Sorry, I didn't get that
January 2, 2017 11:17 AM   Subscribe

Japanese comedians try to use Siri in English and the results are hilarious. Otona no Kiss eigo (大人のKISS英語) is "an 'English conversation variety show' where we get exposed to English through many kinds of variety projects week by week!!" [Episode playlist] The host, Tomohisa Yamashita (山下 智久), is a widely-known actor, singer, and TV host.
posted by AFABulous (33 comments total) 19 users marked this as a favorite
 
As someone who tutors ESL, I think the average person would be shocked at how long we spend just getting one word like this correct. And "alarm" is an excellent word to practice because Ls and Rs are the literal worst. The female contestant really nailed that R! But the repetition helps sooo much. There are words my learner could not even *hear* correctly a year ago, that she now gets right 70% of the time.
posted by greermahoney at 11:37 AM on January 2, 2017 [4 favorites]


Yeah, there's a bunch of really interesting phonetic gaps between English and Japanese, in both directions, and they chose some great simple-seeming specimens to trip folks up with this. "Earth" is a word made up of like three different phonetic moves that just don't show up in Japanese at all basically.

One of the things I thought was really interesting as a wrinkle watching this as an American viewer was that the English-language instructor's own accent sounded non-rhotic when she wasn't making a point of sticking the landing on the r sounds; her example "alarm" had the r all but disappearing a couple times, which if this was Siri set to US English probably wasn't helping all that much. I couldn't quite pin her down, Austrailian or New Zealander maybe?

Very fun watch, thanks for posting it.
posted by cortex at 11:46 AM on January 2, 2017 [2 favorites]


Right, Cortex? The contestant pronounced it better! (Or, I should say, more Standard American.)
posted by greermahoney at 11:52 AM on January 2, 2017


I play this game with Siri all the time, because it's also not a fan of English with a regional accent. It's especially hilarious when I'm driving and asking for directions and it keeps telling me "I can't find a location for [word that fails to understand Norn Irish] Street" and after the third time I end up yelling THATS NOT WHAT I FUCKING SAID AND YOU KNOW IT
posted by billiebee at 11:54 AM on January 2, 2017 [9 favorites]


20-odd years ago in Tokyo I had a friend of a friend (who worked for Motorola Japan) get a spiffy new AV PowerMac, with voice recognition.

But you had to say "Computer . . ." and "KOHMPYUTAH" wasn't close enough . . .
posted by Heywood Mogroot III at 11:55 AM on January 2, 2017 [1 favorite]


The instructor definitely sounded Australian or New Zealander in the second half of the video when she was explaining "earth."

This is hilarious and I kind of want to do the opposite challenge, except I'm not sure if the exercise translates because the Japanese phonetic alphabet seems simpler/fewer phonemes? Maybe? I don't know.
posted by chrominance at 11:57 AM on January 2, 2017 [1 favorite]


I don't know either, but I'd like to try it too. I just need a list of words that commonly trip up non-native speakers.
posted by wanderingmind at 12:03 PM on January 2, 2017 [1 favorite]


I think this is actually kind of a big deal, since the mid-Atlantic accent is, for the time being, regarded as "standard English pronunciation." I suppose over the next few years AI will get better at recognizing regional accents. I know that Google Assistant (Google's version of Siri) is getting better at recognizing and interpreting non-standard pronunciation, and even "non-standard" names, such as my wife's name (she's Japanese, and for the longest time when driving and using hands-free I would have to tell Google to "phone wife", rather than "phone 'wife's Japanese name').

I think for a lot of cultures, though, including Japanese culture, having a standard, elevated pronunciation, where there is a "right way" to pronounce certain words, is pretty normal.
posted by My Dad at 12:04 PM on January 2, 2017


I'm a native English speaking American but I was born hard of hearing and had speech classes for many years. Voice recognition menus (e.g. the cable company) are still hell.

Siri seems to be skewed towards male voices because the longer I'm on testosterone, the better it understands me.
posted by AFABulous at 12:16 PM on January 2, 2017 [2 favorites]


Seeing YamaPi brings me back to my days in the JE fandom. Pi-chan is over 30 now... Time flies etc.
posted by fatehunter at 12:28 PM on January 2, 2017


Are we supposed to believe that the "contestants" are just regular Joes (Jiros?) off the street? Because the two men come across as SNL characters and the woman seems to be from the cast of a different show entirely.
posted by emelenjr at 12:43 PM on January 2, 2017 [1 favorite]


Stephen Fry's QI suggestion for a phrase to baffle the Japanese tongue was "orange-tipped fritillary".
posted by Paul Slade at 12:52 PM on January 2, 2017 [1 favorite]


Are we supposed to believe that the "contestants" are just regular Joes (Jiros?) off the street?

