Is Google Voice Racist? “Hey Negro” Transcription

Posted on January 2, 2010

I CANNOT BELIEVE I just received this transcribed email from Google Voice! At the beginning of my friend’s message she says, “Hey Girlfriend” but Google Voice transcribed it as “Hey Negro”.

Google Voice Screenshot

Google Voice Recording

Here’s the actual recording so you can hear:

This is from my friend Leslie who is confirming that on Sunday, we’re going to get together and hang out. Her son Alec loves my little dog Bluey.

Leslie is White. I’d be concerned if she used the term “Negro” to greet me. Especially since we live in Minneapolis, MN and overt racism is rare here.

Google Voice Transcription

and the original text transcribed by Google Voice:

Hey Negro, Hi, it’s, Leslie in out. Okay so I did get your voicemail or your email if, and I did leave a message for you, but I am going to go with Sunday if, since I hadn’t heard from you. I’m assuming you want to stick to Sunday and I think you’re email said 12 to 2, so maybe we’ll talk later. It is Saturday and Alec and I are getting ready to go out so I just want to make sure that I guess we are in fact first day and somebody’s really looking forward to it. Well actually looking forward to it, so I’ll be fun to see you again and you’re happy face. So Tucker says hi. Happy New Year. Hope you had a nice day. Nearest as did we. You know that I don’t have enough money exciting life. So just anyway. I’ll try and catch you later. On the flip side to confirm for tomorrow, but otherwise. Yeah, we will swing by your place at 12. I don’t if I don’t see. Yeah, or suck to you bye.

As you can see, her greeting sounded nothing like, “Hey Negro”. Bill C notes in the comments that Google could have suggested, “Hey Grill fiend” or “Hey Growl fried” which would be much closer to the actual word. I recently purchased MacSpeech Dictate so I could speak my blog posts versus typing them out and it’s been going really well! It is very accurate but more importantly, they explain how computers translate voice into words based on the expectation that certain words are used together with certain other words like, “Hello, how are you?” is much more likely than, “Hello, I am blue”.

So we must question what’s going on here:

What’s up with Google Voice?
Who is creating the reference material use for transcription?
What other slang and potentially racist terms are being transcribed?
Does Google Voice learn from actual recordings?
If so, are people leaving slang greetings on Google Voice?

In case you missed it, HP got a lot of feedback about this video where Desi and Wanda, coworkers at a computer shop, test out an HP computer’s webcam to show that it doesn’t respond to Black faces:

- 35 Comments - Posted in Adventures in Consulting, Equality, Google, Google Voice Tagged with: Technology racism

< Google Voice Holiday Giveaway Winnners! Dell Discontinues XPS 8000 After Christmas Delivery Delays [VIDEO] >

http://www.mitchellhislop.com mitchellhislop
Adria, as you know since you have been in the technology game so long, voice to text is VERY hard to do. And since you have been using Google Voice, you know that the words that are in the light grey/faded text mean that Google is unsure about them. Your friend said “girlfriend” in a non standard way, making a mis-translation plausable. I promise, it is not a racist plot to destroy you, or anything like that. Its a computer. It took a guess at what your friend said.
Jumping on one of the biggest and best companies as racist just because of what their computer server mistranslated is a bit of a cheap shot…
http://twitter.com/s0l_uk Bill C
In what world does “Girlfriend” SOUND anything like “Negro”??
If it translated it like “Grill fiend” or Growl fried” then OK it's got the words confused due to the sound quality!
But it's supposed to translate the audio, not make things up!
http://www.mitchellhislop.com mitchellhislop
I have seen some fairly messed up google voice transcriptions. With how the “girl” syllable was pronouced, followed by a non-standard pronucination of “friend”, I can see why it messed up.
And even if it was sup-par, Google does not do these by hand. There is no way that you would instantly get an email if there was a human on the other side. There also is no way Googe could employ enough people to do transcriptions. They only have 10,000 employees. No where near enough to do the transctiptions
http://www.mitchellhislop.com mitchellhislop
You may be on to an experiment though. Since GVoice uses the “was this transcription helpful?” voting system, I wonder if you could teach it things…repeated VM's with the same words, voted in a similar way.
http://butyoureagirl.com adriarichards
Mitch,
I know you're White. And you're missing the point.
Can you think about it for a minute why the word “Negro” should even be a
transcript option and get back to me?
This requires empathy, not logic.
http://butyoureagirl.com adriarichards
My point exactly!
John
Mitchell,
I read Adria's question and comments as an honest quiz about something puzzling… although I definitely agree voice-to-text is difficult and the mistranslation is plausibly innocent, I didn't see the post or question as accusatory at all, and so don't see it as a “cheap shot”. There certainly are others who have and will make such accusations, but I just don't see it here.
Agreed, though, the tech is still not good enough to avoid a lot of mistakes; some that can easily appear to be provocative.
http://www.mitchellhislop.com mitchellhislop
And I get that. I intentionally tried to stay out of the race part of it. It sucks that it happened like it did. It really does.
I went the logic route to try and get to an answer.
http://www.mitchellhislop.com mitchellhislop
Should we try an experiment? Not with anything racist, but try to teach Google Voice something? Or apply a new translation to a mispronounced word enough times that it thinks its right?
http://butyoureagirl.com adriarichards
Mitch,
This is about race. You need to address it to be part of the conversation
because otherwise, your comments look empty and superficial.
http://butyoureagirl.com adriarichards
I think there will always be opportunities to “test the system” and we
should do so, sharing our results.
http://allfacetsof80211-technology.blogspot.com/ Veronica
As an African American I am disgusted. Hell to the mo fo they (HP) didn't do this to us again with Google Voice. Yesterday I invited my entire gmail list to try Google Voice. About 45% of them are African American. That makes me part of the conspiracy. Yes, I was appalled during and after watching the HP Web cam racism. How in the heck can a web cam be racist? It didn't program it self. To Mitchellhislop. I'm not trying to hear that playa. I have been in technology for many years starting with the military in the 80's. I know this could have been avoided. These incidents of clan like behavior was AND is intentional. Don't try to defend HP. This is inexcusable. I bet not one reference is mentioned about Honkey, Red Neck, or Cracker. I can go on. You appear to be an intelligent person. Although I don't know you. However, there is no way you can justify your statements. Why aren't you wondering how this could happen. Help to fix this disgusting mistake. Or just be truthful and admit someone messed up horribly. Be a part of helping to stop this mistake. I've supported HP for many year and Have several product one of which I am typing on right now. Holla back with some love and apology from HP for me and my African American family. I've been interested in a MAC for sometime. It maybe time for me to switch my loyalty to a company that appreciations my kind. Mankind.As an African American I am disgusted. Hell to the mo fo they (HP) didn't do this to us again with Google Voice. Yesterday I invited my entire gmail list to try Google Voice. About 45% of them are African American. That makes me part of the conspiracy. Yes, I was appalled during and after watching the HP Web cam racism. How in the heck can a web cam be racist? It didn't program it self. To Mitchellhislop. I'm not trying to hear that player. I have been in technology for many years starting with the military in the 80's. these incidents of clan like behavior was and is intentional. Don't try to defend HP. This is inexcusable. I bet not one reference is mentioned about Honkey, Red Neck, or Cracker. I can go on. You appear to be an intelligent person. Although I don't know you. However, there is no way you can justify your statements. Why aren't you wondering how this could happen. Help to fix this disgusting mistake. Or just be truthful and admit someone messed up horribly. Be a part of helping to stop this mistake. I've supported HP for many year and Have several product one of which I am typing on. Holla back with some love for me and my African American family. I've been interested in a MAC for sometime. It maybe time for me to switch my loyalty to a company that appreciations my kind. Mankind.
http://twitter.com/mypolaropposite T. Lynn Lloyd
This is quite a provocative issue, no doubt the reason Adria has brought it to our attention.
The only way voice transcription software can get from a word starting with a “g” to a word starting with a “n” is by using some assumptions that have everything to do with racism (i.e. translating speech patterns into “ethnic” words when none exist).
This is not happening because of some Google conspiracy that knows the race of all Google Voice users – making an accusation like that would be a “cheap shot”, as mitchellhislop defines it. Rather this is happening because the world of Caucasians continually makes assumptions about how people of color speak. Though I'm African-American, I'll use “we” for the sake of clarity here: we assume that Asians can't pronounce certain letters; we assume that Hispanics speak in a sing-song cadence; and we assume that Black women use the world “girl-frieeend” as did the woman in Adria's message.
The narrow definition of speech suggested by this translation is akin to calling President Obama “articulate”, or the comments I regularly get from people who I meet over the phone who are then surprised to meet a Black woman in person. “You don't look like what you sound like,” or “I thought you'd be Blonde” or “You sound so nice on the phone” are phrases people have actually said to me.
Racism isn't limited to using the “N” word (the big one) or outward racial discrimination: it's the systematic reduction of groups of people to “less than” status due to their racial origin. In this case, the speech cadence attributed to “Negroes” actually spurred the use of the word in a translation. I'm sorry but there's just no other way to spin it.
http://interactivestemlearningprogramsk-12.blogspot.com/ TechniCowl
We understand exactly what you are saying.
I have no doubt that this could become very embarrassing.
However Google Voice users are not supposed to work through this problem.
It should be taken care of on the Corporate end. No questions asked. They Just need to make the adjustments.
At this point no assumption about race is necessary. This problem clearly affects a disproportionate group of people namely African Americans and descendants who just want to be treated no better or worst then other peoples.
demaster
I'm a person who has an ancestry with a strong exposure to sunshine. I'm not sure why Google voice would translate “GirlFRIEND!!!!!!” to “Negro, Hi”, but I have a hard time believing it is a conspiracy. I'm not saying it is cool. I find it a bit unnerving, but not to make me feel as though my rights (Do I have rights?) have been violated. I do believe that an apology is TRULY in order and some SERIOUS software tweaking.
AS a techie, I figure that i threw in the “Ne” part of Negro and mistakenly assumed “girl” was “gro”. Then it took the high pitched “friend” to be “Hi”.
As a fan of Adria… If I were checking my messages and I read this…IT WOULD BE ON!!!!!!!!!!!! So I empathize (sympathize?). I can truly understand this flare of emotions.
As an observer of business…I think a public apology from Google should be in order.
As a person who talks too much…time for me to shut up.
http://interactivestemlearningprogramsk-12.blogspot.com/ TechniCowl
On time . Your words are significant. Thanks.
Time for me to wait and see if a resolution is forth coming.
I've passed the word. Each one reach one. This is too important.
http://interactivestemlearningprogramsk-12.blogspot.com/ TechniCowl
Adria I applaud you for you conviction to bring this forward. There is only one resolution. Thanks
http://butyoureagirl.com adriarichards
*smile*
Thanks folks for speaking up on this. I think each time we question the
status quo, we make progress towards equality.
http://www.jmedwards.net Jamie Edwards
You are asking if a piece of software is being 'racist'? It is a piece of software. Software cannot be racist – just the architects.
Jumping back into rationality, Google Voice is powered by algorithms that learn and refine based on reference data. Perhaps GV's algorithms have noticed that 'negro' commonly proceeds “Hey” in Google Voice conversations.
“Bill C notes in the comments that Google could have suggested, “Hey Grill fiend” or “Hey Growl fried” which would be much closer to the actual word.”
Google Voice, like all advanced transcription algorithms, do not just listen to individual words. They look at the words surrounding the word being listened to, the tone and the usage patterns of that word by their thousands of users.
The word “negro” is in grey there. Is this GV's way of telling you “I am not sure about this word”? That says a lot.
Google Voice making this mistake is not a big deal.
What might be a big deal is:
“# Does Google Voice learn from actual recordings?
# If so, are people leaving slang greetings on Google Voice?”
The answer to the first question is presumably yes. The second question there is the big deal here.
http://www.jmedwards.net Jamie Edwards
When I say:
“Software cannot be racist – just the architects.”
I am not insinuating that Google Voice engineers (or the HP engineers, for that matter) are actually racist. Perhaps it is a design flaw in the software which has produced this offensive result.
It is hard not to get emotional about this kind of thing, but you really do need to step back and consider exactly what is going on – unintelligent algorithms and code trying to process something horrendously complicated and variable.
With regards to the HP and black people's faces situation – consider that the engineers simply got the pixel tone and brightness thresholds a bit wrong.
Simple mistakes, evocative consequences, but I don't think you should go OTT with accusations of racism..
http://butyoureagirl.com adriarichards
Jamie,
I don't expect you to understand what institutionalized racism looks like or
how algorithms setup by people who are not sensitive to diversity can go
wrong.
Here is an example when Walmart's “algorithm” went wrong and suggested
movies like, “Planet of the Apes” when people searched fro movies on Martin
Luther King:
http://www.usatoday.com/tech/news/2006-01-05-wa…
Wal-Mart Stores is shutting down the automated system that creates movie
http://www.jmedwards.net Jamie Edwards
Adria,
