JPH0827799B2 - Free translation dictionary - Google Patents

Description

The present invention relates to a dictionary management system for storing, managing, and retrieving various language data such as morphological information, grammatical information, semantic information, and synonym information in a language processing system. The present invention relates to a free-translation dictionary that realizes synonym search between multiple languages.

(Prior Art and its Problems) A language understanding system such as a machine translation system requires a dictionary for storing headwords (morpheme notation or word notation), morphological information, syntactic information (grammar information), and semantic information. . As a dictionary for machine translation, a dictionary method (Japanese Patent Application No. 59-148138) in which headwords and semantic symbols are used as dictionary search keys and morphological information, syntactic information, and semantic information are linked by pointers is known. The notation is "configuration""constr
The morphological information is character type information such as "two kanji" for "construction" and phonological information such as "starting consonant" for "construct". Furthermore, a semantic symbol is a symbol that expresses the concept of the word and is invariant to the use of the word and the syntactic change. However, if the meaning changes due to the difference in form or syntax, they will be different symbols. Also,
Idioms with varying meanings are also included. For example, "goes" has the same meaning as "go", so the symbols "goes" and "goin"
There is no "g", but the symbol "go". Also, "bank" and "bank", and "interest" and "interesting" have the same expression, or are in the present participle form. , Each symbol exists because the meaning is different. Further, since "take" and "take part in" have different meanings, there are respective symbols.

In this method, vocabularies of almost the same word, which have no difference in form or syntactic information in synonyms or nuance levels, can be related by having the same meaning symbol. That is, synonyms in a narrow sense can be related to each other. However, in a broad sense, it includes differences in aspect information such as movements and states found in the active and passive voices, differences in the degree of adverb modification, and differences in meaning such as relationships between intransitive and transitive verbs and focus. Synonyms of cannot be the same meaning symbol. Setting the same meaning symbol corresponds to describing that there is no semantic difference, and as a result, the meaning information of the difference is lost. This is inappropriate as a dictionary. Also, there is no framework that associates different semantic symbols. Therefore, even if a synonym in a broad sense is not a synonym in a narrow sense, words cannot be associated with each other.

For example, in Table 1, via the semantic symbol "READ",
Because the Japanese noun (reading) and the English verb (READ) are related, the noun phrase "readin" is a translation of the noun "reading".
"g" can be selected. Also, the meaning symbol "SMALL"
Via Japanese to “small” and English to “little” and “smal”
can also be translated as "l". This is because these are synonymous words in a narrow sense and therefore have the same meaning symbols.

However, as shown in Table 2, for Japanese "construction", for English, for "construct" and "consist-of",
Semantic symbols "CONSTRUCT" and "CONSIST-OF" exist separately. Because "construct" expresses a motion, while "consist-of" expresses a state,
Because there is a difference in meaning, the two are not synonymous in a narrow sense. Therefore, the meaning symbols of each are Japanese "composition (ga, o, kara)" and English "construct", and Japanese "composition (ga, o)".
It corresponds to "cosist-of" in English. The former 「CONSTRUC
"T" corresponds to "composition" of "people compose a house from wood, soil, and water", and the latter "CONSIST-OF" becomes "composition" of "wood, soil, and water composes a house." Correspond. Since they have different meaning symbols, they cannot be related to each other as described above. On the other hand, there is a sentence "house is composed of soil, wood, and water" that uses the former "composition (ga, wo, kara)" in the passive voice. The meaning of this sentence is that a certain object (house) is in a certain state (composition) based on a certain resource (soil, tree, and water). There is also a sentence that uses the word ")" in the active voice, "the soil, the tree, and the water make up the house," and this meaning should be the same as the previous sentence. Since it is impossible to express that they are semantically identical, both sentences are interpreted as different meanings even in machine translation, etc., and different translated sentences are generated.

(Effects of the Invention) The present invention solves the above-mentioned drawbacks and provides a translation dictionary capable of considering the difference in meaning and nuance of synonyms, the difference between meaning and focus between active and passive verbs, and intransitive and transitive verbs. The purpose is to propose.

