[liblouis-liblouisxml] Re: Undefined Cells
- From: Michael Gray <mgray@xxxxxxx>
- To: "liblouis-liblouisxml@xxxxxxxxxxxxx" <liblouis-liblouisxml@xxxxxxxxxxxxx>
- Date: Sun, 6 Nov 2016 19:41:23 +0000
No problems with https://github.com/nvaccess/liblouis/tree/backNoUndefined ;
(pull request #213). Up to Christian if to merge it to master.
MRG
mgray@xxxxxxx
________________________________
From: liblouis-liblouisxml-bounce@xxxxxxxxxxxxx
[liblouis-liblouisxml-bounce@xxxxxxxxxxxxx] on behalf of James Teh
[jamie@xxxxxxxxxxxx]
Sent: Thursday, October 27, 2016 10:05 AM
To: liblouis-liblouisxml@xxxxxxxxxxxxx
Subject: [liblouis-liblouisxml] Re: Undefined Cells
In terms of final translation, you're correct; we translate only after the word
is complete. The reason for partialTrans is to provide speech users a "guess"
of what is being typed. For example, if the user types a then b, we say a then
b. If they then press space, the word actually sent will be "about". If they
type a then dots 2 3, we'll say a then bb. If they press space, "a;" will be
sent. However, this is still far better than just echoing dot numbers or saying
nothing at all. It's better than disabling contractions altogether because
things like "er" and "ing" are always correct and the cases that aren't always
correct (such as the earlier examples) are still reasonable enough guesses for
the user to know what they typed.
Sent from a mobile device
On 27 Oct. 2016, at 11:25 pm, Michael Gray
<mgray@xxxxxxx<mailto:mgray@xxxxxxx>> wrote:
I ran the tests before pushing and didn't see any errors
I should have mentioned that the test was set as "expected to fail" and I had
removed that, I apologize for that.
No. That would result in loss of contextual information which is necessary
for reasonable guessing of back-translation.
I feel I have a basic understanding of what you mean. I was thinking that the
program would separate the entire word, or character sequence as UEB calls it,
under the cursor from the previous text and treat it differently until the
cursor had moved out of that sequence.
1. The noContractions mode doesn't seem to be implemented for back-translation.
2. Even if it were, I imagine that noContractions would also disable beginning
word and mid word contractions
The more I thought about noContractions mode and back-translating, the more I
think that it probably isn't necessary and would just cause more problems.
If you don't mind, before merging with the master branch I will like test out
https://github.com/nvaccess/liblouis/tree/backNoUndefined<
https://github.com/liblouis/liblouis/pull/213>
with our programs here since we are soon going to have to deal with this issue.
MRG
mgray@xxxxxxx<
mailto:mgray@xxxxxxx>
________________________________
From:
liblouis-liblouisxml-bounce@xxxxxxxxxxxxx<
mailto:liblouis-liblouisxml-bounce@xxxxxxxxxxxxx>
[liblouis-liblouisxml-bounce@xxxxxxxxxxxxx<
mailto:liblouis-liblouisxml-bounce@xxxxxxxxxxxxx>]
on behalf of James Teh [jamie@xxxxxxxxxxxx<
mailto:jamie@xxxxxxxxxxxx>]
Sent: Wednesday, October 26, 2016 7:00 PM
To:
liblouis-liblouisxml@xxxxxxxxxxxxx<
mailto:liblouis-liblouisxml@xxxxxxxxxxxxx>
Subject: [liblouis-liblouisxml] Re: Undefined Cells
On Wed, Oct 26, 2016 at 11:23 PM, Michael Gray
<mgray@xxxxxxx<
mailto:mgray@xxxxxxx>> wrote:
In the
https://github.com/liblouis/liblouis/pull/211, the rule ;'noback
punctuation " 5' will fail the test in tests/yaml/en-ueb-03-symbols.yaml
That's odd. I ran the tests before pushing and didn't see any errors, and
Travis doesn't show any test failures for my pull request either.
For the partialTrans mode, we have had discussions here (at APH, not the
mailing list) about that as we are going to have to deal with it in the near
future. Is the intension that the partial input is sent to LibLouis by itself
and not with the rest of the previous input.
No. That would result in loss of contextual information which is necessary for
reasonable guessing of back-translation. For example, if you type a number sign
then dot 1, that produces the number 1. If you then type another dot 1, we have
enough context to know that the second dot 1 should produce another number 1.
If you didn't send the previous input, you would get "a".
If it is so, then couldn't the no_contract mode handle these issues?
1. The noContractions mode doesn't seem to be implemented for back-translation.
If I back-translate dots 1 2 3 with en-us-g2 using the noContractions mode, I
get "like".
2. Even if it were, I imagine that noContractions would also disable beginning
word and mid word contractions such as dis, con, bb, cc, er and ing. We still
want those to work in partialTrans, since they're either always correct or are
at least a reasonable thing to report to the user.
Jamie
Other related posts:
