## o ai aaa oa ueui

As ktschwarz pointed out in the comments on yesterday's post "Easy going crazy", Google Translate is disposed to recognize text consisting only of vowels and spaces as Hawaiian, and to hallucinate a coherent if sometimes chilling translation into English.

In order to exercise this option more fully, I wrote and tested a simple R script to generate random messages of this type:

 N = 150
Letters = c("a","e","i","o","u"," ")
cat(sprintf("%s\n",paste0(sample(Letters,N,replace=TRUE),collapse="")))

So for example:

Or again:

One more time:

OK, I can't resist:

This is addicting, but I can stop any time I want to:

More:

iua eo eio oea iiaeaaou eeue uoauoi o oaaoaooioeouaiu oeuaiui aaiao oii eao aaaa u aaiuoiaiueeooaaoo euuue oeoaaeoeaoaoeoeouaio i o auoi ueeouei iaei

you are not sure that you have to keep your eyebrows and you will be able to make sure that you have the right to make a difference

aiauee uiiu ouio ea ao eueaiueoieui i eaoeoi e uoioooaaeoeiua iuo ui ioeuui aaio oiaeauau eii eoio iiioee oui oueoaueuaaiuu uee e oiiae ia uu eao

let's see you while you are here to go to your friends and keep them in touch with you here and then you will be able to keep your eye on it.

aoeeaao aai ooo oooaoooaueaueeeuaeee oieiaa iiuuoi oaeaae u o ou aoeaaaaoo uueuiaeoooauaia auouuoi o o uu a eu eiaaoouaaoeeeiuuo a iaiue ioaiia u

noisy noo noo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo nooo noo

OK, I can take a hint — I'll stop there… Except that if I change the last letter of the first "word" to 'i':

aoeeaai aai ooo oooaoooaueaueeeuaeee oieiaa iiuuoi oaeaae u o ou aoeaaaaoo uueuiaeoooauaia auouuoi o o uu a eu eiaaoouaaoeeeiuuo a iaiue ioaiia u

Or to 'e':

aoeeaae aai ooo oooaoooaueaueeeuaeee oieiaa iiuuoi oaeaae u o ou aoeaaaaoo uueuiaeoooauaia auouuoi o o uu a eu eiaaoouaaoeeeiuuo a iaiue ioaiia u

recreational and productive products and services of minerals and minerals of minerals and minerals in minerals and minerals

There are some interesting results with shorter inputs:

 aoeeaaa aai summer meal aoeeaae aai air conditioning aoeeaai aai number of cities aoeeaao aai breakfast aoeeaau aai lunch aoeeaam aai the temperature aoeeaan aai last year aoeeaav aai tourism

1. ### Jenny Chu said,

February 17, 2018 @ 8:34 am

Hallucination is too gentle a word for it.

2. ### 번하드 said,

February 17, 2018 @ 12:14 pm

I wonder how they would parse Shooby "The Human Horn Taylor", e.g. "Speedo":

3. ### Rebecca said,

February 17, 2018 @ 1:08 pm

I just tried it out from the keyboard and was surprised to find that I had to type quite a bit (maybe 20-30 characters) before it quit thinking it was Japanese and "realized" it was Hawaiian. That said, I'm please to have landed on a very useful phrase, should I ever need it:

ioueoeeeo aa uiouue e eoo a io uouou ioee : 'This is not the case'

4. ### Ed M said,

February 17, 2018 @ 2:29 pm

Translate decides that "aeiou" is Hawaiian, and "a e i o u" is Portuguese. It translates Portuguese "a e i o u" to Hawaiian "a a me oe", and Hawaiian "a a me oe" to "e você": "is that you".

5. ### Brian Ogilvie said,

February 17, 2018 @ 4:07 pm

Using multiple repetitions of "oe" produces some metaphysical reflections on identity, though at a certain point Google Translate gets confused about how well it knows you:

http://tinyurl.com/ya5fcay2

A couple lines:

"it is thou, whether thou, or thou, art thou, even thou, even thou, even thou, even thou, even thou, even thou"

"you are of the same nature as you, or of whom you, even thou, art thou, and who art thou"

[(myl) For posterity, after the Hawaiian translation changes:

]

6. ### Julie said,

February 17, 2018 @ 6:23 pm

It reminds me of the predictive text weirdness you get when you let your phone suggest the next word for a whole sentence.

7. ### Rubrick said,

February 17, 2018 @ 7:50 pm

Seems like it would be an improvement if, when the AI's confidence level for a certain translation is below a minimal threshhold, it just threw up its virtual hands and said "Sorry, I got nothing".

Much less fun for us, though.

8. ### John Laviolette said,

February 17, 2018 @ 11:11 pm

> Seems like it would be an improvement if, when the AI's confidence level for a certain translation is below a minimal threshhold, it just threw up its virtual hands and said "Sorry, I got nothing".