(Structure of the Invention) (a) Morphological notation and morphological information, and a morphological dictionary means having a pointer to a grammatical information description dictionary item corresponding to them, (b) Corresponding to the grammatical information description dictionary item in the morphological dictionary means Syntax dictionary means for holding part-of-speech / grammar information, pointers to items in the morphological dictionary means, and pointers to semantic information description dictionary items corresponding to the morpheme notation and the part-of-speech / grammar information, (c) meaning A symbol dictionary and semantic information, a pointer to the syntactic dictionary item corresponding to them, and a semantic dictionary unit that stores a pointer to a thesaurus item that describes a superordinate concept of the semantic information, (d) Morphological notation in the morphological dictionary unit , And a dictionary search means having the meaning symbol in the meaning dictionary means as a search key, (e) meaning symbol, corresponding to the thesaurus information Meaning dictionary common to the synonyms, a case frame including the case frame of each synonym, and a thesaurus means having a pointer to the meaning information description dictionary item in the meaning dictionary means of each synonym, can get.

(Example) Hereinafter, the present invention will be described in detail with reference to the drawings.
In the following, Japanese and English are taken as examples for ease of explanation.

FIG. 1 is a block diagram showing an embodiment of the present invention. The dictionary search means 0, 4, the form dictionary means 1, the syntactic dictionary means 2, and the semantic dictionary means 3 of the present invention are described in Japanese Patent Application No. 59-148138 (JP-A 61-26176). Since the method is the same as that described, only a brief description will be given here.

The dictionary search means 0 has a function of designating a morpheme notation and outputting information such as corresponding morphological information / part of speech / grammar information / semantic information. The morpheme notation from the outside is input, and the corresponding morphological dictionary item is output to the morphological dictionary means 1 (01).

Further, the meaning symbol input from the dictionary search means 4 is input to the main dictionary search means 0 as information such as corresponding morphological information / part of speech / grammatical information / semantic information from the morphological dictionary means 1 through various dictionary means.

The morphological dictionary means 1 is composed of a set of morphological dictionary items having morpheme notation / morphological information and a pointer to a grammatical information description dictionary item corresponding to the information, and has a function of associating morphological information with grammatical information. When a specific morphological dictionary item is input via the dictionary search means 0, all grammatical information description dictionary items corresponding to the morpheme notation in the morphological dictionary item and the morphological information are output to the syntax dictionary means 2. (12).

Further, when a morphological dictionary item is input via the syntactic dictionary means 2, the corresponding morpheme notation / morphological dictionary is output to the dictionary searching means 0 together with the accompanying part-of-speech / grammar information and semantic information.

The syntactic dictionary means 2 is composed of a set of grammatical information description dictionary items having part-of-speech / grammatical information, a pointer to a corresponding morphological dictionary item, and a pointer to a semantic information description dictionary item. It has a function of associating part-of-speech / grammar information with form information or part-of-speech / grammar information with semantic information.

When a specific grammatical information description dictionary item is input via the morphological dictionary means 1 (12), the morphological information of the incidental input, the part of speech / grammatical information in the grammatical information description dictionary item, and all corresponding semantic information The description dictionary item is output to the semantic dictionary means 3.

When a specific grammatical information description dictionary item is input via the semantic dictionary means 3 (23), the associated input semantic information, the part of speech / grammatical information in the grammatical information description dictionary item, and all corresponding form dictionaries. The item is output to the form dictionary means 1 (12).

The semantic dictionary means 3 comprises a set of semantic information description dictionary items having semantic symbols and semantic information, pointers to corresponding grammatical information description dictionary items, and pointers to thesaurus items. It has a function of associating the semantic information with a superordinate concept on the thesaurus or with a part of speech / grammar information.

When a specific meaning information description dictionary item is input via the syntactic dictionary means 2 (23), the morphological information / part of speech / grammar information of the accompanying input and the meaning information in the meaning information description dictionary item are output to the dictionary search means 4. To do. At the same time, all corresponding thesaurus items are output to the thesaurus means 5 (35).

