 |
Description  |
|
|
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates generally to translating machines, and more
specifically, to a translating machine capable of translating a document
including markup signs for computer typesetting from one language into
another language.
2. Description of the Related Art
Conventional translating machines in practical use include the following. A
conventional translating machine inputs source language documents into a
translation module initially using, for example, a keyboard under the
control of a CPU (Central processing Unit). The translation module
analyzes the input source language text utilizing a group of dictionaries
(such as a basic dictionary stored in memory and a user dictionary
prepared by user registration) and then produces a parsing tree from the
analysis. Similarly, the parsing tree of the source language text is
transformed into a parsing tree in a target language utilizing rules for
transforming tree structures from a source language tree to a target
language tree prestored in memory. An appropriate translation is given to
each word, and then necessary additional parts are supplied to produce a
final text in the target language.
In recent years, systems have been widely developed by which block copies
for printing are produced utilizing small size computers. Therefore,
additional information for printing (such as specifications for
typesetting) is sometimes included in a document text. Such information
includes information for designating a title, the font to be used, the
size of the font, and the words to be employed as index entries.
These pieces of information are conventionally mixed into the text of the
document to be processed in the form of markup signs. By including such
markup signs in the document, the document can be automatically printed
utilizing a format, a font, and a font size according to the markup
information. When index entries are designated, the index can be readily
produced by listing those words or groups of words attached to the text
with such markup signs.
Markup languages have been developed as systems of markup signs. One
example of such a language is the SGML (Standard Generalized Markup
Language) established by the ISO (International Standardization
Organization). SGML is used for designating a logical structure for a
document such as chapters, paragraphs, and itemization. When a document
produced in accordance with SGML is actually printed, a markup language is
often used for more specifically deciding a format. One example of such a
markup language is called TeX.
As the number of documents having designations for printing utilizing
markup languages increased the demand for a technique for translating
these documents into another language has also increased.
A document including markup signs as described above cannot be properly
translated in a conventional translating machine. In some cases, the
document cannot be translated at all. Alternatively, a mistranslation
sometimes occurs because the markup signs are different from the source
language included in the document. Conventionally, it was therefore
necessary to manually check whether or not markup signs were included in
an input text utilizing an editor or the like before inputting the text
into a translating machine. Once all the markup signs were deleted one
after another, the text could then be input into the translating machine.
Accordingly, efficiency in translating a document including markup signs
utilizing a conventional translating machine was very slow.
To overcome such disadvantages, a system for processing documents without
consideration of non-language data (such as format information included in
the document) is disclosed in Japanese Patent Laid-Open No. 4-259057.
According to the system disclosed in this document, only language data is
extracted from document data in which language and non-language data are
mixed, and a prescribed editing processing is performed on the extracted
language data. The language data edited by this editing processing is
compared to the language data in the originally input document data for
determining a corresponding relation between their positions. The language
data of the input document data is replaced with the corresponding
language data after the editing. This permits editing of document data in
which the language data is mixed with format information by ignoring the
presence of the non-language data.
However, various rules are necessary for determining the corresponding
relationship between the edited language data and the input document data.
One cannot immediately judge whether such rules are truly effective rules
or not except by trial and error. And yet an effective corresponding rule
is not necessarily present for every case. Employing such a rule
mistakenly could even degrade the quality of an eventually obtained
translated document.
SUMMARY OF THE INVENTION
It is therefore an object of the invention to provide a translating machine
capable of translating a document including markup signs more efficiently
than a conventional apparatus and of producing a translated document with
equivalent markup signs.
Another object of the invention is to provide a translating machine capable
of translating an English language document including markup signs into
another language more efficiently than a conventional apparatus, and of
producing a translated document with equivalent markup signs.
Yet another object of the invention is to provide a translating machine
capable of translating an English language document including markup signs
into a Japanese language document more efficiently than a conventional
apparatus, and of producing a translated document with equivalent markup
signs.
A translating machine according to the present invention translates an
original text in a first language including one or a plurality of
predetermined markup signs into a text in a second language. The
translating machine includes a separation module for separating the
original text into markup signs and a text main body exclusive of the
markup signs, a memory for storing each markup sign in association with a
specific word or phrase, a module for producing a parsing tree of the text
main body without the markup signs into the second language, and a module
for producing a text in the second language with markup signs inserted
appropriately therein.
