Phase II Log File Documentation

Phase II involved verification of the utterances by native speakers of the individual languages, including:

  1. confirming that each utterance was in fact spoken by a native speaker of that language;
  2. verifying that the caller followed the instructions for that utterance;
  3. judging the callers' accents and dialects, and
  4. providing a brief description of the topic of each caller's one-minute elicited free speech response.
Log File Format

Files ending with ".lg2" are the information (ASCII text) files containing the qualitative judgments. For all the calls in a particular language a native speaker of the language was used to make all judgements.

The log file consists of a two-line header, a body containing utterance-specific judgments and global judgements.

The header: 

	<callnumber> labeler <labeler_name>  
	<callnumber> label_date <date>
The body: 
	<callnumber> <utt_type> <attribute> <values>
where 
<callnumber> -- self-explanatory

<utt_type> -- nlang, clang, htl, htc, room, meal, story-bt, story-at,
              generally in that order
For each <utt_type>, the following attribute-value pairs are present: 
<attribute>        <values>
  speech_type       extemp
                    read
                    recite
                    other: <desc>

  topic             <topic>                                 

  native            yes
                    no

  useful            yes
                    no

  complete          yes
                    no

  partial           yes
                    no

  instructions      yes
                    no

  dialect           standard                 [default]
                    unknown
                    <desc>

  accent            none                     [default]
                    <desc>

  background speech 	none                 [default]
                    	<desc>

  environmental noise 	none                 [default]
			<desc>

  gender            male
                    female
    		    unknown

  age               adult                    [default]
                    child

  intelligibility   poor
                    ok                       [default]

  connection        poor
                    average                  [default]

<topic>             a one-line summary of the topic of the story

<desc>              a one-line description of speech_type, dialect,
                    accent, background speech, or environmental noise.
The <speech_type> attribute only applies to the story portions of the calls. The values for this attribute indicate whether the story was more than half extemporaneous speech (extemp), read speech (read), or recited (recite). If the speech does not fall into any of the above categories, the labeler is required to briefly describe the speech_type of the story.

The <topic> attribute also only applies to the story protions of the calls. The labeler was asked to summarize (in english) what the caller talked about in the story.

When applied to the <nlg> utterance type the <native> attribute has a value of yes if the caller answered with the language name of the language being recorded. If the speaker did not say the language name, <native> is set to no and the labeler makes a judgment whether the speech signal is still useful speech. If so, <useful> is set to yes, otherwise to no. When applied as a general comment, the attribute has a value of yes if the labeler judges that the caller is a native speaker of the language (regardless of how the caller answered the <nlang> question).

The <complete> attribute is only applied to the <dow> and <num> utterance types. A value of yes means that all the days of the week or numbers zero to ten were present in the speech sample. If only part of a list exists then <complete> is set to no and <partial> is set to yes. If there is not even a partial list then <partial> is set to no and the labeler makes a judgement as to whether the speech data is still useful speech. If so, <useful> is set to yes, otherwise to no.

The <instructions> attribute applies to all utterance types except the stories, <nlg>, <dow>, and <num>. If the labeler judged that the caller followed the instructions for each utterance type then the <instructions> attribute is set to yes. When the caller did not follow instructions the labeler judged whether the speech was still useful. If so, <useful> was set to yes, otherwise to no. The attributes <dialect>, <accent>, <background speech>, <environmental noise>, <gender>, <age>, <intelligibility>, and <connection> apply globally. These are judgments made by the labeler about dialect, accent, etc.

Mike Noel
noel@cse.ogi.edu
(503) 690-1309