• Front Page
   
  
• April ‘02 Issue
   
  
• January ‘02 Issue
   
  
• October ‘01 Issue
   
  
• July ‘01 Issue
   
  
• April ‘01 Issue
   
  
• January ‘01 Issue
   
  
• October ‘00 Issue
   
  
• July ‘00 Issue
   
  
• April ‘00 Issue
   
  
• January ‘00 Issue
   
  
• October ‘99 Issue
   
  
• July ‘99 Issue
   
  
• April ‘99 Issue
   
  
• January ‘99 Issue
   
  
• October ‘98 Issue
   
  
• July ‘98 Issue
   
  
• April ‘98 Issue
   
  
• January ‘98 Issue
   
  
• October ‘97 Issue
   
  
• July ‘97 Issue
   
  

English to Japanese—

to What Extent Can Translation Be Accurate?

Introduction

t has always been Man's dream to mechanize the process of translation. Indeed, equipped with the modern technology of today, all we need to do for a word-for-word translation in Microsoft Word is the following: Click on a word, point to Language on the Tools menu, click Dictionary, type a word in the Lookup box and click on Translate. However, the above-mentioned feature is only available in a limited set of languages.

The purpose of this paper is to demonstrate the fact that there are inherent problems in translating from one language to another because of different grammatical resources of the two languages. In other words, a human translator may be armed with some grammatical knowledge of the target language, but may not be able to translate the meaning accurately from one language to another even if s/he were provided with a dictionary. Quite frequently, the nuances of meaning in the sentences get lost as a result of translation from one language to another. Thus, this study discusses the simulation of a computer by a translator who is "pre-programmed" with standard Japanese grammatical rules and structures, notwithstanding the fact that she does not possess the proficiency to deal with exceptional cases or idiomatic phrases. In addition, it discusses the problems that arise from translating English to Japanese and vice versa, often caused by the restrictions of metaphor or particles that exist in one language but not in the other.

Methodology

Five English sentences were selected and translated by the author, who has a limited knowledge of Japanese sentence construction (she has attained Level Four of the Japanese Language Proficiency Test). These sentences were translated by her with the help of a dictionary only. The author then passed on the five English sentences to two native speakers of Japanese, both of whom were teachers of a reputable Japanese school in Singapore. Both were graduates of English, and each was assigned a task: Teacher A was supposed to translate the sentence from English to Japanese, whereas Teacher B was to back-translate the sentence from Japanese to English. This was done so that the semantic change, including any addition or loss of meaning, could be observed and recorded effectively. The results and discussion section comprise two analyses for each sentence: Analysis A consists of the comparison of the two translations, whereas analysis B discusses the interesting effects of the Japanese to English back-translation.

Bearing in mind that the Japanese sentence structure is Subject Object Verb (SOV) order and that the English sentence structure is Subject Verb Object (SVO) order, one can predict that the verb will appear at the end of the sentence after translation from English to Japanese. The English equivalent of the words is indicated under the Japanese words. Both the kanji and romanized versions of the Japanese language are indicated.

Results and Discussion

The children will eat the fish.
 

Native Speaker's Translation:

Analysis A: Comparison of the two translations

In Japanese, there is no definite article equivalent to the article the. Its best alternatives would be kono, sono, and ano. As recommended by the dictionary, these words mean these, those, and those over there respectively. In my translation, the is replaced by kono. Similarly, when performing a translation, the computer, which has limited grammatical knowledge, might first search for the meaning of the in the Japanese lexicon, and upon finding none, replace it with one of the three alternatives. By doing so, the semantic interpretation of the sentence would be drastically changed. For instance, in my translation, I have selected kono to replace the. However, the word kono means these, and the translated sentence therefore reads "These children will eat these fish."

According to my Japanese translator, adding the future morpheme deshyo, the equivalent of will, makes the sentence sound unnatural. The interesting observation is that Japanese does not have this future tense generally, unless it is used for weather forecasts. The Japanese rely purely on context to determine if will eat is to be used, or eat alone is to be used, a phenomenon that does not occur in English. The computer might make the same mistake as the ignorant human translator, i.e., trying to find its equivalent in the dictionary and then substituting it in the correct grammatical form. This results in an imperfect translation—an unnatural-sounding sentence.

Analysis B: Discussion of Back Translation from Japanese to English

Back-Translation: Children eat fish.

