"Human parity" in machine translation
« previous post | next post »
In May of 2015, I gave a talk at the Centre Cournot in Paris on the topic "Why Human Language Technology (almost) works", starting with a list of notable successes, including how well Google and Bing on-line translation did on the Centre Cournot's web site. But my theme required a few failures as well, and I found a spectacular set of examples when I tried a chapter-opening from a roman policier that I was reading (Yasmina Khadra, Le Dingue au Bistouri):
Il y a quatre choses que je déteste. Un: qu'on boive dans mon verre. Deux: qu'on se mouche dans un restaurant. Trois: qu'on me pose un lapin.
Google Translate: There are four things I hate. A: we drink in my glass. Two: we will fly in a restaurant. Three: I get asked a rabbit.
Bing Translate: There are four things that I hate. One: that one drink in my glass. Two: what we fly in a restaurant. Three: only asked me a rabbit.
Should be: There are four things I hate. One: that somebody drinks from my glass. Two: that somebody blows their nose in a restaurant. Three: that somebody stands me up.
These mistakes underline some general remaining difficulties. One: the treatment of pronouns. Two: the treatment of idioms that are not common in the bilingual training material. Three: the lack of common sense.
The French pronoun on can be most appropriately translated as "somebody" or "you" or "we" or "they" in various different circumstances — and sorting out those circumstances is an example of the sort of thing that MT still doesn't do well. Or at least, didn't do well in May of 2015 — and we'll see shortly that things haven't changed much.
The verb se moucher (meaning "vider son nez de ses mucosités") and the idiom poser un lapin [à qqn] (meaning "ne pas aller à un rendez-vous") are not especially obscure, but apparently they don't occur often enough in the sorts of bilingual text that those systems were trained on. Similarly the fact that boire dans X generally means "drink from X", not "drink in X", at least if X is a drinking vessel.
So three and a half years later, have these things improved? Apparently not:
Google today: There are four things I hate. One: that we drink in my glass. Two: we fly in a restaurant. Three: let me have a rabbit.
Bing today: There are four things I hate. One: Drink in my drink. Two: Let's fly to a restaurant. Three: Let me be asked a rabbit.
And unfortunately this is typical. Some aspects of such translations are very good, but the frequent mistakes spoil things. Another (literally) random selection from a recent French novel:
Nous sommes faits comme les rats.
Google: We are made like rats.
Bing: We're made like rats.
Should be: We're trapped like rats.
A slightly longer one:
Là-bas, il le sait, on va tenter de l'infantiliser, de le transformer en papy guimauve. Pas folle la guêpe! Et puis, toutes ces vieilles harpies… Non. Ça ne vas pas être possible. Ras le bol, de ces bonnes femmes!
Google: Over there, he knows, we will try to infantilize him, turn him into grandma marshmallow. Not crazy wasp! And then, all those old harpies … No. It will not be possible. Damn it, these good women!
Bing: There, he knows, we'll try to infantilize it, turn it into Grandpa Marshmallow. Not crazy the Wasp! And then, all those old harpies… Not. It's not going to be possible. I'm sick of these good women!
Should be: There, he knows, they'll try to infantilize him, to turn him into a mushy old codger. He's no fool! And then all those old harpies… No. It's not going to be possible. He's fed up with those biddies.
Or again:
Recevoir un message de son amant en début de soirée, au moment où chaque seconde doit être rentabilisée, cela n'a rien de raisonnable. Pourtant, lorsque son téléphone émet la mélopée caractéristique de la réception d'un SMS, le coeur de Camille prend aussitôt le relais de ses intentions, le contrôle de son cerveau, ne laissant à la raison aucune voix au chapitre.
Google: Receiving a message from his lover in the early evening, when every second must be profitable, this is not reasonable. However, when his phone emits the characteristic chant of the receipt of an SMS, the heart of Camille immediately takes over from his intentions, the control of his brain, leaving no reason for voice in the chapter.
Bing: Receiving a message from her lover at the beginning of the evening, at the moment when every second has to be profitable, it is nothing reasonable. However, when his phone emits the chant characteristic of receiving an SMS, Camille's heart immediately takes over his intentions, controlling his brain, leaving to reason no voice in the chapter.
Should be: Receiving a message from her lover at the beginning of the evening, at a time when every second is precious, there's nothing reasonable about that. Still, when her phone emits the sound characteristic of receiving an SMS, Camille's heart immediately takes the reins of her intentions, the control of her brain, leaving reason no voice in the matter.
