June 1997

This release of CallHome Arabic data consists of both the romanized
and Arabic script versions of the 80 training, 20 devtest, and 20
evaltest transcripts.  

Three subdirectories can be found here:

	doc/
	tools/
	transcrp/


Each of these contains the following:

    doc/    contains ten files that describe the transcript
            corpus:

            ar-trans.doc    - describes Arabic transcripts
            callinfo.doc    - describes callinfo.tbl
            callinfo.tbl    - provides audit information for each channel
            devtest.ids     - provides a list of callids for the devtest
            evaltest.ids    - provides a list of callids for the evaltest
            spkrinfo.doc    - describes spkrinfo.tbl
            spkrinfo.tbl    - provides demographic information
                              on the telephone call originators
            iso-spec.doc    - describes ISO encoding used for
                              Arabic script
            scr2rom.tbl     - describes LDC romanization to Arabic
                              script correspondence
            train.ids       - provides a list of callids for
                              training set


    tools/ contains two sets of tools:

	mule/
		package for viewing and editing ISO 8859-6 Arabic text
		(useful for the ".scr" files in this release)

	scr_conv/
		 tools used to create the Arabic script (.scr) files
		 Includes a simple parser for the romanized transcripts
		 that may be useful for other purposes.


    transcrp/ contains three subdirectories:


	train/	contains two subdirectories:

		roman/	contains the 80 training transcripts
			in romanized (original) form (.txt files)

		script/	contains the 80 training transcripts
			in Arabic script form (ISO 8859-6 Arabic)
			(.scr files)

	devtest/ contains two subdirectories:

		roman/	contains the 20 devtest transcripts
			in romanized (original) form (.txt files)

		script/	contains the 20 devtest transcripts
			in Arabic script form (ISO 8859-6 Arabic)
			(.scr files)

	evaltest/	contains two subdirectories:

		roman/ contains the 20 previously-released evaltest
		       transcripts in romanized (original) form (.txt
		       files)

		script/ contains the 20 previously-released evaltest
			transcripts in Arabic script form (ISO 8859-6
			Arabic) (.scr files)