Shorthand Dictation and other links

I’ve added a few new links on my ShorthandShorthandShorthand site (http://www.geocities.com/shorthandshorthandshorthand/Links.htm).
One of particular interest is http://www.freewebs.com/weloveshorthand/Home.htm which has lots of FREE dictation from 50-120 wpm. 
I finally fixed the error or not listing Cricket’s dictation machine.  (Sorry, Cricket!)
AND, lastly, I found three articles which document attempts to have a computer (PDA) trasncribe handwritten Pitman 2000 back into text.
Marc

(by shorthandmarc
for everyone)

7 comments Add yours
  1. You see I don't think it would be too difficult to get a system that recognizes Gregg.

    For example, The gradient of the line for t is clear distinguishable than that of the line for m or j. The change in gradient in G is higher than in M. The length of t is half that of D. Once the computer identifies all the parts to the outline, it can search the reverse-dict that some guy here posted to find that word. If it doesn't find the word it can try to construct a word out of the parts

    IN fact I think a computer would find the simple strokes used in Gregg very easy to determine simply because they have much less elements to them than a letter.

    The question remains if any of us will find any realy use out of a program like this, and I have doubts.

  2. I think, Michael, if a court reporter was able to take their pen written notes home, drop them into the scanner, and get a draft text of the proceedings without having to type it out, they might think that was a good thing.   I think there's a post or two about that somewhere.   sidhe

  3. The idea is great, I just wonder how to account for the variations in style.  How does one build a translation dictionary?  Proportions vary so much depending on how pressed you are writing at high speed.    I would like to see something like this, I just don't know that the real-time application would work.  Back the CAT was new,  reporters had to build their dictionaries by stroking out a word list.  Scanning a word list written by the stenographer in some way like the Scantron testing forms maybe?   The actual execution seems to be problematic. 

  4. I think with Gregg you'd need more than "letter" level recognition. Shapes are often influenced by the shapes on either side. Loops and hooks especially, T-U vs N-U. Also, some of the shapes are shortened: In S-T-O-R, the O is indicated by retracing the T for a bit, rather than a separate horse-shoe. Straight lines tend to bend near circles. A-M-A can look like A-C-A. (Yes, they shouldn't, but they often do.)

    The Pitman books I've seen could have been printed by a computer. Strokes aren't affected by the strokes on either side. (If two strokes don't fit well, one of the sounds is assigned a second stroke.)

    Also, there are tricks which, while not in the "core", make it easier, such as in Speed Studies Unit 4, where PR and PL begin with a left-ward movement, and BR and BL begin with a downward movement. Subtle, not necessary, but an extra hint to the human reader and an extra challenge for the machine reader.

    Still, we've come a long way with OCR for the normal English.

    Then we have all the things left out of words and phrases, and things that are "spelled" the same but mean different things — quite the linguistic challenge.

    Having said all that, I wonder how hard it would be to create a "computer-readable" version. It would have to include more information than the older versions, and one would have to be even more careful when writing. (Well, until the computer gets really good at it.)

    I also wonder if Gregg could be "machine-generated"? I saw one attempt, but it only combined a few shapes. I'd like to see it do something from the final chapter.

    Cricket

  5. I think Gregg would be more difficult for a computer to read than Pitman, at least looking at the textbooks.

    In Gregg, the same letter often varies, depending on what's around it, even in the texts. In Pitman texts, if you cut a letter out of ten different words, it's always the exact same shape (and if you can't join it nicely, they add a second shape for the same sound).

    Can Gregg can be written that precisely? When I "create" a new word, it takes a few tries to get the angles and curves loop sizes working together. (Fewer tries as I gain experience, but the point is, you need to know what's around it to know exactly how to write it.) If not, the recognition would have to be "smarter".

    Going in the other direction, someone here started a program to "write" Gregg by joining shapes. Has he gone beyond the first few shapes?

    Cricket

Leave a Reply