Typing Chinese like English | Hacker News Reader

Typing Chinese like English

tedclancy.wordpress.com

95 pointsbobajeff73 comments

As a native Chinese speaker and one of the first generation born and grew up with personal computers (born in early 80s), I'd say a machine-learning powered handwriting input method is more convenient. No hard feelings. It's just too hard to reinventing the wheel on this issue where tens of thousands of engineers had worked on this problem throughout the 90s. I remember when I went to college in Beijing in late 90s there were literally 100s of companies were selling Chinese input method softwares and hardwares or both. What I've witnessed along the past two decades is that all converged into a few methods based on pronunciation (variants of Pinyin or voice dictation) or based on shapes (variants of Wubi or handwriting). And I can attest that Pinyin is predominant since it's a mandatory course in schools. There are even newer machine learning powered hybrid input methods emerging.[1]

Having said that, if you don't know Chinese neither any of its input methods, just want input Chinese characters casually, this may be useful but if you do want dive in to learn Chinese from ground up, I am not sure if learning this Wuhou input method is time well spent.

[1]: http://www.xunfei.cn/

> if you don't know Chinese neither any of its input methods, just want input Chinese characters casually, this may be useful

As a casual student of Chinese, pinyin is actually pretty good for me (I'm American). I don't need any new markings on my keyboard, and it helps force me to learn pronunciations anyway, which is good. :)

Also: I saw people using the handwriting-based input in Hong Kong and it seemed like they were typing slower than I do, which is bananas.

> Also: I saw people using the handwriting-based input in Hong Kong and it seemed like they were typing slower than I do, which is bananas.

Speaking as another American who types in Chinese, predictive input doesn't work at all for me when I type in English. It works really well when I type in Chinese, because I'm just not able to produce intricate, idiosyncratic Chinese sentences. It doesn't surprise me that native speakers might take more time to type in their own language; their range of expression is much greater.

Agreed. I type Chinese every day and quite frequently. Pinyin is plenty fast enough for typing. It also seems to be the main input method for Chinese excluding Taiwan/HK.

Pinyin is great, but completely useless when you see a character you can't pronounce, e.g. when using a dictionary. Handwriting input can also be challenging for complete beginners who don't know how to recognize strokes, etc. I can see something like this potentially having some value in those situations. Dictionaries, ...

I'm sure that he – as someone who spent years developing his own input method as a hobby – is acutely aware of all of this.

Still, as a side project, I would consider this a success: he's done his research and he came up with a really cool, well-thought-out tool. It's not going to take the world (or China) by storm, but it doesn't need to.

You're right, of course, about new learners better investing their time elsewhere.

Even if this won't replace current input methods, I think this is clever and a surprisingly intuitive way to type Chinese characters.

I suggest you looking at Wubi input method[1] which is widely used in China for more than 20 years. My first impression of your method is just the same as Wubi.

The reason why Pinyin input method is more popular, I guess, is that Wubi is too hard to learn which need remember a lot of rules.

But quoted from Wikipedia,

> it is true that Wubi is extremely fast when used by an experienced typist.

So Wubi is still very popular among those who need to type a lot of characters.

[1] https://en.wikipedia.org/wiki/Wubi_method

Wait, what's the difference between "wubi" and "wubihua"?

wubi is based on components of a character, which means typing is super fast at the cost of memorizing all the components. wubihua is based on 4 strokes, first 4 and the last one. It's much simpler to use yet not very fast

I spent a few weeks trying to learn Wubizixing about 10 years ago. I found learning the primary 26 component-to-key mappings very easy via staged practise similar to how I learnt touch typing English, i.e. introducing one new component at a time. (The only minor hiccup was because 木 and 目 both have the same sound and tone in Mandarin, I had to remember that completely visually.)

