New HITs forthcoming!

Discussion in 'Requesters' started by fharrell, Oct 18, 2012.

  1. fharrell

    fharrell User

    Joined:
    Oct 10, 2012
    Messages:
    56
    Likes Received:
    0
    I haven't stopped laughing yet! Even funnier, I say the same thing - "thank [deity of your choice]!" Anyway, none of you (from the most recent posts) have anything to worry about...I doubt we will be rejecting any of your HITs (Turkalicious)...especially if you have attempted to communicate with us somehow -- email, through this forum (if I can tell who is who), or even by using the comments field. We are almost ready to upload the email results, then we will begin working on phone. Should have those done today as well.
     
  2. fharrell

    fharrell User

    Joined:
    Oct 10, 2012
    Messages:
    56
    Likes Received:
    0
    We just uploaded results for the email HITs -- rejected 31 out of over 1900. We were rather lenient this time and threw out some of the controls...but we just couldn't overlook misspelling Brenda or Bellsouth. Working on phone now.
     
  3. Turkalicious

    Joined:
    Jun 4, 2012
    Messages:
    1,642
    Likes Received:
    0
    Thank you for all the communication Franklin, you're very awesome I must say. [I'm the Michele that you've emailed with a few times by the way]
     
  4. fharrell

    fharrell User

    Joined:
    Oct 10, 2012
    Messages:
    56
    Likes Received:
    0
    Oh, cool, good to know!

    We just approved the phone results and had to be more strict on these -- wound up rejecting 55 out of +/- 2000. Some of the controls were thrown out, but we based the majority of our decisions off of whether other workers were able to answer it correctly. Now we are looking at names, and I understand why I received so many comments on these HITs...it is going to take us a while to evaluate these results. Based on what I have seen thus far, we will be throwing out several of the controls. Stay tuned.

    --Franklin
     
  5. Kris

    Kris User

    Joined:
    Sep 21, 2012
    Messages:
    801
    Likes Received:
    0
    Looks like 2 of those phone rejects were mine. Just makes me wonder exactly which ones I screwed up. Now to see what happens with the names.
     
  6. Turkalicious

    Joined:
    Jun 4, 2012
    Messages:
    1,642
    Likes Received:
    0
    No rejections for me from yesterday. Now for the name ones I did today. -keeps fingers crossed-
     
  7. ivyleaf

    ivyleaf Member

    Joined:
    Sep 26, 2012
    Messages:
    10
    Likes Received:
    0
    3 of them are mine -- guess I'm done with these!
     
  8. Faiythe

    Faiythe User

    Joined:
    Aug 25, 2012
    Messages:
    393
    Likes Received:
    0
    all of my phone ones are approved, my name ones are pending*crosses fingers*
     
  9. footballchick29

    Joined:
    Aug 6, 2012
    Messages:
    2,150
    Likes Received:
    0
    i had 1.. not complaining at all though considering how many i have completed :)
     
  10. saladpope

    saladpope User

    Joined:
    Sep 14, 2012
    Messages:
    661
    Likes Received:
    0
    Couldn't agree more, particularly considering the fact that I have big fat Fred Flintstone fingers -- makes me very prone to typos :)

    Franklin - I have a couple of "day late and a dollar short" question/comments:

    When I was hitting emails where the caller just said "firstname-dot-lastname" I was putting "substitution required" in the comments so it would be obvious. But a couple of times it occurred to me that from a programmatic perspective (and especially in cases where it was something like "firstname letter P lastname" it might be easier for you if you used some sort of substitution variable scheme - it would not have been any more difficult for me to put %FIRSTNAME%.%LASTNAME%@gmail.com and it would almost have to be easier to deal with on the back-end. There is a chance that this scheme would err on the side of having too much faith in the AMT community generally here :p

    Also, in cases where there was a name mentioned, but it was only and clearly the CSR's name I submitted it as "no name spoken" but put the CSR name in the comments - I assume that differentiating in this manner was the correct thing to do?

    As a complete aside, I are these are going through some sort of speech-to-text before they show up on AMT? It seemed to me that they were, especially since the phone number ones that didn't have a phone number tended to have lots of other numbers (square footage, rent amounts, etc.) and the email ones that didn't have an email were normally of the CSR giving the URL of the community's site. Feel free not to answer that if you think it might compromise future quality :)
     
  11. fharrell

    fharrell User

    Joined:
    Oct 10, 2012
    Messages:
    56
    Likes Received:
    0
    So many to reply to...not sure where to begin. For the most part, the phone digits were evaluated against your peers. If we felt like the control was a fair one to use after listening to it again and if the majority answered the control right, then we had to reject those that didn't match. I don't know what we are going to do with names yet, but I expect we will be very lenient...like last time.

    Kris and Ivyleaf (and anyone else for that matter) -- if you want, email me your worker ID and I will be happy to send you the HITs that were rejected along with the audio file and why they were rejected.

    Saladpope -- thanks for adding that to the comments. Did you also choose "partial/fragmented email" as outlined in the instructions? This will make it much easier for us to find those instances where we need to listen to the full audio to determine the actual email address. Are you a programmer? Nerd? (wink) Because some of your comments make me think that. Anyway, yes, we are using a transcription algorithm to transcribe the recorded audio into text. We then query that text for certain keywords and in the case of phone numbers, we look for a string of 6 or more digits together. Without going into too much more detail, because of the way the SQL statement converts spoken numbers into digits, we wind up catching square footage ranges and price ranges. We have tried looking for a sequence of digits near certain keywords, but the way people say these keywords is seemingly endless and we still wind up missing a lot. That leaves us with just digits. Frustrating for all of you and more costly for us, but at least we know we aren't missing much. Also, I didn't even try to explain in the instructions to try to differentiate between speakers because I thought it would cause too much confusion, but to answer your question, yes, we are only interested in what the customer says, not the agent. A few others caught on to this as well. I would love to talk to you more offline if you want.

    Sorry for the very long reply, but I want to be very clear and open with all of you so that you know where we are coming from and know that we have put a lot of thought into the work we are putting out there. As always, I welcome your comments and feedback, either in this forum or offline. I am in the process of transitioning this project to another colleague so I can focus on the programming and automation. She has worked on the transcription team for several months, so she knows what it is like from your perspective. Enough said for now.

    --Franklin
     

Share This Page