張佑竹
2024-04-02
Taiwanese Tone Group
台語聲調群
Taiwanese has a very important structure which is the tone group. it is unique among natural languages in the world. Other languages, including Chinese, do not have this kind of tone group structure. Although some people say that Taiwanese dialect evolved from Chinese. Some people say that Taiwanese dialect is "southern-Min". But in 1852, what we now call Taiwanese, this term "Taiwanese" appeared in Qing Dynasty documents. The term "Taiwan-wei (台灣話) first appeared in literature in 1874. The so-called "southern-Min" language was only spoken in the late 1930s. In fact, people living in the southern Fujian region of China do not only speak Taiwanese as we speak now.
Of course Taiwanese comes from Zhangzhou (漳州) or Quanzhou (泉州) region. However, as time goes by, Taiwanese and the language spoken in Taiwan have undergone considerable changes. What's more, Taiwanese was influenced by the Spain, the Netherlands, the Pingpu people, and even the English and the Japanese. Taiwanese has become a language with Taiwanese local characteristics. This is the change of language itself. We are not just talking about the increase of vocabulary. Because various natural languages in the world are more or less influenced by foreign languages. It is a very common phenomenon that foreign languages are transformed into local languages. But in modern Chinese, only the third tone has tone sandhi, and no other tones have tone sandhi. Even the major local dialects such as Cantonese and Suzhou dialect in China. Although there are some cases of tone sandhi, it is not as common and regular as Taiwanese. Taiwanese also preserve the ancient pronunciation of many Chinese characters.
In the Taiwanese pronunciation system, there are two ways of expressing pronunciation. One is classical Chinese pronunciation. In the past, there were private Chinese schools (私塾) in Taiwan teaching people to read Chinese. When reading Chinese, there are many Tang poems and Song lyrics that can be read using Taiwanese Wenyin (文音). Not only does it rhyme, but its pronunciation is very similar to ancient Chinese.
For example, Zhang Ji's (張繼) "Night Mooring at Maple Bridge (楓橋夜泊)" is read aloud in Taiwanese pronunciation.
The moon is setting, crows are crowing, and the sky is full of frost.
Jiangfeng and fishing fire is facing melancholy.
Hanshan Temple outside Gusu city.
The midnight bell is ringing for the passenger ship.
Others, such as Tao Yuanming's (陶淵明) "The Peach Colony (桃花源記)", can also be read using Taiwanese Wenyin.
The other is Baiyin (白音) . Baiyin is the pronunciation we use in daily speech. There are many sounds in this Baiyin that should have existed in southern Fujian a long time ago.
That is the local pronunciation that existed before the Chinese pronunciation was introduced to Fujian. Observing the existing Taiwanese words, we can find that many words in Baiyin are different from Beijing dialect. For example, Taiwanese says "chicken mother" and Chinese says "mother chicken". There are hundreds, even nearly a thousand, of these inverted words in Taiwanese. There are many expressions in Taiwanese that are completely different from Chinese. For example, in Taiwanese we have a saying called "Wei Zeng Wei (未曾未)" The usage of "Wei Zeng Wei" is really clever. We don't have to point out the reason before we say, "Wei Zeng Wei, She fell down." We can just say that "Wei Zeng Wei, she was arrested by the police.
There are very few such expressions in Chinese. We can see many other differences in between.
The modern Chinese that is what we say "madalin" in English.
Obviously its tone sandhi structure is completely different from Taiwanese. We can see the changes in Chinese tones from Zhongyuan phonology(中原音韻) and Zhongzhou phonology(中州音韻). This change makes the original rhyme of Chinese characters, which is the pronunciation, almost completely disappear. Modern Chinese cannot establish a structure with tone groups like Taiwanese. In English or Western languages, in order to build a basic corpus, it takes a lot of time to do research and determine the basic units of phrase. A lot of manpower must be used to do different studies. But it is not necessary for Taiwanese to worry about it at all. Taiwanese’s tone sandhi can establish the basic grammatical structure, the tone group. This grammatical structure is also a prosodic structure. It is what is called prosodic features in English. And the tone group also represents a very complete meaning. It is what is called a semantic unit in English. This tone group structure can be said to be innate. That is to say, Taiwanese can distinguish the basic units in a very natural way. In other words, all Taiwanese sentence must be composed of tone groups. No matter how long a sentence is, it can be divided into tone groups. There is even a long sentence in Taiwanese, which is a tone group. Tone group has a definition. That is, all the syllables of the previous words are pronounced sandhi tone, only that last word of a sentence is pronounced lexical tone. So as long as the words that read lexical tone are cut apart, it is a tone group.
