XML-based Natural Language Generation
Practical Exercises

Template-based Generation

• No separate practical exercise. Do Assignment 1: Template-based Generation.

Building with Ant

• Introduction to Ant, from Ant - The Definitive Guide (Jesse Tilly & Eric Burke)
• Exercise 1: Set up the Ant framework.
  • Make a new directory XMLGen for your practical work on XML-based Generation.
  • Copy the Ant build file build.xml into XMLGen
  • Make 4 subdirectories: XMLGen/lib, XMLGen/src, XMLGen/data, XMLGen/xslt
  • Copy these .jar files into XMLGen/lib
  • In XMLGen, run Ant by "ant".
  • Check the difference between "ant help", "ant -help", and "ant -projecthelp"
• Exercise 2: Build a simple XSLT transform program from Java and XSLT (Eric Burke)
  • Copy the program SimpleJaxp.java into XMLGen/src
  • Compile the program by "ant compile"
  • Copy hello.xml into XMLGen/data and copy hello.xsl into XMLGen/xslt
  • Run the program by "ant simpletransform"
  • Remove the compiled program by "ant clean"
  • Recompile and rerun the program by "ant simpletransform"

Getting XML Content

• How to parse CSV files to get XML content, explained in Java and XSLT (Eric Burke)
• Exercise 3: Build a simple CSV-to-XML processor program from Java and XSLT (Eric Burke)
  • Copy AbstractXMLReader.java, CSVXMLReader.java and SimpleCSVProcessor.java into XMLGen/src
  • Compile the programs by "ant compile"
  • Copy the example CSV data file presidents.csv into XMLGen/data
  • Run the compiled programs by "ant csvtoxml" to convert the CSV data to raw XML data
  • Copy csvToHTMLTable.xslt into XMLGen/xslt
  • Run "ant csvtohtml" to convert CSV-to-XML followed by XML-to-HTML
• Now do Assignment 2: Getting XML Content.

Swing XSLT Transformer

• Exercise 4: Build a Swing XSLT Transformer program from Java and XSLT (Eric Burke)
  • Copy SwingTransformer.java, TransformerWindow.java, XMLOutputPanel.java and ErrorListenerModel.java into XMLGen/src
  • As this program uses JDOM, you need jdom.jar in XMLGen/lib
  • Compile the programs by "ant compile"
  • Run the program by "ant swingtransform". This will open a new window "Swing XSLT Transformer"
  • Use the Select XML button to choose an XML file to transform, for example XMLGen/data/hello.xml
  • Use the Select XSLT button to choose an XSLT stylesheet, for example XMLGen/xslt/hello.xsl
  • Click the Transform button, then view XML Output (or Transformation Problems, if there are errors)

Transforming and Validating XML

• No Java programs needed for Exercises 5 and 6 (prove it by doing "ant clean" first),
  but before using Xalan and Xerces, you need optional.jar in XMLGen/lib
• Exercise 5: Transform XML with Xalan XSLT processor
  • Ensure hello.xml is still in XMLGen/data and hello.xsl is still in XMLGen/xslt
  • Run Xalan XSLT processor by "ant xalantransform".
  • Check the output in XMLGen/build/hello.html
• Exercise 6: Transform with Xalan and Validate with Xerces XML parser
  • Ensure liter.csv.xml is in XMLGen/data
  • Ensure your version of liter.csv-xml.xslt is in XMLGen/xslt (from Assignment 2)
  • Run Xalan by "ant litertransform" to make XMLGen/build/liter.xml
  • Run Xerces by "ant litervalidate".
• Now do Assignment 3: Text Planning.

