Schema Specification of RuleML 0.91

An introduction to RuleML is given in our tutorial. Also, the main page's design section discusses the upper layer of the RuleML hierarchy of rules. In that terminology, the system of RuleML XSDs presented here only covers derivation rules, not reaction rules (see the official model).

This is because we think it is important to start with a subset of simple rules, test and refine our principal strategy using these, and then work 'up' to the more general categories of rules in the hierarchy. For this we choose Datalog, a language corresponding to relational databases (ground facts without complex domains or 'constructors') augmented by views (possibly recursive rules), and work a few steps upwards to further declarative rules as allowed in (equational) Horn logic.

Below is a summary of the changes in version 0.91:

For more information, see the Patches and Changes section.

A comprehensive glossary accompanies this specification.

The grammar of RuleML 0.91 (i.e. the content model of each individual tag) is available as a single document called Content Models for RuleML in addition to being encoded within the XML Schemas.

An XML Schema version of RuleML 0.91 generated from Relax NG is being developed at RuleML 0.91 in Relax NG.

A patch to the 0.91 release was applied on May 14, 2011 to correct several of the errors identified during the re-engineering of the schema into Relax NG syntax. The patch details are given at Errata.

Changes in this 0.91 release relative to the previous version 0.9 are detailed below, including examples where appropriate.

It was a RuleML issue up to version 0.9 that the constructor (<Ctor>) of a complex term (<Cterm>) was disjoined, as an XML element, from the user-defined function (<Fun>) of a call expression (<Nano>), although these can be merged by proceeding to a logic with equality. While function-defining <Nano> patterns could contain <Cterm>s but not <Nano>s, obeying the 'constructor discipline', <Nano>s could also be permitted within <Nano>s to legalize 'optimization' rules like reverse(reverse(?L)) = ?L.

Functional RuleML thus conceives both <Cterm>s and <Nano>s as expression (<Expr>) elements and distinguishes uninterpreted (constructor) vs. interpreted (user-defined) functions just via an XML attribute @in={"no","yes"}.

For example, a function first-born(John, Mary) can be uninterpreted, so that first-born(John, Mary) just denotes the first-born child; or, interpreted, e.g. using definition first-born(John, Mary) = Jory, so the application first-born(John, Mary) returns Jory. Such interpreted and uninterpreted functions are both allowed in Functional RuleML, and are marked up here with a discriminating interpreted attribute @in as follows:

   first-born
   John
   Mary
 ]]>

and

   first-born
   John
   Mary
 ]]>

Note that as a result of the above change, <Ctor> is no longer necessary as it is now always <Fun>.

<Entails> is introduced for logical entailment (conventionally indicated by the "|-" symbol) where the first child entails the second, both being <Rulebase>s. When <Assert>ed, this effectively replaces 0.9's <Protect> and renders the explicit <Integrity> tag unnecessary since integrity constraints are just formulas (in the second <Rulebase>) used in an integrity-maintaining way. So instead of

   ic1 . . . icM
   clause1 . . . clauseN
 ]]>

we obtain ("Assert that the knowledge base entails the integrity constraints"):

   
     clause1 . . . clauseN
     ic1 . . . icM
   
 ]]>

More concretely, with 'implicit <And>' assumptions, the inconsistency of a watch being both gold and rusty, in 0.9

   
     
       
         
           gold
           object
         
         
           rusty
           object
                
       
     
   
   
     gold
     watch17
   
   
   
     rusty
     watch17
   
 
]]>

is as follows in 0.91:

   
     
       
         gold
         watch17
       
       
       
         rusty
         watch17
       
     
     
       
         
           
             gold
             object
           
           
             rusty
             object
                  
         
       
     
   
 
]]>

Since <Rulebase>, as an unordered or ordered collection of rules with or without duplicates, is more general than <And> (always unordered, without duplicates), the 'implicit <And>' assumption now becomes an 'implicit <Rulebase>' assumption. So 0.9's

   
     ...
   
 
]]>

is instead

   
     ...
   
 
]]>

in 0.91, which is equivalent to

   ...
 
]]>

because of the 'implicit <Rulebase>' assumption.