The definition of Taiwanese tone group is very clear and there is no need to do phrase Parser or do syntax analysis. We just need to identify the lexical tone and sandhi tone in a sentence. Then it will be easy to establish tone group boundaries. In the same way, think about it from another aspect. If we can build a corpus of tone groups to determine the boundaries of the tone group, it will be clear whether each Taiwanese word should be pronounced lexical tone or sandhi tone. Currently, the biggest difficulty for producing Taiwanese text-to-speech and speech recognition software is that there is no enough corpus of Romanized Taiwanese documents and voice recordings. This is why Google and Microsoft wanted to do Taiwanese text-to-speech and speech recognition a few years ago, but it’s never been completed. Another reason is that the most popular deep learning technology is insensitive to the processing of Taiwanese tone sandhi. It is relatively difficult. So our most important task now is to develop a Taiwanese OCR software. So that we can convert the printed Romanized Taiwanese text accumulated over the past one or two hundred years into digital files that can be used by computers.
In addition, Taiwanese people should be encouraged to learn the reading and writing skills of Taiwanese to establish a modern corpus of Romanized Taiwanese. In this way, we can apply AI technology to improve Taiwan's technological, economic and cultural standards.
I often think that there must be a reason for Taiwanese, a language without writing system, to be spread for thousands of years. Taiwanese's tone sandhi and tone group structure can be said to be unique among the more than , natural languages in the world. Every Taiwanese sentence is a combination of tone groups. I remember Dr. Wang Yude (王育德) once said that Taiwanese tone group is the most worthy of our careful study.
How is the structure of Taiwanese tone sandhi and tone groups established? So far, it has not been able to obtain a satisfactory explanation through Western linguistic theories.
After all, it is because of the structure of Taiwanese tone groups that Taiwanese words have tone sandhi. Or is it because the Taiwanese word itself has tone sandhi that causes the structure of Taiwanese tone groups? There is no empirical answer to these two questions so far. Is it because of the structure of the tone group that the word itself needs tone sandhi to meet the definition of the tone group seems more reasonable. But why do our human minds use the operation of language, especially the structure of Taiwanese tone groups to express grammar, prosody and semantics still an unsolved mystery.
At present, we must first find out the words with fixed pronunciation of sandhi tone and fixed pronunciation of lexical tone. Then we can analyze the reasons that cause these words to use a fixed tone no matter in any context. Although the deep learning method can produce more natural and high-quality speech output, it does not help to solve the questions raised above. In other words, the deep learning method currently cannot display the tone form of words. There is no way to explain why some of these words must be pronounced sandhi tone. There are also some words that need to be pronounced lexical tone. Why is the same word pronounced sandhi tone in one sentence and lexical tone in another sentence? I believe Taiwanese’s research may be one of the keys to opening the door to artificial intelligence.
Taiwanese is a very classical and elegant language. I remember several years ago, I saw a farmer in the countryside of Taiwan. He dig out a piece of soil from the rice field and explain to the reporter that some rice plant was small millet and some was flowering in this piece of soil. Some beginning to form ears of rice, others are fully mature. He said it is all because of fertilizer. Judging from his age, this farmer may not have a high diploma of education. But listening to the Taiwanese he speaks, it's really elegant. Is it possible that today's young people can use small millet, flowering, formed into ears of rice, fully mature such a beautiful Taiwanese words to introduce the process of growing rice?
I often say that Taiwanese is a language of poetry and music. Because the third, seventh and first tones of the Taiwanese tone are just like the "Do", "Me" and "So" of the pentatonic scale. Taiwanese speak Taiwanese as if they are singing. For example, Cheng Hao's(程顥)的"Oucheng(偶成)" in the Song Dynasty is pronounced like this with Taiwanese Wenyin.
It is almost noon, the clouds are clear and the wind is gentle.
The flowers and willows are passing by the front river.
At that time, people did not know how happy I was.
And they would be said that I was taking leisure time to be young people.
Language is not only a kind of culture, it is also the carrier of culture. The death of a language is the disappearance of a culture. The decline of Taiwanese is not only a reflection of the fate of Taiwanese people, but also a loss of human culture.
