Liulishuo Huang (流利說謊);
Or, Lying Fluently

When Bullshit Blows the Cow (吹牛)

You know, I honestly wanted to lay off blowhard programmers who pretend to be language gurus. I thought Duolingo was example enough. Luis von Ahn showed dick for linguistic knowledge in his TEDx talk. Instead, he pulled the con: "I'm a computer genius. I have a vision. Give me money." Then, he removed the only good thing about Duolingo (the part that translated websites). Then, he sold ignorant people a shitty pile of quizzes that were built for a separate purpose. Surely, people won't fall for that again, will they?

Well, fuck my luck! Some Chinese guy copied the exact same bullshit formula. His name is Wang Yi (王翌), and he's not just any language industry fraud. No, no! He worked for Google for not even two years. A project manager, very nice! The project was about language learning, right? I'm betting not. Even if it were, it's not like he mattered to it. Product managers are middlemen between coders and corporate folk. They don't build the software. They don't even come up with the ideas. They just get products out on time and under budget. Sound like he's too focused on business to care about language learning? Seem like he probably doesn't know shit about language education? Bingo! He's yet another a snake oil salesman.

It's all here in an interview he did in March:

"So, that’s something that I’ve sort of got off the plane and started really observing the local market when we saw the need. But then, we thought, 'Okay, mobile is surging.' Pretend you’re in May of 2012 in China, you saw this karaoke app named 'Changba' really surging. At first, I thought, 'It’s stupid. Who would be singing to their cell phones?' But, apparently a lot of people did across different age groups. […]
"And I thought this built-in microphone thing is really unique. It could change people’s behavior. So, we thought, 'Okay, if they were so into singing into their cell phones, maybe they could also practice English. But then, how can we make them stick? So, we thought, Angry Bird [sic] was very popular at the time. Can we gamify it? What would be the key elements of gamification that we could use? Instant feedback. So, what kind of instant feedback? Maybe some feedback on their pronunciation to keep their score, to give them some indication of how good their pronunciation is.' At that time, I called my college buddy Lin Hui who was a research scientist at Google in Mountain View back then, specializing in speech recognition and data mining. I’m like, 'Hey, can you do this?' He said, 'Yeah, of course.' And, I’m like, 'Okay, so why don’t we do something together?'" -- Wang Yi, a program manager at Google for 22 months
Did you catch that? He got the idea from a goddamned karaoke app! He didn't even question himself. He didn't go, "Maybe I should check some literature. I should see if my idea makes any fucking sense." Nope, he shit it out, called it gold, and then called Lin Hui to do the actual work.

Also, in case you were wondering, that idea is shit. Implicit corrective feedback is okay. There's a problem, though. Liulishuo's design doesn't provide corrective feedback. It just judges your output as it pushes you along. That's Wang Yi's "gamification". The came is called, "Guess What My Database Wants To Hear". Confusing report cards with corrective feedback only proves that Wang Yi and Lin Hui are Chinese. Calling report card generation a "game"? That just proves that they don't know what fun is.

Meh, he spun his bio and didn't do any real research. Big deal!

Yeah, what matters is if Liulishuo actually works. Oh, wait.

His Staged Interview Showed That Liulishuo Doesn't Really Work

"She [my mother] is literal [sic] addicted to the app now, and sometimes even enlists my help when she's stuck at a certain level." -- Zara Zhang. an investment analyst at GGV Capital and a former journalist 
This is clearly an unbiased report from an objective former journalist. It couldn't be that GGV Capital got Liulishuo (LAIX) its seed money, and then got GGV podcasters to kiss Wang Yi's ass in an "interview". An objective journalist admits that her own mother gets "stuck" and needs her to make up for Liulishuo's defects. Never you mind that, though! Wang Yi went to Princeton. He worked for Google. He put AI in his app. He shot an 18 on his first golf game. Okay, one of those isn't true. He didn't really put AI in the app. "AI" is just a buzzword for stuff that's existed for decades. The only "AI" in Wang Yi's app comes from his users who grumble, "唉呀……"

More to the point, though, just listen to Wang Yi, himself. He claims that Liulishuo improves English pronunciation. Really? Let's listen to a sample:

Tell me, native English speakers, does his "pronounciation" impress you? Do you think, "Wow! He looks Chinese, but he sounds like he's from Houston!" Or, does it sound like he's using Chinese phonemes to approximate American English ones? (Answer: The latter.)