But when I tried learning the secondary components, I began with the digits 一二三四五六七八九十, I just couldn't remember the keyboard mappings because my mind was already mapping them to the top ten keys in sequence, whereas in Wubizixing they're scattered all over the keyboard based on the first stroke in the character. I gave up on learning Wubizixing after that, and instead looked designing at my own input system.

The most difficult design decision with any such input system is which component-to-key mappings to use. Wuhou uses 92 mappings, Wubizixing uses 26 by overloading many more of the keys. Someone learning such a system must be able to easily learn those mappings.

> My current employer doesn’t let me work on Open Source projects or personal projects without prior permission

That's ... weird. I hope they're paying well for this.

I had the same thought. Do you know if there's a list anywhere of valley / well known companies with this type of policy?

It's the type of thing that'd turn me off for sure during a job search.

The stance on personal after-work projects, as well as whether or not you're to work in an open-space office, are two things I'd really love every job listing mentioned. It would be the primary filter for me.

It's more typical for them to claim ownership of it, rather than to forbid it. Any job where you have to submit a list of prior work with your employment contract (which is most of them in my experience) is usually a sign that there's a clause in there claiming ownership of everything you do while employed there. Personally I doubt it's very enforceable, especially since California is so worker-friendly on everything else I know about, but I don't know how well these clauses work out in practice.

> especially since California is so worker-friendly on everything else

Are there no tech jobs outside California?

He said "valley"

It's illegal under California law, so no "valley" (I assume you mean Silicon Valley?) company can outlaw moonlighting.

California categorically prohibits this so all such provisions are null and void in CA.

Anything you develop during non-working hours, not at your employer's office, and on your own equipment is yours and employers are not legally allowed to have any claim on that work.

And it's really not something I'd have expected from Mozilla.

The previous post suggests that the mention of employer wasn't referring to Mozilla: https://tedclancy.wordpress.com/2015/11/04/rebranding/

Ah, the about page is outdated. Thanks.

(preamble: I'm not a lawyer and this is not legal advice.)

Here (in the UK) it's often in boilerplate employment contracts that any projects worked at any time during employment[0] belong to the company.

I tend to go through and cross it out when I find it but I'm also fairly sure it's one of those "it's in there but it's not legally enforceable" things that are often found in employment contracts so I don't know how well tested it is in the courts and also I don't think a company would damage their reputation enforcing it unless someone was working on a project directly relating to or competing with a product the company made.

It's understandable that an employer wouldn't want someone developing an open source competitor to their own products. The 'social contract' is that the employer should invest in their human capital so that in return the employee gains valuable experience and learning by providing training courses, conferences and the like. How much that happens of course varies greatly in the wild.

[0] During as in the duration of your life where you're employed, not "working hours".

It's not understandable that a company should have the right to limit what an employee does after-hours on his own hardware with his own IP.

If you spend your nights rebuilding your company's software as an open source offering, you don't think that would contravene ethics and your obligations to your employer?

Some might argue, "But existing IP and trade secrets protection already covers this".

In cases such as this it would be difficult and expensive to pursue a criminal case around IP and trade secrets, so it's cheaper and easier to have a contract clause which approximately restates the same thing but as part of the contract, with some stronger statements to provide added protection.

There are plenty of clauses in contracts which might seem odd because they're effectively restating parts of criminal law in the contract, but moving things from criminal law to civil law moves the burden of proof from "Beyond reasonable doubt" to "On the balance of probabilities".

Employment contracts are for the benefit of both parties, but ultimately it is the employers who draft them and they will try to get away with as much as possible.

No, it would only be a violation if you reused IP (which I don't see how you could avoid in your setup). I am not a slave to my employer, and my employer does not own me. I consider this an issue of fundamental human rights. If you do not pay for the work that I do, you cannot be entitled to it.

> My current employer doesn’t let me work on Open Source projects or personal projects without prior permission, so let me be clear that this is something I developed prior to my current employment.

Err wait. How does this work?

Usually, it's a clause in the employment contract saying that the company claims ownership of anything you create while under their employment, under the theory that a salary covers more than just the time you're in the office. Maybe combine that with a company policy against releasing internally-produced source code, or something?

