JPS6329881A - Machine translation device - Google Patents

Abstract

PURPOSE:To process specific expression in each field and to improve the efficiency of translation and a handling by informing information indicating the field of an input text to form a field identifying means to a parts turning means and allowing the parts turning means to instruct the selection of a dictionary to respective parts selector means. CONSTITUTION:The field identifying device 1 checks an input text to identify to which field the input text is related. The information (field information) indicating the field of the input text is sent to the parts tuning device 2 and the input text is transmitted to a machine translation device 3. The device 2 instructs the selection of a dictionary determined by the received field informa tion to respective parts selectors 4-A-4-P. Respective parts selectors 4-A-4-P select the dictionary in accordance with the instruction of the device 2.

Description

[Detailed Description of the Invention] [Summary] In order to enable a machine translation device to process expressions unique to the field when sentences from various fields are input, the system configuration is changed according to the field at the start of translation. This has been changed to allow more appropriate expressions to be output.

[Industrial application field]

The present invention relates to dynamic configuration change of a machine translation device.

The texts input into machine translation devices span a variety of fields. Preparing parts such as dictionaries so that texts in all fields can be translated at the beginning of translation will only place a heavy burden on the device configuration, and only a small portion of it will be used. Therefore, it is necessary to lighten the load by determining the configuration of the device according to the field of the input text at the start of translation.

[Conventional technology]

The translation device performs sentence analysis such as morphological analysis, syntactic analysis, and semantic analysis to generate a conceptual structure, and generates a sentence based on the conceptual structure.

The purpose of morphological analysis is to find the morphemes that make up an input sentence. Morphological analysis is performed using a word dictionary and a connection table in which the possibility of adjacency between Eastern words is entered.

Syntactic analysis reveals how the elements of a sentence are connected and what roles they play. That is, grammatical rules are applied to the elements of a sentence broken down into morphemes to determine the composition of clauses and phrases, and the modification relationships between subjects and predicates. However, at this stage, even if the analysis is grammatically correct, it is not necessarily semantically correct. Therefore, the following semantic analysis is required.

In semantic analysis, the semantically correct one is selected from the grammatically correct analysis results. In syntactic analysis, there are multiple correct results due to the ambiguity of words. This includes some sentences that have completely strange meanings, so if you perform a semantic analysis during the syntactic analysis and filter out those that are semantically strange, you can perform an efficient analysis.

A conceptual structure is a representation of the meaning of a sentence in the form of a semantic network. This conceptual structure is created so as not to depend on the language to be translated, so it is possible to generate sentences in multiple languages from this conceptual structure.

Sentence generation can be divided into syntax generation and morpheme generation.

In the process of generating sentences in the target language (English in the case of Japanese-English translation) from conceptual structures, words are extracted from the dictionary of the target language based on the conceptual symbols that express the concepts.

At this time, since grammatical information is not indexed at the conceptual structure level, it is necessary to select the part of speech that matches the target language. These rules are called production rules.

In production rules, processing starts from the node where the arc of the entry point shown in the conceptual structure enters, and the syntax of that node is determined by looking at the relationship between the surrounding arcs, the nodes ahead of it, and the grammatical attributes of the selected word. Co-occurrence relationships are used to determine phrasing between words for the syntax determined by syntax generation. The co-occurrence relationship determines which word to select when multiple words exist for one concept. In morpheme generation, for words determined through syntactic generation and co-occurrence relationships, the inflection part of the word, which can be determined by information such as tense, person, number, case, gender, etc., is generated by referring to a word dictionary to generate words with meaning. Synthesize into

Dictionaries include word dictionaries, adjacency relationship dictionaries, grammatical rule dictionaries, production rule dictionaries, co-occurrence relationship dictionaries, semantic relationship dictionaries, and the like. Word dictionaries, adjacency relationship dictionaries, grammatical rule dictionaries, production rule dictionaries, and co-occurrence relationship dictionaries exist for each language, and semantic relationship dictionaries are common to each language.

The word dictionary stores information such as headings, conceptual symbols, word symbols, grammatical attributes, production attributes, and frequency of appearance. The adjacency dictionary indicates the possibility of adjacency between words, and is used to extract words in the case of an input string,
Also, in the case of output strings, it is used to output the correct endings, etc.

The grammar rule dictionary stores rules for syntactic analysis and semantic analysis. The production rule dictionary stores rules for syntax generation and morpheme generation.

The co-occurrence relationship dictionary defines words and word usage, and is used for word selection.

A semantic relationship dictionary defines the relationship between concepts and is used to determine meaning.

[Problem to be solved]

In conventional machine translation devices, parts such as word dictionaries, grammatical rule dictionaries, and semantic relationship dictionaries have a single structure. In other words, it was not divided by field.

Additionally, machine translation devices are known that have dictionaries and the like prepared for each field, but this requires a tedious procedure in which a human must manually specify the field and tune the machine translation device each time text translation begins. Ta.