What about his grammar? Not perfect, but forgivable. He's still not mastered English's grammatical number agreements. Chinese expresses grammatical number very differently. It takes appropriate input, feedback, and experimentation to learn it. However, none of Liulishuo's exercises specifically correct it. Worse still, since "feedback" equals grading in this frog's well, he'll never get the feedback that he needs.

We can note, though, that he speaks fluidly and has a strong vocabulary. Of course, he also got his PhD from Princeton. Last I checked, Princeton's doctoral candidates write their theses in English. And, surprise, surprise! His computer science thesis was not on computational or corpus linguistics.

I will say this to his credit: There's real artistry in his con artistry. Of course, any research into his life and work shows that he's not qualified to develop a language-learning app. He's pretty fluent in English, and that's it. Big fucking whoop! I know Mexican farmers more fluent than this guy. The difference is that they can't say (and omit) things like:
  1. "I graduated from Princeton (in an unrelated field)."
  2. "I worked for Google (for 22 months)."
  3. "We have the biggest archive of (badly accented) Chinese speakers of English."
That pales in comparison to this fact:

Wang Yi Hocks His ESL Wares In China

Not Singapore, not Hong Kong, China. Why? He'd probably say it's because he's from the Mainland. But, what's more likely is that those other places check credentials. China and Taiwan are rife with bullshitters like Wang Yi. Wang Yi ironically complains about this, himself. People there blow cash on expensive, ineffective instruction. Wang Yi's response? Get them to blow cash on less expensive, ineffective instruction.

The fact is that East Asia's ESL industry is decades behind. That's reflected in virtually every for-profit ESL company there. In China and Taiwan, alone, VIPKid, Bright Scholar, Wall Street English, 51Talk, TutorABC, SayABC, Alo7, HESS, and many, many more make insultingly bad materials. Also, they treat their employees like garbage. Often, it's hard to say which reeks worse — their disgusting materials, their corporate arrogance, or their false idol worship. All of their founders, like Wang Yi, just wanted to make a quick yuan. They see that they can sell the false promise of a key competence to a naive public. All of their founders, like Wang Yi, know fuck-all about multilingualism. I know this because I've talked with many of them. I've read the bios (in English and Mandarin) of the rest. Half of them aren't even bilingual. Wang Yi is no different. His bullshit is just a different shade of brown. He pretends to have some "AI" secret to "automating" ESL instruction. He doesn't. He just knows that those words make businessmen's dicks hard these days. If East Asians become even partly aware of sensible ESL pedagogy, Wang and the rest will all be bankrupt in a year.

When you want your bad ESL ideas mashed together.
There's a reason why China's main exports are American products, raw materials, and knockoff goods. Neither China's culture nor government prize original thinking. Our Western innovation steers the majority of their economy. When products of that innovation reach them, they reverse-engineer them. Then, they build knockoffs and pawn them on people who don't know or don't care if they're fake. They treat education the same way. Wang definitely did. He saw some popular ideas, tossed them together, and called it "innovation". He then insulted our intelligence with a crap app, a clipped bio, and some big talk. To quote Dennis Miller, "To call him a scumbag would be an insult to bags of scum."

I personally can't wait for the hype to die and his business to fail. He'll have deserved nothing less. Hopefully, exposing him as a fraud will help push the inevitable.


