WebAIM - Web Accessibility In Mind

E-mail List Archives

Thread: Google Automatic Alts?

for

Number of posts in this thread: 14 (In chronological order)

From: L Snider
Date: Wed, Nov 19 2014 11:26AM
Subject: Google Automatic Alts?
No previous message | Next message →

HI Everyone,

Just saw this on another list. Very interesting concept to create
'automatic' alts... I wonder how it will work in practice?

Google building a natural description of images
http://googleresearch.blogspot.ca/2014/11/a-picture-is-worth-thousand-coherent.html

Cheers

Lisa

From: Jared Smith
Date: Wed, Nov 19 2014 11:39AM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

It looks like it may work quite well for generating a *description* of
images. But a description of what an image looks like is most often
NOT a very good alternative text for the image. Of course it is better
than no alternative at all, but I don't think we should view this as a
replacement or solution for alternative text for images, especially
from an authoring or content creation standpoint. From an end user
perspective, it could be useful to request the automated description
for certain images that do not have alternative text assigned to them.

Jared Smith
WebAIM.org

From: Bourne, Sarah (ITD)
Date: Wed, Nov 19 2014 11:47AM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

I had two first thoughts. Well, reactions - I haven't looked into it enough to have thoughts.

The first was that they are descriptions, and a description is only sometimes suitable as an alternative.

The second was that I disagreed with the categorization of their examples. Their first example of "Describes without errors" actually had an error - that's a man riding a dirt bike on a track, not a road. I don't agree with their self-assessed level of success.

But perhaps I'm being too harsh, and I will try to keep Samuel Johnson's words in mind: "A horse that can count to ten is a remarkable horse - not a remarkable mathematician."

sb
Sarah E. Bourne
Director of IT Accessibility
Massachusetts Office of Information Technology
Commonwealth of Massachusetts
1 Ashburton Pl. rm 1601 Boston MA 02108
617-626-4502
= EMAIL ADDRESS REMOVED =
http://www.mass.gov/itd


-----Original Message-----
From: = EMAIL ADDRESS REMOVED = [mailto: = EMAIL ADDRESS REMOVED = ] On Behalf Of L Snider
Sent: Wednesday, November 19, 2014 1:26 PM
To: WebAIM Discussion List
Subject: [WebAIM] Google Automatic Alts?

HI Everyone,

Just saw this on another list. Very interesting concept to create 'automatic' alts... I wonder how it will work in practice?

Google building a natural description of images http://googleresearch.blogspot.ca/2014/11/a-picture-is-worth-thousand-coherent.html

Cheers

Lisa

From: Chagnon | PubCom
Date: Wed, Nov 19 2014 11:57AM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

With Google's face recognition software being so "successful," I'm not
surprised to find them migrating it to Alt-text for photos. Fascinating,
even though it's not ready for prime time.

Agree with Sarah; for now, nothing but a human, professional communicator
can determine that "track" more accurately describes the picture than
"road."

One beautiful facet of language is its massive vocabulary that can describe
subtle nuances, adding important details to Alt-text. So far, it takes a
skilled human to see and use "track" rather than "road."

-BJC
- PubCom.com - Trainers, Consultants, Designers, and Developers.
- Print, Web, Acrobat, XML, eBooks, and U.S. Federal Section 508
Accessibility.
- 508 Workshop: www.workshop.pubcom.com
- US Federal Training: www.gpo.gov/customers/theinstitute.htm
-----Original Message-----
From: = EMAIL ADDRESS REMOVED =
[mailto: = EMAIL ADDRESS REMOVED = ] On Behalf Of Bourne, Sarah
(ITD)
Sent: Wednesday, November 19, 2014 1:48 PM
To: WebAIM Discussion List
Subject: Re: [WebAIM] Google Automatic Alts?

I had two first thoughts. Well, reactions - I haven't looked into it enough
to have thoughts.

The first was that they are descriptions, and a description is only
sometimes suitable as an alternative.

The second was that I disagreed with the categorization of their examples.
Their first example of "Describes without errors" actually had an error -
that's a man riding a dirt bike on a track, not a road. I don't agree with
their self-assessed level of success.

But perhaps I'm being too harsh, and I will try to keep Samuel Johnson's
words in mind: "A horse that can count to ten is a remarkable horse - not a
remarkable mathematician."

sb
Sarah E. Bourne
Director of IT Accessibility
Massachusetts Office of Information Technology Commonwealth of Massachusetts
1 Ashburton Pl. rm 1601 Boston MA 02108
617-626-4502
= EMAIL ADDRESS REMOVED =
http://www.mass.gov/itd


-----Original Message-----
From: = EMAIL ADDRESS REMOVED =
[mailto: = EMAIL ADDRESS REMOVED = ] On Behalf Of L Snider
Sent: Wednesday, November 19, 2014 1:26 PM
To: WebAIM Discussion List
Subject: [WebAIM] Google Automatic Alts?

