SWIlanguageModel n-gram

Note: Information on this feature is provided for backwards compatibility.

You can use n-gram grammars to apply language model weights to SRGS grammars. This is done in the SRGS grammar with a <meta> element that specifies the desired n-gram grammar; for example: 

<meta name="SWIlanguageModel" content="trigramu.xml"/>

When an n-gram grammar is used for this purpose, it is known as a SWIlanguageModel n-gram and its syntax differs from the n-gram grammars described elsewhere. Guidelines for this special n-gram:

  • SWIlanguageModel n-grams use the <lexicon> instead of <vocab>, and (unlike <vocab>) the <lexicon> tag cannot use <ruleref> to import other SRGS rules using the uri attribute.
  • This grammar does not support SRGS pronunciation dictionaries.
  • This n-gram can include the use of a Unigram Class, specified by the <import> tag. See Example (trigramu.xml).
    • The <import> element may only reference unigram n-gram grammars.
    • The imported unigram n-gram grammars cannot use <ruleref> to import SRGS rules or other n-grams.
    • The import attribute must be of the form name#CLASS where name is the name attribute from an import and CLASS matches a rule in the SRGS grammar.
    • The CLASS rule in the SRGS must have the attribute SWIListClass="1".
  • This n-grams ruleref's may only reference alias imports specified with the <import> element.
<?xml version='1.0' encoding='UTF-8'?>
<grammar xml:lang="en-US" version="1.0" root="ROOT"
   tag-format="swi-semantics/1.0"
   xmlns="http://www.w3.org/2001/06/grammar"
   mode="voice">
 <meta name="SWIlanguageModel" content="trigramu.xml"/>
 <rule id="ROOT" scope="public">
  <item repeat="1-">
   <one-of>
    <item>elephants</item>
    <item>like</item>
    <item>eating</item>
    <item> <ruleref uri="#MY_CLASS"/> </item>
   </one-of>
  </item>
 </rule>
 <rule id="MY_CLASS" SWI_listClass="1">
  <one-of>
   <item>apple</item>
   <item>pear</item>
   <item>plums</item>
  </one-of>
 </rule>
</grammar>
<?xml version='1.0' encoding="UTF-8"?>
<N-Gram>
 <!-- SWIlanguageModel uses lexicon instead of vocab -->
 <lexicon>
  <token index="1">apple</token>
  <token index="2">pear</token>
  <token index="3">plums</token>
 </lexicon>
 <tree>
  <node>3 14</node>
  <node>1 2</node>
  <node>2 4</node>
  <node>3 8</node>
 </tree>
</N-Gram>
<?xml version='1.0' encoding="UTF-8"?>
<N-Gram>
 <import uri="unigram.xml" name="uni"/>
 <lexicon>
  <token index="1">-pau-</token>
  <token index="2">elephants</token>
  <token index="3">like</token>
  <token index="4">eating</token>
  <!-- Import a Unigram Class -->
  <token index="5"><ruleref import="uni7#MY_CLASS"/></token>
  <token index="6">-pau2-</token>
 </lexicon>
 <tree>
  <node>4 210</node>
  <node>1 2 110</node> <!-- "-pau-" -->
  <node>2 1 100</node> <!-- "-pau- elephants" -->
  <node>3   50</node> <!-- "-pau- elephants like" -->
  <node>2 1 100</node> <!-- "elephants" -->
  <node>3   50</node> <!-- "elephants like" -->
  <node>3 1 50</node> <!-- "like" -->
  <node>4 1 50</node> <!-- "like eating" -->
  <node>5   25</node> <!-- "like eating MY_CLASS" -->
  <node>4 1 50</node> <!-- "eating" -->
  <node>5 1 25</node> <!-- "eating MY_CLASS" -->
  <node>6   25</node> <!-- "eating MY_CLASS -pau2-" -->
 </tree>
</N-Gram>