I'm going to assume Japanese comedy/reality show viewers are just as down with the concept of editorial kayfabe as media consumers everywhere else.
posted by cortex at 12:59 PM on January 2, 2017 [4 favorites]


Are we supposed to believe that the "contestants" are just regular Joes (Jiros?) off the street?

The premise, AFAICT, is that they're going around a TV studio asking celebrities to speak English. So these are people who're used to being on camera, and also explains the "SNL character" vibe from those two - they're comedians, being comedians.
posted by wanderingmind at 1:03 PM on January 2, 2017 [4 favorites]


There are definitely Japanese words that you can get close enough in context of a sentence for people to understand, but still be wrong. It's kind of a tough test to say a single word "correctly" enough. (I was saying 好き wrong forever and still can't quite hear if I'm doing it right or not)
posted by ctmf at 1:26 PM on January 2, 2017 [1 favorite]


Sometimes when my field assistant and I get bored in the forest, we try to trick eachother with words which are hard to pronounce. When he starts laughing at me for not being able to speak his tonal language properly, I ask him to say "Slightly," and hilarity ensues.
posted by ChuraChura at 1:54 PM on January 2, 2017 [7 favorites]


I think the hardest Japanese words for English-speakers to pronounce are ones with long vowels, and also words containing "ryu/rya/ryo" sounds, like "ryuugaku" (留学). They might also have trouble with "tsu" at the beginning of a word like "tsuite" (ついて). And words where "r" comes right after "n" like "benri" (便利).

This is assuming the player knows the basics of romanized Japanese words. If someone really has zero knowledge of Japanese then quite a lot of words will be mispronounced just from being misled by the spelling--e.g. one might look at "samidare" (五月雨) and think it's pronounced "say-mih-dayr" when it should be pronounced "sah-mee-dah-ré" and similarly karaoke is not "carry-oh-kee" but "kah-rah-oh-ké"

There are also words that contain "ei" in the spelling but are pronounced more like "eh" in standard dialect. It took a while for me to get Siri to recognize "keizai" (経済).

Also there are some words where the emphasis matters. I had trouble in the past (and still do) with differentiating between "ame" (雨/rain) and "ame"(飴/candy). When I speak to Siri it always assumes I'm saying rain, but I'm not sure if that's because of my pronunciation or if it's just favoring inquiries about the weather. If I use the microphone function on google translate, it will also assume rain, unless I put it in context--it correctly interpreted me saying "ame tabetai" as 飴食べたい ("I want to eat candy").
posted by picklenickle at 2:31 PM on January 2, 2017 [3 favorites]


ones with long vowels

And double consonants. A double 't' in English sounds about the same as a single, so if you learned the word by reading, you might be pronouncing it wrong. But I think it would be easier to imitate a speaker if you had to, like in the video, because the sound actually exists in English.
posted by ctmf at 2:40 PM on January 2, 2017


This is silly.





(I'll show myself out)
posted by randomkeystrike at 3:25 PM on January 2, 2017 [1 favorite]


I play this game with Siri all the time, because it's also not a fan of English with a regional accent. It's especially hilarious when I'm driving and asking for directions and it keeps telling me "I can't find a location for [word that fails to understand Norn Irish] Street" and after the third time I end up yelling THATS NOT WHAT I FUCKING SAID AND YOU KNOW IT

Reminds me of this bit about a voice activated elevator in Scotland!
posted by sively at 3:28 PM on January 2, 2017 [8 favorites]


Is that weird, lopey male idol stride something that people are born with, or must it be learned?
posted by evidenceofabsence at 4:03 PM on January 2, 2017


For people with non-American accents, Siri and Google Now really suck. And not quite enough that we give up using them; just enough that every time we hope it will be slightly quicker than typing, and then it isn't.

I made my phone think it was a pixel recently so I could install Android Nougat and the new voice assistant, because I was thinking about getting a Google Home but didn't really believe it would be good enough at voice recognition to be really usable. And sure enough, I've had to go back to the older Google Now, which at least understands me about 50% of the time. (My husband, whose NZ accent is less broad than mine, and who has a deeper voice, gets understood almost 100% of the time, which makes me grumpy because technically English isn't even his first language.)

What the new voice assistant couldn't handle, was my pronunciation of "Google" itself, even after lots of training. So that meant it wouldn't even trigger unless I was super careful to make sure the final "l" didn't become a vowel instead, as it naturally doesn't in my accent. So to get it to trigger I'd usually have to say, "okay google" about three times, increasingly carefully (and with a rising note of frustration).

