鳳

被慈愛的機器關照

I like to think (and	我想像：
the sooner the better!)	（越快越好！）
of a cybernetic meadow	模控草地
where mammals and computers	哺乳類與電腦
live together in mutually	活在互助
programming harmony	程式和諧裡
like pure water	如純水貼近
touching clear sky.	澄藍天際
I like to think	我想像：
(right now, please!)	（現在就要！）
of a cybernetic forest	模控森林
filled with pines and electronics	松樹電路繁茂
where deer stroll peacefully	鹿群悠然
past computers	漫步電腦旁
as if they were flowers	它們像花朵
with spinning blossoms.	旋轉綻放
I like to think	我想像：
(it has to be!)	（必定來到！）
of a cybernetic ecology	模控生態
where we are free of our labors	解放人世辛勞
and joined back to nature,	回歸自然
returned to our mammal	同哺乳類
brothers and sisters,	兄弟姐妹一道
and all watched over	被慈愛的機器
by machines of loving grace.	關照

 

Reclaiming Our Language

(Excerpt from our upcoming journal article on the First Year of g0v.tw.)

---

The resounding success of the December 2012 gathering prompted a successive event. The “1st Hackathon of Homesteading Commons” was held on January 27 at three cities across Taiwan, attracting one hundred on-site hackers and even more online participants.

2012 年 12 月的聚會非常成功，下一場活動於是接踵而至。「第壹次公地放領黑客松」於隔年 1 月 27 日在臺北、臺中、高雄等三個城市舉行，現場吸引了上百位黑客，遠端參加的人更多。

One of the online participants was Ping Yeh, a long-time open source advocate who immigrated to the US in 2011. In a blog article posted the night before the Hackathon, he called upon g0v hackers to liberate the Revised Ministry of Education Dictionary from the Ministry’s archaic website into “an open API, with full index and search capabilities, free for all individuals and companies to use.”

其中一名線上參與者是葉平，他是開放源碼的長期倡導者，在 2011 年移居美國。他在黑客松前夕發布了一篇部落格文章，呼籲零時政府的黑客把重編國語辭典修訂本從教育部的老舊網站系統解放，以便用開放的文字 API 釋放出來，加上索引和搜尋的功能，讓任何想加值的個人或公司都可以使用。

With over 160 thousand entries, rich etymologies, and full references to classical texts, the MoE Dictionary has been the authoritative definition of the Traditional Chinese language since it was first published in 1945. The digital-only Revised edition launched in 1996 as website and attracted nearly 200 million visits—even with almost no upgrades—over the next seventeen years.

教育部國語辭典擁有超過 16 萬筆詞彙，包含豐富的歷史用法，充分引用經典文本作為書證。自 1945 年首次出版以來，它一直是繁體中文的權威定義參考。在 1996 年推出的網頁版重編國語辭典修訂本，即使幾乎未曾升級，卻在接下來 17 年吸引了近 2 億人次使用。

What would it take to bring the Revised MoE into the modern web of permanent URLs, Unicode text, and mobile devices? Yeh argued that the Ministry of Education ought to provide unencumbered dictionary data and let citizens take care of the presentation. Since the Ministry took the opposite stance by explicitly disallowing redistribution, Yeh pledged to personally design the required data structure so civic hackers could scrape the dictionary data to “reclaim our language.”

如何才能把這部辭典發展成擁有固定網址的現代網站、使用 Unicode 字碼，甚至適用於行動裝置？葉平認為，教育部應提供開放的辭典資料，讓民眾自行決定如何呈現這些內容。由於教育部持相反立場，明確禁止重製，葉平決定親自動手設計所需的資料結構，讓公民黑客可以擷取辭典資料以還文於民。

More than 30 participants on the MoeDict team—formed during the Hackathon—worked overnight. The next day, the team published not only complete data in JSON and SQL formats, but also prototyped websites and apps for all major platforms. With the recent memory of Real Price Map’s closure, we forestalled copyright issues with two legal devices:

1. By claiming Fair Use of government-produced data under Article 50 of Taiwan’s Copyright Act, we took a non-infringing position by adhering to strict not-for-profit principles.
2. By disclaiming all property rights using the CC0 public domain declaration, we made sure the Ministry of Education could incorporate all our contributions at any time.

超過 30 人參與的萌典團隊在黑客松期間組成，並且連夜趕工。翌日，該團隊不但以 JSON 和 SQL 格式發表了完整的內容，而且還完成了適用於各大平台的網站和應用程式雛型。由於當時民間版實價登錄地圖關站事件記憶猶新，我們搶先透過兩個法律程序避開版權問題：

1. 根據著作權法第 50 條，我們聲稱對政府生成的資料進行合理使用，並嚴格堅守非營利用途的原則，採取非侵權的立場。
2. 透過 CC0 公眾領域貢獻宣告，我們放棄所有著作財產權，確保教育部可在任何時候納入我們的所有貢獻。

After months of deliberation, the Ministry eventually agreed to our fair use claim and collaborated with us to convert their CC-licensed Taiwanese Holo & Hakka dictionaries into a cross-referenced, multilingual platform at moedict.tw. The website would go on to serve millions of visits per month, with over 100 thousand regular mobile app users by the end of 2013.