In the translating machine, the separation module separates an original
text into markup signs and a text main body. The system stores the markup
signs in memory. Each markup sign is associated with a particular work or
phrase. The module for producing a parsing tree performs a prescribed
grammatical processing on the original text to produce a parsing tree in
the second language. The module produces a text in the second language
based on the parsing tree in the second language and the storage contents
of the memory. In the produced text, a markup sign equivalent to each
markup sign in the original is attached to the translation by the
producing module.
In a preferred embodiment, the memory includes a plurality of term memories
for respectively storing a word included in the original text, and a
plurality of markup sign memories for storing markup signs attached to
words stored in corresponding term memories. At least one of the markup
sign memories is provided for each of the plurality of term memories.
Each word and markup sign (attached to the word) are stored in association
with each other on a word-by-word basis. Therefore, any markup signs can
readily be associated with a translation of these words as the text in the
second language is produced from the parsing tree in the second language.
In the preferred embodiment, the first language is English and the second
language is Japanese.
The foregoing and other objects, features, aspects and advantages of the
present invention will become more apparent from the following detailed
description of the present invention when taken in conjunction with the
accompanying drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a representation schematically showing the concept of machine
translation;
FIG. 2 is a representation schematically showing the structure of a
translation machine for performing translation using the method;
FIG. 3 is a block diagram showing a translating machine according to one
embodiment of the invention;
FIG. 4 is a block diagram showing in detail a translation module 5 as shown
in FIG. 3;
FIGS. 5 to 9 are representations schematically showing the storage contents
of buffers A, B, C, D and E, respectively;
FIG. 10 is a flow chart for use in illustration of tag sign processing;
FIG. 11 is a representation schematically showing one example of a tag sign
in SGML language;
FIG. 12 is a representation schematically showing an input original text as
stored in buffer A;
FIG. 13 is a representation schematically showing the storage contents of
buffer F;
FIG. 14 is a representation schematically showing the storage contents of
buffer B;
FIG. 15 is a representation schematically showing the storage contents of
buffer B after a dictionary lookup processing;
FIG. 16 is a representation schematically showing the storage contents of
buffer B after attaching tag sign information;
FIG. 17 is a flow chart for use in illustration of a subroutine program in
production processing; and
FIG. 18 is a representation schematically showing the storage contents of
buffer E after production processing.
DESCRIPTION OF THE PREFERRED EMBODIMENTS
A translating machine according to one preferred embodiment of the
invention will be described in conjunction with accompanying drawings. An
English-Japanese translating machine for translation of documents from the
English language to the Japanese language will be described for purposes
of illustration. The invention is however not limited to this machine but
has applicability to machines for translating between other languages.
Moveover, markup signs used by the markup language TeX (referred to as
"tag sign(s)") are taken for purposes of illustration only. The invention
is not limited to this language and is generally applicable to any
language utilizing markup signs.
Before describing the embodiments, the concept of machine translation will
be summarized. Referring to FIG. 1, an analyzing process performed by
machine translation goes through various analysis levels. In machine
translation, when a source language sentence (as displayed at the upper
left in FIG. 1) is input, processing at various levels are sequentially
performed and a target language sentence (as displayed at the upper right
in FIG. 1) is eventually obtained. After, for example, a source language
sentence is input, the analysis proceeds through selected steps from
levels L1-L10 including a dictionary lookup processing at level L1, a
morphological analysis processing at level L2, then a syntactic analyzing
processing at level L3, . . . , and finally a morphological producing
processing at level L10 is performed, thereby generating a target language
sentence.
The methods for machine translation are roughly divided into the following
two kinds depending upon the level at which the analyzing processing is
performed. One method is the pivot method by which the analysis is
performed up to the intermediate language (so called "interlingua")
displayed in level L6, and a target language sentence is produced
therefrom. The other method is the transfer method by which the analysis
is performed through levels L2-L5 to obtain the internal structure of a
source language sentence, then the obtained internal structure is
converted into the internal structure of a target language at the same
level as the internal structure of the source language. Thereafter, the
target language sentence is produced.