My employer has something similar to that contract clause, but I've never heard of them actually invoking it.

There is already wubi typing method in China, which was widely used by my parents' generation. This is essentially reinventing wubi method.

It's way easier than Wubi. At least, that's the intention.

Hey, I don't know who you are, but thanks for posting about my input method here! Most of my referrals are coming from here.

Edit: Hacker News is now complaining that I'm commenting "too fast" and won't let me leave new comments. Feel free to leave comments on my Wordpress blog.

Someone upthread said they couldn't figure out how to type the 金 radical. It's just the 金 key. 钱 is just "金戋". Couldn't be easier. Make sure you have simplified mode (简体) selected at the top of the page since that's a simplified character.

Not sure if it is a new system or an implementation of an existing one. There are already a few existing ones [0] (no English translation). They are great for people who can't speak the official Chinese language properly, which is more than 80% of the population in China (or > 90% world wide), to whom pronunciation based input systems are impossible to grasp.

[0] https://zh.wikipedia.org/wiki/字形输入法

The efficiency of such input methods depend much on one's proficiency, and also the compression ratio of the encoding. After some practice, these methods beat pronunciation base methods easily, as the codes are shorter and with less degeneracy.

On the other hand, all the cell phones have the basic Wubihua [1] that uses only five keys to encode the order of how you would write a character.

[1] https://en.wikipedia.org/wiki/Wubihua_method

As you indirectly suggest, it really depends who the user is.

Especially when you say these methods beat pronunciation-based methods, I think it's important to qualify that as something that depends on the specific user.

For a touch typist and an only semi-literate (emphasis on semi) non-native speaker, the ability to go straight to touch typing of pinyin with absolutely no learning curve is a HUGE advantage over other systems. By contrast, this system does not leverage my existing typing ability at all, so its ultimate speed is gated by the fact that I will never bother to use and learn it. For users like me, the same is true of other such systems.

The point is just that theoretical speed comparisons are moot for some people if practical matters make one system more useful.

Speaking of moot, it seems to me that all our points may be moot as voice input is going to be an ever growing proportion of how text is entered. Especially in Chinese languages.

shameless plug

And for those who can speak Chinese, i.e. know the grammar but are limited to pinyin and can't type in Chinese characters there's this handy tool that combines pinyin input with a dictionary (and frequency table) and makes it easy to write in Chinese. Basically just type the word in pinyin and then choose the right chinese character from the suggested word list. The English definition of the word helps to choose the correct Chinese character. There is support for traditional and simplified.

https://github.com/ensisoft/pinyin-translator

In fact popular IMEs like Sogou do this and also get updated to recognize names of famous people, movies, etc.

You can even type abbreviations for common phrases, like "cflm" and it'll figure out you meant chi-fan-le-ma 吃飯了嗎

Oh, that's interesting. Back in the day when I wrote that tool all the IME tools I found and tried that were able to understand pinyin expected you know the correct character but lacked support for any kind of verbal cue (definition, dictionary) about which is the correct character with the intended meaning.

Current input methods are based on large corpora and use context to disambiguate. They are surprisingly smart.

I'm a non-native/hobbyist Chinese learner, and I'm liking it so far. I wonder if this could end up suiting the purposes of people in my niche, even if it can't supersede other methods among native speakers? Personally, I strongly dislike pinyin because using roman letters engages the intrusive "thinking in English" part of my brain - something I (and I think, a lot of other learners) try and avoid when engaging a new language.

I've spent a decent amount of time trying wubi, and this seems to be easier. Wubi tries to limit itself to 26 keys, so you get a lot of non-intuitive grouping of seemingly unrelated strokes onto the same key (Or at least, going back to my first point, non-intuitive to non-native speakers. Can't speak to how the groupings are perceived by people who learn to read/write via the standard pedagogy.).