經數月審議，教育部終於同意這項合理使用聲明，並與我們合作，把以創用 CC 授權釋出的臺灣閩南語常用詞辭典與客家語常用詞辭典整合進可交叉引用的萌典多語文應用平台。這個網站目前每月提供數百萬人次的服務，至 2013 年年底，也有超過 10 萬用戶將萌典應用程式安裝在自己的行動裝置上。

文章造天下，功業還蒼生。

(This is a Chinese translation of my previous blog post: Our paroqial fermament, one tide on another, written in mixed cn-tw vocabularies.)

（這是我前一篇部落格的中譯，標題原本引 Joyce，實在沒法譯，只好改周德偉的對聯，是當年辦藝立協時，每周在牆上看到的。）

Adina Levin 要我講講中文推特的資訊密度與英文有何不同，我答「且等咱們社算表（原圍紀算表）運行速度上的瑕疵都修好了再說。」

昨天我除完了蟲，這下只好勉力用英語編程了。

我首先想說的是，Ken 當年創發 UTF-8 編碼（見 Rob Pike 的現場實錄），實在已正確反映了英美字母和中日韓字元間的資訊差異。

想想，這通用字符集從 16 位元被迫擴充到了 21 位元，豈不正是「多如繁星，萬碼奔騰」的中文字害的麼？即便硬生生搞個漢同文，滅了無數重複字元，到頭來雙位元組仍是不敷使用。

假設推特的限制是 140 個 UTF-8「位元組」，那我們寫中文推特的感覺，和寫英文或許相去不遠。因為每個中文字占 3 個位元組——有時我用些生僻古字，超出了基本多語面，那就要占 4 個位元組了。

但歷史的意外，讓推特算的是 140 個「字符」。（不是「語素」，這我親身求證過了。）如此一則中文推特，就有 420 個位元組可用，相當於一篇博客短文了。

...且慢，還沒完呢！

要知道，中文有兩種寫法：Vernacular (白話) 和 Literary (文言)。

這白話呢，通常用兩個字符，來代表一個英文詞兒。像「網絡」，就是「Network」的意思。

在文言裡，每個字表示一組「概念」，像「網」，英文的「net」，可以是「網路」、「漁網」、「連網」、「網羅」的意思，全依上下文脈絡決定。

寫文言文時，字與字間既沒有空白，標點句讀也省略不少，這樣就有 140 組「概念」可用，換成英文得用 200 個詞來表示，也就是 1kb 的信息量。

理論如上，介紹完畢，接下來看看我最近的三則推特，小心求證一番。

第一則完全是大白話，第二則半文半白，第三則（最近的一則）則幾乎全是文言。

在白話到文言的過程裡，信息密度應該要遞增，這可以用英文譯本的長度來計算。

第一則是引陳映真（1936 年生，台灣當時影響力最大的作家及運動者之一）最近說過的話：

RT 陳映真：「文學退化，影像、聲音成為『當下世界』的符碼，」他說：「托爾斯泰生在今日，大部頭的作品也會喪失大量讀者。」現在的創作流行輕薄短小，又以自 我為中心，他因此形容這一代青年創作者「是脫光了衣服站在鏡子前面，凝視鏡中自己的身體與慾望… 他們讀的不多，不能成就崇高的文學」。(139 字)

谷歌將它譯成：

RT Chen Ying-chen: "Literature degradation, images, sounds become codes of present world," he said: "Tolstoy was born in today, voluminous works of the loss of a large number of readers will be."

Now the creation of popular thin and light short, Youyi self-centered, so he described this generation of young artists "is stripped of clothes and stand in front of the mirror, staring in the mirror his own body and desires ... they read much, does not develop high literature."

(475 字)

這譯本大致無誤，但小錯比比皆是（更別提對青年創作者身體的性別岐視了）。而最後一句更是整個譯錯：「他們讀得少」竟然譯成「他們讀得多」！

但至少這譯本還能達意，我們不妨除除看：中文與英文字元之比，是 3.4 倍。

接著第二則是我對陳的回應：

Re 陳映真：但背了整本維基，行遍無數國度，亦不成就崇高的文學。歐巴馬「以父之名」成不了作家，只好化藝術為行動、為現實。上一輩在高壓下，被迫濃粹經驗為符碼，而我們讀遍之後，融合歷史視域，還原此在而為 Hacktivism，竊自以為然。 (116 字)

谷歌的譯本這下亂了套，幾乎沒法看懂了。比例是 3.8 倍：

Chen Ying-chen Re: But the back of the whole wiki, line countless times a country, nor the noble achievements of literature. Obama, "Name of the Father" can not become a writer had to arts-based action, into reality. The last generation under high pressure, was forced to experience concentrated Intrade codes, and we read times, the integration of history, depending on the domain, restore this in the for Hacktivism, stolen from that it does.