That reminds me: does anyone know why Bing translate, at least on Twitter, sometimes identifies English phrases as a particular language and asks if you want to translate it to English, and it almost always responds with "Could not translate"? It seems that it would at least try to translate the word it identifies as being another language…

9. ### John Swindle said,

February 17, 2018 @ 11:26 pm

For this non-speaker of Hawaiian, the results "summer meal," "breakfast," and "lunch," all from long inputs ending in "aai," call to mind Hawaiian ʻai, "eating." Is this a coincidence, or is that pretty much unknowable?

10. ### John Laviolette said,

February 17, 2018 @ 11:31 pm

Example:

https://i.imgur.com/wV9F7cu.png

(Easiest way to find one was to look through tweets about Star Trek, because science fiction proper nouns seem to confuse it the most… but it happens with tweets on other topics, too.)

11. ### John Swindle said,

February 17, 2018 @ 11:41 pm

I should have added that, in the very last example, Hawaiian ʻoe, as in Liliʻuokalaniʻs famous song "Aloha ʻOe," is of course "you," so the many occurrences of "you" and "thou" in the output do have a rational connection with the "oe oe oe …" business in the input.

12. ### Kaleberg said,

February 18, 2018 @ 12:05 am

Does this work for the vowel deprived languages of eastern Europe? I tried typing in a few examples, but just got GIGO. Still, it is interesting the way the vowels seem to have drifted to the mid-Pacific while the consonants cluster half a world away.

13. ### Paul A Sand said,

February 18, 2018 @ 6:49 am

For what it's worth, a Perl version:

perl -e "print (map {(split(//,'aeiou ')) [rand(6)] } 1 .. 150);"

I got something … almost profound from Translate:

"you know how much you know and how much you know about how much you know about how to use your"

Well, right up until the end.

14. ### Jarek Weckwerth said,

February 18, 2018 @ 6:50 am

@Kaleberg: Does this work for the vowel deprived languages of eastern Europe?

It probably won't work with random garbage, because those languages are not in fact vowel deprived. There are occasional words spelled without so-called "vowel letters" in e.g. Czech and Croatian, but quite few. As a result, "vowellessness" is probably not sufficiently indicative.

But for many Indo-European languages (at least*), if you remove the vowels from actual (non-garbage) text, GT will detect the language correctly, simply because "consonant letters" constitute the majority of text.

You can "even" try it with English, for example: http://bit.ly/2o7RvWA.

(*) I would even risk saying, for many languages with alphabetic writing systems, or even most languages in general.

15. ### Dan Lufkin said,

February 18, 2018 @ 2:58 pm

I wonder what would result from an input text in which the first and last letters of words are preserved but the interior letters are randomly rearranged. I recall a recent study that found that human readers are not even slowed down much in reading such text.

[(myl) The "research" was nonsense, alas — or more precisely, it never existed as depicted in the media, and the conclusions foisted on credulous readers were completely false. See "Invaralent Jumumble at the Guardian", 10/23/2003, and the links therein. My observation in 2003:

A lot of journalistic commentary is like this: a few scraps of false rumor, social stereotype and personal anecdote, eked out with enough conventional wisdom to fill the measure.

]

16. ### James Wimberley said,

February 19, 2018 @ 5:42 am

ARM have just announced a line of microprocessor designs specifically for AI applications running on your phone or tablet, without the need for a fast connection to the cloud. The selling point for ARM is that the processors will be really cheap, single-digit dollars each. So there is much more AI goodness to come!

17. ### Ollie said,

February 19, 2018 @ 2:33 pm

For what it's worth, here's a quick Python version:

 import random letters = ['a','e','i','o','u',' '] def makePhrase(): phrase = '' for i in range(250): phrase += random.choice(letters) print(phrase)

 

# Now simply call makePhrase() each time you want a new phrase! Copy and paste this into Google Translate for an odd translation. 

18. ### Emily said,

February 20, 2018 @ 12:44 am

If you type enough e's Google thinks it's Maori… and "byteactive and byteactivating the point of view"
https://imgur.com/a/2N6pU
It gives up at around 300 e's, though, and just returns the input.

19. ### Emily said,

February 20, 2018 @ 12:52 am

Oh, and 119 u's is Hawaiian for "alliuuuuouiouiuouiouiuoui." Whereas 7 of them is Maori for "skeleton."

[(myl) And more:

]

20. ### Dan said,

February 21, 2018 @ 5:35 pm

Oh my goodness, this is fun!

aaoouueeii eeii aaoo u
ououououououououououououououououououououou
uiuiuiuiuiui uiuiuiuiuiuiuiuiui oioioioioioioioioi

set up the frequency
do not get married again
telecommunication webcams online

21. ### Andy Averill said,

February 21, 2018 @ 6:15 pm

Gertrude Stein would be so proud