Small things that would make this better for me:

* The fact there's more radicals behind the shift key is really not obvious. You might want to add a note/tooltip pointing that out?

* A "reverse lookup" going from character to key compositions. Right now I'm struggling to figure out how to type 钱 because I can't figure out how to write the 金 radical.

* Is there just the one, canonical way of writing a character? Or can you "compose" them up from any "correct" series of strokes? If it's the latter, how hard would it be to make it the latter?

> Personally, I strongly dislike pinyin because using roman letters engages the intrusive "thinking in English" part of my brain

If this really bothers you, use zhuyin, or double pinyin.

> Is there just the one, canonical way of writing a character? Or can you "compose" them up from any "correct" series of strokes?

There is always one canonical way (the "stroke order") of writing a character. I find it a little surprising that a hobbyist learner wouldn't already know this? How are you studying?

My introductory text for writing characters was https://www.amazon.com/Learning-Chinese-Characters-Ms-Zhang/... , and it worked well enough that I can generally input unfamiliar characters in bihua.

On your phone, pleco (which you should have if you're studying Chinese) has a $5 stroke order addon, and skritter (meh) is all about practicing stroke orders.

> If this really bothers you, use zhuyin, or double pinyin.

Double pinyin is still pinyin. Zhuyin works, but I suppose, on reflection, I also want the benefit of reinforcing my knowledge of how to write a character by hand.

> There is always one canonical way (the "stroke order") of writing a character. I find it a little surprising that a hobbyist learner wouldn't already know this? How are you studying?

I'm well aware; I was talking about in this input system. Ie, a given radical can be further decomposed into smaller radicals or individual strokes. If I input "stroke 1" + "stroke 2" instead of "radical comprising stroke 1 and stroke 2" is it meant to still work? (I discovered several situations where that wasn't the case, so another way of stating my question is: Is this intentional or a bug?).

> Double pinyin is still pinyin. Zhuyin works

Double pinyin is zhuyin on a keyboard.

>I also want the benefit of reinforcing my knowledge of how to write a character by hand.

In that case you should just use handwriting recognition software such as that built into your smartphone.

wubi and pinyin are for two different kinds of people. pinyin is for ones who read the character out in their mind as typing. And wubi is for people who visualize the character. For native speaker, speaking is natural, that's why pinyin is gaining more popularity among native speakers.

Pinyin is a good tool with lasting benefits. You'll get past the early stages faster with it than you will without it.

Things like this make me appreciate Hangul(한글) more and more. It isn't immediately obvious to me how I can enter basic characters like "国". (I still can't figure it out.)

Couldn't figure it out either. I went as far as 玉 with Q-G-E. I would have assumed that the box around was "F" but that didn't work.

because you're using the simplified version.

the traditional version of that character is: 國

type "F" for 囗, and "z" for 戈, if you were using cangjie you'd have to add the extra "口" and "一" but i'm guessing this system proposes that 國 is just 囗 + 戈 because that's good enough to identify what you're going for.

edit: the problem seems to be you need to switch to simplified. "F-q-g-e" got me the character.

but some simplified characters are encoded (like 号). I also can't type words like 回 (which is weirdly not FF)

It's Ff (capital F, lowercase F). Don't know why there's a difference between big and small 口, though.

Ah, so 国 is Fqg

What I enjoy about pinyin input is I can just touch type.

Without remembering how the character is written.

And then when I see the result, I can check it and (assuming I sort-of-know the character well enough to do so) I confirm it and go on. Sometimes confirming is not an extra step; it's simply continuing to type.

Also I can type a long bunch of characters in sequence, and they give each other context so the characters get disambiguated, meaning I have less to check and fewer things to choose or correct.

This new method seems like a good system for someone who doesn't know how to say the words they want to type. Such as some of the people the OP described, so good job on that front! Maybe not so good if you are learning or aiming to learn spoken Chinese, though, because it keeps you away from learning or reinforcing the pronunciations, doesn't build on existing knowledge of the keyboard, and doesn't create a very portable skill as pinyin typing would.

