Superchecker Duties

Guide for Superchecking Overheard Speech Files

Getting Started

1. Blabshare

   1. Follow the path: /Volumes/Fas-Phyc-PEB-Lab/OvSpeech/SubjectFiles/Seedlings/ overheard_speech

   2. Drag the assigned file from annotations-in-progress to annotations-to-be-superchecked folder

2. Open the Clickup task with the corresponding file name

   1. Move the task to "Superchecking In Progress" section and set the clip’s subtask status to “In Progress” for the specific clip you’re actively working on

3. Supercheck one clip at a time by following the below checklist:

MinCHAT

• Using the Grid view in Annotation mode, review for typos, nonsensical utterances, and format for all speakers

  • Before fixing the typo, listen to the clip to make sure that the "typo" wasn't intentional

  • Make sure there aren't any blank codes

• Check that all annotations on the CHI & EE1 tiers contain "0."

  • Make sure there aren't sneaky spaces before the annotations -- when you merge two annotations, ELAN sometimes inserts a space

• Wherever the annotator marked speech as unintelligible (xxx), listen to the clip (plus neighboring clips) to try to determine what the speaker said

• Run file through Minchat checker

  1. Export superchecked file as tab-delineated text

  2. Fix any new issues that have emerged

  3. Re-export as tab-delineated text and run through checker until all errors are resolved

Segmentation

• Verify whether the clip is "silent" or "speech" ; if the clip contains speech, segment appropriately

  • For background noise/voices: segment if you can clearly identify a sentence boundary regardless of whether you can make out the specific utterance

Speaker Consistency

• Listen to all clips for each speaker, one speaker at a time

• If you're unsure: read through the cast of characters (either located within the same corresponding OvS file folder or in the specific seedlings SubID folder)

Subtier Coding (VCM/XDS/CDS)

• Make sure there aren't any blank codes

• Wherever the speaker categorized vcm, xds, or cds as unknown (U), listen to the clip (plus neighboring clips) to try to determine the appropriate category

• Check that all xds@EE1 annotations are coded as "O"

• Check that the xds tier for each speaker does not contain a T* (Note: This is only applicable to ACLEW or VIHI files that were also used for Overheard Speech)

After File Completion:

1. Mark the task complete in ClickUp by sliding the file to the "Complete + not merged" section

2. Run the superchecking script

3. Push the checked .eaf, .docx., and .txt to Overheard_Speech github repo