An Interlingua (as used in the pivot method) is a conceptual intermediate
language which does not depend on a source language or a target language.
Accordingly, once the interlingua of a sentence is obtained by a single
analyzing processing of the source language, a plurality of languages can
be produced from this interlingua, which is advantageous for translating
between several languages. According to such a pivot method, however, it
is uncertain if the interlingua which is the essential concept of the
method can really be obtained.
The transfer method is a compromise approach which accommodates possible
inaccuracies in the interlingua associated with the pivot method, and
today, many systems use the transfer method. The following description
concerns the transfer method, and a translating machine according to a
preferred embodiment which employs the transfer method.
Now, the content of each analyzing processing level shown in FIG. 1 will be
described.
(L1 and L2) Dictionary Lookup, Morphological Analysis
In these processing levels, the source language is divided into morpheme
strings (word strings) utilizing, for example, the longest coincidence
method while referring to a dictionary which stores morphemes. Then,
grammatical information such as the part of speech of each of the obtained
words and a translation for each word are provided. The words are analyzed
for determining the tense/person/number, etc. of each word within the
sentence.
(L3) Syntactic Analysis
This processing level involves constructing and determining the structure
(parsing tree) of the sentence based on the relationship between words
using the part of speech variant. In this processing level, a
determination of whether or not the obtained structure of the sentence
represents the correct meaning is not performed.
(L4) Semantic Analysis
This level determines what is correct and not correct in terms of meaning
from among a plurality of parsing trees obtained from the syntactic
analysis processing for adopting the correct meaning.
(L5) Context Analysis
In the context analysis processing level, the topic of the input sentence
is examined to remove any ambiguity and supply any omitted parts included
in the input sentence.
In the translation module of the translating machine according to one
embodiment of the invention which will be described below, it is assumed
that analysis processing is performed as far as level L3. More
specifically, the translation module of the translating machine according
to a preferred embodiment has a structure as shown in FIG. 2. The
translation module as shown in FIG. 2 includes a separation unit 10 for
separating an original source language sentence into tag signs and text, a
storage unit 17 for storing the tag signs associated with the words to
which the signs are attached, a dictionary lookup/morphological analysis
unit 11 for performing a dictionary lookup/morphological analysis
processing on the text, a syntactic analysis unit 12 for performing a
syntactic analysis on the input sentence after it has been morphologically
analyzed, a transformation unit 13 for generating the parsing tree of a
target language by transforming the results of the syntactic analysis, and
a translated sentence producing unit 14 for producing a translated
sentence in the target language (with inserted tag symbols) based on the
parsing tree of the target language generated by transformation unit 13
and referring to the contents of storage unit 17. Processings performed in
units 10-14 will be described in more detail in conjunction with the
following embodiments.
FIG. 3 is a block diagram showing a translating machine according to one
embodiment of the invention. Referring to FIG. 3, the translating machine
includes a main CPU (Central Processing Unit) 1, a bus 7 to which the main
CPU 1 is connected, a main memory 2 connected to bus 7, a display unit 3
formed of a CRT (Cathode Ray Tube) or an LCD (Liquid Crystal Device)
connected to bus 7, a keyboard 4, a translation module 5 connected to bus
7, a memory 6 connected to translation module 5 for storing a knowledge
base such as dictionary/grammatical rules and tree transformation
structure rules for translation.
When text in a source language is input, translation module 5 outputs a
target language text by translating the text using a prescribed procedure.
Referring to FIG. 4, translation module 5 includes a translation CPU 15
for translating a text in a source language (English language in the
embodiment) input through bus 7 into a text in a target language (Japanese
language in the embodiment) based on a prescribed translation program and
for outputting the result to bus 7, a translation program memory 16
connected to bus 7 for storing the translation program executed at
translation CPU 15, a buffer A for storing the input source language
original text on a word-by-word basis, a buffer B for storing information
such as the part of speech, and translation of each word referring to a
dictionary included in memory 6 for every word stored in buffer A, a
buffer C for storing information related to the parsing tree of the source
language text, a buffer D for storing the parsing tree of the target
language text which is generated by transforming the parsing tree of the
source language text stored in buffer C, and a buffer E for storing a
sentence rearranged in a Japanese style by adding appropriate particles
and auxiliary verbs to the parsing tree of the Japanese text stored in
buffer D, and a buffer F for storing tags, the number of words to which
the tags are attached, etc. which are used in a tag removing processing
which will be described later. These buffers correspond to storage unit 17
(FIG. 2).
