JPH04112367A - translation device - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a translation device, and more particularly to a machine translation system. For example, it is applied to natural language processing technology.

In the case of Chikuken's previous English-Japanese machine translation system, some adverbs were sometimes translated as particles or auxiliary verbs, but in addition, the target independent word also retained other particles and auxiliary verb information. When I was doing this, I was in a state where I couldn't handle it very well. Furthermore, machine translation is still a difficult technology.

For example, when considering machine translation from English to Japanese, there are various issues. English sentences must be analyzed and converted to generate Japanese sentences. Even if we look at generation, we are required to create more natural Japanese sentences.

Generally speaking, there is a phenomenon that English parts of speech and Japanese parts of speech do not correspond. There are many adjectives in English, but there are not necessarily equivalent adjectives in Japanese. In most cases, an adjective verb or a translation of the verb is given. In this case, the words are still the same, so there is no major inconvenience, but sometimes the corresponding parts of speech may be far apart. In such cases, there was a problem that the translation of the Japanese text was not very good.

Purpose The present invention was made in view of the above-mentioned circumstances.
To provide a translation device that can translate a second language more naturally even if the expression forms of the first language and second language are different, such as translating an adverbial expression in English as an auxiliary verb expression in Japanese. It was done for the purpose of

! --In order to achieve the above object, the present invention provides (1) an input device for inputting an original text in a first language into a processing device, a display device for displaying information input by the input device, and a display device for displaying information input by the input device; In a machine translation system equipped with a processing device for processing and an output device for outputting a second language, an adverbial expression in the first language is translated as a particle or auxiliary verb expression in a second language as necessary. , furthermore, (2) the auxiliary verb expression translated in the second edict may become a compound auxiliary verb expression together with other mote information held by the main predicate; and (3) the above. It is characterized in that the particle expression translated in the second language can become a compound particle expression together with other case information held by the main nominal. Hereinafter, the present invention will be explained based on examples.

FIG. 1 is a block diagram for explaining one embodiment of a translation device according to the present invention, in which 1 is a cathode ray tube (CRT), 2 is a keyboard, 3 is an optical character reader (OCR), and 4 is a Input document, 5 is a spell check section, 6 is a pre-editing section, 7 is a translation main section, 8 is a post-editing section, 9 is a dictionary, 10 is a grammar rule,
11 is an output document, and 12 is a printer.

The input sentence obtained by file input, keyboard input, or OCR input is stored as an input document 4, and then the spell check unit 5 checks the spelling with reference to the dictionary 9, and uses the pre-editing unit 6 to Can process. The translation main unit 7 performs translation using a dictionary 9 and grammar rules 10, and the resulting output sentence can be edited by a post-editing unit 8 using translation information. The edited output sentence is stored in the output sentence M11. The input and output sentences can be printed using printer 12.

FIG. 2 is a diagram showing the four processes of the translation main unit in FIG. 18 is a generation section, 19 is an output section, and 20 is a grammar rule. The four main processes of this translation main body are: ■ The morphological analysis unit 15 performs dictionary lookup of the input text; ■ The syntactic analysis unit 16 obtains information on individual words and parses them according to grammatical rule 2o; Creating a tree structure from the analysis results; ■ Transforming the tree structure of the input language into the main structure of the output language in the conversion unit 17; ■ Translating the obtained main structure node by node in the generation unit 18. be.

The present invention pertains to a converter and a generator. Here, the input text is English. Adverbial expressions that require special processing in the present invention are called modal adverbs. After the syntactic analysis is completed, the syntactic conversion section converts this modal adverb into
When relating to a predicate (verb, adjective), it transmits the mode information it holds to the predicate, which is the node to which it relates.

It only conveys information and does not have its own translation.

The words receive and receive the information they convey. Finally, when the expression is translated in the generation unit, the mode information is translated as a particle dynamic expression. Also, this modal adverb can be used as a nominal (
When a node relates to a node (noun), it transmits the mode information it holds to the node to which it relates, which is the nominal. It only conveys information and does not have its own translation. Vocal language receives and receives the transmitted information. Finally, when the nominal expression is translated in the generation section, the mode information is translated as a particle expression. Furthermore, when this mode adverb relates to an adverb, the structure conversion unit does not transmit mode information. The generator generates its own translation.

