W3C home > Mailing lists > Public > public-xg-htmlspeech@w3.org > October 2011

Re: <reco> examples

From: Olli Pettay <Olli.Pettay@helsinki.fi>
Date: Thu, 27 Oct 2011 18:30:56 +0300
Message-ID: <4EA97930.5080807@helsinki.fi>
To: Michael Bodell <mbodell@microsoft.com>
CC: "public-xg-htmlspeech@w3.org" <public-xg-htmlspeech@w3.org>

Just two random comments.

On 10/27/2011 02:46 PM, Michael Bodell wrote:
> I updated a simple reco example in the latest web api document and
> tried to add more text about how binding would work to make more
> concrete the proposal.  Glen also sent out some reco examples at
> http://lists.w3.org/Archives/Public/public-xg-htmlspeech/2011Oct/0048.html
> that are well worth reviewing and while it does require flash I quite
> like Charles's examples from
> http://lists.w3.org/Archives/Public/public-xg-htmlspeech/2011Oct/0055.html
> and think it would also be a clear place where markup can solve the
> task.
>
> I know there has been some discussion the past few weeks about a
> "Speech IME" that Satish and some others suggested may negate the
> need for the<reco>  tag (because the UA will use a Speech IME to
> speech enable all input fields) but I fundamentally disagree.  The
> model of a speech IME in the UA means the web application author does
> not have the abiltiy to select either the grammars nor the
> recognition service.  In essence the web application author is at the
> mercy of the UA for the speech web service.  That is fundamentally
> unexceptable.  The web application author needs the control over
> choosing which recognition service is in place.
>
> There have been some other discussions about decalative versus
> procedural and here to I agree with Glen.  There are a wide variety
> of users.  Some only know mark up and want to copy and paste mark up
> from view-source and want it to work.  Some know mark up plus very
> basic JS to write conditions and handlers in attributes of elements.
> Some live and breath a dynamic JS web where all of their elements are
> constructed in JS.  We should endevour to make the easy truely easy
> (and for many people that means mark up with a minimum of JS) while
> leaving the hard possible (and for some things that does mean you
> need to go to scripting).
>
> Here is my take on the 8 examples from Glen's original Oct 4 mail
> http://lists.w3.org/Archives/Public/public-xg-htmlspeech/2011Oct/0000.html
>
>  Example 1: Using<reco>  with a simple field
>
> <form> <reco/> <input id="in1" type="text"/> </form>
>
> The reco inside a form is automatically is associated with the next
> input.  The default grammars work with the text and the default
> assignement is what we want.  If you wanted to be explicit with the
> for attribute and also wanted to make sure the value was always
> overwritten you could do:
>
> <form> <reco for="in1" onresult="in1.value =
> this.item(0).interpretation"/> <input id="in1" type="text"/> </form>
>
> Note that with the semantics that are in the current proposal these
> two will be slighlty different in the case that there was some text
> already in the input.  Consider that the user typed "this is speech"
> and then highlighted the "is" before speaking and that the
> interpretation of the speech is the word "cool".  The first markup
> would now have the value of the input in1 as "this cool speech" while
> the second one would be just "cool".
>
> Example 2: Using CSS to position reco <form> <reco
> style="position:relative; left:-30px;"/> <input id="in2"
> type="text"/> </form>
>
> CSS still works the same, although there is no guarentee the UA will
> work with the style the way you are expecting as the UA rendering of
> the reco element could be producing a button in the chrome that
> wouldn't be styled like this.  But in any case, normal CSS rules
> would work.

If we define that there is a reco element, we must also define how it 
affects to the layout of the page (if it affects at all. Perhaps it is 
just a hidden element and then UA can add something to input element or
browser chrome)


>
> Example 3: Overwriting and then submitting the form: <form
> action="http://google.com/search" method="get"> <reco
> onresult="q.value=this.item(0).interpretation; this.form.submit()"/>
> <input id="q" type="text"/> </form>
>
> No problem to define the simple 2 JS statements as an attribute on
> reco.
>
> Example 4: Using<reco>  to append continuous text to<input>  field
> <form> <reco continuous="true"/> <input id="in3" type="text"/>
> </form>
>
> Here the smart default binding will be appending to the input
> anyways.
>
> Example 5: Using<reco>  to append continuous text to<textarea> <reco
> continuous="true" for="ta"/> <textarea id="ta" cols="50"
> rows="2"></textarea>
>
> Again the binding by default appends.
>
> Example 6: Using<reco>  to append continuous text to<span> <span
> id="spn" style="border:solid red;"></span> <reco continuous="true"
> onresult="spn.innerHTML+=this.item(0).interpretation"/>
>
> This time no for (neither implicit nor explicit) but the tag still
> works (assuming the UA will do something smart with the default
> grammar and with the visual rendering of this reco control).
>
> Example 7: Using<reco>  to append continuous text to<div> <div
> id="dv" style="border:dashed blue;"></div> <reco continuous="true"
> onresult="dv.innerHTML+=this.item(0).interpretation"/>
>
> Same as above.
>
> Example 8: Using<reco>  to select from a menu <form> <reco/> <select
> id="sel"> <option value="asparagus">asparagus</option> <option
> value="broccoli">broccoli</option> <option
> value="cauliflower">cauliflower</option> <option
> value="spinach">spinach</option> </select> </form>
>
> Here the reco inside the form will be implicitly set to the select
> element in the form, the default assignment will be if the option
> matches to set the selectedness to true.

This sounds strange. <reco> element may behave very differently 
depending on the context it is used.
APIs should be consistent. No special cases, please.

>  The one caveat that should
> be handled better would be the building of the grammar as right now
> it will just use builtin:select where it should probably do something
> like
> builtin:select?value=asparagus&value=broccoli&value=cauliflower&value=spinach
> or some other way of communicating the select options.  If I use a
> little more non-default behavior I can use the sample grammar Glen
> did and get: <form> <reco for="sel"
> grammar="http://example.com/menugrammar.xml"/> <select id="sel">
> <option value="asparagus">asparagus</option> <option
> value="broccoli">broccoli</option> <option
> value="cauliflower">cauliflower</option> <option
> value="spinach">spinach</option> </select> </form>
>
> In general I think these type of mark up bindings are both simple and
> also quite powerful.  HTML 5 has already gone to the trouble to
> define a whole bunch of types, define what are valid text values or
> not, and say how to convert from text to the various types.  I think
> we'd be foolish to not leverage this.  If we do another pass or two
> we may get all the bindings specified perfectly, but even if we
> don't, I think an editors note in the final report would be valuable
> and the working group that fully standardizes our proposals could
> complete the details to make sure the i's are dotted and the t's are
> crossed correctly.
>
Received on Thursday, 27 October 2011 15:31:25 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Thursday, 27 October 2011 15:31:26 GMT