CastingWords Turker Blog

Transcribing on dial-up

Just because I'm on dial-up doesn't prevent me from accepting CastingWords HITS. All I do is open a window and let the audio file download. Sampling the audiofile before accepting the HIT gives me a headstart on the download. While it's downloading, I simply open another window and accept other quick HITS for a penny or so. Otherwise, I try transcribing the first few minutes of the downloaded file, which works for me because by the time I finish transcribing that snippet, a new snippet is available for me to do. To save on dial-up costs, as soon as the audiofile is fully downloaded, I right-click the file, save it on my desktop, and disconnect. I then upload the audiofile on my Express Scribe and continue transcribing. I keep the HIT window open on my computer -- the timer is still running even if I am disconnected so I get to pace myself. When I'm done editing my work, I log on again and send the file. At least three hours shaved off my dial-up internet account usage! Soon, I hope to make enough money from this to afford a DSL account.

Posted by Cabecou

Increasing Productivity

There are many things you can do to increase your productivity while doing podcasts. The easiest thing that I've found is slowing the audio down. This can easily be done using a winamp plugin (such as "Slow Me Down" for winamp ). Using these slows down the talking without affecting the pitch of the speaker, giving you more time to type.

Another thing you can do is use hotkeys. Instead of using your mouse to pause, rewind, and play the audio file. Set up your global hotkeys to do these so your hands don't have to leave your keyboard. Even better is if you happen to have a set of joystick-pedals from a racing or driving game that plug into your computer. There are programs out there to convert their movement into controlling the speed and direction of an mp3 file in winamp. Remember, keeping your hands typing will speed up your transcribing because your thought processes doesn't have to stop when you need to adjust the playback!

Posted by Andrew F

How Accurate Should a Transcript Be?

First of all, this is my own opinion and not anything officially approved by Casting Words. Over the past few months, I've had a chance to see many transcripts and do some myself. When I first started transcribing, I tried to make the text an exact replica of everything that the speakers said, recording every ummm, ah, you know, and partial sentence start. Over time, I've come to realize that readability can be greatly improved by abandoning that approach. We speak and write differently and I think that has to be taken into account when transcribing. Here are three simple things that you can do to make your transcripts much more readable:
1) Eliminate "you knows".
2) Eliminate false sentence starts.
3) Repair sentence fragments when possible.

So, unless the notes say otherwise, I think it's best to favor readability over 100 percent accuracy by following the rules above.

Posted by bjs

Learn From Your Mistakes

If you're like me, the thought of receiving a grade for your work can be scary. I still shudder at the memory of receiving papers back in grade school with red editing marks all over the place. However, you owe it to yourself (and to the Turkers who will edit your work) to check in at the turkers.castingwords.com site and review the feedback that your Turking peers have provided. Armed with your Worker ID at the site, you'll have access to the status of your CastingWords' HITs as well as the feedback generated by the "grading" HITs completed by your fellow Turkers.

Yes, some of the graders can be unnecessarily harsh. Don't take the criticism personally. Look for trends that will help you correct things that you may not even realize you're doing. For example, I didn't realize that the rules for labeling speaker names had changed until I read the comments pointing out my error. If you're still in doubt after reviewing the comments, double-check the style guide or post a question at Turker Nation. Don't worry. We're all human and with the Mechanical Turk's "artificial artificial intelligence," mistakes will happen. The key is learning from your mistakes and the only way to do that is to swallow your pride and read your feedback.

Posted by qs933

Common Errors

Most of us are familiar with the its - it's and there - their errors but here are some less common ones that I've run into.

First, there's effect - affect. The hint that helped me most on this one is to remember that effect is almost always a noun and affect is almost always a verb.

Then there's punctuation of quotes. I often see transcript errors in that area. At the beginning of the quote, preceding the quote sign, there should be a comma. The first letter after the quote sign should be capitalized. The punctuation (usually a period or comma) at the end of the quote should be before the last quote sign. That has always looked funny to me but that's the way it is.

The word Internet is used in many transcripts. Should it be capitalized? I don't know. Many stylebooks and dictionaries say that the words "Internet" and "Web" (when referring to the World Wide Web) should always be capitalized because they're proper nouns. There is only one Internet and only one World Wide Web. But language referring to the Web is evolving and some people don't bother to capitalize internet or web thinking that they've become such common words that they should be in lower case. I think the important thing is to be consistent within one document. If you capitalize it at the start, make sure it's capitalized at the end.

How about words like email and website and so on? Should they be one word, two words, or hyphenated? Again, technical language is evolving. I like using one word for the sake of economy. I think the important thing here, too, is consistency.

Posted by bjs

Don't drop a stitch!

Casting Words is a bit like knitting. Plain one (easy to hear), purl one (a bit garbled) makes for the basic rib of the podcast. Then the fancy cable stitch (concentrate or you miss a crucial word and XX is a dropped meaning and too much time lost.) There’s the right-handed word caster (U.S English) and the left –handed word caster (U.K. English) and what a muddle when one edits the other! “Two nations (forever) divided by a common language,” George Bernard Shaw (nevtnorth.tec.ok.us/etiquette/eeurope.html)

At first, I found completing HITs for Casting Words to be as difficult as it was to learn to knit. But then I found Turker Nation, the most friendly and helpful forum around. (I don’t actually need to grovel, but just in case any of the regulars read this, it’s a really, really, really useful board!)