In the correct Japanese version, the is omitted in both "The children" and "the fish." Therefore, when the Japanese sentence is translated back into English, the sentence becomes "Children eat fish." Similarly, if the computer were to perform the task by merely omitting the and not replacing it with kono, sono, or ano, the meaning of the sentence changes again. (a) "Children eat fish" is a generic statement which means that children can eat fish, without specific reference to any particular children or fish, (b) "These children will eat these fish" means that the speaker was referring to the children and the fish that were both near the speaker, during the time of the speech. Thus, the identification of the children and the fish is definite, and the presence of the children and the fish is specified. (c) "The children will eat the fish" however, means that the speaker knows which children and fish he is referring to, but the children and the fish need not be present during the time of the speech. Thus, although the identification of the noun phrase is definite, the presence of the noun phrase is unspecified. Clearly, the original sentence (c), translated, would have the meaning of either (a) or (b) but its original meaning could not be retained. Hence, this is a problem inherent in translation itself in this particular language combination, not only in machine translation.

It is important to realize that a one-to-one correspondence in translation is indeed hard to achieve. As shown in the examples above, one either replaces the with kono, sono or ano, or one omits the completely. Either way, it is difficult to obtain a perfect translation for "The children will eat the fish." A similar problem probably arises in machine translation, where the computer might either replace the with the three alternatives or omit the completely.

Sentence Two:

Send the professor a letter from your new school.

Analysis A: Comparison of the two translations

In my translation, the preposition from precedes the noun phrase your new school, whereas in the correct translation by native speakers, the same preposition follows the noun phrase "new school." This translation problem arises from the fact that in English, the preposition from always precedes the noun phrase, whereas in Japanese, the preposition from (kara) always follows the noun phrase. Similarly, although the computer may be able to change all English words into Japanese words, and analyse the sentence in a default SOCV structure, it may not be able to deal with more complex problems pertaining to the position of prepositional particles before or after a noun phrase.

It is interesting to observe that once the correct translated sentence is in the imperative form, the need for the possessive pronoun your disappears. This is due to the fact that the imperative form (indicated by nasai attached to the Japanese verb for send) assumes that the new school belongs to the addressee. Thus there is no need for anatano, which means your. However, the computer may regard it as a necessity to translate the sentence, word-for-word, so that every single English word is replaced by its Japanese equivalent (or near-equivalent, which was discussed earlier in the case of kono, sono, and ano). Note that the correct translated version employs morphemes such as "nasai," which renders the use of "your" unnecessary, according to native speakers of Japanese.

Analysis B: Discussion of Back Translation from Japanese to English

In English, the above sentence is actually ambiguous, for it can have two semantic interpretations:

1. Send a letter from your new school to the professor (i.e, "have a letter written by your new school and send it to the professor").
2. Send from the new school a letter to the professor (i.e., "write a letter and send it from your new school to the professor").

It might not have been possible to translate the sentence while retaining its ambiguity, because interestingly, the native Japanese translator chose to do so such that the translated Japanese sentence has only one of the two interpretations, i.e., the meaning of (II). When asked to do the translation for the other meaning, the native translator found it quite impossible. It therefore follows that when the other Japanese translator translated the sole Japanese sentence back to English, the results were such that the meaning was that of (II), and not (I).

This problem of syntactic ambiguity is not a serious problem in human life if contextual cues are present. For instance, suppose a professor wants my new school to inform him of my academic achievements, my friend's utterance of this sentence Send the professor a letter from your new school would automatically assume the meaning (I), i.e., "Send a letter from your new school to the professor." But if I had nothing better to do at my new school and I complained to my friend about my boredom, her utterance of the same bolded and italicized sentence would automatically assume the meaning of (II), i.e., "Send from your new school a letter to the professor."

The crux of the problem in machine translation, however, is that contextual cues are not always present. The disadvantage that computers have would be the lack of knowledge of the world, which human beings, in contrast, can automatically access.

Sentence Three:

The fish will be eaten by the children.

My Translation

Native Speaker's Translation

Analysis A: Comparison of the Two Translations

This time, it is essential to include the future tense morpheme deshyo, unlike the translations involved in Sentence (1). The reason is that the original sentence is in the future passive form, If deshyo were to be omitted, and only the passive rareru were to be used, then the translated sentence will mean The fish is eaten by the children, rather than The fish will be eaten by the children. Hence, the use of deshyo is governed by the presence of a passive. Such a rule, though taken for granted by the Japanese language, does not exist in the English language. Unless the computer is highly advanced, the inappropriate employment of the morpheme deshyo would pose yet another serious problem to MT systems.

Armed with a limited knowledge of Japanese grammar, the author is inclined to translate the Japanese sentence in a basic SOV (Subject-Object-Verb) order. However, she was not aware of the order of future and passive morphemes and had therefore arranged it such that the Japanese future morpheme preceded the passive morpheme, which was in fact, the classic English verb order. However, the correct translation is the other way around—the Japanese passive morpheme rareru has to precede the future morpheme deshyo. In addition, when followed by a Japanese passive morpheme rareru, the future tense morpheme deshyo can coexist with it without resulting in an awkward sentence structure.

