- Coders rate odd and even patient numbers.
- We start with first and last session, then backfill other sessions when time allows
- Sessions are already broken into chunks of 5 minutes.
When analyzing a chunk our goal is to code exchanges which contain turns:
- A turn is one person speaking for minimum of 3 seconds. If longer than 5, cut at the first natural break after 5 seconds.
- A exchange is two adjacent turns.
- Turns within two different exchanges can't be adjacent (i.e. there has to be a gap of at least one turn)
- One we have coded an exchange, we scan for and sample the next exchange in which the order of speakers is reversed.
- Any number of exchanges can be coded for a chunk
We will have three tiers defined in Praat used to code turns and exchanges. These are:
- Patient
- Therapist
- Comments
In patient and therapist tiers we will mark turns, and include transcripts of what is said in each turn.
In the comments channel we will use lower case letters and numbers to encode these variables:
- Who is speaking (p or t: patient or therapist)
- Exchange number within the chunk (1, 2...)
- Turn within the exchange (always 1 or 2)
- Turn type (s,e,c: start, end or complete sample from the utterance)
- Backchannel present or not (b,n)
For example:
- "t12cn" means: therapist speaking, Exchange=1, Turn=2, Turn type=complete, no backchannel.
- "p11cb" means: patient speaking, Exchange=1, Turn=1, Turn type=complete, backchannel present
After the code, we will put a colon, followed by comments on the turn, e.g. recording what backchannel was heard. A complete example would be:
Patient tier: [empty]
Therapist tier: t11cn: "So how do you feel today?"
Comments tier: t11cn:
The following turn might be:
Patient tier: p12cb: "Really miserable, my cat died"
Therapist tier: [empty]
Comments tier: p12cb: backchannel "mmmm"
Our final spreadsheet will have a row per turn, with columns something like this:
-
PatientID
-
TherapistID
-
Session# (1, 2... 24)
-
Chunk (1, 2...)
-
Exchange (1,2...)
-
Turn (1 or 2)
-
TurnType (s, e, c)
-
Who (p or t)
-
BackchannelPresent (1/0)
-
Comments (string)
-
TurnStart (ms)
-
TurnDuration (ms)
-
Transcription (string)
-
Prosodic variables