HI Everyone,

Just saw this on another list. Very interesting concept to create
'automatic' alts... I wonder how it will work in practice?

Google building a natural description of images
http://googleresearch.blogspot.ca/2014/11/a-picture-is-worth-thousand-cohere
nt.html

Cheers

Lisa
messages to = EMAIL ADDRESS REMOVED =
messages to = EMAIL ADDRESS REMOVED =

From: deborah.kaplan@suberic.net
Date: Wed, Nov 19 2014 12:03PM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

And for the forseeable future, it takes a page-context-sensitive human to understand that on this page, the image should be described as "biker on a track," on this other page it should be described as "Danica Patrick at the finish line of the 14th Fun Bike Race," and on yet a third page it should be described as "'Biker as Dawn,' by Ansel Adams."

Deborah Kaplan

On Wed, 19 Nov 2014, Chagnon | PubCom wrote:
> One beautiful facet of language is its massive vocabulary that can describe
> subtle nuances, adding important details to Alt-text. So far, it takes a
> skilled human to see and use "track" rather than "road."

From: Don Mauck
Date: Wed, Nov 19 2014 12:26PM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

I just don't see how Google could ever get this right. Too many variables and too subjective in my opinion.

-----Original Message-----
From: = EMAIL ADDRESS REMOVED = [mailto: = EMAIL ADDRESS REMOVED = ]
Sent: Wednesday, November 19, 2014 12:03 PM
To: WebAIM Discussion List
Subject: Re: [WebAIM] Google Automatic Alts?

And for the forseeable future, it takes a page-context-sensitive human to understand that on this page, the image should be described as "biker on a track," on this other page it should be described as "Danica Patrick at the finish line of the 14th Fun Bike Race," and on yet a third page it should be described as "'Biker as Dawn,' by Ansel Adams."

Deborah Kaplan

On Wed, 19 Nov 2014, Chagnon | PubCom wrote:
> One beautiful facet of language is its massive vocabulary that can
> describe subtle nuances, adding important details to Alt-text. So far,
> it takes a skilled human to see and use "track" rather than "road."

From: John Foliot
Date: Wed, Nov 19 2014 12:43PM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

Jared Smith wrote:
>
> It looks like it may work quite well for generating a *description* of
> images. But a description of what an image looks like is most often NOT
> a very good alternative text for the image.

A huge +1 for that.

As I noted yesterday
(https://www.facebook.com/johnfoliot/posts/10152873832544784?pnref=story)
what they are producing is actually closer to @longdesc than it is to @alt
(but of course very few will split that particular pedantic hair). We might
also consider using that output with aria-describedby, or the
not-quite-ready-for-primetime aria-describedat
(https://dvcs.w3.org/hg/aria-unofficial/raw-file/tip/describedat.html)


I will continue for the next short while however to beat the drum about
understanding the difference in the ARIA mapping to Accessibility APIs
specification (http://www.w3.org/WAI/PF/aria-implementation/#mapping)
between an "accessible name" and an "accessible description", which as their
names suggest are different ideas altogether:

Accessible name: John Foliot
Accessible description: always wears a cowboy hat, and has a long
moustache

For images, the mappings are:

Accessible name: can be one of @alt, @label, @labeledby
Accessible description: can be one of @longdesc, @aria-describedby,
@aria-describedat

Due to legacy reasons, and authoring issues, we often find authors using
@alt as a hybrid, where the value of alt="{something}" becomes closer to the
description than the name, and "we've" sort have accepted that in the past,
as *something* is always better than nothing. Authors need to remember as
well however that the alt texts should be as short and succinct as possible,
as the values there (unlike my poor beleaguered @longdesc) are voiced
without user-interaction in screen readers, so a 60 word @alt text will be
cumbersome for many.

JF

PS: If you've never looked at the W3C Mappings document, you should, at
least once. Yes it is traditional, dry, W3C spec language, but it is worth
the 20 minutes or so it would take to read through it once.

From: John Foliot
Date: Wed, Nov 19 2014 12:46PM
Subject: Re: Addendum Google Automatic Alts?
← Previous message | Next message →

In my previous email, I neglected to add @title as a possible means of
providing the accessible name.

If you only are interested in this piece of the mapping discussion, here's a
more focused URL:
http://www.w3.org/WAI/PF/aria-implementation/#mapping_additional_nd

JF

From: Olaf Drümmer
Date: Wed, Nov 19 2014 2:11PM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

On 19 Nov 2014, at 20:26, Don Mauck < = EMAIL ADDRESS REMOVED = > wrote:

> I just don't see how Google could ever get this right. Too many variables and too subjective in my opinion.