I hope that one day more people will learn Taiwanese. There are more scholars doing research on Taiwanese. More writers and writers use Taiwanese to write their works. You see, Korea, Vietnam, and Indonesia all have their own languages. Although our population in Taiwan is not as large as theirs, Taiwanese people have very chic and ambitious personalities. I hope everyone will come to speak Taiwanese and learn to read and write Taiwanese. We should not be afraid that one day Taiwanese will disappear. As long as we start learning Romanized Taiwanese, Taiwan will definitely become a country with diverse cultures and languages.
台語有一個很重要的結構就是聲調群。在全世界的自然語言當中可說是獨一無二的。其他的語言，包括中文都沒有這種構造。雖然有人說，台灣話是從中文演化而成。台灣話就是閩南語。但是1852年，我們現在所說的台語這個名詞就在清朝的文獻裡出現過。台灣話這個語詞，最早在文獻出現的時間是1874年。所謂的閩南語大約是在1930年代的後期才有人使用。其實住在中國閩南地區的人所說的話，也不是只有我們現在說的台灣話。
台灣話當然是從漳州或泉州這些地區傳過來，但是經過多年，台灣話和原鄉所說的語言已經有相當的變化。更何況受到西班牙，荷蘭，平埔族，甚至英文，日文的影響，已經變成一種具有台灣本土特色的語言。這就是語言本身的變化，我們說的不僅是詞彙的增加。因為世界上各種自然語言或多或少都有受到外來語的影響而轉化成本地的語言，這是很普遍的現象。但是現代的中文，只有第三聲有變調，其他聲調都沒有變調。即使中國的廣東話，蘇州話這些主要的地方方言，雖然有部分變調的情況，卻不像台語有這麼普遍而規律性的變調，並且保保留許多漢字的古音。
台灣話的語音系統裡面，有兩種語音的表示方法。一種就是文言音。以前在台灣有漢語書塾在教人讀漢文。讀漢文的時候有很多像是唐詩宋詞之類都可以用文音去讀。不僅有押韻而且和古漢語的讀音很類似。比如說張繼的楓橋夜泊，用文音來朗讀就是這樣;
月落烏啼霜滿天，江楓漁火對愁眠，姑蘇城外寒山寺，夜半鐘聲到客船。
其他像是晉朝陶淵明的桃花源記，這類文言文，也可以用文音來讀。另外一種就是口音。口音就是我們日常說話用的語音。這種口音裡面有很多應該是很久以前在閩南地區就已經存在的語音。那是漢語的文音傳入福建以前就有的本地語音。從現有的台語語詞觀察，就可發現很多口音都和北京話不同。譬如台灣話說"雞母"中文就說"母雞"。這種倒裝的語詞，在台語裡就有幾百個，甚至近千個。台灣話裡面還有很多用語的表達方式和中文完全不同。譬如說我們台灣話有一句叫做"未曾未"。"未曾未"的用法實在很奇巧。我們說"她未曾未就跌一跤"之前都不用指出什麼原因，就可以直接說她未曾未就怎樣。"未曾未就被警察抓去"。中文就很少有這種說法。其他不同之處還有很多。
我們說現代的中文，英文叫做 madalin。很明顯的它的變調結構就和台灣話完全不同。我們從中原音韻和中州音韻就可以看出中文聲調的變化。這種變化使得漢字原有的入聲韻就是促音，差不多完全消失。現代的中文就無法建立像台語這樣具有聲調群的結構。英文或是西方的語言為了建立基本的語料庫。必須用很多時間去做研究，把phrase的基本單位定出來，一定要用很多人力去做不同的研究。但是我們台灣話根本就不必大費周章。台語的變調可以建立基本的語法結構，就是聲調群。這個語法結構也是一種韻律結構，就是英文所說的prosodic features。而且聲調群也代表一個很完整的意思，就是英文所說的semantic unit。這個聲調群結構可以說是是天生的。就是說台灣話可以用很自然的方式把基本的單位區分出來。換句話說，所有的台灣話一定是由聲調群組合而成。不管多長的一句話都可以被切成聲調群。甚至有很長的一句台灣話，就是一個聲調群。聲調群有一個定義，就是前面所有的語詞音節都讀變調，只有最後那個語詞讀本調。所以只要把讀本調的語詞切開，就是一個聲調群。
