2

I am trying to connect the dialogue acts of the Switchboard Dialogue Acts Corpus with the word alignment timing information available here.

I can see how it is possible to match up every utterance with every word. For example, this utterance is labeled with the conversation number:

| swda_filename            | ptb_basename | conversation_no | transcript_index | act_tag | caller | utterance_index | subutterance_index | text                                                                       |
|--------------------------|--------------|-----------------|------------------|---------|--------|-----------------|--------------------|----------------------------------------------------------------------------|
| sw00utt/sw_0001_4325.utt | 4/sw4325     | 4325            | 3                | +       | A      | 3               | 1                  | What kind of experience [ do you, + do you ] have, then with child care? / |

And can be aligned word-by-word to the information available in individual ISIP files, e.g.:

sw4325A-ms98-a-0005 4.495250 4.705000 what
sw4325A-ms98-a-0005 4.705000 4.908250 kind
sw4325A-ms98-a-0005 4.908250 5.008750 of
sw4325A-ms98-a-0005 5.008750 5.669625 experience
sw4325A-ms98-a-0005 5.669625 5.801375 [silence]
sw4325A-ms98-a-0005 5.801375 5.889125 do
sw4325A-ms98-a-0005 5.889125 6.122625 you
sw4325A-ms98-a-0005 6.122625 6.321625 do
sw4325A-ms98-a-0005 6.321625 6.471625 you
sw4325A-ms98-a-0005 6.471625 6.801625 have
sw4325A-ms98-a-0005 6.801625 7.168375 then

I am sure this alignment would contain a lot of errors and one-off misalignments/exceptions. Is this sort of utterance-level information available with word-by-word annotation, so that I do not need to duplicate an existing effort?

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.