JP2000305930A - Language conversion rule preparing device, language converter and program recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To surely convert an inputted voice sentence into an objective language even when an unlearned part exists in the inputted voice sentence and a mistake is made in a part of voice recognition and further to automatically prepare a phrase dictionary required for conversion and rules between phrases without any manual work to the utmost. SOLUTION: This device is provided with a language rule preparing part 2 to automatically and statistically learn an integral or semantic regulation rule for a partial word or a word string from a translation copath and to describe the rule in a form that sentences of an objective language part and an original language part are answered to each other a voice recognition part 7 to perform the vice recognition of voice sentence of the original language by using the prepared language rule and to output a recognized result, and a language conversion part 9 to convert the sentence in the original language into the one in the objective language by using the same language rule.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a language conversion device for converting input speech or input text into another language or another stylistic type and outputting the same, and a language conversion rule creating device for creating the conversion rules.

[0002]

2. Description of the Related Art The prior art will be described below with reference to an example of an apparatus for translating an input voice into another language (hereinafter referred to as an interpreter), which is one of the language conversion apparatuses.

[0003] The interpreting apparatus recognizes speech for converting an utterance sentence input as an acoustic signal into an output sentence displayed as a word text string, and inputs a sentence displayed as a word text string and translates the sentence into another language sentence. Interpretation is realized by sequentially executing language translation. The language translation unit further includes a language analysis unit that analyzes a syntactic or semantic structure of the input sentence, a language conversion unit that converts the input sentence into another language based on the analysis result, and an output that generates a natural output sentence from the translation result. And a sentence generation unit.

However, when the speech recognition unit misrecognizes a part of the utterance sentence, inserts or rephrases the sentence, or ends the utterance without completing the sentence.
If the utterance itself is not syntactically or semantically unnatural,
Even if the speech recognition result is input to the language analysis unit, the analysis fails,
As a result, there is a problem that a translation result is not output.

[0005] In order to solve this problem, the phrase is divided into phrases, the inside of the phrase and between the phrases are regularized, and incomplete utterances are analyzed using only the rules within the phrase, and the analysis result can be output. That is, it is configured.
(For example, Takezawa, Morimoto: IEICE Transactions D-II, Vo
l.J79-D-II (12)). FIG. 14 shows an example of a conventional intra-phrase and inter-phrase rule. In this example, for the corpus example 301 “Please book a single room tonight”,
The intra-phrase rule is described in a tree structure like the intra-phrase rule 302 based on a grammar rule common to written words, and the inter-phrase rule is described by an adjacent probability between phrases in the learning corpus. For example, an inter-phrase rule is described as an inter-phrase rule 303.

When analyzing an input sentence, rules within a phrase are applied sequentially from the beginning of the sentence, and at the end of the phrase, the input sentence analysis is performed while connecting the phrases such that a phrase candidate having a high adjacent probability is adjacent to each phrase. Done. With such a sentence analysis method, even if a part of the sentence is misrecognized and the analysis of the entire sentence fails, the phrase analysis of the part that does not include the misrecognition is performed correctly, so only the analyzed partial phrase is used. Is a framework that can partially output the translation result.