Now, referring to FIGS. 2-9, an operation of an English to Japanese
language translation performed by the translating machine according to a
preferred embodiment will be described. A translation program executed in
translation CPU 15 will be later described in detail.
A read original text is segmented into morphemes by a morphological
analysis, and separated into tag signs and text by separation unit 10
(FIG. 2). The text body is stored in buffer A as illustrated in FIG. 5.
The tag signs are stored in buffer F.
Then, the part of speech for each word in the original text stored in
buffer A is determined using the dictionary lookup/morphological analysis
unit 11 as illustrated in FIG. 2. The resulting information is stored in
buffer B. The part of speech information for each word is stored as
illustrated in FIG. 6. More specifically the word "this" for example has a
plurality of parts of speech associated with it, i.e., two parts of
speech, pronoun and demonstrative adjective. The part of speech for the
word "is" is verb. Similarly, the parts of speech for the letter "a" and
the word "pen" are stored in buffer B.
Since the word "this" has a plurality of parts of speech, the part of
speech to be employed in the sentence is uniquely decided by a processing
performed by the syntactic analysis unit 12. In the processing performed
by the syntactic analysis unit 12 in the translation program, a parsing
tree showing the relationship between the words is performed (as
illustrated in FIG. 7) based on the dictionary and grammatical rules
stored in memory 6. The results of the syntactic analysis are stored in
buffer C.
The parsing tree is produced as follows. Among the grammatical rules stored
in memory 6, grammatical rules for translating to the English language
include the following rules:
Sentence .fwdarw. Subject, Predicate
Subject .fwdarw. Noun Phrase
Predicate .fwdarw. Verb, Noun Phrase
Noun Phrase .fwdarw. Pronoun
Noun Phrase .fwdarw. Article, Noun
The parsing tree is decided based on these grammatical rules. Such
grammatical rules are also prepared for translating to the Japanese
language, and correspondence is preestablished between the English
language grammatical rules and the Japanese language grammatical rules.
In the translation program, in a processing corresponding to the
transformation unit 13, the structure of the parsing tree (see FIG. 7) of
an input English language text is transformed into the structure of a
parsing tree corresponding to a Japanese language text illustrated in FIG.
8. In this transformation, as in the case of the above-described syntactic
analyzing unit 12, the tree structure transformation rules stored in
memory 6 are used. This transformation corresponds to shifting from level
L3 to the level L9 of FIG. 1. The result is stored in buffer D. The
example text used in the description "This is a pen" will be transformed
into a Japanese character string " " by this transformation.
In the translation program, the translated sentence producing unit 14 (in
FIG. 2) adds an appropriate particle "" or an auxiliary verb to the
resultant Japanese character string " " to form a grammatically correct
Japanese text as shown in FIG. 9 and the same is stored in buffer E. This
processing corresponds to a conversion from level L9 to level L10. The
resultant Japanese text " " is output from translation module 5, stored
in main memory 2 and displayed in display unit 3.
Now, referring to FIGS. 10-18, the process for translating an original
source language text including tag signs to a target language including
tag signs will be described in detail. This processing removes the tag
signs included in the original text before translation. The original text
without the tag signs is input to the translation module. The removed tag
signs are stored and associated with the particular word of the original
text to which the tag sign was attached. Subsequently, the tag signs are
replaced after the text is translated.
FIG. 11 is an example of tag signs used in an SGML document. Each tag sign
is formed of a pair of tags, i.e., a start tag and an end tag, the group
of words between the start tag and the end tag are considered to be marked
up.
Tag signs <term> <.backslash.term> indicate that a group of words between
the tag signs is used for a specific purpose, for example, words used as
index entries. Tag signs <emph> <.backslash.emph> indicate that a group of
words between the tag signs is to be emphasized for printing. For example,
the group of words between the tag signs may be printed in bold face or
the like.