With the addition of the <Retract> performative (syntactically identical to <Assert>), rulebase assertions can be "removed". For example, if John no longer likes Mary (after an argument, perhaps) then this fact could be retracted:

   
     likes
     John
     Mary
   
 ]]>

The <Data> element, already introduced in version 0.89 to support XML Schema built-in datatypes, now also allows arbitrary XML (in addition to unrestricted text) when a datatype is not specified using the 'xsi:type' attribute. For example, this XML about some kind of object can now be treated as data within an <Atom>:

   
   
     115897
     John Doe
   
 ]]>

Also introduced in 0.89, the <Reify> element was initially too permissive in that it allowed any valid XML instead of being restricted to valid RuleML (within the specified sublanguage). This has been corrected, so that, e.g., while

 <Reify><Ind>John Doe</Ind></Reify>

   
   
     115897
     John Doe
   
 ]]>

The new @mapMaterial attribute has been introduced to complement the (renamed) @material attribute (previously @kind) in the same way as @mapClosure and @mapDirection. (See Attribute Mapping for more information on this "mapping" convention.) So a "first order" (classical) rule from 0.9

   ...
 ]]>

is represented as follows in 0.91:

   ...
 ]]>

Conversely, "logic programming" (kind="lp" in 0.9) implication is now indicated by material="no".

An equation (<Equal>) can now have an @oriented attribute for indicating whether it is oriented (directed) or unoriented (symmetric). For example, the (oriented) defining equation home(father-of(John)) = Mexico City appears as follows:

   
     
       home
       
         father-of
         John
       
     
   
   
     Mexico City
   
 ]]>

Note also that the <side> role tag from 0.9 has been split into separate <lhs> and <rhs> tags in 0.91.

Since interpreted functions are often set-valued (non-deterministic), a @val attribute is introduced in Functional RuleML with values including "1" (deterministic: exactly one) and "0.." (set-valued: zero or more). For example, the function children(John, Mary) can be interpreted in a set-valued manner using a definition children(John, Mary) = {Jory, Mahn}, so that the application children(John, Mary) returns {Jory, Mahn}. In 0.91, the example is marked up as follows:

   children
   John
   Mary
 ]]>

Numerous sample RuleML documents have been prepared and maintained; some exemplify features of RuleML and are useful didactically while others are mostly for testing puposes. Four new examples (ackermann.ruleml, functional.ruleml, reify.ruleml and ssn.ruleml) accompany the 0.91 release and can be found with all the others in the Examples directory.

Examples from previous versions of RuleML are also maintained, e.g. 0.9 examples).

Continuing the practice begun in 0.86, an XSLT stylesheet named 090-to-091.xslt has been created to automatically update RuleML 0.9 documents to 0.91. For instance, taking the discount example from 0.9 as input, the stylesheet outputs the following: discount_91.ruleml. (See a comparison of these two files generated using HTML Diff, noting that the differences are minor.) Additional translation output samples are found in the Upgrader directory. In the comparison directory a comparison between manual and automatic updates of a few files from the 0.9 exa directory can be found.

It should be noted that -- since there are changes in RuleML 0.91 that limit the content of some elements compared to RuleML 0.9 -- some RuleML 0.9 documents cannot be automatically converted to valid RuleML 0.91 files. In such cases, the 0.9-0.91 upgrader can still be used; however, some manual post-editing is required to make the files valid with respect to the RuleML 0.91 XSD schema. Namely, the relevant changes between RuleML 0.9 and RuleML 0.91 are the following:

• Slot names can no longer contain <Skolem>, <Var> or <Reify>
• <Reify> now correctly allows only valid RuleML instead of any valid XML

Upgrader transformation can be accomplished using a web-based XSLT transformation tool provided by W3C. Instructions for this process can be found in Appendix 4.

The stylesheet has also been tested using XML Spy.

An XSLT processor which may be used to perform these transformations on a whole directory at once is Saxon, using the following command:

The glossary for RuleML 0.91 contains descriptions for every tag in RuleML 0.91.

The most current model of the modularization of RuleML is always documented at http://www.ruleml.org/modularization/#Model.

A stable XML Schema specification of RuleML 0.91 has been created using an approach that is consistent with that used in the (version 0.85) DTDs.

Starting with RuleML 0.91, Schematron is used for additional constraints that are either difficult or impossible to enforce using only XML Schema. This approach has the benefit of reducing complexity in the XSDs, making them more maintainable by avoiding dependence on "XSD tricks" to achieve context-sensitivity.

As explained in the official overview, "The Schematron differs in basic concept from other schema languages in that it not based on grammars but on finding tree patterns in the parsed document. This approach allows many kinds of structures to be represented which are inconvenient and difficult in grammar-based schema languages." A Schematron article also points out that XML Schema was not designed to meet all XML validation needs in the first place; in fact, the specification states: "However, the language defined by this specification does not attempt to provide all the facilities that might be needed by any application. Some applications may require constraint capabilities not expressible in this language, and so may need to perform their own additional validations."

The use of Schematron in RuleML allows, for example, enforcing that a defining equality (i.e. an <Equal> in the head of an <Implies>) must have an interpreted left hand side (i.e. <Fun in="yes">...<Fun>):

  
    A defining equality must have an interpreted left-hand side.
  