XML Transformation Pipeline

• Exercise 7: Build an XML transformation pipeline
  • Copy NLGPipeline.java into XMLGen/src
  • Ensure liter.xml is in XMLGen/data
  • Ensure your version of liter.content.xslt is in XMLGen/xslt (from Assignment 3)
  • Run "ant nlgpipecontent" to do content determination only
  • Ensure your version of liter.discourse.xslt is in XMLGen/xslt (from Assignment 3)
  • Run "ant nlgpipediscourse" to do content determination and discourse planning.
    This should show the text plan created by the two text planning stages
  • Run "ant nlgpipelexical" to do lexicalization after text planning
    This should fail, warning that the stylesheet liter.lexical.xslt is not in XMLGen/xslt
• Now do Assignment 4: Microplanning.
• Exercise 7 (continued)
  • Ensure your version of liter.lexical.xslt is in XMLGen/xslt (from Assignment 4)
  • Run "ant nlgpipelexical" to do lexicalization after text planning
  • Ensure liter.aggreg.xslt is in XMLGen/xslt (from Assignment 4)
  • Run "ant nlgpipeaggreg" to do lexicalization and (null) aggregation after text planning
  • Ensure your version of liter.refexp.xslt is in XMLGen/xslt (from Assignment 4)
  • Run "ant nlgpiperefexp" to do text planning, lexicalization, aggregation and referring expressions.
• Now do Assignment 5: Surface Realization.

Generation Algorithms

• These exercises require Sicstus Prolog.
• Exercise 8: Van Noord's Bottom-Up Generator (BUG) algorithm
  • Copy BUG1.pl into your directory and study it.
  • Start Prolog and load the generator by the shell command prolog -l BUG1.pl
  • Run the test suite by the Prolog query ?- test. (don't forget the fullstop)
  • Stop Prolog by ?- halt.
• Exercise 9: Phillips' bag generation algorithm
  • Copy ChartGen.pl into your directory and study it.
  • Start Prolog and load the parser/generator by prolog -l ChartGen.pl
  • Run the test parse by the Prolog query ?- test_parse.
  • Run the test generation by the Prolog query ?- test_gen.
• These examples are intended to show the basic algorithms in the simplest possible form.
  If you know Prolog, try extending the examples to generate some new sentences.

Natural Language Generation and Speech Synthesis

• Exercise 10: Build a complete NLG pipeline
  • Copy a new Ant build file build2.xml into XMLGen
  • Ensure NLGPipeline.java is in XMLGen/src
  • Ensure liter.xml is in XMLGen/data
  • Ensure your version of liter.content.xslt is in XMLGen/xslt (from Assignment 3)
    Run "ant -f build2.xml nlgpipecontent" to do content determination only
  • Ensure your version of liter.discourse.xslt is in XMLGen/xslt (from Assignment 3)
    Run "ant -f build2.xml nlgpipediscourse" to do content determination and discourse planning.
    This should show the text plan created by the two text planning stages
  • Ensure your version of liter.lexical.xslt is in XMLGen/xslt (from Assignment 4)
    Run "ant -f build2.xml nlgpipelexical" to do lexicalization after text planning
  • Ensure liter.aggreg.xslt is in XMLGen/xslt (from Assignment 4)
    Run "ant -f build2.xml nlgpipeaggreg" to do lexicalization and (null) aggregation after text planning
  • Ensure your version of liter.refexp.xslt is in XMLGen/xslt (from Assignment 4)
    Run "ant -f build2.xml nlgpiperefexp" to do text planning, lexicalization, aggregation and referring expressions.
    This should show the text specification created by the text planning and microplanning stages
  • Ensure your version of liter.realize.xslt is in XMLGen/xslt (from Assignment 5)
    Run "ant -f build2.xml nlgpiperealize" to do text planning, microplanning and surface realization.
    This should show the natural language text generated by the complete NLG pipeline
• Exercise 11: NLG and speech synthesis
  • Copy PipeSpeak.java into XMLGen/src
  • Copy these .jar files into XMLGen/lib
  • Copy this speech.properties file into your home directory
  • Ensure liter.xml is in XMLGen/data
  • Ensure all the XSLT stylesheets from Exercise 10 are in XMLGen/xslt
  • Run "ant -f build2.xml nlgpipespeak"