Now that's interesting - how of all things can Google's mechanism be **subjective**? I always that that's reserved for human beings….

Olaf

From: Patrick H. Lauke
Date: Wed, Nov 19 2014 2:14PM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

On 19/11/2014 21:11, Olaf Drümmer wrote:
>
> On 19 Nov 2014, at 20:26, Don Mauck < = EMAIL ADDRESS REMOVED = > wrote:
>
>> I just don't see how Google could ever get this right. Too many variables and too subjective in my opinion.
>
> Now that's interesting - how of all things can Google's mechanism be **subjective**? I always that that's reserved for human beings….

Choosing a correct alt (which encapsulates the author's intent) is
*subjective*. So an automated mechanism could not do this job correctly.
That's what Don meant.

P
--
Patrick H. Lauke

www.splintered.co.uk | https://github.com/patrickhlauke
http://flickr.com/photos/redux/ | http://redux.deviantart.com
twitter: @patrick_h_lauke | skype: patrick_h_lauke

From: Mallory van Achterberg
Date: Thu, Nov 20 2014 2:29AM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

On Wed, Nov 19, 2014 at 12:26:20PM -0600, L Snider wrote:
> HI Everyone,
>
> Just saw this on another list. Very interesting concept to create
> 'automatic' alts... I wonder how it will work in practice?
>
> Google building a natural description of images
> http://googleresearch.blogspot.ca/2014/11/a-picture-is-worth-thousand-coherent.html


I can see something like this being *very* useful for places where
images are uploaded via automation and scans of images where the
human scanner has no good idea wtf the images are (I've done this
as a library volunteer).

Selective use of something like automated image descriptions would
help the problem that lots of times the description isn't what's
needed, but actual alt text. When images need real alts, don't use
this thing.
When dressing for a formal party, don't call the service that helps
you dress "street".

_mallory

From: deborah.kaplan@suberic.net
Date: Thu, Nov 20 2014 8:00AM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

Actually, it would be incredibly useful in publishing, especially
educational publishing. While on the one hand the correct thing
to happen in publishing is for publishers to require content
creators to provide meaningful alt text, in practice that
continues not to happen, year after year, and probably never will
unless there is money in it. So a way to automate detection of
undescribed images and describe them would be a vast improvement
over the status quo, if nothing else.

Deborah Kaplan

On Thu, 20 Nov 2014, Mallory van Achterberg wrote:
> I can see something like this being *very* useful for places where
> images are uploaded via automation and scans of images where the
> human scanner has no good idea wtf the images are (I've done this
> as a library volunteer).
>
> Selective use of something like automated image descriptions would
> help the problem that lots of times the description isn't what's
> needed, but actual alt text. When images need real alts, don't use
> this thing.

From: Mallory van Achterberg
Date: Thu, Nov 20 2014 12:44PM
Subject: Re: Google Automatic Alts?
← Previous message | Next message →

Depends probably entirely on how good it is, otherwise you get
the alt version of YouTube automatic "craptions". :P

_mallory

On Thu, Nov 20, 2014 at 10:00:32AM -0500, = EMAIL ADDRESS REMOVED = wrote:
> Actually, it would be incredibly useful in publishing, especially
> educational publishing. While on the one hand the correct thing
> to happen in publishing is for publishers to require content
> creators to provide meaningful alt text, in practice that
> continues not to happen, year after year, and probably never will
> unless there is money in it. So a way to automate detection of
> undescribed images and describe them would be a vast improvement
> over the status quo, if nothing else.
>
> Deborah Kaplan

From: L Snider
Date: Thu, Nov 20 2014 4:59PM
Subject: Re: Google Automatic Alts?
← Previous message | No next message

Hi everyone,

Loved the comments, thanks! For me, I don't think anything 'automatic' can
do what a human can do, at least one who understands why we describe
images. I also wonder if it will make developers say, hey why do we have to
do this when it is done for us-that is my big concern.

Thanks again, great discussion as always!

Cheers

Lisa

On Thu, Nov 20, 2014 at 1:44 PM, Mallory van Achterberg <
= EMAIL ADDRESS REMOVED = > wrote:

> Depends probably entirely on how good it is, otherwise you get
> the alt version of YouTube automatic "craptions". :P
>
> _mallory
>
> On Thu, Nov 20, 2014 at 10:00:32AM -0500, = EMAIL ADDRESS REMOVED =
> wrote:
> > Actually, it would be incredibly useful in publishing, especially
> > educational publishing. While on the one hand the correct thing
> > to happen in publishing is for publishers to require content
> > creators to provide meaningful alt text, in practice that
> > continues not to happen, year after year, and probably never will
> > unless there is money in it. So a way to automate detection of
> > undescribed images and describe them would be a vast improvement
> > over the status quo, if nothing else.
> >
> > Deborah Kaplan
> > > >