Google Now is better because it actually has a NZ English setting, although it still is far from perfect for me. But the closest the new assistant has is Australian, which I think is the main issue.
posted by lollusc at 4:32 PM on January 2, 2017 [1 favorite]


(Also the front rounded vowel in "google" is one of the ones AusE and NZE really differ on, so "okay google" is one of the worst possible trigger phrases for more than one reason.)
posted by lollusc at 4:35 PM on January 2, 2017


and after the third time I end up yelling THATS NOT WHAT I FUCKING SAID AND YOU KNOW IT

Yeah, I have to say that 90% of the time my conversations with Google Now go, "Okay Google...". "Okay GoogEL", "OKAY GOOGLE", "Fuck you, Google."
posted by lollusc at 4:39 PM on January 2, 2017


"Okay Google...". "Okay GoogEL", "OKAY GOOGLE", "Fuck you, Google."
posted by lollusc

I think you may have found the new trigger for using Google..
Can imagine trying to teach people in a shop that one.... "Fuck you, Google"
posted by Merlin The Happy Pig at 6:07 PM on January 2, 2017


I think the comedians were set up because the woman, with her perfect Rs, probably speaks clear, fluent English. I remember being in a bar in Japan, and ordered a beer. The Japanese word for beer is pronounced something like 'beerdu' with a very soft 'd', so pronouncing the English 'beer' with the non-Japanese 'r' is probably very difficult.

The bartender had a neat little trick, though. He pronounced it like a Bostonian, dropping the r altogether. "One bee-ah coming up!"

Other languages have sounds that we are not so familiar with as native English speakers. I remember someone trying to get me to pronounce a French word with one of the many 'u' sounds that we don't have and I was like these comedians, trying over and over and never getting it right.

And I remember some Dutch students having fun with my inability to say some Dutch words that no one outside of the Netherlands, even the Germans, can pronounce correctly. I retailated with the word 'horror', which was quite a challenge for them.
posted by eye of newt at 7:38 PM on January 2, 2017


An amusing one my Uruguayan ex and I discovered was "red railroad." In spanish it's "ferrocarril rojo". Each is basically impossible for a non-native speaker to say.
posted by juice boo at 8:35 PM on January 2, 2017 [1 favorite]


@eye of newt: Japanese renderings of "beer" vary between ビール (what you heard as "beerdu", with the ル landing between an "R" and an "L" and sounding to some like a "D") and ベーア depending on context. Basically two well-recognized methods for dealing with a sound that actually doesn't exist in the target language.
posted by oheso at 4:58 AM on January 3, 2017


@chrominance: Back in the day I was spinning a scifi tale where one subtheme was using Japanese for computer voice recognition based on the limited set of phonemes and the more standardized pronunciation thereof. (Dialects in Japanese are more about differences in colloquialisms than about pronunciation in the way that, e.g., Georgia differs from Bronx.)

Two things have happened in the interim to put me off this particular subtheme: Computer voice recognition overall has gotten better, and my experience with Japanese voice recognition is that my pronunciation is still not recognizable to machines despite being recognizable to most people I speak with.
posted by oheso at 5:03 AM on January 3, 2017 [1 favorite]


Building off what oheso said, the take away for me of the first video is a commentary about the state of voice recognition.

First, yes I found the video humorous and amusing.

Second:

To my ears, as a biracial native English speaker who (admittedly) has many native Japanese and Japanese-American friends (and many mixes of CJK-native speakers), the first two "contestants" to articulate the word "alarm" were perfectly and unmistakably saying the word "alarm" to my ears.

So, to my mind, the problem is the disjunct between the obviously recognizable if somewhat inflected articulation of the word and the voice recognition algorithm which performs so poorly that it cannot recognize obviously human pronunciations of "alarm" as "alarm".

In other words, it's not the speakers who are at fault but the bad voice recognition algos.

To that point, even native speakers of English can have a hard time having their speech properly transcribed into machine-readable text streams by such software, and don't even get me started on machine transcriptions of non-English vocalizations by native English speakers. I mean, try getting Siri to recognize many common Japanese names and correctly transliterate them.
posted by mistersquid at 9:21 AM on January 3, 2017 [1 favorite]


I retailated with the word 'horror', which was quite a challenge for them.

There was a hilarious video of German speakers all trying to say the same word, but I forgot what it was. I'm pretty sure it was the name of a common animal. Owl, maybe?
posted by AFABulous at 11:18 AM on January 3, 2017





Reminds me of this bit about a voice activated elevator yt in Scotland!
posted by sively at 3:28 PM on January 2 [8 favorites −] Favorite added! [!]


I thought of that as well, mainly because I've just been in Edinburgh for the first time in 45 years. When I was a kid, I was known for my mean Scots dialect, but then I forgot it. Going back, it took me minutes to recover, and the elevator sketch was perfect. English is many languages and Siri seems to have a problem getting this.
posted by mumimor at 12:07 PM on January 4, 2017


« Older Stories of Irish suffering opened wounds that had...   |   Airlines Take Note Newer »


This thread has been archived and is closed to new comments