KnowBrainer Speech Recognition
Decrease font size
Increase font size
Topic Title: correction choices are all capitalized
Topic Summary: Correction choices in correction window are all capitalized
Created On: 10/10/2012 01:39 PM
Status: Post and Reply
Linear : Threading : Single : Branch
 correction choices are all capitalized   - Degrantphd - 10/10/2012 01:39 PM  
 correction choices are all capitalized   - Lunis Orcutt - 10/10/2012 06:45 PM  
 correction choices are all capitalized   - Degrantphd - 10/11/2012 12:48 PM  
 correction choices are all capitalized   - Alan Cantor - 10/11/2012 01:42 PM  
 correction choices are all capitalized   - Degrantphd - 10/12/2012 11:43 AM  
 correction choices are all capitalized   - R. Wilke - 10/12/2012 01:47 PM  
 correction choices are all capitalized   - Chucker - 10/12/2012 06:12 PM  
 correction choices are all capitalized   - edmart - 10/13/2012 03:17 AM  
 correction choices are all capitalized   - Chucker - 10/13/2012 06:15 AM  
 correction choices are all capitalized   - Graham - 10/13/2012 07:17 AM  
 correction choices are all capitalized   - R. Wilke - 10/13/2012 09:12 AM  
 correction choices are all capitalized   - Meinhard - 10/13/2012 10:51 AM  
 correction choices are all capitalized   - Chucker - 10/13/2012 08:05 AM  
 correction choices are all capitalized   - Graham - 10/13/2012 11:36 AM  
 correction choices are all capitalized   - edmart - 10/13/2012 12:55 PM  
Keyword
 10/10/2012 01:39 PM
User is offline View Users Profile Print this message


Degrantphd
New Member

Posts: 10
Joined: 07/28/2007

I am using Dragon v 12 with an Olympus DS-5000 voice recorder and generally use Dragon one of two machines.  The machine I am currently using is a Dell E7600 3.06Ghz, 3.25 GB ram on a Windows XP SP3 system.  After I transcribe my audio file and go through the document correcting the errors, the corrections choices will be all capitalized.  For example, if I say {caps on} "Estimates {caps off} of premorbid intellectual functioning," but the program transcribes this as "Tangents of morbid intellectual functioning" when I bring up the correction box, the choices may be 1) Estimates of Morbid Intellectual Functioning, 2) Estimates of Premorbid Intellectual Functioning, 3) ESTIMIATES OF PREMORBID INTELLECTUAL FUNCTIONING,  etc. 

I have noticed this on various computers, so I do not think that it is specific to the one I am currently using.  I have two laptops which run Windows 7, and have much better components (the system met the specifications advised by Chuck in one of his posts last fall on the support forum).  Is there a way to change the way in which words appear in the correction window?

 

Thanks

Doug



-------------------------

Douglas

 10/10/2012 06:45 PM
User is offline View Users Profile Print this message

Author Icon
Lunis Orcutt
Top-Tier Member

Posts: 26152
Joined: 10/01/2006

Ver. 12 has changed the standard correction formula. In the event that NaturallySpeaking doesn't offer additional choices, the 2nd to the last choice will appear in Title Case with the last choice appearing in UPPERCASE. What this means, in your case, is that NaturallySpeaking didn't see the correct option so you will have to manually correct it. There is no option to change the correction box behavior.



-------------------------

Click KB 2014 to take KnowBrainer 2014 on a 30 day test drive

Click SpeechStart+ to take SpeechStart+ on a 15 Day test drive



 10/11/2012 12:48 PM
User is offline View Users Profile Print this message


Degrantphd
New Member

Posts: 10
Joined: 07/28/2007

Thanks Lunis. I am seriously considering going back to v11.5. I have not noticed any improvement in accuracy or speed with 12 and I find that some of the changes are more irritating than helpful, with the correction box being one. Do you know how many documents is needed to train dragon to a particular writing style. I write long reports with very exact phrasing from one report to another. In my initial post, I have uttered Estimates of premorid intellectual functioning over 300 times in the course of the past year, and I still have to correct that statement and Dragon cannot find the phrase in my vocabulary. When I created a new profile in Dragon 12, I had it scan 350 of my reports, but I still have the problem of it understanding phrases that I routinely use. Any suggestions on how to improve this? BTW, I have tried commands, but most of the time when using a digital recorder and transcribing the report, the commands do not launch.