When a specific semantic information description dictionary item is input via the thesaurus means 5 or the dictionary search means 4 (34, 3
5), outputs all corresponding grammatical information description dictionary items to the syntactic dictionary means 2 together with the semantic symbols and the semantic information in the semantic information description dictionary item (23).

The dictionary search means 4 has a function of outputting information such as morphological information / part of speech / grammar information / semantic information corresponding to the semantic symbol. A semantic symbol from the outside is input, and the corresponding semantic information description dictionary item is output to the semantic dictionary means 3 (34). Further, the morpheme notation input from the dictionary search means 0 is input to the main dictionary search means 4 as information such as corresponding morphological information / part of speech / grammatical information / semantic information from the semantic dictionary means 3 through various dictionary means. .

The thesaurus means 5 has a hierarchical structure based on the semantic hierarchical relation, has semantic information including the contents of the semantic information description dictionary items below it, and case frame information along with the semantic symbols, and all subordinate meanings. It consists of a set of thesaurus items with pointers to information description dictionary items. It has a function of associating synonymous meaning dictionary items with each other.

When a specific thesaurus item is input via the semantic dictionary means 3 (35), all the corresponding semantic information description dictionary items are output to the semantic dictionary search means 3 together with the item contents (2).
3).

Here, an embodiment of the operation of the present invention will be described with reference to the drawings, taking as an example the case where "a house is composed of soil, wood, and water" and "very large" are input as input data.

Here, the input data “a house is composed of soil, trees, and water” will be described. First, for the “structure”, the meaning information description dictionary item ID1 is output to the meaning dictionary means 3 through the dictionary search means 0, the form dictionary means 1, and the syntax dictionary means 2. From the morphological dictionary means 1, the morpheme notation “structure” and morphological information (characters: 2 kanji), and from the syntactic dictionary means 2, POS (verb) and grammatical information (verb form: ga, kara) etc. I want it. Here, the verb form is a set of particles that the verb easily takes. Furthermore, by the semantic dictionary means 3,
As shown in Fig. 2, "Semantic symbol: S # CONSTRUC
T, semantic information: action, case frame: AGENT, OBJECT, SOURCE "
Is found. On the other hand, the meaning information is changed to the state by the Japanese feature that "was made" and "becomes state when the action is made passive", and the case used from the particle "ga, kara" is only "OBJECT, SOURCE". Become. The semantic information in this example is a verb aspect, and the case frame conforms to Fillmore's deep case. (For Fillmore's deep case, see "Principle of Case Grammar: Meaning and Structure of Language" written by Fillmore (Translated by Harumi Tanaka, Michio Funaki) Sanseido Publishing, 1975.) Therefore, the meaning of "composed" input data “Semantic symbol: S # CONSTRUCT, semantic information: state, case frame: OBJECT, SOURCE” is obtained.

In the conventional method, it is only possible to select the translated word “construct” that succeeds this ID1, and the expression such as “beconstructed” was selected. In the present invention, according to the semantic dictionary means, ID1 and TD1
Upper thesaurus item TD using the upper / lower relation of
Get one. Furthermore, according to the thesaurus means, in the subordinate term of TD1
Item ID2 can be referred to as a synonym for ID1. When the semantic information of this ID2 and the content of the case frame match the information from the previous input sentence, this semantic information description dictionary item is selected.

Furthermore, the morpheme notation “consist of” connected from ID2 via the semantic dictionary means, the syntactic dictionary means, and the morphological dictionary means
Is the translated word corresponding to the input data.

Similarly, in the input data "very large", the meaning information description dictionary item ID4 is selected from [large] and the meaning information is changed from [very]. The meaning information here is a degree, and the meaning of the input data is “S # large, meaning information: strong”. Also in this case, by referring to the meaning dictionary item ID3 by the thesaurus means and the meaning information contents are matched,
As a result, the translated word "huge" can be selected.

In the input data “small fish”, the meaning information description dictionary item ID5 is selected from [small] and the meaning information is changed from [small]. The meaning information here is the size, and the meaning of the input data is "S # FISH, meaning information:
Small size ". In this case as well, the thesaurus means refers to the meaning dictionary item ID6 and the meaning information contents match, so that the translated word "small fish" can be selected.