The present invention was created in view of this point, and aims to provide a machine translation device that has features such as being able to process expressions unique to a field, being able to perform translation efficiently, and being easy to use. It is said that

[Means for solving problems]

FIG. 1 is a diagram showing the principle of the present invention. In FIG. 1, 1 is a field identification device, 2 is a component tuning device, 3 is a machine translation device main body, 4-A to 4-P are component selectors, A1
or A, is a field-specific dictionary belonging to dictionary group A, B
1 to B1 are field-specific dictionaries belonging to dictionary group B, P
1 to P indicate field-specific dictionaries belonging to dictionary group P, respectively.

The field identification device 1 examines human-written text and identifies which field the input text relates to. Information indicating the field of input text (called field information)
is sent to the component tuning device 2, and the input text is sent to the machine translation device main body 3. The component tuning device 2 instructs each component selector 4-A, 4-B, . . . that a dictionary determined by the received field information should be selected.
・Instruct 4-P. Each parts selector 4-A, 4-B
.

..., 4-P selects a dictionary according to instructions from the component tuning device 2.

〔Example〕

FIG. 2 is a diagram showing an example of a component selector and dictionary. In the figure, 4-A is a word dictionary selector, 4-B is a grammar rule selector, 4-C is a semantic relationship selector, A is a word dictionary for the chemical field, A2 is a word dictionary for the mechanical field, and A7 is a word dictionary for the computer field. Word dictionary, B1 is a dictionary of grammatical rules for letters, B2 is a dictionary of grammatical rules for children's sentences, B1 is a dictionary of grammatical rules for manual sentences, C is a dictionary of semantic relationships in the field of chemistry, C2
indicates a semantic relational dictionary in the mechanical field, and Cp indicates a semantic relational dictionary in the computer field.

Examples of grammatical rules for letters are as follows:

In Japanese letters, the words “I” and “you” are not used very often, but in English letters, “I” and “YOU” are not used very often.
It is common to use the word ``■'' or ``YOU.'' The grammar rules for letters are that if there is a letter that does not include the words ``I'' or ``you,'' you should use the words ``■'' or ``YOU.'' It is used to generate sentences with words. For example, the grammar rules for manual sentences are as follows. Manual texts are written in the passive tense. The grammatical rules for manual sentences are so that when a Japanese manual sentence is manually input as input text, it is translated into English in the passive form.

For example, in the field of computers, semantic relational dictionaries are
Although it is called cess memory, there is no expression called access memory in the field of psychology.

FIG. 3 is a diagram for explaining the present invention in detail. Third
Figure (a) shows the operation when the input text ``Omitted...Please send samples of polyvinyl chloride and melamine resin.'' is input. In the case of FIG. 3(a), the field identification device 1 outputs field information indicating that the text is in the field of chemistry and is a letter. Upon receiving this field information, the component tuning device 2 instructs the word dictionary selector 4-A to select the connecting line a1, and instructs the grammar rule selector 4-H to select the connecting line a1. Then, it instructs the semantic relation selector 4-C to select the connecting line C1. As a result, a word dictionary A1 in the field of chemistry, a dictionary of grammatical rules for letters B1, and a semantic relationship dictionary C8 in the field of chemistry are selected.

FIG. 3(1)) shows the operation when the input text ``768 KB of memory is required, and the user should follow the instructions on the console panel'' is input. In the case of FIG. 3 (bl), the field identification device 1 outputs field information indicating that it is a computer field and a manual text. Upon receiving this field information, the component tuning device 2 selects a word dictionary selector 4-A. Instructs the grammar rule selector 4-B to select the connecting line a, and instructs the semantic relation selector 4-B to select the connecting line a.
-Instructs C to select the bond line cp. As a result, there is a word dictionary in the computer field, a grammar rule dictionary B for manual sentences, and a semantic relationship dictionary C in the computer field.
2 is selected.

〔Effect of the invention〕

As is clear from the above explanation, when translating texts from many different fields, if a huge number of components such as dictionaries and rules covering each field are integrated at the beginning of translation, the entire system becomes huge. However, according to the present invention, there is no need to integrate a huge number of components such as dictionaries and rules covering each field at the start of translation, and the components can be switched depending on the text to be translated. All you have to do is select the most suitable one and operate it. This reduces the load during system operation. Furthermore, since the selection of optimal parts can be automatically performed from the input text, the burden on the user can be reduced.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the principle of the present invention, FIG. 2 is a diagram showing an example of a component selector and a dictionary, and FIG. 3 is a diagram for explaining the present invention in detail. 1...Field identification device, 2...Parts tuning device, 3...Machine translation device main body, 4-A or 4-P...
・Part selector, A1 to A, ... Word dictionary by field, Bl to B1 ... Grammar rule dictionary by field, C1
Or C2...Semantic relationship dictionary by field.

Claims (1)

[Scope of Claims] Field identification means (1) for identifying the field of input text; a plurality of dictionary groups each composed of a plurality of field-specific dictionaries of the same type; each dictionary group having a one-to-one relationship with the dictionary group; A plurality of component selector means (4-A, 4, B, ..., 4-P) that correspond to each other and select one dictionary in the corresponding dictionary group, and a component tuning means (2). The field identification means (1) is configured to notify the component tuning means (2) of field information indicating the field of the input text, and the component tuning means (2), according to the received field information, Each component selector means (4-A, 4-B,...
, 4-P) to instruct dictionary selection.