Thanks.
Doug

-------------------------

Douglas

 10/11/2012 01:42 PM
User is offline View Users Profile Print this message


Alan Cantor
Top-Tier Member

Posts: 2009
Joined: 12/08/2007

Have you added phrases like "Estimates of premorbid intellectual functioning" to your vocabulary?

 

If so, have you tried specifying a spoken form to disambiguate the pronounciation by breaking up premorbid? "Estimates of pre morbid intellectual functioning." 

 

Another approach -- less desireable -- is to use a completely different phrase for the spoken form, e.g., "Insert estimate of intellect function."

 10/12/2012 11:43 AM
User is offline View Users Profile Print this message


Degrantphd
New Member

Posts: 10
Joined: 07/28/2007

I have not tried to use a different spoken form for this particular phrase, but have tried it with other phrases. I have mixed results with that approach. I do have completely different phrases to get Dragon to transcribe what i want, but like you said, that is not a very desirable approach. What I would prefer, and this is a general criticism, not specific to this example, but for Dragon to learn my corrections, learn my writing style from importing reports, and somehow look at the frequency in which my reports group words together over the frequency of words that occur together in the general language. I get recognition errors that I cannot believe are based on frequency distributions in either my own reports, or the general language. In some cases, I have corrected a name I commonly use over 50 times, just to have Dragon inconsistently correctly recognize the word.

-------------------------

Douglas

 10/12/2012 01:47 PM
User is offline View Users Profile Print this message

Author Icon
R. Wilke
Top-Tier Member

Posts: 5295
Joined: 03/04/2007

In some cases, I have corrected a name I commonly use over 50 times, just to have Dragon inconsistently correctly recognize the word.


The first thing to understand here, if correcting once or twice doesn't provide the desired result, correcting 50 times won't do that also, so all it will do is waste your time and add to the frustration. Whenever that happens, you should read it as a sign that something else is needed to achieve what you want. The same is true for feeding documents, if Dragon doesn't learn the things from it that you expect it to right from the start, feeding even more documents won't work also. There may be a solution, more or less useful, for any given issue but you need to look at them individually.

For instance, teaching Dragon to prefer a capitalized form over a non-capitalized form, when it is not a common way of using it anyway, as in "Estimates", by correcting it over and over again, is an exercise in futility, because that won't teach Dragon anything, just the same as it won't teach Dragon anything if, for instance, you go on correcting "Mark" to "Marc" if that's what you prefer. So you need to do something about it, such as:

delete estimates, and add Estimates
say "cap estimates" every time
delete Mark, and make sure that Marc is present
add Marc as written form, along with Marc with a C as spoken form, and say it like this

Etc.

Rüdiger



-------------------------



 10/12/2012 06:12 PM
User is offline View Users Profile Print this message

Author Icon
Chucker
Top-Tier Member

Posts: 11304
Joined: 10/10/2006

Douglas,

Rüdiger has covered all the bases extremely well. You would be well advised to follow his explanations and understanding in this case. However, I will add one additional minor point. That is, when you are dictating into a Digital Voice Recorder, avoid using the "caps on/caps off" commands. If you just want to capitalize one word in a phrase concentrate on your dictation style and just use the suggestion that Rüdiger gives for saying "cap estimates". When using a DVR, the more types of commands (i.e., Dictation commands or otherwise) the more it can tend to throw off your dictation style. You'll also get better results when correcting transcribed results.

As regards to Mark vs. Marc, or similar proper names that have many variations, I don't worry about how Dragon transcribe such. Nor do I worry about correcting such words because as Rüdiger points out, it's basically a waste of time. I just dictate, transcribed, and correct these types of instances only worrying about the current document rather than what Dragon will do next time around. Otherwise, Rüdiger suggestion if you need that to come up a certain way every time is what generally works best.

Chuck

"Many of the things you can count, don't count. Many of the things you can't count, really count." Albert Einstein

-------------------------

 10/13/2012 03:17 AM
User is offline View Users Profile Print this message

Author Icon
edmart
Top-Tier Member

Posts: 492
Joined: 10/07/2006

I cannot add anything useful to what Rudiger and Chuck have said, but would comment:-

 

* I do find Dragon rather perverse sometimes. Say I dictate a document over three pages and the word 'Stuart' appears twenty times. It might be transcribed fourteen times as 'Stuart' and six times as 'Stewart'. Ok, so that's easy enough to cope with, but given how brilliant Dragon is in other respects, I sometimes wonder whether its intelligence might say: "I've heard this word very recently. It's a homonym. So chances are that it makes sense to spell it the same way I did last time". Clearly that is NOT how it works, but it seems a shame.