[Note that changing the gender-neutral name Camille to e.g. Marie doesn't affect Google Translate's choice to use masculine pronouns throughout…]
Last example (about a car accident):
Ses mains s'agrippèrent au volant. Tout autour, le décor valdinguait.
Google: His hands gripped the steering wheel. All around, the decor was worthwhile.
Bing: His hands grasped at the wheel. All around, the decor valdinguait.
Should be: His hands gripped the steering wheel. All around, the scenery went flying.
In this context, consider Hany Hassan et al., "Achieving Human Parity on Automatic Chinese to English News Translation", arXiv 6/29/2018:
Machine translation has made rapid advances in recent years. Millions of people are using it today in online translation systems and mobile applications in order to communicate across language barriers. The question naturally arises whether such systems can approach or achieve parity with human translations. In this paper, we first address the problem of how to define and accurately measure human parity in translation. We then describe Microsoft’s machine translation system and measure the quality of its translations on the widely used WMT 2017 news translation task from Chinese to English. We find that our latest neural machine translation system has reached a new state-of-the-art, and that the translation quality is at human parity when compared to professional human translations. We also find that it significantly exceeds the quality of crowd-sourced non-professional translations.
The paper is credible and impressive — and it describes a set of systems that work significantly better than what Microsoft now offers via online Bing translation, but are some time away from being generally deployed.
However, it's important to keep in mind that those new systems were trained on a large collection of carefully screened parallel texts (18 million sentence pairs), plus 7 million monolingual sentences on each side; and then tested on (withheld) examples from the same sources.
Applied to a different sort of material — conversational transcripts, or novels, or legal contracts, or scientific papers — the system would encounter new vocabulary, new constructions, and new concepts. We can confidently predict that substantial new training would be needed to approach "human parity" in those new domains. And we can wonder how well the methods would do with (for example) ambiguous pronoun choice in cases where this translation is harder than it typically is in news text.
Jerry Friedman said,
November 6, 2018 @ 10:17 am
Well, they're… close.
In case anyone else was left in suspense by that roman policier:
Quatre : rester là, à ne rien foutre, dans mon bureau minable au fond d’un couloir cafardeux où les relents des latrines et les courants d’air adorent flûter.
The only reason I can do better than Google or Bing on that is that I know when to use a dictionary.
(Four: just staying there, not doing a damn thing, in my pitiful office at the end of a gloomy hallway where the reeks of the restrooms and the drafts love to flutter. Corrections welcome.)
Dominik Lukes said,
November 6, 2018 @ 12:36 pm
I read that paper some time ago and did not find it all that credible. In particular their definition of parity does not actually include human conducted evaluation but rather questionable measures.
Benjamin E. Orsatti said,
November 6, 2018 @ 12:58 pm
Jerry,
I wonder if there isn't something about the French language itself, or, more specifically, its relationship to English, that is particularly confounding to AI. That is to say, the two languages are so, so similar in terms of cognate vocabulary, that the idiomatic divergence over the past thousand years or so from that initial state of semantic near-identity will stymie a brain crammed with 1's and 0's. In a word, there's a reason the term "faux amis" is a French one.
And the "argot" keeps changing so fast! Par example, I'm someone who, by all accounts, should have no problem reading any French text you might have — studied the language and literature since 7th grade, majored in French in college (Penn!), and studied at U. Paris VII while living in Paris for 6 months. Here's my first reading of the quoted text: "Quatre : rester là, à ne rien foutre [not familiar with any sense of "foutre" beyond…], dans mon bureau minable [never seen "minable" before, and no convenient etymology springs to mind] au fond d’un couloir cafardeux [also no clue here rg. "cafardeux"] où les relents [huh?] des latrines et les courants d’air adorent flûter."
Would like to comment further, but have to foutre le champ to go pick up my meuf, whose bagnole is en panne outside her boulot.
Mick O said,
November 6, 2018 @ 1:31 pm
We see that machine translations are cognizant (for lack of a better term) of numbers, as the repetition of weird characters can vastly change the output. So one possibility that springs to mind here. The translations were tripped up by a list of four things the writer hates only having three items.
[(myl) I wish. Adding the fourth item yields (from Google Translate):
There are four things I hate. One: that we drink in my glass. Two: we fly in a restaurant. Three: let me have a rabbit. Four: stay there, do not miss anything, in my shabby office at the end of a roaring corridor where latrines and drafts love to flutter.
