What you will receive
- You will receive a folder of .aiff or .wav. sound files
- Each file is a 6 to 10 minute conversation between two people
- You will need to open these files in a program called Praat that is already downloaded on the lab computers (you can download it, but Mac users need to download version 5.3.14)
http://www.fon.hum.uva.nl/praat/
Getting set up in Praat
- Each sound file needs to be associated wtih what Praat calls a “textgrid” file
- These textgrid files are where boundary markings are entered and recorded. Critical!
- See detailed instructions below and watch the following YouTube video:
Criteria of what speech utterances should be bounded
There are five major utterance types that you will be coding:
- Conversational turns: “u”
- Backchannels: “*” and/or “$”
- Extended backchannels: “u*”
- Failed floor attempts: “#”
- Garbage: “^”
Complete conversational turns, that is, when a speaker holds the conversational floor. The duration of such turns can be long or quite short. A conversational turn ends when the speaker’s partner takes the floor (thereby becoming the new speaker). For these type of boundaries, type the letter “u.”
Backchannel responses occur when a listener (the “non-floor-holding” participant) uses an utterance designed for social or meta-conversational purposes. These are utterances like “right,” “yes,” “ok,” “for sho.” These occur when a speaker (the “floor-holding” participant) has the floor, and does not constitute an attempt by the listener to take the floor. Please mark such occurrences with an * symbol. Another type of backchannel that needs to be marked is laughter. Please mark such occurrences with a “$” symbol.
Extended backchannel response are a slight derivation of a backchannel. Again, these occur when a listener says something while the speaker has the floor, but the contribution is not meant to take the floor. However, the contribution is more than a simple “hm hm” or “yes,” but might be a few words or more. Please mark such occurrences with an u* symbol.
A failed floor attempt is when the listener attempts to take the fllor but the speaker will not relinquish the floor. This is a failed floor attempt and should be marked with an # symbol.
Garbage occurs mostly when participants talk to the experimenter or address someone other than their partner. People also may make weird grunting sounds, clear their throats, etc. If this happens, mark these boundaries but make sure to also code them with a ^ symbol.
Keyboard shortcut keys.
- Use TAB to control playback
- Command + N to zoom into selected region
- Command + O to zoom out incrementally
- Command + A to show the entire timeline
- Option + delete to remove errant backchannel boundary-markings
Detailed Instructions for boundary marking with Praat
Begin by watching the following Youtube tutorial videos:
Written instructions:
- Open an audio file: in left Objects window, Open -> Read from file…
- Create a new TextGrid: Annotate -> To TextGrid …
A dialogue box will appear, type:
- For “All tier names:” type in “A B” for speakers A and B (do not include quotation marks, make sure A B is space delimited)
- For “Which of these are point tiers?:” leave it blank
- Hit OK
- Select both Sound and TextGrid then hit View & Edit
- Click and drag to select a portion of a waveform where conversation begins
- Command + N to zoom to selection (Command + N)
- Play the selected portion by clicking on a gray rectangle specifying segment duration, located either above the waveform or below the label text box (= tier). Also, to start and stop playback, press “TAB”
- Adjust boundaries
- Specify an interval in an appropriate text box
- Position the cursor at the initial boundary then click
- Click inside the circle in the tier specified for the speaker of the utterance
- Boundary is created when red double lines appear
- Repeat for the end boundary
- When finished, close editing window and go to:
- Save in Objects window -> Save as text file
Record your impressions, thoughts, etc.
At the end of each session, or as you are boundary marking, jot down whether anything weird happened in the conversation. That is, did people fail to agree with each, did they really go off topic, was the conversation very awkard, did one person seem more dominant/persuasive? Just record your thoughts and impressions. I’m looking for things that might compromise the analysis.
Tips
RAs have also found that coding one person at a time makes things easier. If doing this with head phones, you can just listen to one channel at a time. You may also want to do one pass of boundary marking, and then a second pass where you place the conversational codes.