Language Without Metalanguage 4:
Visualization Over Analysis

Where "Picturing" And "Thinking" Part

When I work with students, the hardest thing to teach them is to think less. It's to teach them not to stop and not to check themselves. Output, wrong or not, needs to come out. For an easy analogy, I often quote William Forrester:

"The first key to writing is to write, not to think." -- William Forrester
This is exactly how speaking is. Fluent speech, just like fluid writing, is automatic. Sure, we pause our speech for various reasons. We don't think ever about where we should pause, though. We don't ask ourselves how fluent our speech is. We speak first. We revise our speech afterwards.

But, why do students struggle with this? Two major reasons come to mind.

The First Barrier: Perfectionism

Students are often afraid of being wrong. They want to do everything right the first time. It's part of our psychology. To be wrong is to admit deficiency. It's a sign of weakness, and we dislike vulnerability. We're conditioned to be proud of rightness and ashamed of wrongness.

Unfortunately, foreign languages are too complex. You won't speak them perfectly on your first try. Your brain has to rightly coordinate phonological, semantic, and syntactic information. Meanwhile, you don't fully know what right or wrong speech is. You just know that you don't know. Worse yet, you have to risk being foolish to remove all doubt.

This is one likely reason why toddlers progress in natural languages faster. They aren't conditioned to feel shame in being wrong. Also, they're virtually deaf to direct grammatical correction:
"The evidence from the experimental language acquisition literature is very clear: Parents, despite their best intentions, do not, for the most part, correct ungrammatical utterances by their children."
However, that barrier is more more easily overcome. It just takes some humility and some thick skin. Drop your pride and take criticism lightly, and you'll be fine. Besides, the second barrier is much more troublesome.

The Second Barrier: Analytical Bias

Starting your first year in school, you're coached in analysis. You're taught to examine facts. You're taught to dissect complexity. You're taught to compose reasoned thoughts. You're taught to cite references. You're taught to fit the institutional ideal. From stickers on worksheets to high GRE scores, people mainly judge your intelligence on this skill. Yet, it's just one domain of mental activity.

Such emphases on analysis impart a cognitive bias, as well. In psychology, it's called "the law of the hammer". Analytical hammer in hand, we'll try to pound every foreign subject. What's more, natural language looks just like a nail for it. It's bound by rules. It has guidelines. Experts can judge it as right or wrong. We can pinpoint errors. But, what good, deep analysis of language acquisition shows is that this is misguided.

Analysis of language is like analysis of music. Sure, elements of language are its terms, its phrasal organization, its agreement rules, and such. Elements of music are its notes, tempos, and such, too. But, just like music is not the application of music theory, neither is language the application of a language theory. Theory comes later to explain what arises naturally. We can hum tunes and speak just fine without theory. The idea that, like learning a new genre of music, learning a new language requires this theoretical knowledge, is just plain false. The facts are in. It does more harm than good. Even its advocates only support "judicious" and "developmentally ready" uses of it.

If analysis doesn't help, then what does?

What helps language acquisition is a method that is informed by sound theory. What helps more is to remember that this doesn't imply teaching the theory to you. You don't need to become a biochemist for antibiotics to work. Likewise, you don't need to become a linguist to learn foreign languages.

To answer you more directly, I'll analyze analysis for you! What are some key features to analysis? What features are its opposites? Answering these questions, we find something anti-analytic, like so:
  • Analysis is taxonomic (about members and sets).
    Therefore, a sound method must be meronomic (about parts and wholes).
  • Analysis is computational (about derivations from rules).
    Therefore, a sound method must be creative (about creations without set rules).
  • Analysis is deliberative (about organizing concepts).
    Therefore, a sound method must be autonomic (about raw observation).
There is sound research in SLA and in "literacy therapies" (for ASD and dyslexia sufferers) that yields such a method. It's often called "visualization". The easiest way to summarize visualization is with two words — "directed imagination". The headword, "imagination", takes its etymological meaning. Imagination "makes an image" for us. All normal humans have this capacity. We can dream vividly. We can picture hypothetical scenarios. We can even recall memories. If we can see, these images are mainly visual, and our sound method will exploit this fact. Second is being "directed". Visualization isn't just random flashes of nonsense. Our images create meaningful, sequential scenes. Each of us gets a front-row seat in our own Cartesian theater. Again, our sound method will exploit this part of ourselves.

