dbdc3

Submission format for dialogue breakdown detection results

This page shows the submission format for dialogue breakdown detection results. The evaluation script can only accept this format. Refer also to this section for the information about how to submit your runs at the formal-run.

Naming convention for input/output files

The input dialogue file for dialogue breakdown detection is <dialogue-id>.log.json. The output result file will be <dialogue-id>.labels.json. For example, if an input dialogue file is 1408278293.log.json, then the output result file will be 1408278293.labels.json.

Format of the output result file

The output result file should be in a JSON format．An example is given below:

{
  "dialogue-id" : "1408278293",
  "turns" : [ {
    "turn-index" : 0,
    "labels" : [ {
      "breakdown" : "O",
      "prob-O" : 0.8,
      "prob-T" : 0.2,
      "prob-X" : 0.0
    } ]
  }, {
    "turn-index" : 2,
    "labels" : [ {
      "breakdown" : "O",
      "prob-O" : 0.75,
      "prob-T" : 0.15,
      "prob-X" : 0.10
    } ]
  }, {
    "turn-index" : 4,
    "labels" : [ {
      "breakdown" : "X",
      "prob-O" : 0.0,
      "prob-T" : 0.0,
      "prob-X" : 1.0
    } ]
  }, {
         ・・・
  }, {
    "turn-index" : 18,
    "labels" : [ {
      "breakdown" : "T",
      "prob-O" : 0.2,
      "prob-T" : 0.6,
      "prob-X" : 0.2
    } ]
  }, {
    "turn-index" : 20,
    "labels" : [ {
      "breakdown" : "O",
      "prob-O" : 1.0,
      "prob-T" : 0.0,
      "prob-X" : 0.0
    } ]
  } ]
}

Description of each field

• dialogue-id ： This ID should match that of the input dialogue file.
• turns ： This field contains all the dialogue breakdown detection results for all turns.
• turn-index ： This field denotes the index of a turn and should match that of the input dialogue file.
• labels ： This field contains the dialogue breakdown detection results for a turn.
• breakdown ： This can be one of O or T or X, corresponding to not a breakdown, possible breakdown, and breakdown, corresponding to the hard-decision made by a dialogue breakdown detector.
• prob-O ： Probability that the dialogue breakdown detection result is O (not a breakdown).
• prob-T ： Probability that the dialogue breakdown detection result is T (possible breakdown).
• prob-X ： Probability that the dialogue breakdown detection result is X (breakdown).