 

* I do wish that Dragon understood that if you say '19 may' or indeed '19th may' you mean the word 'May'. It seems determined to be obtuse in this respect. I find that I get 'Mark' rather than 'mark' (when I want it, and the context suggests it) more frequently than I get 'May' rather than 'may'. Question: does each of those words have a probability rating attached to them somewhere in the bowels of Dragon as to how often they should  be capitalised? Taking my example, my experience might suggest that 'Mark' had a rating of 15% whereas 'May' had a rating of say 3%. I have no reason to suppose that they do, but if they did, is it possible to find out what the rating is for a particular word, and adjust the way one works round the individual problem on an individual basis, taking that rating into account?

 

Ed



-------------------------

DNS 12.5 Pro, Windows 7 Professional, SP1, i7-2700K 3.50 Ghz Sandybridge, 16Gb DDR3 PC3-160000C9 2000Mhz dual channel RAM, SpeechMike 5276 and Samson Airline 77

 10/13/2012 06:15 AM
User is offline View Users Profile Print this message

Author Icon
Chucker
Top-Tier Member

Posts: 11304
Joined: 10/10/2006

Ed,

And when I'm dictating and I get hungry and put up two fingers, I want Dragon to understand that it should go out and get me two Big Macs and a large order of fries.

Chuck

Be careful what you ask for. What you get might not be what you expect, or want. - Aesop (620 BC - 700 BC)

-------------------------

 10/13/2012 07:17 AM
User is offline View Users Profile Print this message

Author Icon
Graham
Top-Tier Member

Posts: 809
Joined: 10/01/2006

Chuck,

If you are so rude as to put up two fingers I'm not surprised that Dragon will ignore you!

Ed,

If you want 19 May why not change the properties of "may" to be capitalised ("May") after numbers? In that case you may never see the incorrect date (just testing). Using DNS 11.5 if I use "th" I always seem to get 19th of May.

Graham



-------------------------

 


www.itspeaking.co.uk

 10/13/2012 09:12 AM
User is offline View Users Profile Print this message

Author Icon
R. Wilke
Top-Tier Member

Posts: 5295
Joined: 03/04/2007

If you want 19 May why not change the properties of "may" to be capitalised ("May") after numbers? In that case you may never see the incorrect date (just testing).

 

If dictating "19 may" doesn't come out as "19 May" automatically, quite obviously it is just because Dragon won't recognize it as dictating a date. So therefore it is anything but wrong doing it like this, but rather logical. As regards the date formats, just like any other format handling, the rules are pretty much fixed, and we all should not only accept that for a fact, but rather appreciate it.

 

The same is true for homonyms, or just any other ambiguities for that matter, having the same probabilities by default that we cannot access or change, or rather a fixed set of probabilities shared between them, because they may be the same, or they may be shared by and kind of proportion between them. Therefore, if you get "Stuart" a certain amount of times, but then you may also get "Stewart" any other time, which may seem randomly to you, this is just logical because, by default, both are probable to their own degree as long as you don't provide any other kind of information for Dragon to judge by. The same would be true for a human transcriptionist, as long as he or she has no way of either knowing what you want, or reading your mind.

 

Just imagine what would be the consequence if, after using a homonym a certain amount of times, any other alternative would be sort of deleted internally by reducing its probability to zero, and then if you want it to come out, it just won't happen?

 

So you have to make up your mind about such cases, and how to deal with them more effectively. As I stated in a previous post, trying to overcome them by correcting is not the way to go, but a waste of time instead, and that was the main point. The thing is not a lack of intelligence on the side of Dragon, however, and I am saying this because it has been brought up by Ed.

 

Rüdiger

 

 

 

 



-------------------------



 10/13/2012 10:51 AM
User is offline View Users Profile Print this message

Author Icon
Meinhard
Junior Member

Posts: 39
Joined: 10/12/2012

Just imagine what would be the consequence if, after using a homonym a certain amount of times, any other alternative would be sort of deleted internally by reducing its probability to zero, and then if you want it to come out, it just won't happen?

May I try and sort out the issue for myself (and only maybe for others): the concept underlying Ed's wish seems to be that Dragon might assign a higher priority (relative probability) to one spelling of a homonym versus another on a per document basis, based on recent corrections.

Although it might well be possible to code this, there would be severe downsides to such an approach. Dragon can never know for sure for which homonyms such a "temporary predominance" of one spelling over another should be enforced, because, obviously, there are a lot of homonyms for which such a behavior is not desirable. "May" and "may" seems like an excellent example to me – and also good evidence for the superiority of the approach Dragon actually takes, namely deciding by context, not by document.