In order to solve this problem, language analysis is not performed according to the conventional grammar, but a corresponding source language sentence is extracted from an utterance example including utterances that cannot be analyzed by the conventional grammar. Extract the bilingual phrase of the target language sentence,
There has also been proposed a method of creating a bilingual phrase dictionary in which the phrase pairs are described in a generalized form as much as possible, and performing language analysis and language conversion using the dictionary (for example, Furuse, Sumida, Iida: Information Processing) Journal 35, no3, 1994-
3). FIG. 15 shows a conventional language conversion rule creation device. Before translating, a bilingual phrase dictionary is created from the utterance bilingual corpus in advance. Also in this case, the utterance sentence example is divided for each phrase in consideration of the fact that some words are erroneous or omitted, and rules within phrases and dependency rules between phrases are created. First, the morphological analysis unit 360 performs a morphological analysis of the source language sentence and the target language sentence, and converts each sentence into a morphological sequence. Next, the phrase determination unit 361 divides the morpheme examples of the source language and the target language into phrases, and creates rules within the phrases and rules for dependency between the phrases. At this time, the phrase unit is determined manually by taking into account that the phrase unit is a partial sentence whose correspondence is clear in the bilingual translation, in addition to a semantically unitized unit. For example, "I'd like to request a room reservation""I'd like toreserve a
(a) "Reserving a room" and "reserv
ea room ", (b)"I'd like to ask, ""I'd lik
e to "(a) (b)
Dependencies such as "(b) to (a)" and "(b) to (a)" are regularized. The bilingual phrase is a bilingual phrase dictionary 362
In the meantime, translations between phrases in a bilingual form are stored in the inter-phrase rule table 363. Such processing is performed for all utterance sentences included in the bilingual corpus. Since the phrase division and dependency depend on factors such as the semantic information of the sentence and the degree to which the sentence is not broken grammatically, it is difficult to automatically determine each sentence. Has been determined.

[0008]

However, in the sentence analyzing means in the first conventional example, the phrase handled is a language-dependent phrase that depends only on the source language.
It often does not match the phrase unit of the target language. Therefore, there is a problem in that even if a correct phrase is input to the language conversion unit in the source language, it is often not accepted after all. Although the framework of the first conventional example is a framework that can be used even using language-independent phrases, it is necessary to manually create an analysis of language-independent phrases, and development takes time. A new problem arises in that the fluctuations in the manual production standards distort the rule performance.

Further, in the method of creating a bilingual phrase dictionary in the second conventional example, since there is no means for automatically analyzing the semantic information and grammatical information of an utterance sentence, it must be manually created. For this reason, there is a problem that development takes time, and the fluctuation of the creation standard manually distorts the rule performance. For example, if the target task of the interpreter is changed, or if the language type of the source language and the target language is changed, the rules that have been constructed cannot be applied and rules must be created from the beginning. The development efficiency is poor and it takes time.

The phrase dictionary 362 and the inter-phrase rule 363 determine the phrase unit with emphasis on the correspondence of the bilingual corpus, and determine whether the phrase unit is appropriate for the speech recognition unit 364 to recognize. Has not been evaluated. It is difficult to determine the phrase unit while manually determining whether the phrase is appropriate for speech recognition, and there is no guarantee that the recognition rate can be ensured when using the determined phrase for recognition. I have.

An object of the present invention is to solve the above-mentioned problems and to make it possible to always convert an input speech sentence to a target language even if there is an unlearned part or a partial error in speech recognition. An object of the present invention is to provide a language conversion device that can automatically create a phrase dictionary and an inter-phrase rule required for conversion with as little effort as possible.

[0012]

In order to solve the above-mentioned problems, a first aspect of the present invention (corresponding to claim 1) is to provide a sentence (hereinafter referred to as an original) to be subjected to language conversion inputted by voice or text. A sentence, which is called a language sentence, a sentence that is correspondingly converted in language is called a target language sentence) and a learning database (hereinafter, referred to as a bilingual corpus) in which the target language sentence is paired. Storage means for storing language rules obtained by learning grammatical or semantic constraint rules; and performing speech recognition of input speech using the stored language rules, and performing recognition on a sentence to be subjected to language conversion. And a language conversion unit that converts a sentence to be subjected to language conversion into a language-converted sentence using the same language rule as used in the voice recognition unit. It is a language conversion device characterized by the following.

Further, a second aspect of the present invention (corresponding to claim 2)
Is that the language rule divides a sentence to be subjected to language conversion and a converted sentence into a part (referred to as a form-independent phrase) that forms a semantic unit, and The language converter according to the first aspect of the present invention is characterized in that the language rule and the language rule between the form-independent phrases are separately ruled and made.

