Counting 2 segment deletions as 2 points off?

[marybeckman]

The current script only has the option of 1 point off for either or both segments deleted. As a result, the deletion of both the /m/ and the /p/ in the target /mp/ cluster at the end of /fI.k6.taemp/ is counted as prosodically no worse than the deletion of just the /m/ in the target /mb/ cluster at the end of /kE.d6w6mb/. (Example from first child to be transcribed in current exercise.) When we change to automate this, do we want to keep that scoring or count the removal of each target segment position separately?

[tjmahr]

I am under the impression that prosody scoring measures the general prosodic structure of the child's repetition using a list of criteria. Double sound deletions each get scored in the segment scoring with 0's, so docking them an extra prosody point seems redundant.

[patrickreidy]

I think we also need to consider how to treat the following possibilities: 1) double sound deletion, and 2) one sound deletion and one 'Unclassifiable'. Since deletion and 'Unclassifiable' are scored as 0 segmentally, then, if each deletion is not docked its own prosody point, then producing nothing is equivalent to producing something that is 'Unclassifiable'. This seems to go against the motivation for including 'Unclassifiable' as a segmental transcription.

[janroslynedwards]

I agree with Tristan. Jan

Sent from my iPhone

On May 2, 2014, at 1:31 PM, tj mahr notifications@github.com wrote:

I am under the impression that prosody scoring measures the general prosodic structure of the child's repetition using a list of criteria. Double sound deletions each get scored in the segment scoring with 0's, so docking them an extra prosody point seems redundant.

— Reply to this email directly or view it on GitHub.

[marybeckman]

Jan, you seem to be forgetting that "the general prosodic structure" is an only-recently added aspect of the "prosody transcription", and that the the original purpose of this judgment (back when we had only one prosody point) was to capture aspects of the local phonological pattern that could not be captured in the transcription of the three segmental features. Pat has identified one such aspect when he points out that not being able to identify any of the features of the target segment is different from noting that the leaf node for the target segment has been deleted entirely.

To be sure, this business of whether we want to count two Omitted segments separately or just once is not something that we need to decide now, since whichever way I set this up in the automatic calculation of this part of the approximated prosody score now, we can recover the other way when Pat builds the R script for exploring different ways of scoring different types of errors.

So I will revise the script to count this the easier way, and focus our discussion of prosody scoring on other issues, and especially those issues that are about interdependencies between segmental transcription and the prosodic parse that cannot be captured at all in the way that the prosody transcription is currently set up. These issues, again, are not ones that we should try to decide a priori now, but instead catalog and discuss this month, ready to try to finalize on the basis of some actual experience of trying to resolve inter-transcriber disagreements that we notice as we transcribe the first 15 children.

[marybeckman]

This issue will be moot after we have revised to accommodate to the new way of handling the prosody transcription / scoring, with one point for each target segment being in its target prosodic position (e.g., docking 1 if a following consonant has been deleted so that the target C in a VC sequence has been re-syllabified to be the onset of the following syllable) and one point for the frame as a whole not having been made "easier" by being shortened. So ...

Reframing issue as need to modify to accommodate.

[marybeckman]

This issue was addressed in the modifications that Franzo made in October, to implement the new prosody transcription protocol.