After observing the effects of human translation, it is clear that MT systems should not merely be equipped with the ability to translate based on word-order alone, for this will result in imperfect translations. The systems must be empowered so that they will include morpho-syntactical rules in their programs, such as whether the particle indicative of future tense of a language follows that of the passive tense in a verbal phrase, as illustrated by Sentence (3) above.

Analysis B: Discussion of Back Translation from Japanese to English

I think those fish will be eaten by those children.

Apart from its function as a future tense morpheme, deshyo has the added connotation I think. This expresses one of the cultural differences between the Japanese and English language; the Japanese language has the additional connotation of "hesitation" expressed by some of its morphemes. Whether or not this hesitation will be expressed by machine translation is yet another problem faced by computers.

Sentence Four:

Who is the person that is hugging the dog?

My Translation

Native Speaker's Translation

 
Analysis A: Comparison of the Two Translations

One problem faced by the computer (or a translator who has minimal knowledge of the English Language) would involve the participle verb hugging. We realize "working" = "work" + "ing" and "kissing" = "kiss" + "ing." Interestingly, however, "hugging" is neither "hug" + "ging" nor "hugg" + "ing." Hence, not all words have regular morphology. The primary difficulty exists if each root word in the lexicon is separated from its prefixes and suffixes. However, if sophisticated morphology algorithms were to be applied (Goshawke et.al., 1987), then the computer would be able to perform the splitting of "hugging" into "hug" + "ing." This enables the splitting to occur systematically, resulting in the Japanese word "hug"—daikishime and the Japanese continuous form teiru. A related problem, however, is that the computer may not be able to merge the two forms morpho-syntactically. It may be similar to a human translator with limited knowledge, in that it merges the full root form of the Japanese "hug"—daikishime and its continuous teiru. However, such a merger is not correct because "hugging" in Japanese is simply dai-teiru, i.e., kishime is omitted. The portion dai alone from the full root is sufficient to merge with teiru to form dai-teiru.

Another interesting observation that could be made is that in the correct translation, there is an absence of the usual SOV order. Instead, the order is OVS, with a few grammatical manipulations. The dog (object) precedes hugging (the verb) which precedes who (the subject). The computer therefore faces a problem because Japanese sentences need not always follow the default SOCV structure, and hence, it has to search through several structures (which must already be pre-programmed) in order to get the grammar correct.

Analysis B: Discussion of Back Translation from Japanese to English

Who is hugging that dog?

The above question also demands a specific answer, as "Who is the person that is hugging that dog?" presupposes that the hugging process was engaged by only one human being and the speaker was at the scene. Instead of expecting a banal answer such as "An old woman," the speaker clearly expects more information such as "Jane, the kennel owner."

However, when translated into Japanese, the question becomes Who is hugging that dog? Now, this question has a slightly different meaning inasmuch as it does not presuppose that the speaker already knows there is only one person hugging that dog. The answer to this latter sentence could very well be "A group of children," which could not be the answer to the original, untranslated question "Who is the person that is hugging that dog?" Note that it is the opinion of the native Japanese translator that including the person (English to Japanese translation) is redundant and inappropriate.

Sentence Five:

The spirit is willing, but the flesh is weak.

My Translation

  Native Speaker's Translation

Analysis A: Comparison of the Two Translations

There is one problem that the computer might face in translating the above well-known sentence. (a) In English, the word spirit already has various lexemic properties. It could mean part of a person's mind, a certain force and even alcohol. Carried to the extreme, the computer may, for example, randomly pick alcohol to mean spirit and meat to mean flesh. The translation may even read: The alcohol is willing, but the meat is weak. To go one step further, one may cite Sampson (1992): the translated Russian version is The vodka is good but the meat is rotten.

Another problem that the computer faces is that even if a perfectly translated sentence were to carry the semantic equivalent of each lexical item, the translated sentence would not make pragmatic sense. The root of the problem lies with the fact that the above is a fixed idiomatic expression. This means that the sentence makes communicative sense only if its structure and its words are fixed in English. As this is a fixed expression, one cannot change the structure of the idiom to, for instance, "Willing is the spirit, but weak is the flesh." In addition, one cannot attain its meanings simply by replacing the lexical items word-for-word.

Analysis B: Discussion of Back Translation from Japanese to English

The mind is mighty, but the meat body lacks strength.