My point is that it could well be a simple case of “Ah, we didn't think of this”. The correlation here (the end result) circumstantially looks like racism. However, correlation does not always equal causation.
That is all I am trying to say.
You cannot expect people to create code to “Catch anything that may be potentially offensive to anyone” – this includes racially offensive, politically offensive, and so on. Why can't we expect that? Because offence is entirely and utterly subjective and personal. What one person takes offence to, another person may not.
To fit this into your Walmart example – yes, this is incredibly unfortunate and racist. Maybe it was done on purpose by a rogue Walmart employee.
Or, maybe (and far more likely), Walmart's system works like Amazon's. The system looks at all the other films and books people who watch or read anything to do with 'Martin Luther King', and found that 'Planet of the Apes' was popular with those customers.
How would you code to prevent 'offence' from *ever* being caused by automatic suggestion engines?
On another hand, and going back to Google Voice, it is a shame people have been using words like 'Negro' casually in conversation, which I presume has caused the algorithm training error. I am not part of the politically correct brigade, but I am very aware of why causal use of words like that is a problem and is not acceptable.
Pingback: I'm Just Sharing » Blog Archive » Subtle Tech Racism
md20737
I understand how those who are not of color dont understand how a machine could be racist. They think oh it wasnt a person so how can it be racist? But as a person of color I am asking why is google referring to black people as Negro's in 2010? The fact the term is in its vocabulary raises flags for me, and thats why it is considered racist.
beejayski
Hi, I am an african and i dont think there is anything racist about the word negro.Negro means black and if you are black then there is nothing racist in calling you black.
from Wikipedia
The word Negro was used to refer to a person of Black ancestry prior to the shift in the lexicon of American and worldwide classification of race and ethnicity in the late 1960s. The appellation was accepted as a normal and was used by those of Black African descent as well as those of non-African black descent during the eras prior to the Civil Rights movement.
http://butyoureagirl.com adriarichards
Beejayski,
Thanks for your opinion on this.
I personally don't want to be called Negro. I correct people when they
refer to me as “Mulatto” because I'm not. I'm biracial.
You can pull up a dictionary term up but that doesn't stop the sting when
you hear something like this.
John Rourke
Actually, that's why it is 'racist' – to you. I read further on this and, in the case of the Census Bureau, THEY use the word 'Negro', because some people – apparently mostly older people – prefer the term and, in fact, write it in themselves.
So, I would expect that is why the word 'even exists at all in the Google Voice vocabulary'; would you prefer that it not be there, effectively silencing those who prefer that word in identifying themselves?
Having said that, I think it is clear that Google needs to improve their transcription service… but as for actually removing the word 'Negro' from their vocabulary, it seems there are plenty of people who do not agree that it is a 'racist' term.
Marla
Excuse me, but “my kind” is NOT mankind – I am not a man, are you?? That is offensive, please use “humankind” or “human beings” or “human race”. There are plenty of other descriptors that are more accurate, inclusive and not offensive.
md20737
People describe themselves as Negroes as usually older black people. But as black person growing up in a time where I know I am not inferior to white people or any other race I will not allow myself to describe by a term that carries the connotation that Negro does. If never knew better I might call myself a Negro too, but I do know better.
Blacks were Negroes while fighting for civil rights, and living in fear under Jim Crow if you were in some places. Negro to me is unacceptable. Negro deserves to be left in time period it was used it. There is also plenty of people who dont think N*gga is racist either, should that be in Google's vocabulary as well?
John Rourke
Actually, that's why it is 'racist' – to you. I read further on this and, in the case of the Census Bureau, THEY use the word 'Negro', because some people – apparently mostly older people – prefer the term and, in fact, write it in themselves.
So, I would expect that is why the word 'even exists at all in the Google Voice vocabulary'; would you prefer that it not be there, effectively silencing those who prefer that word in identifying themselves?
Having said that, I think it is clear that Google needs to improve their transcription service… but as for actually removing the word 'Negro' from their vocabulary, it seems there are plenty of people who do not agree that it is a 'racist' term.
Marla
Excuse me, but “my kind” is NOT mankind – I am not a man, are you?? That is offensive, please use “humankind” or “human beings” or “human race”. There are plenty of other descriptors that are more accurate, inclusive and not offensive.
md20737
People describe themselves as Negroes as usually older black people. But as black person growing up in a time where I know I am not inferior to white people or any other race I will not allow myself to describe by a term that carries the connotation that Negro does. If never knew better I might call myself a Negro too, but I do know better.
Blacks were Negroes while fighting for civil rights, and living in fear under Jim Crow if you were in some places. Negro to me is unacceptable. Negro deserves to be left in time period it was used it. There is also plenty of people who dont think N*gga is racist either, should that be in Google's vocabulary as well?
Hastur
Oh my god that is kind of horrible, but it did make me laugh!!!
Especially “Yeah, or suck to you bye. ” that is awesome.
anon
I read comments here… hell, you’re crazy. Racist google, lol. You’d better go and read books.
EM
Adria Richards takes my breath away.
That is all I can say. Sorry.