Incredibly, Casting Words HITs have managed to unite IT people, homemakers and odd-bods across the globe. Arise, “you have nothing to lose but your chains” Karl Marx (zompist.com/quotes2.html) And, somehow, the offspring of the American dream (Mechanical Turk ex Amazon) has crossed with 60s hippy ideal of love and sharing (peace man, Turker Nation rules) to produce a really efficient way for podcasts to be transcribed quickly, cheaply, and accurately. Best of all, I do the HITs because I find them fun, because I’m learning all kinds of new things and because I’m Welsh!

Pob lwc i pawb a chroeso i’r byd Casting Words!

Posted by Welsh Wannabe

Eye of the storm part II

Eye of the storm, part II, or how I learned to love the cheaters...

"In eye of the storm", I was giving you some tips on editing problematic transcripts that looked good and were in fact awful. It makes one wonder how some of those actually got passed the grading system. Alas, no system is perfect, and most graders proceed with at best a sampling of the audio track. Only when the word density is very low do we get a hint that there may be a problem. Others get passed because the grader scanned, but did not read the transcript. If you want to start grading transcripts, you should have a good eye for details, and being a quick reader sure helps. I don't mean to say you just read the whole text start to end everytime, but you should be able to sample the text in an accurate manner and spot errors and serious mistakes quickly.

I usually start by the end: I look at the numbers of [sp] and [xx] posted in the bottom notes. If the numbers are quite high, I look for a note from the transcribers explaining the reason for this. If there is no note, I will sample the audio at various point to see if the quality of the source was good or not. I have no qualms at flunking a bad transcript: If the audio is good, there is no reason to have a high number of [xx]. The editor is not there to redo a transcript - he should NOT, in fact, listen to the audio at all. The editor should edit the text transcribed, period. Compound that with serious structural problems like bad or no punctuation, serious deficiency in grammar, anything like that - and I will flunk you for certain. My position is that, with the number of excellent transcribers working for CW, those producing substandard work should not get paid: The hard, good workers should be getting that job. If you just did a hard transcript and it has many [xx], add a note. I have yet to disagree with such a note, and in those cases I will send it to CW for final verdict. Grading should be hard. This is not a place for compassion. Passing a bad transcript thinking "this poor soul just worked so hard, I'll give him or her a break" means that the editors will have to do triple the work for a quarter the pay, and that's utterly unfair. Either that or the end result will be of poor quality, CW will lose clients -- and then, all Turkers working for CW will be penalized by fewer HITs being listed. If you must feel compassion, please direct it at those dedicated transcribers who produce amazingly good jobs from definitely rough sources. Those, I applaud, because at times the results are nothing short of a miracle. I know my betters when I meet them.

MTEQC.COM

Eye of the storm

It happens to the best of us: You accept an editing job, thinking it looked fine and should be a quick one. It did look fine, at a glance. However, as soon as you start working, you realize that the transcript was done by a 300 pounds gorilla wearing boxing gloves. And then corrections we typed by a python using its tongue.

Following are three quick tips I found useful to finish the job within deadline and with my sanity mostly intact

1. MASS REPLACEMENT
Luckily, typos are usually consistent within one transcriber's work: He or she will keep on making the same mistakes, over and over again. Your friend here is "REPLACE ALL" - at least as long as the error is obvious. Someone writing "SCUZZY" for "SCSI" in a technical paper, for instance. So scan the transcript, and do a mass replace. Repeat and rinse.

2. IT'S'N'ITS
"its" instead of "it's" is probably one of the worse offenders, made worse because there is little to do but review all of them one by one. Impossible to mass replace! What I do to accelerate the process is copy a "it's" and keep that in "pasting memory" (whatever your operating system calls it). I then do some FIND, and FIND NEXT, using keyboard commands. If "its" is right, I do FIND NEXT and move on. When "ITS" should be "IT'S", I do a paste, again using key commands, and correct it instantly. Then, on to the next one.

3. THE ROCK
You've seen this one: One BIG block of text running on and on. With luck, at least some of the punctuation will be correct. To correct it, and do a quick reflow, I look for patterns in the speaker's dialect: Speakers have recognizable speech patterns, and often like to start introducing discreet ideas using their pet transitional expressions. Look for breaks in concepts or topics, and try to define the "starting block." If the speaker starts topical section by using some "And then,", or something like that, a REPLACE ALL using " And then," will produce instant results. Your big block of text will already be broken in more logical sections. You can then delete some of those "And then," to make the text easier to read. In most cases these expressions are just redundant speech patterns anyway. Just make sure to review the text - in some cases you will have to reverse the process. But from experience I'd say the result is valid 75% of the time.

Posted by MTEQC.COM