These systems are not "cognizant" in any meaningful sense of numbers, they just react differently to things of different lengths, especially in cases where the input is not really constraining the output much (which is what generates the Elephant Semifics examples).]
Zerbie Hynson said,
November 6, 2018 @ 1:56 pm
DeepL does a bit better:
There are four things I hate. One: Let's drink from my drink. Two: blow your nose in a restaurant. Three: stand me up.
Still not exactly right grammatically, but much closer to communicating the actual meaning.
Chandra said,
November 6, 2018 @ 3:37 pm
I find the translation of "mouche" the most surprising – it's a verb confused with the more common but unrelated noun, translated into the equivalent English noun and then used as a verb???
ktschwarz said,
November 6, 2018 @ 3:43 pm
@Mick O, a list of four things the writer hates only having three items: cool guess, but that's not it. DeepL, Google, and Bing all handle only one sentence at a time, with no knowledge of other sentences. Try changing "quatre choses" to "trois choses", or adding a fourth sentence, or putting them in random order, or interspersing other sentences — none of that makes any difference, every sentence is independent. That's also why none of them gets the syntax right: they don't know that the paragraph is a list of "things" that are parallel.
(MYL, your Google translation in the comment just above is different from the one in the main post; did you do it on a different platform, or different day? I've seen different results on Android vs. desktop vs. mobile.)
[(myl) It's the same as the "Google today" translation in the original post, which is slightly different from (but not better than) the 2015 version given at the start of the post.]
Gendered pronouns are a good example of the difficulty when the target language encodes information that the source doesn't. Changing "Camille" to "Marie" is a great test. I tried Google, Bing, and DeepL on the sentence about Camille's phone, changing "Camille" to 10 unambiguously male names and 15 unambiguously female ones and checking the three pronouns (her/his phone, her/his intentions, her/his brain). All of the male names got 100% "his" pronouns. For the female names, DeepL is the best by far, with "her" for all the names I tried except Victoire. Google is the worst, with "her" only for Sarah, "his" for all the others. Bing gets "her phone" right for almost all the girl names, but also commits the faux pas of mixed pronouns for some names: "her phone"/"his intentions" for Jeanne, Adele, Marie, and Catherine.
Tommi Nieminen said,
November 6, 2018 @ 4:42 pm
There were multiple problems with Hany Hassan et al., the most obvious being that the results didn't make much sense if you had a hard look at them. The evaluation methods they chose were direct assessment on a scale of 1-100 and a manual error analysis. The supposed professional translators that the MT was compared to totally bombed on the direct assessment test, getting an average score of less than 70. The error analysis was performed only on the MT output, and the amount of errors was nothing like you'd expect from MT better than professional translators. This should have raised some red flags for the authors.
So apparently the human translators in the study weren't particularly competent, which is confirmed in https://arxiv.org/pdf/1808.10432.pdf. This paper also points out some other shortcomings, such as the fact that half the test set was not text originally written in Chinese, but translations from English (which favors the MT).
Also, https://arxiv.org/pdf/1808.07048.pdf shows that even these poor human translations were preferred to MT when the translations were evaluated as documents instead of isolated sentences.
KevinM said,
November 6, 2018 @ 4:50 pm
Tripped up by 4 vs. 3 items? Then how will it do with Proverbs 3:18-19? Or the Spanish Inquisition sketch?
Nica said,
November 6, 2018 @ 5:18 pm
DeepL does substantially better on all the examples. I'd say machine translation has clearly improved since 2015.
Kaleberg said,
November 6, 2018 @ 10:54 pm
You would think that they might teach ML how to use a dictionary. People get stumped by idioms, but they can work from context or look the darned thing up. Before I loaded a dictionary into my phone, I'd often go for a paragraph or two before realizing what was probably meant by some weird idiom. (She fell like an apple. According to Newton, we all fall like apples. Oh, she's recovering and people are asking her if she's OK. Guess.)
Machine translation has gotten better. For a long time it was nearly worthless. Now it's good enough to for simple commercial web sites and news headlines. Whenever it starts getting weird, go back and look for an idiom.
ajay said,
November 7, 2018 @ 5:42 am
[also no clue here rg. "cafardeux"]
"Cafard" sort of made it into English, courtesy of people like PC Wren: it's the deep-desert equivalent of "cabin fever". It's what you get when you've been stuck in Fort Zinderneuf for too long with nothing happening.