]]>

This constraint makes the following rule invalid, even though it is valid with respect to the XSDs:

  
    greaterThanOrEqual
    
      this-year
    
    1993
  
  
    
      
        age   
        
          first-born 
          John
          Mary
        
      
    
    
      
        subtract
        
          this-year
        
        1993
      
    
  
]]>

The Schematron constraints for RuleML 0.91 are embedded in the relevant XSD modules as annotations (e.g. the one above is in the equality module) but are also available collected into a single schema (ruleml.sch). A public XSLT stylesheet called XSD2Schtrn.xsl can be used to create ruleml.sch by extracting all embedded Schematron constraints from the XSDs.

Schematron's ease of use, being based on XSLT, has led to numerous implementations. For RuleML, we use Academia Sinica Schematron 1.5. All essential files are available in the Schematron directory, along with a batch file (validate.bat) that performs both XSV and Schematron validation on a RuleML document (after extracting Schematron constraints embedded in the XSDs).

See Appendix 4 for instructions on validating an example with Schematron using W3C's Online XSLT 2.0 Service.

To ensure validation stability, the 0.91 XSDs are tested (using corresponding instance documents/examples) with the W3C XML Schema Validator (XSV) (currently version 2.10-1). If using the locally-installed version of XSV, a simple batch file called xsvall.bat can be downloaded and used to validate all examples in the current directory, displaying the results in a conveniently summarized format.

All 0.91 examples and schemas validate perfectly except for the binary sublanguages. There was a related known bug in XSV 2.7-1 affecting the binary sublanguages that may still be causing problems. The issue has been (re-)reported to the xmlschema-dev mailing list.

Note that the "Recognised as RDDL, but no W3C XML Schema resource found" warning can safely be ignored; it only appears because Schematron rules are embedded within the XSDs.

Full compliance to the RuleML 0.91 specification involves both XML Schema and Schematron validation, the latter acting as a kind of "semantic validation". A batch file (validate.bat) is available for performing both XSV and Schematron validation on a RuleML document.

See Appendix 3 for instructions on validating an example against its XML Schema and Appendix 4 for instructions on validating an example with Schematron using W3C's Online XSLT 2.0 Service.

There are two kinds tags in RuleML: method-like 'role' tags and class-like 'type' tags. The two are easily distinguished by their initial letter, much like the convention used in Java: role tags start with a lower-case letter, whereas type tags start with an upper-case letter. For example, <head> is a role tag and <Ind> is a type tag.

Furthermore, type and role tags alternate in RuleML. For example, notice the alternation in this fact asserting that John sells XMLBible to Mary:

   
     sell
   
   
     John
   
   
     Mary
   
   
     XMLBible
   
 
]]>

This is known as the "striped syntax" of RuleML, and is important for compatibility with RDF. Because this is quite verbose, all role tags can also be skipped (in the spirit of Sandro Hawke's StripeSkipping), leaving a more compact form. For example, the above fact can be shortened to:

   sell
   John
   Mary
   XMLBible
 
]]>

Note that there is no loss of information, because all role tags can be reconstructed based on the remaining type tags. In fact, this reconstruction (from the latter positional form to the former slotted form) can be done automatically via XSLT, as with the Normalizer in 0.89.

This gives the syntax of RuleML a lot of flexibility because it allows either a compact or expanded form (or any combination thereof), where the compact form is more user-friendly and the expanded (normalized) form is more compatible with other languages.

There are three pairs of attributes in RuleML that have names which differ only by a 'map' prefix: closure and mapClosure, direction and mapDirection, and material and mapMaterial. The naming convention reflects attribute 'mapping' through XML trees, where

. . .. . .. . .]]>

basically acts like

. . .. . .. . .]]>

Appended below are a simple example rulebase (Appendix 1), the XML Schema for the Datalog sublanguage of RuleML to which the rulebase conforms (Appendix 2), instructions for how to validate the example against the schema (Appendix 3) and instructions on validating the example with Schematron (Appendix 4).

Source: own.ruleml

  
    
      own
      person
      object
    
  
  
    
    
      
        buy
        person
        merchant
        object
      
      
        keep
        person
        object
      
    
  





  
    
      buy
      person
      merchant
      object
    
  
  
    
      sell
      merchant
      person
      object
    
  

 
 

 
    
      sell
      John
      Mary
      XMLBible
    

 

 
    
      keep
      Mary
      XMLBible
    
  


]]>

Source: datalog.xsd

	
		
			XML Schema for the RuleML sublanguage "datalog".
			File: datalog.xsd
			Version: 0.91
			Last Modification: 2006-07-19
		
	

	

	

	
	
	
	
	
	
	
	

]]>