(444 字)

我的翻譯如下，中英比率是 5.7：

Re Chen Ying-Zhen: But having an entire Wikipedia-backed memory, and having traveled to countless countries, still won't make high literature out of me.

Consider Obama who, having failed to launch a writer's career with "Dreams from my Father", is forced to project his art into activism and into reality-shaping.

My earlier generation, under tremendous Fascist pressure, is forced to distill their subjective experience to highly compressed literary code.

When my generation finished deciphering those codes to achieve a fusion of horizons, we decompressed them into the here-and-now as Hacktivism. This kind of adaption is IMHO natural and quite justified.

(662 字)

緊接著看第三則，近一步解釋前面的看法，大致用文言寫成：

舉實例言，《金盾工程》（即「功夫網」、「資訊長城」），無異吾儕之柏林牆。我十年前譯寫自由網，而至其後 Tor、無界等武裝，無非保持對話、以獨促統之意。與 Beijing.pm 嘗言：「功網未散，何以為族？」。凡此種種，生自文事，衍為武功，皆此代之共業。 (127 字)

谷歌譯出 509 字來，長則長矣，意義盡失（「促進意大利之統一」？？），故不在此引述。茲譯如下：

As a concrete example, for our contemporary people, the "Golden Shield Project" (a.k.a., the "Gong-Fu Web", "Great Firewall of China") is no different from the Berlin Wall.

Ten years ago I translated and coded for the Freenet project; along with its follow-ups such as Tor and Wu-Jie, they're nothing but cyberspace armaments designed with the sole purpose of keeping the conversation flowing.

This way we maintain our independent identity, in the hope of accelerating a fusion of horizons with the Chinese government.

I've told Beijing.pm: "Without the dissipation of the Great Firewall, how can we make one people out of us?"

All these circumstances, born out of the literary world, has derived into the cyberspace hacktivism, and affected us in creating a software-shaped reality. This is a factor in the shared karmic setting of this generation.

(848 字)

正如所料，密度比例達到了 6.67 倍。切記，上述範例並非刻意作成：中文推特寫手，確實活在 2~8 倍於英文寫手的資訊密度當中。

Our paroqial fermament, one tide on another.

(For a Chinese translation of this blog entry, see 中譯本：「文章造天下，功業還蒼生。」)

(Alternate title: "Chinese Twitter users live in a density 2x to 8x their English counterparts; here's why.")

I promised Adina Levin a treatise on the information density of Chinese characters on Twitter "after all SocialCalc (nee wikiCalc) performance bugs are fixed".

As I've fixed them yesterday, let's try coding some English...

I'll begin by saying that Ken's in(ter)vention of UTF-8 (as narrated by Rob Pike) accurately reflects the relative information density between ASCII and CJK characters.

After all, UCS was extended from 16-bits to 21-bits precisely because so damn many Chinese characters need to be encoded, even after the controversial decimation from the Han unification effort.

Hypothetically, if Twitter had set its limit to 140 UTF-8 bytes, then our experience when tweeting Chinese would be on par with tweeting English, because each Chinese character would then take 3 bytes — and since I occasionally venture beyond the BMP, sometimes 4 bytes.

By a historical accident, though, Twitter counts in characters. (Not graphemes, as I've clinically proved.) So that gives a Chinese tweet 420 effective bytes to work with, which is already sufficient for a short blog post.

...but wait, there's more!

You see, there are two modes of Chinese: Vernacular (白話) as well as Literary (文言).

Typically, the vernacular mode takes two characters to encode a English word, e.g. "網路" for "network".

While in Literary mode, one character encodes a concept, e.g. "網" for "net" can stand for "network", "fishnet", "to connect", or "to capture", depending on surrounding context.

So by writing in Literary mode, which conveniently also elides spaces and only uses minimal punctuation, gives me 140 concepts to work with, which is about 200 English words, or nearly 1kb of English text.

With that brief intro, let's use my three most recent tweets for a concrete demonstration.

The first one is entirely in vernacular mode, the second one is in mixed mode, while the third (most recent) one is written almost completely in Literary mode.

We shall see how the information density increases with the shift of modes, as compared with their English translations.

