Data processing: speech signal processing – linguistics – language – Linguistics – Translation machine
Reexamination Certificate
1999-08-10
2001-09-11
Edouard, Patrick N. (Department: 2644)
Data processing: speech signal processing, linguistics, language
Linguistics
Translation machine
Reexamination Certificate
active
06289302
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates to a Chinese generation apparatus for machine translation, which utilizes statistic data instead of a large number of semantic and syntactic rules.
2. Description of the Related Art
In the 20th-century, people have to keep learning to prevent themselves from getting disconnected from the society. However, for most of the new knowledge are from foreign countries, document translation is important in the efficiency of foreign document reading. In order to improve the quality and efficiency of document translation, a recent trend is to use computer instead of human for a translation job. Such translation apparatus is commonly called a machine translation apparatus. In such a machine translation apparatus, the input language that is to be translated is known as the source language, while the output language that has been translated from the input language is known as the object language. For example, the source language of a Japanese-to-Chinese machine translation apparatus is Japanese, while the object language of the same is Chinese. Furthermore, the translation format used in the machine translation apparatus may be the direct form, the intermediate converting form or the pivot form, depending on the characteristic of the language to be translated. Generally, the intermediate converting form is the one which is commonly used.
Referring to
FIG. 8
, a conventional machine translation apparatus employing the intermediate converting form includes a source language parsing unit
1
, an intermediate structure converting unit
2
, an object language generating unit
3
and a dictionary unit
4
. However, the quality of machine translation depends on whether the input sentence is correctly parsed in the source language parsing unit
1
, whether the difference between the source language and the object language is eliminated in the intermediate structure converting unit
2
(e.g. solve the difference in syntax or meaning, or the selection of the lexicon item translation), and whether an object language is correctly generated in the object language generating unit
3
in accordance with the syntactic rule of the object language.
However, a Chinese sentence will have different meanings as the locations of the lexicons in the sentence are changed. For example, in the sentences, [
] (He is jumping on a table.) and [
] (He jumps onto a table.), since the location of “
” (on a table) in the former sentence differs from that in the latter sentence, the two sentences have different meanings. Therefore, the arrangement of some lexicons in a Chinese sentence has a given sequence, and unless it is so, an incorrect Chinese sentence may be generated. The following is an example, wherein the time lexicon must be placed before the location lexicon. (correct Chinese sentence)
. (Literally: He*yesdaty**at school*ate) (He ate dinner at school yesterday.) (incorrect Chinese sentence)
. (Literally: He*at school*yesdaty*ate)
On the other hand, the sequence of some lexicons in Chinese sentence is unrestrained. The following is an example, wherein the time lexicon may be placed before or after the subject. (the time lexicon is placed before the subject)
. (Yesterday he went to school.) (the time lexicon is placed after the subject)
. (He went to school yesterday.)
Therefore, if the object language of a machine translation apparatus is Chinese, the most important problem to be solved is how to correctly determine the arrangement sequence of the lexicons in a Chinese sentence. Referring to
FIG. 9
, R.O.C. Pat. Publication No. 324804 discloses a Chinese generation apparatus for machine translation.
A preprocessing unit
200
of the Chinese generation apparatus in
FIG. 9
recovers the subject node for the sub-structure which omits the subject in the Chinese sentence dependency structure that is input as shown in FIG.
10
A and that is an intermediate structure with the use of dummy node. Next, a basic item spreading unit
300
generates a basic sentence structure including basic item as shown in
FIG. 10B
in accordance with the basic sentence pattern stored in the basic sentence pattern memory unit
350
with the use of the verb classification code of the main item (verb or adjective) of each of the sub-structure as the searching key.
An unrestrained item spreading unit
400
retrieves the surface case marker of the phrase head, the surface case marker of the phrase tail and the sentence item slot in accordance with the sentence item information memory unit
450
with the use of the case marker of each unrestrained item, the surface case marker of the source language, the semantic dominating code and the semantic code of itself in the dependency structure as the searching key, and generates the sentence structure of each unrestrained item of
FIG. 10C
in accordance with the corresponding location of the sentence item slot location in the sentence structure.
A special sentence pattern generation unit
500
generates the special sentence pattern sentence structure of
FIG. 10D
in accordance with the special sentence pattern attribute of each verb or adjective. As shown in
FIG. 10E
, an item location adjusting unit
600
orderly retrieves the item arrangement sequence limitation in each sentence item slot from the sentence formation item sequence memory unit
650
and adjusts the item arrangement sequence in each sentence item slot in the sentence structure. Afterwards, a post processing unit
700
performs the generation of other accessory item and punctuation on the sentence structure and lines the sentence structure. An output unit
800
outputs the translation result “
” (I put the book in the car.). A buffer unit
900
is used for temporarily storing the output from the basic item spreading unit
300
, the unrestrained item spreading unit
400
and the item location adjusting unit
600
.
The drawbacks that are associated with the aforementioned conventional Chinese generation apparatus for machine translation are as follows:
1. The verb or adjective of Chinese language has a plurality of Chinese basic sentence patterns that are possibly generated, for example, the verb “
” may be generated into the basic sentence patterns as follows. (Wherein S represents subject, V represents verb, O represents direct object or indirect object, and C represents complement)
SVOO:
. (I gave a book to him.)
SVOOC:
. (I gave him a book as a souvenir.)
SVOC:
. (I saw him home.)
SVO:
. (He will get killed.)
Therefore, the conventional Chinese generation apparatus for machine translation cannot solve the problem of the difference in the basic sentence patterns by the verb classification code. This problem must be solved by heuristic method and thus, the translation quality cannot be ensured.
2. The translation quality cannot be improved since the location of the unrestrained item is assigned in accordance with the content of the sentence item information memory unit but not in accordance with the associated item state. For example, if the location of the time lexicon “
” (today) is assigned to 2, the conventional Chinese generation apparatus only can generate the sentence “
.”(I graduated today.), but cannot generate the sentence “
” (Today I graduated.) that emphases “
” (today).
3. Since the adjustment of the relative location among the unrestrained item in the same slot is related with the content of the sentence item sequence memory unit, a strange or incorrect Chinese sentence may be generated when the content of the sentence item sequence memory unit is incomplete.
SUMMARY OF THE INVENTION
Therefore, the main object of the present invention is to provide a Chinese generation apparatus for machine translation which can overcome the drawbacks that are associated with the aforementioned prior art.
According to the present invention, a Chinese generation apparatus for machine translation which utilizes statistic information instead of a large number of semantic, grammatical and phrase structural rules and which converts
Edouard Patrick N.
Greenblum & Bernstein P.L.C.
Matsushita Electric - Industrial Co., Ltd.
LandOfFree
Chinese generation apparatus for machine translation to... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Chinese generation apparatus for machine translation to..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Chinese generation apparatus for machine translation to... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2538746