Thank you! I'm working for a computer company in Kaohsiung. There are no Mandarin classes outside my working hours. I'm doing what I can with private tutoring and self-study, but it's going to take decades before I can read basic signs or have a conversation. I thought about making an app like yours, but also got quite intimidated by the scale.

Is there some way to see your character composition database? Lots of words that look like 包 sound like "bao" or "pao", but I haven't yet seen any patterns in the few characters I've actually learned.

It won't take decades. I have not been the most diligent of students in my four years working in Shanghai and it was over s year before I even tried to learn to read a bit. It's really easy to get by in Shanghai with very, very little Chinese too.

You won't be able to get cheap one on one tutoring in Taiwan but you can get pretty cheap tutoring over italki.com or another language exchange website. Putonghua and Guoyu aren't that different anyway.

For learning to read the best way in the long run is to hand write it but it's a massive pain. Skritter is almost as good and takes care of revision for you. It has a great spaced repetition system inbuilt. You should also get Pleco. The free version is excellent, the paid version is cheap if you actually use it. It also has a flash card SRS built in for practicing vocabulary.

There are a number of good simplified Chinese graded readers. I'm not familiar with any for traditional Chinese but if you look through tokenadult's comment history you'll find some recommendations.

Tangent: One of the nice things about OSX is that it has a pretty robust set of input method APIs. I used them to make a transliterating Russian keyboard[1] (though I only scratched the surface) and I think they would work really great for something like this as well. Typing in a browser (in my case, using translit.ru) works alright, but nothing beats native support!

[1]: https://github.com/archagon/cyrillic-transliterator (warning: not very polished!)

Finally! Looking forward to sending this to my parents after it develops a bit more. They've struggled with communication for years (decades?) now because they can't type using phonetics.

That's exactly my goal! I'm very much interested in supporting people who speak minority dialects. I don't feel that technology should be pushing Mandarin on people.

Have they tried voice dictation?

I really want to like this. Maybe what the developer can do is to build a simple app that translates a Chinese character into a series of keys. That will ease some of the frustrations on not being able to figure out how to type using this input method. For example, I am trying to type 瓦 but for the life of me I can't figure out how.

edit: removed "shift doesn't work"...does seem to work.

Ok, I figured it out...瓦 = qRre

I would be interested to read more about the history of this idea and how it compares to existing input methods.

Ok, trying to write 国家栋樑. Unfortunately I only got the first two characters..国 is Fqge, 家 is ei(, but haven't figured out how to do the last two. Any idea how to get the last two?

Quick suggestion: when shift key is not down, should show lower case letters. When shift key is down, show upper case letters.

Think I'll stick with the pinyin input method.

I spent about 20min trying to figure out how to write 白 and still cannot.

It's 丿日.

Did you figure out that you have to press Shift to see some of the radicals? Apparently that's not obvious. My bad.

A basic lookup table would be very nice to get started with this input method. I am currently trying to find out how to type "武后".

It's very similar to how you would look up characters in a paper dictionary and as such most dictionary apps iknow of have a similar feature.

Start at the top left and go counter clockwise and look for compounds.

Try typing 一戈 to get 武

I like this a lot. I didn't even read how it works, and was able to input some Chinese characters I know. Incredibly intuitive.

I actually had the opposite impression. Tried to write my Chinese name but couldn't figure out which pieces to put together. There seem to be a lot of shorthand radicals but not enough basic strokes.

It is really awesome as a side project. Super curious how he build the big (19K) lookupMap, anybody knows?

Nowadays voice input in pure Chinese is already fast enough for daily usages.

Is it actually a new system or is it an implementation of an existing one?

New system.

How do I write "日月"? I can only write "明"

space after each character.

so 日 <space> 月 <space>, or c <space> v <space>.

Reply on news.ycombinator.com