Re: PC Speech Recognition App's on a Mac, Pros and Cons

Mark Conrad wrote:
Start of raw uncorrected text -
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Semantic technologies are based on models that explicitly encode the meaning of information to avoid ambiguity and support automated reasoning. ..... ...... These include: controlled vocabularies, the
sorry, taxonomies, ontologies, object oriented models and database schema. .......

The fourth mistake, "the sorry" instead of the correct "thesauri" is a serious limitation of all present speech recognition applications, even the very expensive Dragon NaturallySpeaking Pro versions 9.00 and 9.5

Namely, they can not handle homonyms properly.

The pronunciation of the word "thesauri" is "thu-sorr-eee" ...

... which Dragon mistakes as "the sorry".

There is no practical way to avoid such mistakes. In the distant future, speech recognition app's will be able to avoid such mistakes, however right now we have to live with these limitations.

There are at least three ways to avoid such mistakes. (Let's not have
an argument over the meaning of "practical.")

1. Semantic technologies. If the Dragon's Dictionary included parts
of speech, and if those had been consulted, the program would "know"
that 'sorry' is not a noun and therefore not likely to be preceded
by 'the.' And analyzing the text for hints to meaning would make it
possible to know that "sorry" is isn't at home as the middle member
of a list of nouns of that sort.

2. "Training." Dragon, like ViaVoice, "learns" during correction.
Statistical methods store information from which can be computed
the probability for a certain word to be in the vicinity of certain
other words. This is similar to the Bayesian classification at the
core of any spam filter that actually works. ViaVoice (and I suspect
Dragon also) allows one to pre-train by scanning files of text
already composed. It also allows multiple user profiles, which
could be used to separately store the probability sets for different
authors.

3. Put "thesauri" in the Dragon Dictionary. :-)

Bayesian classification, like that done for spam, has been used to
assist in identifying the authors of unattributed text. Scan all works
known to be by Joe Blow, and you can generate a probability that the
next file is by him.

Google is currently working on using a variation of this to improve the quality of machine translation:

http://www.theregister.co.uk/2007/03/29/google_translator/

The basic idea has been around for a while:
http://tinyurl.com/25qkon (Google Scholar)
http://tinyurl.com/yp8rlz (Google Popular)

As for whether "thesauri" or "thesauruses" is better, you'll just have
to _guess_ on my opinion. Hint: I fully agree with Silver Han's usage
rules, as stated in the middle of http://tinyurl.com/37afqf

--
Wes Groleau

In any formula, constants (especially those obtained
from handbooks) are to be treated as variables.
.

Relevant Pages

  • Re: BLINK - a brief book review
    ... ## certain way and dropping your opponent with trained reactions rather than ... ## against mistakes. ... One achievable option is to avoid "Condition White" - ... She ended up with a broken arm, a fast trip to the ...
    (rec.guns)
  • Re: adapting getline
    ... the Grave Hand of kid joe Did Inscribe: ... I would avoid this sort of thing - don't put arbitrary limits in your ... char *buf = NULL; ... Mistakes are a part of being human. ...
    (comp.lang.c)
  • Re: Mass. Lawmakers OK Mandatory Health Bill
    ... I been meaning to drive up for that free ... dinner you promised. ... And we all made some mistakes in our youth. ... no matter how much the land appreciates. ...
    (soc.culture.haiti)
  • Re: Skolem Again
    ... the theory T1 given by just one axiom A1: ... For example, consider M1 as follows: the domain of T1 is the real world; the interpretation of Pis "x is a good philosopher", as understood by common English speakers. ... Now I say that "Some people are not good philosophers" is part of the meaning of T1. ... in fact, is it true in M1, so ....'it is terribly difficult to avoid' that conclusion. ...
    (sci.logic)
  • Re: Transverse Velocity of Light
    ... But it's not clear that your statements carried the meaning accurately, ... And the precision I added is necessary to make it a ... I was puzzled why you used the phrase "transverse velocity" in the ... And will keep making mistakes like this. ...
    (sci.physics.relativity)