A third aspect of the present invention (corresponding to claim 3)
Wherein the language rule is created by regularizing co-occurrence or connection between the grammatical or semantic rule in the type-independent phrase and the type-independent phrase. A language conversion device according to the invention.

A fourth aspect of the present invention (corresponding to claim 4)
The speech conversion device according to the first invention, further comprising: a speech synthesis unit for speech-synthesizing the language-converted sentence using the same language rules as those used in the language conversion unit. It is.

A fifth aspect of the present invention (corresponding to claim 5)
Is a language rule group in which language rules having the same target language sentence are grouped into the same category among the language rules. Rule calculating unit for calculating a distance between rules, and creating an optimum rule for optimizing the rule group by merging language rules having the calculated distances close to each other to increase the recognition level of speech recognition A language conversion device according to any one of the first to fourth inventions, comprising:

The sixth invention (corresponding to claim 6)
Calculates the adjacency frequency of words or parts of speech in the bilingual corpus and the source language sentence and target language sentence in the bilingual corpus, and connects the frequently-used words and parts of speech to form a semantic unit (hereinafter referred to as a semantic unit). And a phrase extracting unit for extracting corresponding phrases from the phrase extracted by the phrase extracting unit, and determining a corresponding phrase by examining a relationship between phrases in a source language and a target language. And a phrase dictionary for storing the corresponding corresponding phrase, the phrase dictionary is used when performing language conversion, and the language conversion is performed when an input sentence is input when a source language sentence is input. A language conversion rule creating apparatus for performing language or stylistic conversion by collating with the corresponding phrase stored in a phrase dictionary. A.

The seventh invention (corresponding to claim 7)
The phrase conversion unit according to the sixth aspect of the present invention, wherein the phrase determination unit determines a corresponding phrase by examining a co-occurrence relationship between phrases in a source language and a target language.

An eighth aspect of the present invention (corresponding to claim 8)
Is a morphological analyzer for converting the source language sentence of the bilingual corpus into a word string, and using the result of the morphological analyzer, replacing some or all of the words in the source language sentence and the target language sentence with a part of speech name. The language conversion apparatus according to a sixth aspect of the present invention, further comprising a part-of-speech conversion unit that creates a bilingual corpus, wherein the phrase extraction unit extracts a phrase from the bilingual corpus that is part-of-speech converted by the part-of-speech conversion unit. It is a rule making device.

A ninth aspect of the present invention (corresponding to claim 9)
Has a bilingual word dictionary of a source language and a target language, and the part-of-speech unit converts the words associated with the bilingual word dictionary and words whose source language is a content word into parts of speech. A language conversion rule creating device according to an eighth aspect of the present invention.

According to a tenth aspect of the present invention (corresponding to claim 10), a morphological analyzer for converting a source language sentence of the bilingual corpus into a word string, and a semantic analysis using the result of the morphological analyzer. Based on a table in which similar words are regarded as being in the same class and the words are classified into the same class, and words in the same class are given the same code (hereinafter referred to as “classified vocabulary table”), part or all of the source language sentence and the target language sentence Further comprising a semantic encoding unit that creates a bilingual corpus in which the words of the words are replaced by the codes of the classification vocabulary table, and the phrase extracting unit extracts a phrase from the bilingual corpus replaced by the code in the semantic encoding unit. 7. The language conversion rule creating device according to claim 6, wherein:

The eleventh invention (corresponding to claim 11) has a bilingual word dictionary of a source language and a target language, and the semantic coding unit is associated with the bilingual word dictionary. A first feature in which only words are converted into meaning codes.
0 is a language conversion rule creation device according to the invention.

According to a twelfth aspect of the present invention (corresponding to claim 12), the phrase extracting unit stores in advance a word or a part-of-speech sequence that is to be preferentially regarded as a phrase by pairing the source language and the target language. A sixth aspect of the present invention is a language conversion rule creating apparatus according to the sixth aspect, wherein a phrase is extracted by also using a phrase definition table.