Athel Cornish-Bowden said,
November 7, 2018 @ 7:57 am
I have noticed for a long time that Google Translate does a remarkably bad job with French and Spanish (though it seems to be improving), much worse than it does with Hausa (not that I can understand Hausa, but I can tell when the English translations make sense).
When I read Yasmina Khadra's book Le Dingue au Bistouri I found it very surprising that it was apparently written by a woman, so I checked. In fact "Yasmina Khadra" is a pseudonym, based on his wife's name, of a former officer in the Algerian Army. There are of course examples of women writing under men's names (George Eliot and the Brontë sisters spring to mind) but I can't think of another example of a man writing under a woman's name: are there any? (Not that I disapprove: I admire him for it.) Despite having a somewhat androgynous name myself I have never pretended to be a woman.
ajay said,
November 7, 2018 @ 8:24 am
I can't think of another example of a man writing under a woman's name: are there any?
"Anne Rice" is a pseudonym used by the horror author Howard O'Brien, but she is nevertheless a woman.
There seem to be quite a few male romance authors who use female pen names, according to the answers here: https://ask.metafilter.com/107743/How-many-men-have-written-work-under-a-womans-name
Jason M said,
November 7, 2018 @ 8:42 am
@BE Orsatti- I don’t claim quite the French chops you have, but I read modern stuff and watch some stuff and go to France (or Quebec) once a year on average, but my translations were pretty foutues. I got the Chandleresque noir feel and imagery of the bureau bit and guessed thus more or less correctly for cafardeux and minable and foutre in the sense employed. I knew “relents” but totally missed “fluter” which I took in the sense of piping as if from a flute. But then I also got confused in a quick read about Camille, going with he/his in my head and not revising when we got down to where her name was. And I knew nothing about posing rabbits or crazy wasps but should have guessed “se moucher” because I know what a ”mouchoir” is.
Slightly apropos of this discussion for francophiles with day jobs. I’ve been enjoying Netflix’s “Au Service de la France” with the French subtitles on. Good for early 60s slang anyway.
George said,
November 7, 2018 @ 9:17 am
Recevoir un message de son amant en début de soirée, au moment où chaque seconde doit être rentabilisée, cela n'a rien de raisonnable.
Maybe it's just me, but I read that as a general principle and spontaneously went for one's rather than either his or her.
languagehat said,
November 7, 2018 @ 9:24 am
And the "argot" keeps changing so fast!
That may or may not be the case, but it's irrelevant here, since none of the argot cited is new (and, no offense, but given your remarks — "not familiar with any sense of "foutre" beyond…," "never seen 'minable' before" — you don't seem ever to have read any informal French at all); "poser un lapin," for instance, goes back almost a century (e.g., "Vous vous rendez compte, lui expliqua le relieur, qu'il n'est pas question une seconde que vous me posiez un lapin": Jules Romains, Les Hommes de bonne volonté, 1932).
George said,
November 7, 2018 @ 11:21 am
@languagehat
Indeed. And "poser un lapin" is actually a fair bit older than that. In the 19th century it meant 'not paying for the services of a prostitute'. It's easy enough to see how this could slip into the general idea of 'not keeping one's side of a bargain' and in turn into 'not turning up when one is expected to'. In the example you cite it possibly hasn't quite completed that journey.
Antonio L. Banderas said,
November 7, 2018 @ 12:02 pm
I also immediately thought about the cursive style (for example, Introduction to Chinese Cursive Script, by Fang Yu-Wang).
The arabic style which uses ligatures heavenly (Unicode Arabic Presentation Forms A) might feel distinctly as well for those only used to the Latin script.
Incidentally, for those interested, the following article adds a bit more to the topic:
Stroke systems in Chinese characters: A systemic functional perspective on simplified regular script (by Xuanwei Peng).
Peter S said,
November 7, 2018 @ 1:43 pm
Google: Four: stay there, do not fuck, in my shabby office. Bing: Four: Standing there, not doing anything, in my lousy office.
I think Bing wins on the fourth round.
(Original French, *Quatre : rester là, à ne rien foutre, dans mon bureau minable …*)
Chandra said,
November 7, 2018 @ 2:44 pm
@ajay – Anne Rice is her legal name, not a pseudonym (legally changed first name + married surname), though she was indeed given the name Howard O'Brien at birth.