FIG. 12 shows one example of an English language original text to be
translated containing tag signs. The original text is first stored in
buffer A. In the original text shown in FIG. 12, several words or group of
words are between tag signs. The pair of tag signs, <term>
<.backslash.term> respectively surround the word(s) "Tag" and "table
chart". These word(s) can be entries in producing an index for a document
including the sentence. The locations of the portions appearing between
the tag sign pairs are stored as the appearing page of each word group in
the index. The words "produce" and "specifying" between the tag sign pairs
<emph> <.backslash.emph> are to be printed in a typeface different from
the other parts, for example in bold face.
FIG. 10 is a detailed flow chart for use in illustration of tag sign
processing performed by separation unit 10 and dictionary lookup
morphological analyzing unit 11 of translation module 5. Referring to FIG.
10, an input text as shown in FIG. 12 is stored in buffer A in step SO1.
In step SO2, a word position pointer indicating the position of a word to
be processed in the original text stored in buffer A is set to 0. In the
example, the original word pointed by the word position pointer is "One".
In step S3, a determination is made whether or not the head of the word
presently pointed by the word position pointer is a start tag. If it is
not determined to be a start tag, the processing proceeds to step SO4. If
it is determined to be a start tag, the processing proceeds to step SO9.
If it is determined that it is not a start tag in step SO3, a determination
is made whether or not the word pointed by the word position pointer is an
end tag in step SO4. If it is an end tag, the processing proceeds to step
SO8, and if it is not an end tag, the processing proceeds to step SO5. If
the word pointed by the word position pointer is neither a start tag nor
an end tag, the word presently pointed by the word position pointer is
stored in buffer B (see FIG. 4) in step SO5, and processing proceeds to
step SO6. If it is determined to be an end tag, the end tag is deleted and
the word pointed by the word position pointer is stored in buffer B in
step SO8, and the processing proceeds to step SO6.
Meanwhile, if the head of the word pointed by the word position pointer is
determined to be a start tag in step SO3, the start tag is stored in
buffer F in step SO9.
In step S10, the word position in the text at which the start tag is
detected (in other words the value of the word position pointer) is stored
in buffer F in association with the start tag stored in SO9.
In step S11, a determination is made of whether or not an end tag is
present. If an end tag is present, the processing proceeds to step S12,
and otherwise, the processing proceeds to step S14.
In step S12, a word number "1" is stored in buffer F in association with
the start tag stored in step SO9.
Further in step S13, the start tag and end tag attached to the word in the
input original text designated by the word position pointer are deleted
and the word is stored in buffer B and the processing proceeds to step
SO6.
Meanwhile, if the processing proceeds to step S14, a processing for
counting the number of words to the end tag is made. The number of words
counted is similarly stored in buffer F in association with the start tag
stored in buffer F in step SO9.
Further in step S15, the start tag attached to the word pointed by the word
position pointer is deleted, and then the word is stored in buffer B. The
processing then proceeds to step SO6.
In step SO6, it is determined whether or not the text word is present in
buffer A. If the next word is present, the processing proceeds to step
SO7, and otherwise the processing proceeds to step S16.
In step SO7, a processing of incrementing the word position pointer is
performed and the processing returns to step SO3. Thereafter, the
above-described processing will be repeated. When there is no longer a
word to be processed in buffer A , the processing proceeds to step S16.
In step S16, dictionary lookup processing is performed for every .entry
stored in buffer B. As a result, information on the part of speech and
number for every entry is stored in buffer B.
Further in step S17, information on tag signs stored in buffer F is added
to the result of the dictionary lookup processing in buffer B. The
information representing the tag signs which have been attached to entries
with tag signs are added to a corresponding dictionary lookup results.
Now, the process shown in FIG. 10 when performed for the original text
shown in FIG. 12 will be described. The word "One" is pointed by the word
position pointer by setting the word position pointer to zero. Since the
results of the determination in steps SO3 and SO4 are both "NO", the word
"One" is stored in buffer B in step SO5. In step S6, the answer becomes
YES due to the presence of the next word, "of" and therefore the
processing proceeds to step SO7. Since the value of the word position
pointer is incremented by 1, the next word "of" will be pointed by the
word position pointer.
Hereinafter, the same processing as described above is repeated for the
words "of" through "this" and these words are sequentially stored in
buffer B together with their word positions.
