F
Oracle XML Grammar Subset

This appendix contains information about the Oracle XML Grammar Subset.

F.1 Oracle XML Grammar Subset

The Oracle Grammar Subset (OGS) is a subset of the XML Form of the W3C Speech Recognition Grammar Format (SRGS). The primary aims of the subset are:

• make OGS grammars self-contained so they can be transformed to other formats without resolving references to other grammars

• exclude features such as multiple entry points and recursive rule references that may not be supported by all gateways

• exclude syntax-like double-quoted-delimited tokens

• exclude features such as weights, and repeat probabilities whose semantics are not well defined by the Speech Recognition Grammar Specification (SRGS)

• restrict the use of semantic interpretation

• rule out some features of both XML Form grammars and semantic interpretation whose semantics are more likely to vary between implementations (such as ambiguous grammars), and references to undefined identifiers

Grammars in the OGS are not standalone XML documents; they have no XML header or DOCTYPE, but consist only of the root <grammar> element and contents. Such a <grammar> element and contents are in the subset if and only if they are legal according to the DTD of the W3C Speech Recognition Grammar Format, and satisfy the following additional restrictions:

• The root attribute of the <grammar> element is defined, and is a reference to one of the <rule>s in the grammar (which must, therefore, define at least one <rule>). The rule referred to by the root attribute of the <grammar> will be called the root rule.

• Any scope attributes defined have the value private.

• All <ruleref> elements have a uri attribute whose value begins with a #; that is, they are references to other rules in the same grammar. No <ruleref> has a type, alias, or special attribute.

• There are no loops in rule references. In other words:
  • No rule contains a <ruleref> that refers to itself.

  • There is no sequence of rules r1, ..., rn such that ri contains a <ruleref> that refers to ri+1 for all i = 1, ..., n-1, and rn contains a <ruleref> that refers to r1.

• There are no double-quote-delimited tokens; all multi-word tokens are delimited by <token> tags.

• The only entity references occurring in the grammar are &, <, >, ", and ', and those represented using decimal or hex codes.

• There are no <example>, <alias>, <meta>, <metadata>, or <lexicon> elements.

• No element has a weight or repeat-prob attribute.

• The grammar is not ambiguous, as defined in the Speech Recognition Grammar Specification.

• Any utterance recognized by a rule includes at most one <ruleref> occurring in that rule with a given uri attribute value (In other words, no rule can recognize an utterance that includes more than one <ruleref> that refers to the same rule).

• All <tag> elements contain text in a subset of the format described in W3C Semantic Interpretation for Speech Recognition (SISR), and satisfy the following additional restrictions:
  • The content of every <tag> is one of the following:
    • a double-quote-delimited string element

    • a grammar rule reference attribute, that is, a $ followed by the id attribute of a <rule>

    • a property of a grammar rule reference attribute, that is, a grammar rule reference attribute followed by a period (.) and a property name.

    • an AssignmentList as defined in the Semantic Interpretation specification (a semicolon-delimited list of assignments), in which the left-hand side of each assignment is an Identifier as defined in the Semantic Interpretation specification, and the right-hand side is one of the above three types of expression (a double-quote-delimited string element, a grammar rule reference attribute, or a property of a grammar rule reference attribute).

      For example, all of the following lines are allowed content of a <tag>:

      "sausage"
      $gender
      $animal.species
      what = "animal"
      where = "New York City";
      ; when = "now"
      who = "me"; why = "because"
      question = "how" ; how = "by hook or by crook";
      ;;
      food = $order
      diameter = $order.size
      what = "pizza"; size = $order.diameter
      
      

  • If a rule is not the root rule, it falls into one of the following categories:
    • the rule contains no <tag>s (we will call such rules void rules)

    • each utterance recognized by the rule includes exactly one <tag> in that rule, that <tag> is the last thing in the rule that is included in the utterance, and the content of the <tag> is a double-quote-delimited string, a grammar rule reference attribute, or a property of a grammar rule reference attribute (we will call such rules string rules).

    • each utterance recognized by the rule includes exactly one <tag> in that rule, that <tag> is the last thing in the rule that is included in the utterance, and the content of the <tag> is an AssignmentList (we will call such rules structure rules).

  • Every <tag> in a structure rule must assign to the exact same set of identifiers (that is, the exact same identifiers must appear as left-hand sides of assignments in the every <tag>'s content). The set of identifiers assigned to by the <tag>s in a structure rule will be called the signature of the rule.

  • If the content of a <tag> includes a property of a grammar rule reference attribute, the referenced rule must be a structure rule, and the property must be in the referenced rule's signature. For example, if a <tag> has content:

    $y.z 
    

    or

    x = $y.z 
    
    

    then the rule with id y must be a structure rule, and all of the <tag>s in rule y must include an assignment to identifier z.

  • If the content of a <tag> contains a grammar rule reference attribute with no property, the referenced rule must be a string rule. For example, if a <tag> has content:

    $y 
    

    or

    x = $y 
    
    

    then the rule with id y must be a string rule.

  • Every <tag> in the root rule contains an AssignmentList (the term structure rule applies only to non-root rules, so the root rule is not subject to the above restrictions on structure rules).

  • If an utterance is recognized by a rule (root rule or not), then the collection of <tag>s in that rule that are included in that utterance do not contain more than one assignment to the same identifier (not even if the expression the identifier is assigned to is the same).

  • If an utterance recognized by a rule (root rule or not) includes a <tag> in that <rule>, and that <tag> contains a grammar rule reference attribute $id (standalone or as part of a property), then that utterance must include a <ruleref> in the same rule as the <tag>, occurring before the <tag> in the utterance, with uri= #id.