T6.REGION and T2.MONTH1=T4.MONTH1 andT2.MONTH1=T6.MONTH1 and T2.PRODUCT=T3.PRODUCT2 and T4.MONTH1=T6.MONTH1 andT3.PRODUCT1=T6.FINISHED_PRODUCT and T3.SEASON=T4.SEASON and T3.SEASON=T5.SEASON andT4.SEASON=T5.SEASON and T6.ID=tt1.ID and T6.ID=tt2.ID and tt1.ID=tt2.ID)union(select T6.ID x3,T2.PRODUCT,T1.NAME,T2.AMOUNT x4,T2.PRODUCT fromT1,T2,T3,T4,T5,T6,T6 tt1,T6 tt2 whereT1.NAME=T2.NAME and T1.REGION=T6.REGION and T2.MONTH1=T6.MONTH1 andT2.PRODUCT=T6.FINISHED_PRODUCT and T6.ID=tt1.ID and T6.ID=tt2.ID and tt1.ID=tt2.ID)) group by T6.FINISHED_PRODUCT,T6.ID,tt1.ID,tt2.ID,x3) whereT6.ID=tt2.ID and tt1.ID=T6.ID and T6.FINISHED_PRODUCT=x7 and T6.ID=x8 and tt1.ID=x8 andtt2.ID=x8 and T1.NAME=T2.NAME and T1.REGION=tt2.REGION and T2.MONTH1=T4.MONTH1 andT2.MONTH1=tt2.MONTH1 and T2.PRODUCT=T3.PRODUCT2 andT3.PRODUCT1=tt1.FINISHED_PRODUCT and T3.PRODUCT1=tt2.FINISHED_PRODUCT andT3.SEASON=T4.SEASON and T3.SEASON=T5.SEASON and T4.MONTH1=tt2.MONTH1 andT4.SEASON=T5.SEASON and T6.ID=x6 and tt1.FINISHED_PRODUCT=tt2.FINISHED_PRODUCT andtt1.ID=tt2.ID and tt1.ID=x6 and tt2.ID=x6order by x6,T2.PRODUCT,T1.NAME,T2.AMOUNT,x5;

63


• It would be difficult to write the SQL query on the previous slide by hand, or to manually reconcile it with the business knowledge specified in the rules.

• How do we know that the automatically generated SQL yields results that are correct with respect to the business rules ?

The concern is eased by the fact that we can get step-by-step business level English explanations

64


• Could a programmer write more readable SQL by hand ?

Yes, but we would need to add comments in English to help people to

reconcile the hand-written query with the business knowledge

By their nature, the comments would not be used during machine processing,

so the correctness of the hand written-SQL would rely on lengthy,

and perhaps error prone, manual verification

Comments are sometimes not kept up to date when the code that they

supposedly document is changed

• The situation with SPARQL is similar

65










• Summary

66

Google indexes and searches applications that are written in English Search: for estimated demand that-id fraction of the order

67

Google indexes and searches applications that are written in English

Search: for estimated demand that-id fraction of the order

Result:


68

Google indexes and searches applications that are written in English


Result:


The executable English rulesand facts that define the application

A paper that describesthe application

69

Summary • The World Wide Database vision

– all the data in the world as one database


– OwlResearchOnt -- Bob does research into Semantic Web








70

1. The NIST / UMBC paper listed in the presentation can be downloaded from : http://www.mel.nist.gov/msidlibrary/publications.html

2. What a reasoner should do:Backchain Iteration: Towards a Practical Inference Method that is Simple Enough to be Proved Terminating, Sound and Complete. Journal of Automated Reasoning, 11:1-22.

3 . Video about interactions between drugs www.reengineeringllc.com/ibldrugdbdemo1.htm

4. Video about energy independence www.reengineeringllc.com/EnergyIndependence1Video.htm

5. The English inferencing examples OwlResearchOnt SemanticResolution1 Oil-IndustrySupplyChain1 Oil-IndustrySupplyChain1MySql1

(and many other examples provided) can be run, changed, and re-run as follows:

1. Point Firefox or IE to www.reengineeringllc.com2. Click on Internet Business Logic3. Click on the GO button4. Click on the Help button to see how to navigate through the pages5. Select OwlResearchOnt

6. You are cordially invited to write and run your own examples. Shared use of the system is free.

Links

http://www.reengineeringllc.com/