Go to Dan Dennett for the analysis.
One more thing we must consider is this: Visualization is pre-linguistic. That is, before we ever had words, we had the ability to direct our imaginations. Babies and dogs dream. If they had no such abilities, basic recognition (of mothers or masters) would be impossible. The corollary?  Materials in a sound method have to be pre-linguistic. A learning session demands structure. That structure, though, can't force a specific form of language. It also can't encourage parroting already heard language. What causes both of those is the presence of linguistic input. Learners are too tempted to copy or paraphrase what they immediately hear or read. We're too tempted to accelerate past or skip the visualization and go straight to language. Remember, though, we want to create language (express thoughts individually) beyond just producing it (forming sentences). That, in turn, demands focus on our pre-linguistic state.

Such materials exist, but are not in language textbooks. Instead, they're in…

Dialogue-Free Media

This Buni Comic is a good example:

A comic is worth 5,000 words.
Now, the method I use takes this story piece by piece. I ask students questions about the image. My questions' order is based on orders in logic, and they proceed like this:

  • Zero-Order Questions (Characters and Objects)
    • "What do you see?"
    • "Who is there?"
    • "What is that?"
  • First-Order Questions (Actions and States)
    • "What is that first rover doing?"
    • "Can you describe the parachute?"
    • "What does the second rover look like?"
    • "How the first rover feel about 'her'?"
  • Second-Order Questions (Action in Setting)
    • "Where do these rovers meet?"
    • "When did the second rover land?"
    • "How many times does 'he' drive around 'her'?"
  • Third-Order Questions (Transitions)
    • "How did the second rover land on Mars?"
    • "Why are there hearts around 'him'?"
    • "What is 'he' driving around 'her' for?"
  • Fourth-Order Questions (Opinion and Conjecture)
    • "What do you recommend that the rover do to win 'her' love?"
    • "How would things have unfolded if the second rover were 'male'?"
In most sessions, first-order questions lead to first- and second-order responses, second-order questions lead to second- and third-order responses, etc. Either way, I just follow their level. I just need to be sure that my questions can be answered. Fourth-order questions allow for more liberty in the responses. There, coherence matters more than truth. Along the way, I give complete, native sentences reflecting what they said. Part by part, the learner describes the whole.

I then remove the images. They must visualize their summary. They're not reciting it. They're creating it. They're saying what they're confident they can say. When they're stuck, I ask a question to help them recall their images. Then, at the very end, they get a transcript of a corrected summary. That's their input. The learners don't need vocabulary drills. They don't need a grammar lecture. They know what they said. They just then see how to say things more clearly.

Finally, above all else, I remind myself:

Watch Those Eyebrows!

Thinking hard.
Hardly thinking.
Analysts have obvious tells. They furrow their brows and focus their gazes.

Visualizers have opposite tells. They raise their eyebrows and look askance.

Of course, some people are just mean-looking, so a baseline is important. Once you find it, though, you must switch that habit. It's frustrating at first. Some learners don't feel like they're learning unless they're deep in thought. Others assume it makes no difference, and so do what's habitual. That's where William Forrester's talk with Jamal is again relevant:
William: Is there a problem?
Jamal: No, I'm just thinking.
William: No, no, no. No thinking. That comes later.
Thinking comes when you can teach yourself. You just need to save your analysis for the end. If you're not sure if you're ready, try The Pink Panther in your target language. 100% confidence in 95% of your output, that's your goal. Anything less, and you'll need a native to guide you.