According to a thirteenth aspect of the present invention (corresponding to claim 13), a sentence complexity calculator for calculating perplexity (sentence complexity) of a corpus is provided. Nos. 6 to 13 characterized in that phrases are extracted using the class adjacency frequency and the sentence complexity.
A language conversion rule creation device according to any one of the inventions.

A fourteenth aspect of the present invention (corresponding to claim 14) relates to a language conversion device or a language conversion rule creating device according to any one of the first to thirteenth aspects. A program recording medium storing a program for executing a function in a computer.

[0026]

Embodiments of the present invention will be described below with reference to the drawings.

(First Embodiment) First, a first embodiment will be described.

In the first embodiment, as an example of the language conversion device, an explanation will be given using an interpreter device for performing conversion between different languages as in the conventional example. FIG. 1 is a block diagram of the interpreter of the present embodiment.

In the interpreting apparatus of this embodiment, first, before interpreting, the language analyzing unit 2 uses the learning database 1 having a bilingual corpus, a bilingual word dictionary, etc., in advance, for the source language of the utterance sentence and the target language. Learn language rules. FIG. 3 shows an example of learning a language rule.

The language rule creator 2 converts the content words of the source language sentence and the target language sentence into parts of speech using, for example, a bilingual corpus to which a part of speech tag is attached. Further, when the phrase in the source language and the phrase in the target language correspond as a group, the group is delimited as a body-type independent phrase. That is, if the body-dependent phrase in the source language and the body-dependent phrase in the target language correspond as a unit,
The set is defined as the boundary of the body-type independent phrase. If the target language's body-dependent phrases corresponding to the source language's body-dependent phrases do not correspond as a group, the body-type dependent phrases are linked and the phrase boundaries are corrected until the corresponding parts exist as a unit, and the body shape is independent. A phrase. In Figure 3, the text of the bilingual corpus "I want to reserve a room tonight""I'd like to room
-reservation tonight "26
In this case, the part-of-speech is expressed as "<normal noun> | <normal noun><sa-variantnoun> | "<Ordinary noun>", "<ordinary noun> of <ordinary noun>",
The boundaries are separated as a body-independent phrase such as "I want to do it." Next, for each type-independent phrase, the mixed sequence of part-of-speech and words, the word name of the part represented by the part-of-speech, and the frequency of appearance of each type-independent phrase in the bilingual corpus are determined according to rule 3 in the type-independent phrase.
Described as Write the above rules for the entire text of the bilingual corpus. In FIG. 3, the above-mentioned contents are described in 3 by the description 31 of the rule in the phrase. In 3 of FIG. 3, rule 1 is that Japanese is “<ordinary noun>”
English is "<noun>". The parts of speech are "tonight" for Japanese and "tonight" for English. If it appears in the bilingual corpus, "tomorrow", "tomorrow", etc. are also described in rule 1.

Further, the co-occurrence relation of the rules within each phrase is described as a rule 4 between body type independent phrases. For example,
When regularizing the co-occurrence relationship as a phrase bi-gram,
The adjacent frequency of each type-independent phrase is described.

The above description means that the description 32 of the rule between phrases describes 28 in FIG.
28 is an example of the phrase bi-gram. The rule number pair is, for example, “(rule 1) (rule 2)”, and its appearance frequency is 4. This means that in the process of learning from the bilingual corpus, rules 1 and 2 appear four times in the sentence. In the example of Rule 28, Rule 2 and Rule 3 appear six times in a sentence.

Further, the syntactic structure between each type-independent phrase is also described in rule 4 between type-independent phrases. This means that the description 32 of the rule between phrases describes 29 in FIG. That is, the description 32 of the rules between phrases
However, since the order in which the body-type-independent phrases appear in Japanese and English is different, the language structure is made into a tree shape at 25 in order to establish a correspondence in order relation.