聲調群的定義非常的明確，完全不須做phrase的Parser或是去做語法分析。我們只要把一句話裡面的本調和變調變區分出來，就可以很容易地建立聲調群分界。相同的道理，從另外一方面思考，如果我們可以建立聲調群的語料庫，來決定聲調群的頭尾，這樣每一句台語裡面的語詞，到底要念變調還是本調就很清楚了。目前製作台語語音輸出和語音辨識最大的困難，就是欠缺台語的羅馬字文檔和錄音的語料。這也是Google和Microsoft雖然在幾年前就想做台語的語音輸出和辨識，但是一直都無法完成的緣故。另外一個原因是，現在最流行的Deep Learning 技術對處理台語文的變調比較困難。所以我們現在最重要的工作，就是要開發一個台語OCR軟體，將過去一兩百年累積的台語羅馬字書面資料，轉為電腦可以使用的檔案。此外，還要鼓勵台灣人學習台語羅馬字的讀寫能力，來建立台語羅馬字的語料庫。這樣我們才能應用AI的技術來提升台灣的科技，經濟和文化水準。
我常常在想，台灣話這種沒有文字的語言，能夠流傳幾千年一定有它的道理。台語聲調的變化和聲調群的結構，在世界上六千多種語言當中，可以說是獨一無二的。每一句台灣話都是聲調群的組合。我記得王育德博士曾經說過台語聲調群是最值得我們用心研究的所在。
台語的變調和聲調群的結構，到底是如何建立的? 到目前為止，還無法經由西方的語言學理論得到圓滿的解釋。到底是因為台語聲調群的結構，致使台語語詞有變調，還是因為台語語詞本身有變調，才會造成台語聲調群的分界。這兩個問題到目前為止，還沒有一個經過實證的答案。是因為聲調群的結構造成語詞本身需要變調，來滿足聲調群的定義?這種想法看起來好像比較合理，但是為什麼我們人類的頭腦會利用語言的運作，特別是台語聲調群的結構來表現語法，韻律和語義，依然是一個未解的謎團。
目前要先將固定讀變調和固定讀本調的語詞先找出來。然後才能分析是什麼原因，導致這些語詞不論在任何語境，都是用固定的聲調。Deep learning的方法雖然能產生比較自然和高品質的語音輸出，但是對上面所提出的問題並沒有什麼幫助。換句話說Deep learning的方法目前不能將語詞的調型，就是tone form顯示出來， 也沒辦法解釋為什麼這些語詞有些一定要讀變調，還有一些語詞就要讀本調。為什麼同一個語詞在一句話裡面讀變調，而在另一句話又是讀本調。我相信台語的研究，也許是打開人工智能大門的鑰匙。
台灣話是一種很古典又文雅的語言。我記得好幾年前，在台灣鄉下要看到一個正在種田的人從稻田裡挖出一塊禾土，向記者說明這塊禾土裡有的稻株剛在出青，有的正在開花，有的開始結穗，也有的已經飽水了。他說這都是因為肥料的關係。從他的年紀看來，這位種田人可能沒有很高的學歷。
但是聽他在說的台灣話，實在有夠文雅。現在的年輕人是否能用出青，開花，結穗，飽水，這麼漂亮的台灣話來介紹稻米成長的過程?
我常說，台灣話是一種詩歌和音樂的語言。因為台語聲調的第三聲，第七聲和第一聲，剛好類似五聲音階的“Do” “Me” “So”。台灣人說台灣話，就像在唱歌一般。譬如宋朝程顥的"偶成"，用台語文音來讀就是這樣;
雲淡風輕近午天，傍花隨柳過前川，時人不識余心樂，將謂偷閒學少年
語言不僅是一種文化，它也是文化的載體。一種語言的殞落，就是一種文化的消失。台灣話的沒落，不僅是台灣人命運的觀照，也是人類文化的損失。希望有一天，有更多的人來學台灣話。有更多的學者來做台語的研究。
更多的文學家和作家，用台語來書寫他們的作品。你看韓國，越南，印尼都有他們自己的語言。我們台灣的人口，雖然沒有他們那麼多，但是台灣人的個性很瀟灑，又有志氣。希望大家都來講台灣話，都來學習讀寫台語。我們不要怕有一天台語會消失。只要我們開始來學習台語羅馬字，台灣一定會成為一個具有多元文化和語言的國家。
[ 正進行計畫 |
回主畫面 | 已發表論文
| 台語語音筆記本
| 台語作品
]
This website is sponsored
by VIKON Corp.
Copyright July 1996.
All rights reserved.
Last update : 2024-04-02
有什麼建議嗎 ? 來信請寄:
vikontony@gmail.com