FIG. 3 is a flowchart showing the absorption of modal adverb information in the conversion unit of the translation device according to the present invention. below,
Each step will be explained in turn.

step 1; Set the pointer to the top note at the top of the tree structure.

5tep2 Determine whether there is an i-node. That is, the nodes are examined one by one starting from the top node. When one node is finished, move to the next node recursively using the following steps. If there is a child node (a branch one level below itself), move to the child node. If there are no child nodes, move to the second node (the node next to you and at the same height as you). If there is no node, the process ends.

' ste 3.4.5; Check the part of speech of each node, and if it is a modal adverb, check the part of speech of its parent node (the node one level higher than itself).

5te6 7s; Determine whether the part of speech of the parent node is a particle, an adjective node, a noun node, or an adverb node.

If it is a verb, an adjective node, or an adverb node rather than a noun node, leave it as is.

If it is a verb, adjective node, or noun note, give mode information to the parent node and delete your own node.

n↓yyu; Move the point to the child node and then to the th note, and return to step2.

This processing is performed for all nodes, and the processing of one sentence is completed.

FIG. 4 is a flowchart showing the translation of mode adverb information in the generation unit of the translation device according to the present invention. below,
Each step will be explained in turn.

Set the pointer to the top node at the top of the tree structure.

1 construction y5; Determine whether there is a node. That is, nodes are examined one by one starting from the top node.

If the node absorbs modal adverb information, look at the part of speech of the node.

"Yoyoyo!"−; If the part of speech of the node is a verb node, add other mode information to the verb and translate the mode adverb information.

If the part of speech of the node is an adjective node, other mode information is added and mode adverb information is translated.

ste 11. ．． 12, 13; If the part of speech of the node is a noun node, the word order with case information is determined, and mode adverb information and case information are generated.

d; Move the point to the child node and then the th node, 5t
Return to ep 2.

This processing is performed for all nodes, and the processing of one sentence is completed.

Next, we will take only' as an example of a modal adverb. An example sentence using this is shown below.

1, I only ate it.

2, I was only 1ucky.

3. I know it only recently
ly.

4. I hear only the resu
lt.

The parse tree created as a result of parsing the above four sentences is as follows.

(Left below) 1, I only ate it+(,), <
period> +--(V) ate <verb> +-(S) I <noun> +--([)) only <adverb> +-(○) it (noun> 2, I was only 1ucky+(,)
, <period> +-(V) as <verb> +-(S) I <noun> +--(C) 1uky <adjective> +--(D) only <adverb> 3, I know it only recen
tly.

(,), <period> +--(V) knee (verb) +--(S) I <noun> +--(0) it (noun) +--(D) recently (adverb) +-- (D
) only <adverb> 4, I hear only the resu
lt.

(,), <period> +--(V) hear <verb> +--(S) I <noun> +--(o) result <noun> +--(A)
only <determiner> +--(A) the <article> Note that in the above parse tree, the leftmost period is a top note, and the child note to the right is a verb node. Multiple noun notes exist as child notes under the verb note. The noun notes are called sibling notes, with the upper one being the older brother and the lower one being the younger one.

Perform syntax conversion from the above results. At this time, the meaning of the mode information given to the part of speech to which the moat adverb 'only' is dependent may differ depending on the part of speech to which it depends. This appears as a difference in the translated words when they are generated. In addition, in the case of 'only', the limiting side relates to a noun.

In this case as well, the mode information is absorbed by the noun to which it depends, as a mode adverb is used in a broad sense. Modal adverbs are pragmatics (verbs,
(adjectives), it gives mode information to the verb it depends on, and its own branch (note) disappears. When a modal adverb relates to a nominal word (noun), it does not give mode information to the noun to which it is dependent, and maintains its own branch (node) as it is.

Therefore, the tree structure obtained by syntax conversion from the above syntax analysis results is as follows.

1, I only ate xt.

+-,-+=> aje verb verb - mode left 0NLY
<-only+-5-+=> I noun is +-Q- company> It noun is 2, I was only 1ucky.

+-,-+=> yBsBs verb - Mort E past 0NL
Y <-vas<-onl, y+-8-+=> I noun is 3, I kneti it only re
century.

+-1-+:> kne verb no mode: past +-5-
+=> ■Noun +-〇-+=) ij noun +-I -+=> recently adverb muju-old +=
＞Onln adverbless 4, I hear only the r
esult.