The sentence generation rule 5 describes a target language rule that is lacking in the language rules 3 and 4. For example, in the case of Japanese-English translation, articles and indefinite article rules, third person singularization rules, and the like are described as the contents.

The intra-phrase language rule 3 and / or the inter-phrase language rule 4 are examples of storage means of the present invention.

At the time of interpreting, first, the uttered source language voice is input from the microphone 6 and input to the voice recognition unit 7. In the speech recognition unit, for example, a mixed sequence of parts of speech and words described as a body-independent phrase language rule 3 and a phrase as a body-independent phrase language rule 4
With the bi-gram, recognition word candidates are predicted sequentially in time series. An acoustic score and a phrase bi-g based on the distance value between the previously learned acoustic model 8 and the input speech
The sum of the ram and the language score is used as the recognition score, and Nbest-s
A continuous word string that is a recognition candidate is determined by earch.
The continuous word string determined in this way is input to the language conversion unit 9. Intra-phrase language rule 3, Inter-phrase language rule 4
In, the source language and the target language are ruled in advance while corresponding to each other. The language conversion unit 9 converts the continuous word string into a phrase string in the target language using the above rules and outputs the converted word string. At this time, if the input source language phrase sequence applies to the syntax structure between the phrases that have already been learned, the target language phrase sequence is output after being modified along the syntax structure.

The output target language sentence is input to the output sentence generator 10 and corrects grammatical unnaturalness. For example, the addition of definite or indefinite articles, optimization of pronouns and verbs such as third personization, pluralization, and past tense are performed. The corrected target language translation result sentence is output, for example, as text.

In the above embodiment, when learning the language rules used in speech recognition, the source language and the target language are ruled in units of a unit having both meanings. By performing recognition based on the constraints of the above, it is possible to solve the problem that even if there is an unlearned part in the input speech sentence or a part of the speech recognition makes an error, the translation result for the entire sentence is not output at all and correct recognition A language conversion device that can output an appropriate translation result can be realized for the part that has been set.

In this embodiment, an interpreter has been described as an example of a language conversion apparatus. However, this is not the case with other language conversion apparatuses, for example, in which a spoken utterance is converted into a text sentence such as a written word. The same can be used in a language conversion device for converting to.

(Second Embodiment) Next, a second embodiment will be described with reference to the drawings. In this embodiment, as in the first embodiment, description will be made using an interpreter. FIG. 2 is a block diagram of the interpreter of the present embodiment.

In the interpreting apparatus of this embodiment, first, before interpreting, the language rule creating section 11 reads the phrases of the source language and target language of the uttered sentence from the learning database 1 having a bilingual corpus and a bilingual word dictionary in advance. The internal language rules 12 and the inter-phrase language rules 13 are learned. The rules to be learned are the same as those for the language rules in the first embodiment. Next, the learned language rules are optimized. Figure 4 shows an example of optimization
Shown in

First, in the learned body type independent phrases, phrases having the same target language phrase are put together in the same category. In FIG. 4, reference numeral 12 denotes a language rule. Rules 1, 2, and 3 are in the same category because the target language rule is the same as "I'd like to". Rule 4 is that the target language rule is “please”
Therefore, the rule is classified into a different category from the rules 1, 2, and 3. Next, the acoustic distance between the source language phrases included in the same category is calculated by the rule distance calculating unit 14. In FIG. 4, reference numeral 15 denotes an example of calculating the acoustic distance between source language phrases. In Rule 15, the distance between Rule 1 and Rule 2 is 7, and the distance between Rule 1 and Rule 3 is 2
It has become.