This approach works very well in most cases, but, in my opinion, it can't regarding proper names – simply because there are no specific contexts assigned to the spelling of such names. Mr. Stewart and Mr. Stuart tend to have just the same preferences regarding their favorite words in their vicinity.

Fortunately for forums, the complexity of the probabilistic rules governing Dragon's behavior, allow enough leeway for speculation and a never-ending discussion.

I bet there are a lot of people out there who are under the impression that Dragon typically prefers the spelling most recently corrected. Sometimes, I am under this impression, too, but then again, I'm not.



-------------------------

Dragon NaturallySpeaking 13 Legal (GER/ENG), SpeechMike Premium 3500, Win 8.1 64 Bit

 10/13/2012 08:05 AM
User is offline View Users Profile Print this message

Author Icon
Chucker
Top-Tier Member

Posts: 11304
Joined: 10/10/2006

Graham,

Perhaps I should of put a after what I said so that everyone would understand that I was joking while trying to make a point. The point being that speech recognition doesn't understand what words mean (probability or otherwise) and it can't use common sense the way people do, which is often the misconception that users have. In addition, Dragon doesn't know what you meant to say it only knows what you said. That comes from the training script, which hasn't changed significantly over the last 12 years. Nevertheless, I got your attention, which was what my original intent was. You see how easy it was for you to misinterpret what I meant. How do you expect Dragon to do that, Dragon being even less capable than you and I.

How many times do we have to make this point clear before users understand what Dragon can and can't do, or more simply what speech recognition can and can't do?

I apologize if anybody took offense. I just thought that was the easiest way to point out the difficulty with the assumptions that Ed was making.

Chuck

"Assumption is the mother of all screw ups" - Unknown



-------------------------



 10/13/2012 11:36 AM
User is offline View Users Profile Print this message

Author Icon
Graham
Top-Tier Member

Posts: 809
Joined: 10/01/2006

Chuck - Sorry I was joking too! I'm not that easily offended!

Graham




-------------------------

 


www.itspeaking.co.uk

 10/13/2012 12:55 PM
User is offline View Users Profile Print this message

Author Icon
edmart
Top-Tier Member

Posts: 492
Joined: 10/07/2006

.....and, Chuck, happily neither am I.

 

But I do wonder that you think I have not assimilated your previous postings on the subject of Dragon not having a human brain. It worries me that you think my level of comprehension could be that low....

 

Those of us who have merely worked with Dragon for a while (fifteen years, I think, for me - anyway, since version 4) keep trying to understand more. In my case, I understand that there are some words and phrases that are hardcoded by DNS to have certain properties and obey certain rules. Sometimes, as with 'apostrophe ess' the particular properties appear pretty clear. At other times (I cannot currently think of an example) they have properties and obey rules which are pretty arcane - to me anyway. Yet those involved with the software (particularly you with your experience) often know those rules, and can help us by elucidating them.

 

In this instance I was wondering whether there were any rules or properties attaching to 'Mark' and 'mark' over and above the contextual. It would seem that there are not: is that right? Thing is, if I don't ask, I never know what I might be missing!

 

And as for Stuart and Stewart, it just struck me that conceivably there might be a heavier probability rating attached to such homonyms being spelled the same way if they were, say, dictated within ten minutes, one of the other. Again, it would seem that there is not. Again, is that right?

 

If I am right about both those points, then let us rejoice, (or let me, anyway....), because it means I know now for certain what previously I could only guess at. And that is always good news (even if I do not have a natty epigram as my tagline to make the point in a different way!)

 

GRAHAM: Yes, your idea would work fine for 19 May, but I love ordinals. I rather imagined that 'after numerals' in the properties box would not include 'after ordinal numerals': but maybe it does. Do you happen to know, or do I need to experiment?

 

Thanks one and all.

 

Ed



-------------------------

DNS 12.5 Pro, Windows 7 Professional, SP1, i7-2700K 3.50 Ghz Sandybridge, 16Gb DDR3 PC3-160000C9 2000Mhz dual channel RAM, SpeechMike 5276 and Samson Airline 77

Statistics
28628 users are registered to the KnowBrainer Speech Recognition forum.
There are currently 0 users logged in.
The most users ever online was 2028 on 04/05/2013 at 07:36 PM.
There are currently 85 guests browsing this forum, which makes a total of 85 users using this forum.

FuseTalk Standard Edition v4.0 - © 1999-2014 FuseTalk™ Inc. All rights reserved.