Replicate Clark and Wilkes-Gibbs (1986) in your chosen population and in english.
The parameters of your experiment are:
Use 6 somewhat similar or hard to describe images (tangrams or otherwise).
Make people take turns talking about them.
Measure something to do with linguistic alignment and/or turn-taking.
Show the images 3 times each in the experiment.
Develop an experimental procedure (see bottom of handout for C&WG procedure)
Decide how to present your pictures (How will you determine their arrangement? Do all pictures appear on all trials? Do both participants have all of the same information?)
Decide how many trials to present per pair of people.
Determine if there are any other pieces of information you want to collect from the participant, such as their first language (L1) or age. Collect only information that you think might be important.
Determine any other details that are necessary to answer your research question.
In developing your procedure, do consider any obvious sources of experimental error. For example, you might balance the location of images across trials.
Find one pair of people willing to serve as subjects in your experiment. Don’t ask someone to be a participant if you’ve already told them about the project. (This means you may have to keep the hypothesis of the experiment a “secret” from your roommates, etc., until after you’ve had them do it.)
Collect your data!
Get informed consent: Ask each person if they agree to participate in a paired picture describing experiment that will take about 10 minutes. Verbally get their agreement to be recorded.
Conduct the experiment. Record audio (for a spoken language) or video (for signed language) on your phone/computer. Depending on your question of interest, you may also want to time the conversational turns. You can absolutely conduct this over Zoom.
After conducting the experiment, you can ask some follow-up questions (e.g., “What do you think the purpose of this study was?”).
Debrief your participant: Tell them what the study was about.
Tabulate your results in a spreadsheet. Pay close attention to the formatting and to the example below!
Each row should represent one response. It should show any participant-level data, what was in the trial, the role each participant had in each trial, the participant’s turn number in the trial, and what each participant said. You can use codes for each picture (‘PicID’ — T1, T2, etc). Participant-level data includes an ID number for each participant (not their name!) and each pair of participants, along with any other info about that participant that you collected (such as their age or native language).
Information that is the same across multiple trials of the experiment (such as the participant info for multiple trials done by the same participant) should be duplicated across the corresponding rows.
Some definitions here: a TRIAL is all the pictures that are presented together (all 12 tangrams in Clark & Wilkes-Gibbs), a PICTURE is all the descriptions related to one picture, and a TURN is the tradeoffs between director and matcher. Each trial in C&WG has 12 pictures, and each picture will typically have 2 turns (director, and matcher), and possibly many more.
Transcribe the results of both participants.
Listen to/rewatch your recording. Write down the descriptions used on each trial, including filled pauses like ‘uh’ and ‘um’. Transcribing requires lots of rewinding, so do this carefully and systematically so that you record precisely what was said.
You may use an AI service like otter.ai if you wish to give a *first pass* to your transcription– you need to listen through and check it for accuracy, adding filled pauses and disfluencies as you hear them. Please cite this service in your references if used.
Figure out how to code some dependent measure, perhaps per describer, per trial, per picture, and/or per turn. Just code ONE: this is quite time consuming.
Some options include: number of words per picture on the first turn, number of turns per picture or per trial, amount of time per picture or per trial, number of ‘um’s, number of ‘a/an’s (these are all things that have been shown to change over reference development). You can also come up with something else to measure if you would prefer to do so.
Then, figure out what parts of your transcript file need to go in your coded data file. This will depend on whatever you coded—add only what you need in order to mark the ‘unique level of observation’—like these 3 examples that each could go with the transcript file above.
Write it up! Follow a structure of four labeled sections: Introduction, Methods, Results, and Discussion. You can add references at the end but don’t need to do outside research. Aim for a length of 3 to 5 pages (double spaced). Don’t stretch to make the maximum page limit: Short and clear will earn you a better grade in this class than long and wordy.
Introduction should describe the problem and provide small amount of background. You can use the textbook, lecture, and the Clark & Wilkes-Gibbs paper as your sources; you don’t need to do outside reading. This will probably be 1-2 paragraphs. It’s good practice to include any references in a short reference list at the end.
Methods should show the stimuli you used (pictures in arrays) and give a brief explanation of why you decided to use those particular ones. It will also describe your procedure in enough detail that we can clearly tell what you did. This will probably be 1-2 paragraphs.
Results should include your findings. Include at least one figure or table of your results, appropriately captioned. This will probably be 2-3 paragraphs.
Discussion should include what hypothesis you think the data support, note any potential problems with the way you did the study, and list any ideas for future work. You can also mention any anecdotal information you found that may be important (e.g., “My participants were a couple and they used some inside jokes”). This will probably be 2-3 paragraphs.
References come at the end, and can be in any standard format that you like. The reference list on Canvas is in APA style and you can use that as a starting place.
Procedure used in Clark & Wilkes-Gibbs (1986)
-Assign ‘director’ role to one participant and ‘matcher’ role to the other.
-Participants are seated across from each other with dividing screen so they can’t see each other’s pictures. (Some nice low-budget dividers: cardboard box, 3-hole binder, a stack of books)
-On each of 6 trials:
-Give the director a set of images in a particular order (they used 15×20 cm cards in 2 rows of 6). The experimenter arranged the image cards each time for the director.
-Then give the matcher a set of 12 image cards in a stack.
-Director has to get matcher to arrange cards in the same order as they have them in.
-Experimenter checks arrangement for errors.
-Each of the trials has different arrangements of pictures, making it a bit more challenging.
-Experimenter then codes the data!