At word position 8, the answer to the determination in step S3 becomes YES
and the processing proceeds to SO9. In step SO9, the detected start tag
(in this case <term>) is stored in buffer F, and then in step S10, its
word position is similarly stored in buffer F. Since the answer to in step
S11 is YES, the word number "1" is stored in buffer F in step S12. As a
result, tag sign information related to the term "Tag" stored in buffer F
takes the form of the first line in FIG. 13. In step S13, the start tag
and the end tag are deleted and the word "Tag" is stored in buffer B.
The presence/absence of a start tag and an end tag is determined for each
word while incrementing the word position pointer as described above, and
each word is stored in buffer B after removing a tag sign. Information as
shown in FIG. 14 is provided at buffer B. Information related to tag signs
as illustrated in FIG. 13 is obtained at buffer F.
Referring to FIGS. 13 and 14, the word at word position 8, in other words
the word "Tag", has a tag sign <term> attached, and as can be seen, the
word between the tag sign is only this word. Similarly, the word "produce"
at word position 14 has a tag sign <emph> attached. The group of words
"table chart" is between the tag sign pair <term> and <.backslash.term>.
The word "specifying" at word position 25 is between the tag sign pair
<emph> and .backslash.<emph>.
The stored contents of buffer B becomes as illustrated in FIG. 15 after
dictionary lookup processing performed in step S16. More specifically,
each entry is supplemented with information related to its part of speech
and number produced by the dictionary lookup processing. It is noted that
in FIG. 15 specific contents of parts of speech and numbers attached to
respective entries are omitted for the sake of simplification of the
figure.
In step S17, the contents of buffer F shown in FIG. 13 is added to
respective entries in buffer B. This processing is performed by attaching
a tag sign stored in buffer F to a word stored in buffer B corresponding
to the word position in buffer F in FIG. 13. More specifically, the eighth
word "Tag" has a tag sign <Term> attached. Similarly, the word "produce"
at word position 14 has a tag sign <emph> attached. The two words "table
chart" starting from word position number 15 have each attached with a tag
sign <term>. This is because the information of buffer F shown in FIG. 13
indicates that a tag sign <term> will be attached to each of the two words
starting at word position 15. The word "specifying" at word position 25
also has a tag sign <emph> attached.
Then, a usual translation processing will be conducted by syntactic
analysis unit 12 and transformation unit 13 in FIG. 3 based on the entries
stored in buffer B, and their respective part of speech and number, etc. A
translated sentence is finally produced by translated sentence producing
unit 14.
At that time, if an inputted word to be produced has tag information,
translated sentence producing unit 14 produces a start tag and an end tag
as indicated by the contents of buffer B shown in FIG. 16 before and after
the translation of the word, respectively. If the corresponding word does
not have such tag information, only a translated word is produced as
usual.
This production processing will be described below in conjunction with
FIGS. 17 and 18. The processings of steps S21-S27 shown in FIG. 17 are
conducted for each word belonging to the "leaf" parts of the parsing tree.
First in step S21, for a word, the column labeled "tag" corresponding to
the word in buffer B is checked (see FIG. 16), and the presence/absence of
tag information is determined. If the information is present, the control
proceeds to step S22; otherwise, the control proceeds to step S26.
In step S22, a processing is performed for storing the start tag in buffer
E which has been stored in buffer B.
Subsequently, in step S23, the stored translation corresponding to the word
undergoing processing is stored in buffer E from buffer B.
Then in step S24, a processing of storing an end tag in buffer E from
buffer B is performed.
In step S25 following step S24, an appropriate particle to be attached to
the translation is selected, and stored in buffer E. After step S25, the
control proceeds to step S27.
Meanwhile, if it is determined that no tag information is present in buffer
B at step S21, the control proceeds to step S26. In step S26, a processing
concerning tag information is not performed. Thus a processing similar to
steps S22 and S24 is not performed, and the translation is simply stored
in buffer E from buffer B. After step S26, the control proceeds to step
S25.
In step S27, a determination is made as to whether or not the current word
is the last word of the parsing tree to be processed. If it is not the
last word, the control returns to step S21, and steps S21-S27 are
repeated. If it is reached, the production processing is completed.
The final contents of buffer E is illustrated in FIG. 18. The transl | | |