Geoff said,
November 7, 2018 @ 4:32 pm
@George: I was wondering how such a random phrase as 'poser un lapin' could come to have its modern idiomatic meaning. Presumably the original idea was 'offer a worthless gift instead of paying properly'?
Jerry Friedman said,
November 7, 2018 @ 4:59 pm
Ayant bien réfléchi, the Google Translate version might be better than mine where I used a dictionary, though overall I think mine is better. Also, I guessed that "flûter could be "flutter", but it isn't in WordReference or Wiktionary, so I wasn't very sure till I looked at Google Translate's version.
languagehat: no offense, but given your remarks — "not familiar with any sense of "foutre" beyond…," "never seen 'minable' before" — you don't seem ever to have read any informal French at all
Not to speak for Benjamin E. Orsatti, but "any at all" is a bit of an exaggeration. I've read a few books' worth of informal post-1932 French and I didn't know the same words he didn't know. Maybe they didn't happen to be in my reading, or maybe I saw them and forgot them in the 20-some or 30-some years since then. I'm afraid my conversational French would suffer from my having read too much Racine and such, though.
Benjamin E. Orsatti said,
November 7, 2018 @ 5:10 pm
@Jason M
"Slightly apropos of this discussion for francophiles with day jobs. I’ve been enjoying Netflix’s “Au Service de la France” with the French subtitles on. Good for early 60s slang anyway."
Thanks! Will definitely check out.
@languagehat
"you don't seem ever to have read any informal French at all"
Probably true. But I _did_ know "poser un lapin"; like I said, I was in France in _college_, so dealing with female rejection in translation was certainly not an option.
But I still can't shake the notion that French (as distinguished from, say, Italian), is a great deal more idiomatic than English and other Romance languages. There has to be some sort of corpus study that the CL guys can gin up to arrive at something like a "literal / figurative" language use ratio, no(n)?
George said,
November 7, 2018 @ 5:26 pm
@Geoff
I'm afraid I don't know offhand how it came to mean what it meant in the 19th century. Like you, I'd found the expression bizarre and asked someone years ago when I was living in France (which was in the '90s). Nowadays I could Google it, of course, and pretend that I already knew! So maybe I will… just without the pretending bit.
Paul Kay said,
November 7, 2018 @ 7:51 pm
@ Athel Cornish-Bowden
"On est toujours trop bon avec les femmes" (We always treat women too well) was written by Raymond Queneau under the pseudonym Sally Mara. Queneau also wrote "Le journal intime de Sally Mara". (Queneau is probably best known to anglophones as the author of "Zazie dans le metro".)
ajay said,
November 8, 2018 @ 9:04 am
I was wondering how such a random phrase as 'poser un lapin' could come to have its modern idiomatic meaning. Presumably the original idea was 'offer a worthless gift instead of paying properly'?
Or if the original sense was to do with leaving without paying for services, perhaps the idea is that you're running off with rabbit-like speed.
(Original French, *Quatre : rester là, à ne rien foutre, dans mon bureau minable …*)
Best translation would surely be: "Four: stay there, doing fuck-all, in my shabby office…"
languagehat said,
November 8, 2018 @ 9:18 am
Or if the original sense was to do with leaving without paying for services, perhaps the idea is that you're running off with rabbit-like speed.
Very possibly; compare the venerable Russian expression ехать зайцем [ekhat' zaitsem] 'to travel without paying for a ticket' [literally 'to travel like a hare'].
ajay said,
November 8, 2018 @ 9:35 am
Presumably because you run the risk of ending up jugged. (While we're on the subject of synonyms for prison…)
Jerry Friedman said,
November 8, 2018 @ 2:51 pm
ajay:
Or if the original sense [of "poser un lapin"] was to do with leaving without paying for services, perhaps the idea is that you're running off with rabbit-like speed.
Wiktionary relates it the use of the rabbit as a symbol of fertility and hence wealth, so its absence means poverty, but I don't get that.
To me the explanation at expressio.fr is much more believable. It says that faire poser was slang for making someone wait, and lapin referred to the rabbits on some kind of merry-go-rounds, which could never be reached.
(Original French, *Quatre : rester là, à ne rien foutre, dans mon bureau minable …*)
Best translation would surely be: "Four: stay there, doing fuck-all, in my shabby office…"
"Staying there", right? Since it's a thing. Anyway, it depends on your dialect. "Fuck-all" might work in British English and related varieties, but it doesn't work in American English. Also, I have the impression that "fuck" is good deal stronger in American English than "foutre" is in French, but I don't really know, much less how strong it is in British, Australian, etc., English.