Furthermore, in the input data "read thoroughly", [rea
Semantic information description dictionary item ID7 is selected from [d], and [thorou
Semantic information is changed from [ghly]. The meaning of retribution here is the degree of detail, and the meaning of the input data is "S # READ,
Semantic information: High level of detail ". Also in this case, the thesaurus means refers to the meaning dictionary item ID8, and the meaning information contents match, so that the translated word "deep reading" can be selected.

(Effects of the Invention) According to the present invention, a dictionary structure is provided in which a thesaurus has a case frame having a common meaning between synonyms, and the difference in meaning between individual words has a difference in nuance between the meaning dictionary items. Can be realized. As a result, a translated word that reflects the nuances of the source language can be selected when the translated sentence is generated. Also,
Selection of transitive verbs and intransitive verbs, depending on the difference in focus of topic / expression,
Although there is a difference between the active form and the passive form, the selection can be made according to the language according to the present invention, and it is possible to carry out free translation while retaining the intent of the original sentence.

Also, Japanese Patent Application No. 59-148138 (JP-A-61-26176)
In the invention described in Japanese Patent Publication No.), synonyms (synonyms) having a common meaning symbol can be associated with each other. However, words having different semantic information such as aspect / modality / degree are not synonymous words. The words having different meaning information cannot be used as a common meaning symbol because of the difference in meaning. If they are shared,
Since the semantically different expressions are interpreted as the same, an error occurs when they are used for translation.

In the present invention, the symbols are combined via the upper symbols on the thesaurus, and the semantic information corresponding to each semantic symbol is described at the semantic information description dictionary item level. Therefore, it is possible to realize the correspondence between the synonymous expressions in a wider range because the semantic difference and the correspondence in the higher rank are realized.

For example, Japanese Patent Application No. 59-148138 (JP-A-61-2617)
(Publication No. 6) as a synonym search function described in "festival, fa
"ir, bazar" can be handled by standardizing symbols. However, as for "construct, consist of", the former works and the latter has different state and aspect information.
The existing method cannot be associated. Also, "giant, ve
Since “ry large” also has different degree information, the meaning symbols are different and it is impossible to associate them. On the other hand, in this method, the words are associated with each other through “T # composition” and “T # large” on the thesaurus item, so that correspondence between words is possible.

Taking the example of translation, "construc"
“Large” is combined with “t” and “large”. Regarding the translation of the expressions "constructed" and "very large", since the existing methods have a limited vocabulary that can be combined, they can only be translated as "be constructed" and "very large". On the other hand, in this method, an optimal expression can be selected from a wider range of words, so that appropriate translations of "consist of" and "giant" are possible.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, and FIG.
It is a schematic diagram explaining the relationship between the semantic dictionary means in the figure and the thesaurus means. 0,4 ... Dictionary search means, 1 ... Morphology dictionary means, 2 ... Syntax dictionary means, 3 ... Semantic dictionary means, 5 ... Thesaurus means, IDi ...
Semantic information description dictionary item, TDj ... Thesaurus item.

Claims (1)

[Claims] 1. A morphological notation and morphological information, and a morphological dictionary means having pointers to corresponding grammatical information description dictionary items, and part-of-speech / grammatical information corresponding to the grammatical information description dictionary item in the morphological dictionary means. A syntax dictionary means for holding a pointer to an item in the morphological dictionary means, and a pointer to a semantic information description dictionary item corresponding to the morpheme notation and the part-of-speech / grammatical information; A semantic dictionary unit that stores a pointer to the corresponding syntactic dictionary item and a pointer to a thesaurus item that describes a superordinate concept of the semantic information, a morpheme notation in the morphological dictionary unit, and a meaning in the semantic dictionary unit A dictionary search means that has a symbol as a search key, a meaning symbol, and a meaning common to the synonyms corresponding to the thesaurus information. A free-translation dictionary having information, a case frame including the case frame of each synonym thereof, and a thesaurus means having a pointer to a meaning information description dictionary item in the meaning dictionary means of each synonym.