The first one is a quote from Chen Ying-Zhen, b. 1936, one of the greatest Taiwanese author and activist of his generation, who recently uttered:

RT 陳映真：「文學退化，影像、聲音成為『當下世界』的符碼，」他說：「托爾斯泰生在今日，大部頭的作品也會喪失大量讀者。」現在的創作流行輕薄短小，又以自我為中心，他因此形容這一代青年創作者「是脫光了衣服站在鏡子前面，凝視鏡中自己的身體與慾望… 他們讀的不多，不能成就崇高的文學」。(139 characters)

Google Translate renders the above as:

RT Chen Ying-chen: "Literature degradation, images, sounds become codes of present world," he said: "Tolstoy was born in today, voluminous works of the loss of a large number of readers will be."

Now the creation of popular thin and light short, Youyi self-centered, so he described this generation of young artists "is stripped of clothes and stand in front of the mirror, staring in the mirror his own body and desires ... they read much, does not develop high literature."

(475 characters)

The translation is subtly wrong on multiple regards (not to mention having a sexist default). It's especially wrong on the last sentence, where Chen actually said "they don't read much"!

But it's at least comprehensible, and will serve as a good compression-ratio example: the Chinese/English information density ratio here is 3.4x.

Now the second one is from me, a reply to Chen:

Re 陳映真：但背了整本維基，行遍無數國度，亦不成就崇高的文學。歐巴馬「以父之名」成不了作家，只好化藝術為行動、為現實。上一輩在高壓下，被迫濃粹經驗為符碼，而我們讀遍之後，融合歷史視域，還原此在而為 Hacktivism，竊自以為然。 (116 characters)

The Google Translation is much weaker this time, bordering on incomprehensible, with a ratio of 3.8x:

Chen Ying-chen Re: But the back of the whole wiki, line countless times a country, nor the noble achievements of literature. Obama, "Name of the Father" can not become a writer had to arts-based action, into reality. The last generation under high pressure, was forced to experience concentrated Intrade codes, and we read times, the integration of history, depending on the domain, restore this in the for Hacktivism, stolen from that it does.

(444 characters)

My own translation would be something like this, with a 5.7x ratio:

Re Chen Ying-Zhen: But having an entire Wikipedia-backed memory, and having traveled to countless countries, still won't make high literature out of me.

Consider Obama who, having failed to launch a writer's career with "Dreams from my Father", is forced to project his art into activism and into reality-shaping.

My earlier generation, under tremendous Fascist pressure, is forced to distill their subjective experience to highly compressed literary code.

When my generation finished deciphering those codes to achieve a fusion of horizons, we decompressed them into the here-and-now as Hacktivism. This kind of adaption is IMHO natural and quite justified.

(662 characters)

This leads us to the third tweet, a follow-on elaboration composed entirely in Literary mode:

舉實例言，《金盾工程》（即「功夫網」、「資訊長城」），無異吾儕之柏林牆。我十年前譯寫自由網，而至其後 Tor、無界等武裝，無非保持對話、以獨促統之意。與 Beijing.pm 嘗言：「功網未散，何以為族？」。凡此種種，生自文事，衍為武功，皆此代之共業。 (127 characters)

The Google translation at 509 characters is entirely incomprehensible ("promoting unification of Italy"??), so I'll not even bother quoting here. My translation:

As a concrete example, for our contemporary people, the "Golden Shield Project" (a.k.a., the "Gong-Fu Web", "Great Firewall of China") is no different from the Berlin Wall.

Ten years ago I translated and coded for the Freenet project; along with its follow-ups such as Tor and Wu-Jie, they're nothing but cyberspace armaments designed with the sole purpose of keeping the conversation flowing.

This way we maintain our independent identity, in the hope of accelerating a fusion of horizons with the Chinese government.

I've told Beijing.pm: "Without the dissipation of the Great Firewall, how can we make one people out of us?"

All these circumstances, born out of the literary world, has spawned into cyberspace as hacktivism, and affected us in creating a software-shaped reality. This is a factor in the shared karmic setting of this generation.

(848 characters)

Here the ratio is 6.67x. Bear in mind that this is not an edge-case of maximum compression rate; Chinese Twitter users constantly live in a density anywhere from 2x to 8x of their English counterparts.

鳳たんです!

Last month when I visited Tokyo, I was very flattered (and pleasantly surprised) when fellow Japanese hackers -- Miyagawa and Takesako in particular -- referred to me as "ōtori-tan", literally the "phoenix girl", where phoenix is my Chinese chosen name.

The fascinating thing is that this sounds almost exactly the same as the Japanese romanization of "Audrey Tang", my English chosen name. Miyagawa wondered if I knew about this strange coincidence -- I really did not. :-)

Wikipedia has some more information about the -tan suffix, as seen in e.g. the OS-tan phenomenon. After some ego-googling, it seems that this usage started from lolipop's blog, then subsequently propagated to Rocco's and other places.  Lovely!