Andrew (not the same one) said,
November 8, 2018 @ 3:22 pm
William Sharp alias Fiona Macleod is a classic example of a man writing under a woman's name.
James Wimberley said,
November 8, 2018 @ 4:03 pm
Languagehat: could the absconding rabbits have the same source as "bistro", the Russian officers quartered in Paris in 1814-15? I dare say they led active social lives.
George: With "son amant", the problem isn't the pronoun but the noun, which is masculine in form. I'm been away from France for a decade, but I feel that "petite amie" would be more usual as the feminine.
George said,
November 8, 2018 @ 4:58 pm
@James Wimberley
I think you've misunderstood what I wrote. My point was that if what's being expressed is a general principle, then Camille isn't necessarily referring to her lover, but to anyone's lover, in which case neither his nor her would capture it, whereas one's would.
Andrew Usher said,
November 8, 2018 @ 8:32 pm
Perhaps we shouldn't be so quick to think of French as 'idiomatic' without thinking about our own language. 'Poser un lapin' is strange, no doubt, but the English equivalent 'stand (someone) up', which everybody has used, is probably just as unpredictable if a little less weird.
And the 'new' uses of 'foutre' are paralleled by our use of 'fuck', which is famously productive. 'Fuck-all' certainly works, 'fuck around' also came to mind … enough details.
George:
Certainly "one's" could work there, though probably it wasn't the intention. The French pronoun is of course completely ambiguous since it simply agrees with its noun.
k_over_hbarc at yahoo.com
ajay said,
November 9, 2018 @ 5:32 am
Wiktionary relates it the use of the rabbit as a symbol of fertility and hence wealth
Fertility, yes, but wealth? Are there any other examples of this link? In English rabbits are only associated with speed ("off like a rabbit"), with fecundity and amorousness ("at it like bunnies", "breeding like rabbits") and with timidity ("If Mary Whittaker were to marry, Miss Climpson thought, she would marry a rabbit"). Rabbit is a poor man's food, and historically regarded as vermin rather than game.
George said,
November 9, 2018 @ 7:29 am
@Andrew Usher
Without further context, I don't think we can say which of the two interpretations is the most probable. Full disclosure: I'm about as near to being perfectly bilingual in English and French as it is possible to be without having grown up in a French-speaking country or having a parent who's a native French speaker (I can 'pass' unless I'm very tired or tipsy).
Stuart Brown said,
November 10, 2018 @ 6:35 pm
I agree with George that the first sentence of the novel extract with “son amant” in is a generalisation, and therefore the translation should be “one’s lover” or “your lover”.
With regard to the bit of roman noir, I don’t like “doing fuck all”, which jars on this aged BrE speaker, because to me that’s much stronger than the French. In my idiom “doing sod all” or “doing bugger all” might work. Of course, it depends who you are translating for. This Brit finds “restroom” quaintly euphemisti anyway, and surely “latrines” suggests something primitive (GB slang might suggest “the bog”).
I knew several meanings for the noun “cafard” : cockroach, informer, “the blues”, but I’d never come across cafardeux. But to me “gloomy” just suggests poor lighting: I’d suggest “depressing”. For “minable” here I’d suggest “crummy”, but maybe that is very BrE?
Finally, there is no way that flûter means to flutter! it would be a terrible image anyway: stench and draughts don’t flutter. Flûter must come from flûte, which of course is a flute! It’s still a bit of a mixed metaphor, but I get the sense that the draughts are whistling down the corridor, carrying the stench from the toilets with them.
ktschwarz said,
November 10, 2018 @ 7:23 pm
Even if Google's Hausa-English translations sound good in English, you can't trust them to represent the original. For example, this headline from a Hausa news site is translated by Google as: "The English-speaking people in Nigeria are about one million and twenty." But it really means: "The Hausa-speaking population in Nigeria has reached 120 million." "Hausa" is often translated as "English", probably for the same reason as the old systems that used to translate Austria to Ireland, or London to Paris.
(Note: 120 million is wildly inflated. Ethnologue puts Hausa at ~60 million worldwide, including L2 users.)
George said,
November 11, 2018 @ 5:13 pm
…. and I agree with Stuart Brown that "doing fuck all" is a lot stronger than "ne rien foutre".
However, as possibly the only person here who has actually spent the night in an Algiers police station (not, I hasten to add, in custody), I can live with 'latrines'!