How computers translate human language - Ioannis Papachimonas

譯者: Ivy Wang 審譯者: Gentian Pan

How is it that so many intergalactic species in movies and TV

為何電影、電視中星際間的不同物種

just happen to speak perfect English?

恰巧能講一口流利的英語？

The short answer is that no one wants to watch a starship crew

答案是：沒人想看太空船員在影片中

spend years compiling an alien dictionary.

花費數年來編撰外星人字典。

But to keep things consistent,

但為保持一致性，

the creators of Star Trek and other science-fiction worlds

「星際迷航」和其他科幻小說創作者

have introduced the concept of a universal translator,

引進「萬能翻譯器」的概念：

a portable device that can instantly translate between any languages.

一種攜帶式裝置，可即時翻譯任何語言。

So is a universal translator possible in real life?

那麼，「萬能翻譯器」可能存在於現實嗎？

We already have many programs that claim to do just that,

已有很多個程式宣稱做得到：

taking a word, sentence, or entire book in one language

從一種語言中選取單字、句子，或整本書，

and translating it into almost any other,

幾乎可以將它們翻譯成任何語言，

whether it's modern English or Ancient Sanskrit.

不論是現代英語，或是古梵語。

And if translation were just a matter of looking up words in a dictionary,

如果翻譯只是在詞典中查找單字，

these programs would run circles around humans.

那麼，這些程式早就普及了。

The reality, however, is a bit more complicated.

然而，現實複雜許多。

A rule-based translation program uses a lexical database,

基於「規則」的翻譯程式使用字彙資料庫，

which includes all the words you'd find in a dictionary

包含字典找到的單字、

and all grammatical forms they can take,

套用的文法型式、

and set of rules to recognize the basic linguistic elements in the input language.

以及「辨認基本語言元素」的規則。

For a seemingly simple sentence like, "The children eat the muffins,"

這個看似簡單的句子為例：「孩子們吃松餅」，

the program first parses its syntax, or grammatical structure,

程式首先分析「語法」或「文法結構」，

by identifying the children as the subject,

辨識出「孩子們」為主詞，

and the rest of the sentence as the predicate

剩下的句子為「述語」－ 由動詞「吃」構成。

consisting of a verb "eat,"

and a direct object "the muffins."

和直接受詞 「松餅」。

It then needs to recognize English morphology,

程式需要辨識出「英語構詞學」，

or how the language can be broken down into its smallest meaningful units,

也就是將該語言拆分成 有意義的最小單元，

such as the word muffin

例如單字 「松餅」

and the suffix "s," used to indicate plural.

及字尾加「s」表示複數型。

Finally, it needs to understand the semantics,

最後，程式還需要理解「語意」－ 各別部份所表達的意思。

what the different parts of the sentence actually mean.

To translate this sentence properly,

為了正確翻譯句子，

the program would refer to a different set of vocabulary and rules

程式會參考不同語言的字彙與規則

for each element of the target language.

來處理目標語言的每個元素。

But this is where it gets tricky.

這卻是棘手的地方。

The syntax of some languages allows words to be arranged in any order,

某些語言允許單字以任何順序排列，

while in others, doing so could make the muffin eat the child.

但在其它語言，這樣做會出現 「松餅吃孩子們」的句子。

Morphology can also pose a problem.

「構詞學」也有同樣問題。

Slovene distinguishes between two children and three or more

「斯拉維尼亞語」可區分是 兩個、三個、或更多孩子－

using a dual suffix absent in many other languages,

「雙字尾」的用法未見於其它語言中。

while Russian's lack of definite articles might leave you wondering

而 俄語 則缺少「定冠詞」，你可能會困惑

whether the children are eating some particular muffins,

孩子們是在吃某種特定的松餅，

or just eat muffins in general.

還是泛指一般松餅。

Finally, even when the semantics are technically correct,

最後，即使「語意」技術上正確，

the program might miss their finer points,

程式也可能遺失細微部分，

such as whether the children "mangiano" the muffins,

例如，孩子們是在「吃」松餅，

or "divorano" them.

還是在「吞」松餅？

Another method is statistical machine translation,

另一種方法是基於「統計」的機器翻譯，

which analyzes a database of books, articles, and documents

該方法分析「已翻譯的書籍、文章、文件」 所建立的資料庫。

that have already been translated by humans.

By finding matches between source and translated text

從「原文」與「譯文」之間， 尋找非偶然的匹配模式，

that are unlikely to occur by chance,

the program can identify corresponding phrases and patterns,

程式就可以辨識出對應的片語和句型，

and use them for future translations.

以便使用在未來的翻譯上。

However, the quality of this type of translation

然而，這種翻譯的品質

depends on the size of the initial database

決定於資料庫的大小

and the availability of samples for certain languages

以及能否應用於特定語言或 寫作風格的翻譯上。

or styles of writing.

The difficulty that computers have with the exceptions, irregularities

電腦的困難：會遇到異常、非常規情況、

and shades of meaning that seem to come instinctively to humans

以及無法呈現人類「直覺本能」可以了解的意函－

has led some researchers to believe that our understanding of language

這些令研究者相信「語言的理解能力」

is a unique product of our biological brain structure.

是我們大腦生理結構的獨特產物。

In fact, one of the most famous fictional universal translators,

實際上，小說中最著名的萬能翻譯器之一，

the Babel fish from "The Hitchhiker's Guide to the Galaxy",

出自《星際大奇航》的 「寶貝魚」，

is not a machine at all but a small creature

根本就不是機器，而是小生物－

that translates the brain waves and nerve signals of sentient species

是一隻能透過心靈感應，翻譯腦波和 神經信號的 「有感知」的生物 。

through a form of telepathy.

目前傳統的語言學習

For now, learning a language the old fashioned way

will still give you better results than any currently available computer program.

仍然會優於利用電腦程式的翻譯。

But this is no easy task,

但這不是簡單的任務，

and the sheer number of languages in the world,

世界上語言的數量，

as well as the increasing interaction between the people who speak them,

和人與人之間逐漸增加的語言互動，

will only continue to spur greater advances in automatic translation.

都會繼續激發「自動翻譯」的進步。

Perhaps by the time we encounter intergalactic life forms,

也許，遇到星際間的其他生物時，

we'll be able to communicate with them through a tiny gizmo,

我們已經能夠透過小裝置來溝通，

or we might have to start compiling that dictionary, after all.

也或許最終，我們還是得著手編寫那部字典。