conversational turns in CLAN?

[Janet Bang]

Hello, 

We are trying to see if there's a way to automate a count of turns between a child and an adult within a transcript. We saw that with CHIP, we can get the number of responses to a child's utterance, but this value counts multiple utterances as separate responses, rather than one turn. 

For example, for the following interaction we would count 5 turns: 

MOT-CHI

CHI-MOT (MOT has 4 utterances, but count as 1 turn)

MOT (same MOT as one above)-CHI

CHI-MOT

MOT-CHI

*MOT: come here !

*CHI: xxx .

*MOT: come here let's sing the A_B_C_s@s .

*MOT: ready ?

*MOT: a@l b@l c@l d@l . 

*MOT: you don't want to sing ?

*CHI: no .

*MOT: are you sure ?

*CHI: no .

Is there some way to combine MLT and CHIP to get a value that represents turns as above? If that is too difficult, we could also go without the responses serving as the initiation of the next turn, meaning that the this would be 3 turns as in:

MOT-CHI

MOT(1 turn, 4 utterances)-CHI

MOT-CHI

Apologies if I've missed this in another function, 

Janet

--

Janet Y. Bang, Ph.D.

Postdoctoral Fellow

Department of Psychology

Stanford University

jb...@stanford.edu

[Brian MacWhinney]

Dear Janet,

    CHIP is designed to track patterns of feedback to the child, not turn-taking.  As I understand it, you are trying to actually count the "turns" as defined by a change of speaker.  If that is the emphasis, then you could create a rather complex cross-tier COMBO search patterns that looks for instance of *CHI followed in the next utterance by *MOT.  For your example, that would give you 3 such sequences.  You could create another pattern to search for the opposite sequence with *MOT followed right away with *CHI.  Perhaps you could read the part of the manual that talks about cross-tier COMBO for the details.

-- Brian MacWhinney, Teresa Heinz Professor of Cognitive Psychology, CMU

-- 
You received this message because you are subscribed to the Google Groups "chibolts" group.
To unsubscribe from this group and stop receiving emails from it, send an email to chibolts+u...@googlegroups.com.
To post to this group, send email to chib...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/chibolts/DM5PR02MB3275D80094B08E54CA78CCC5D7B80%40DM5PR02MB3275.namprd02.prod.outlook.com.
For more options, visit https://groups.google.com/d/optout.

[Nan Bernstein Ratner]

Wouldn't that be Mean Length of Turn? (MLT program?)

N

Nan Bernstein Ratner, F-, H-ASHA, F-AAAS, ABCLD

Professor

Hearing and Speech Sciences

University of Maryland

0100 Lefrak Hall

College Park, MD 20742

nra...@umd.edu, 301-405-4217

Co-director: FluencyBank (www.fluency.talkbank.org);  http://languagefluency.umd.edu/  

President, International Fluency Association (IFA; http://theifa.org)

Director, University of Maryland Autism Research Consortium (UMARC), www.autism.umd.edu

Faculty, Language Science (languagescience.umd.edu; Neuroscience & Cognitive Neuroscience (NACS, nacs.umd.edu), Developmental Science Field Committee

http://hesp.umd.edu/facultyprofile/bernstein%20ratner/nan

My PubMed Bibliography

[Brian MacWhinney]

Nan,

Yes, you are right.  MLT will do this.  I tested on this small sample from Janet and it reports 3 turns for MOT and 3 for CHI.

-- Brian

To view this discussion on the web visit https://groups.google.com/d/msgid/chibolts/CAAFocx5U0ENWSULgmRCyBFjmq8Brnmt-fJybUpbDg0Qv5E5P9w%40mail.gmail.com.

[Janet Bang]

Hello, 

We looked into MLT and what it provided was the number of continuous utterances by a single speaker, but not whether it was contingent on a particular speaker. Apologies, my example did not include the case for multiple speakers as in the following so we would want: 

MLT would provide: 

3 turns for MOT

1 turn for GMA

3 turns for CHI

Whereas what we are interested in is 5 turns if we count each response as an initiation (specifying the child and MOT/GMA as adults):

MOT-CHI

CHI-GMA

MOT-CHI

CHI-MOT

MOT-CHI

Or 3 turns between an adult and child if we always take the utterances before as an initiation and do not count each response as an initiation (this is similar to how LENA calculates their automated conversational turn measure):

MOT-CHI

MOT-CHI

MOT-CHI

*MOT: come here !

*CHI: xxx .

*GMA: come here and sit with me. 

*GMA: let's get you up here.

*MOT: come here let's sing the A_B_C_s@s .

*MOT: ready ?

*MOT: a@l b@l c@l d@l . 

*MOT: you don't want to sing ?

*CHI: no .

*MOT: are you sure ?

*CHI: no .

Basically, we are interested in trying to get an automated conversational turn measure, that is similar to LENA's, but using transcriptions that we have already vetted as being child-directed speech with our adult speakers of interested already identified. 

Janet

To view this discussion on the web visit https://groups.google.com/d/msgid/chibolts/B353A61B-1E37-46D2-8458-F00DB6A7308C%40cmu.edu.