The back translation from Japanese to English informs us of how the translated Japanese sentence would appear to the Japanese. As observed, the idiomatic/metaphorical meaning of the sentence is lost. The original meaning of the sentence is "One may be determined to achieve something, but one's poor health may forbid him to do so." Instead, the meaning has changed to "The mind is mighty but the meat body lacks strength." To them, this sounds unnatural and makes as much sense as "Colourful green ideas sleep furiously" by Chomsky. This syntax is perfect but there is no way one can make coherent sense of the sentence.

The problem of translation could be alleviated, however, if the idiom is stored in a separate lexicon from ordinary open-choice words. In this manner, the computer could perhaps translate the meaning of the idiom into the target language (in this case, Japanese).

An Insight into the Japanese Government Project

An additional problem that stems from machine translation between these two languages with vast linguistic differences would be that English is fundamentally a "do"-language, while Japanese is essentially a "be"-language. Consider, for instance, a well-known phrase such as "I love you." In English, although "love" may be classified as a mental state, it confers upon a meaning that has connotations of the active participant being "I" and the passive participant being "you." The object of love, namely "you," is the entity upon which the mental action "love" is showered. In contrast, the Japanese language decrees that one expresses one's love by saying "ai-shitteiru". The root verb "ai" means love, whereas "shitteiru" is a general suffix that can be attached to many other verbs, and simply means in the process of. (For instance, "benkyo-shitteiru" means in the process of studying whereby "benkyo" is the root verb for study). The pronouns "I" and "you" are simply redundant in the Japanese language. The Japanese do not have to say "watashi (I) wa anata (you) ga ai-shitteiru (love)" to mean "I love you" simply because adding pronouns such as I and you contributes to redundancy and is stylistically inappropriate in Japanese. Yet, in English, a one-to-one translation of this phrase from Japanese to English means "in the process of loving," and this certainly does not sound romantic at all when said by one native speaker of English to another. This perhaps illustrates why English is understood as a DO-language, whereas Japanese is known as a BE-language. Nonetheless, transformations of this kind are incorporated into the generation grammar, to produce more natural English expressions. However, the stylistics transformation part of the process is still limited. For a detailed study, refer to Slocum (1985) on Machine Translation Systems.

The GRADE system is a hybrid of the probabilistic approach and the rules-based approach. The annotations required are based on heuristic probability, while there are certain rules governing the sequence of words as they are parsed. For instance, its transfer grammar has a set of default rules that translate English prepositions into Japanese postpositional case particles. However, the default rules are often violated, as English often requires more specific prepositions. For example, in Sentence (2), atarashi gakko kara kyuju ni tegami o dashinasai is back translated into "Send the letter from your new school to your professor." This is a case where ni can be translated into to. However, ni can also mean the following prepositions: at, on, in, to, for, into, onto, and by. This poses a problem to the GRADE system, which the researchers overcame by coding the Transfer Dictionary for nouns (Slocum, 1985). However, one can easily tell that a huge amount of programming needs to be involved for such coding, and more money, time and effort would have to be invested for the system to make this extra step, which human translators take for granted.

Conclusion

If it is not too much to hope for, the computer can perhaps reach an ultimate point where it can translate the meaning of a sentence such as "The spirit is willing but the flesh is weak" so that it matches the meaning of any idiom stored in the lexicon of the target language. The end result would be a Japanese idiom, which does not necessarily have to contain the words spirit, willing, flesh, and meat. Instead, the translated sentence would have to be as close in meaning to the English idiom as possible. Such is perhaps one of the best forms of translation that we could hope to achieve for MT, one of the goals we can perhaps set for ourselves in the new millennium of Corpus Linguistics.

Biguenet, J., and Schulte, R. (1989). The Craft of Translation. Chicago: The University of Chicago Press.

Frawley, W. (1984). Translation: Literary, Linguistic and Philosophical Perspectives. Newmark: University of Delaware Press.

Gorlee, D. L. (1994). Semiotics and the Problem of Translation. The Netherlands: Amsterdam-Atlanta.

Goshawke, W., Kelly, I. D. K., Wigg, J.D. (1987). Computer Translation of Natural Language. United Kingdom: Sigma Press.

Sampson, G. (1992). Computing in Linguistics and Phonetics: Machine Translation. London: Academic Press.

Slocum, J. (1985). Machine Translation Systems: The Japanese Government Project. London: Cambridge University Press.

Wilss, W. (1982). The Science of Translation: Problems and Methods. Germany: Gunter Narr Verlag Tubingen.

Yoshida, M. and Nakamura, Y. (1996). Kondasha's Furigana English-Japanese Dictionary. Tokyo: Kondasha International Limited.

Yoshida, M. and Nakamura, Y. (1996). Kondasha's Furigana Japanese- English Dictionary. Tokyo: Kondasha International Limited.