QA Rejection Guidelines
Our objective is to improve the quality of transcription work and maximize the accuracy of the transcript that goes to QA. The 2-Step QA plays an integral role in this process. If you come across a poor chunk, it is important to not only fix the mistakes but also flag struggling transcribers. To do this, reject poorly transcribed chunks. The benefit of rejecting chunks is that, collectively, we can remove poorly performing transcribers from the system and thus significantly reduce the time and effort it takes you and other QAs to complete a session.
DO reject for:
You MAY reject for:
DO NOT reject for these if the rest of the transcript is good:
How to Reject
Do I reject before or after I correct the chunk?
Rejecting Sessions for Poor Transcription
When to Reject
The decision to reject largely boils down to a judgment call. A good guideline is to use common sense. You can usually tell when a transcriber has put forth sincere effort but may have made an honest mistake versus one who rushed through and did not do their job properly, thus creating more work for you.
DO reject for:
- Incorrect, capitalized, or misspelled tags ([guess?], [Laughter], [phone ringing], [cross-talk], [inudiabel], etc.)
- If any comments are inserted into a file (e.g., [I can't hear this], [this file is inaudible])
- If timestamps or speaker IDs are inserted in the text (Note: any orange timestamp boxes would have been added by another QA, not the TR, so should not be rejected.)
- Three or more typos/misspelled words
- Beginning and/or ending a chunk with dashes or ellipses
- Use of [inaudible] or [crosstalk] instead of striving to discern what was said
- Lack of research despite clear context
- An incomplete chunk where you must transcribe the missing portion
- Several errors of any kind that show a general lack of care
- Using BE spelling in a US file or US spelling in a BE file
- Nonsense sentences that indicate lack of proofing or a poor grasp of English
- Missing capital letters and/or punctuation throughout the chunk
- Multiple misheard, missing, or mixed up words
You MAY reject for:
- Misheard words
- Missing words
- Not following the Basic Style Guide
- Improper punctuation, spelling
If you are rejecting for the above, please always take into consideration the quality of the audio. Never reject a chunk for just one or two minor mistakes. On the other hand, numerous missing or misheard words in crystal clear audio would be a cause for concern and should be rejected.
If you come across an incomplete chunk or one that is very poorly transcribed and contains multiple errors of any kind, please fill out the form found in your respective QA team's Watercooler group.
DO NOT reject for these if the rest of the transcript is good:
- Spacing issues
- One misspelled word
- A few [inaudible] tags when the audio is somewhat difficult
- A few minor punctuation issues
- Not adhering to the Advanced Style Guide, e.g. not punctuating the tags according to the Advanced guide
How to Reject
To reject one individual chunk of transcription, click on the Reject Selected Block link. It is very important that you make sure the chunk you want to reject is highlighted blue (i.e., your cursor is in that chunk) when you click the link.
Click Reject Selected Block, and a box will open displaying different rejection reasons. Once you select the primary reason for the rejection, the box will close, and a pop-up will appear informing you that the file was rejected and the transcriber will not be paid. If you find that you are not able to reject a chunk, this means it has already been rejected by another QA or Ops member. After rejecting, please correct the chunk and continue with your session.
Missing Words: This means the transcriber left out several words. This can also be used for when a transcriber inserts [inaudible] for a full chunk, but you can make out words.
Misheard Words: Use this if the chunk has too many misheard words or the audio is clear and the mishears, therefore, strange. There is no set amount here, but always take into consideration the audio quality. Around 10 to 15 percent of the words misheard can be a good indicator of a poor transcriber. This can also apply to 'nonsense mishears' where the text makes no sense in context and Guess tags were not used.
Syntax and Grammar: This reason should be used for chunks with a large number of typos, misspelled words, or cases where the transcriber used the wrong homophone, e.g. "I herd their was work too due today."
Punctuation: Remember that transcribers do not have full context, and context makes a huge difference in transcription. For this reason, please do not reject for minor punctuation issues, especially near the chunk boundaries. But leaving sentences with no punctuation or inserting improper quotes should be rejected.
Style-guide Adherence: You should always reject if there are major style guide errors. However, please keep in mind that the transcribers are working from the Basic style guide. Any style details contained in the Advanced sections should be applied to the transcript as you edit, but should not be considered when deciding whether to reject the chunk.
Do I reject before or after I correct the chunk?
It doesn't matter. Whatever works best for you is fine. The transcriber will see both the rejection and your changes regardless. It's usually best to reject as you're making the edits so you don't forget.
Rejecting Sessions for Poor Transcription
Sometimes you might run into a file that has a number of chunks with unusually poor transcription. As a rule, you always want to check a few chunks at the beginning, middle and end of a file before you start making any changes. This will give you an overall idea of the transcription quality of the file. If you find that a large percent - say, 10-20% - of the chunks have been poorly transcribed or are incomplete, you need to reject the problem chunks then reject the full session for Terrible Transcription. This way, we save valuable QA time and effort for sessions that have already received a best effort from the transcription team, rather than redoing their work.
When you reject the session for Terrible Transcription, all the chunks that you rejected will return to the TR queue to be transcribed again. Once the chunks have been re-transcribed, the session will again be ready for the next available QA. The chunks you do not reject will not be affected.
Note that rejecting a full file for Terrible Transcription without rejecting any of the poorly transcribed chunks only returns it to the QA queue with no edits. So please make sure if you are going to reject a full file that you first reject those poorly transcribed chunks.
Also note that rejections are tracked. QAs who are rejecting sessions for Terrible Transcription that are of acceptable quality for QA will be contacted with a warning.