+-,-+:> haard no verb +-8-+=> I noun is +-〇-+:> resul noun mode: THE O
NL'W <-the<-only In addition to mote adverbs, there are some parts of speech that are absorbed. For example, be as a copula, the auxiliary verb article class, and the auxiliary verb class.

Next, a translation is generated. Generation is performed in order from the older node of the node to the node (top and bottom of the same height), and from the child to the parent (nodes of different heights, child on the right and parent on the left). Therefore, the main verb of the main cylinder is generated last.

The first thing to be generated is its eldest node (and if there are child notes, its eldest node).

Therefore, in number 1, ' I ” it” ate(+
only)'.

In the second verse, ' I '' 1ucky (+ was, o
Translate in the order of 'nly)'.

In the third verse, ' I ” it” only” rece
Translate in the order of ntly''knesu'.

In number 4, ' I '' result (+ the,
only)” Translate in the order of head’.

The mote information, which is information absorbed by other notes, has different translation forms depending on the part of speech of the node of the absorbed light.

When absorbed into a pragmatic (verb, adjective), it generates ゛ only' as an auxiliary verb expression connected to the pragmatic. At this time, this 'only' is generated by reflecting other mode information absorbed by the predicate.

That is, for example, if you have absorbed ``past'' information, it becomes ``only''. Also, since it includes information on the copula's be auxiliary verb, when it follows an adjective, it becomes only the conjunctive form of the adjective, 10'. When it is absorbed into a nominal word (noun), it generates ``only'' as a particle expression connected to the nominal word. At this time, this 'dake' is generated by considering the origin of the case information that has been absorbed. In this case, this is determined by the linguistic relationship between the mode information 'only' and the absorbed case information. In other words, each mode information and case information holds ranking information for determining word order. For example, if the noun ``result'' absorbs the information ``only ° and ′,'' then ``only'' is generated before ′, and ′
Only the result will be ``.

Therefore, No. 1 says, ``All I had to do was eat it.''

" is generated. However, in the case of verbs, it is easier to understand Japanese by changing the main verb itself rather than changing the auxiliary verb expression. Therefore, it is easier to understand if the verb is placed directly in the past, as shown below, rather than the above translation. Rather, it will be like this. "I just ate it.

” Number 2: ``I was just lucky.''" is generated.

Number 3 is generated as "I only recently learned about it."

Number 4 is generated as "I only heard the results."

As can be seen from this result, the translated form of the modal adverb 'only' is used in the following three ways. Furthermore, in Table 1, onl
An example of dictionary information for y is shown.

1. It is only an auxiliary expression for a predicate. 2. It is an adverb that modifies an adverb. Finally, it is 3. It is only a particle expression for a nominal. Although I have taken only 'only' as an example of a modal adverb, there are other examples. There are also adverbs such as 'even' and 'also'.

Effects As is clear from the above explanation, according to the present invention, the first
Since adverbial expressions in a language are translated as particles and auxiliary verbs in the second language as necessary, the first and second language
Even if the language expressions are different, a more natural second language can be translated.

[Brief explanation of drawings]

FIG. 1 is a block diagram for explaining an embodiment of the translation device according to the present invention, FIG. 2 is a block diagram for explaining the processing of the translation main unit in FIG. 1, and FIG. FIG. 4 is a flowchart showing the absorption of modal adverb information in the generating part. FIG. 4 is a flowchart showing the translation of modal adverb information in the generating part. 1...Cathode ray tube (CRT), 2...Keyboard,
3. Optical character reader (OCR), 4... Input document,
5... Spell check department, 6... Previous editorial department, 7...
・Translation main unit, 8... Post-editing section, 9... Dictionary, 10.
- Grammar rules, 11. Output document, 12. Printer. Figure 2 Figure 2

Claims (1)

[Claims] 1. An input device for inputting an original text in a first language to a processing device;
In a machine translation system comprising a display device that displays information input by the input device, a processing device that processes the information, and an output device that outputs a second language, the first
A translation device characterized by translating an adverbial expression in a language as a particle or auxiliary verb expression in a second language as necessary. 2. The translation device according to claim 1, wherein the auxiliary verb expression translated in the second language can become a compound auxiliary verb expression together with other mode information held by its main predicate. 3. The translation device according to claim 1, wherein the particle expression translated in the second language can become a compound particle expression together with other case information held by its main nominal.