The acoustic distance of the source language phrase in the category rule is calculated as follows. First, the same word is applied to the part-of-speech part of the mixed sequence in all target language phrases in the category if the part-of-speech is the same, and all the mixed sequences are converted into a word sequence. Next, in order to check whether the pronunciation of each word string is similar, the distance to the difference in the character string of each word string is calculated using (Equation 1) and described in the inter-rule distance table 15. Phrase X = [x composed of n words
1, x2, x3, ... xn] (x is each word) and the phrase Y = [y1, y2, y3, .. ym] consisting of m words is D (Xn, Ym) As

[0044]

(Equation 1)

Next, in the optimum rule creating section 16, only the rule having the largest number of occurrences of the phrases whose distance value is within a certain value is left, and the other rules are deleted. For example, in the example of FIG. 4, when the above-mentioned constant value is set to 2, in 33, the distance between rules 1 and 3 in the same category is 2 and is equal to or less than the above-mentioned constant value 2. Therefore, the rule 1 having the higher appearance frequency of the two rules is adopted, and the rule 3 is deleted from the rules. The number of appearances is rewritten accordingly.

After the above-mentioned optimal rule is applied to all rules written in the intra-phrase language rules 12, only the language rules that have not been deleted are stored as the intra-phrase optimal language rules 17. In accordance with the optimized rule, the rule between the phrases 13 is rewritten with the rule that has been removed, and the number of appearances is also corrected. In FIG. 4, rule 3 is deleted by the optimum rule creation 16, and rule 1 is unified as rule 1. Accordingly, the number of appearances of Rule 1 is 15, which is the sum of Rule 3 and Rule 3, which has been deleted, such as 17.

The sentence generation rule 5 describes a target language rule that is lacking in the language rule created from the corpus. For example, in the case of Japanese-English translation, articles and indefinite article rules, third person singularization rules, and the like are described as the contents.

At the time of interpretation, first, the uttered source language voice is input from the microphone 6 and input to the voice recognition unit 7. In the speech recognition unit, for example, a mixed sequence of part-of-speech and column words described as body-independent phrase-independent language rules 17 and phrase adjacency frequency as body-independent phrase-to-phrase interlanguage rules 18 are sequentially used in time series. Recognized word candidates are predicted. The sum of the acoustic score based on the distance value between the previously learned acoustic model 8 and the input speech and the linguistic score based on the phrase bi-gram is used as the recognition score.
A continuous word string that is a recognition candidate is determined by best-search. The continuous word string determined in this way is converted to a language conversion unit 9.
Is input to In the language rules 17 and 18, the source language and the target language are ruled in advance while corresponding to each other. The language conversion unit 9 converts the continuous word string into a phrase string in the target language using the above rules and outputs the converted word string. At this time, if the input source language phrase sequence applies to the syntax structure between the phrases that have already been learned, the object phrase sequence is output after being modified along the syntax structure.

The output target language sentence is output from the output sentence generator 10.
To correct grammatical unnaturalness. For example,
The addition of definite articles and indefinite articles, optimization of pronouns and verbs such as third personization, pluralization, and past tense are performed. The corrected target language translation result sentence is output, for example, as text.

In the above embodiment, when learning the language rules used in the speech recognition, the source language and the target language are ruled in units of a unit having both meanings. If the source language phrases that have the same regularized target language part are acoustically similar,
By adopting only the most frequently occurring rules from among similarities and eliminating the remaining rules, it is possible to reduce the number of rules by using a body-independent phrase as a unit without reducing the performance of language rules as much as possible. An object of the present invention is to realize an interpreting device that enables high-performance recognition and language conversion.

In the present embodiment, an interpreter has been described as an example of a language conversion apparatus. However, this is not the case with other language conversion apparatuses, for example, in which a spoken utterance is converted into a text sentence such as a written word. The same can be used in a language conversion device for converting to.

(Embodiment 3) In the present embodiment, as an example of a language conversion device, an interpreter that performs conversion between different languages will be described as in the conventional example. FIG. 5 is a block diagram of the interpreter of the present embodiment.

In this embodiment, the bilingual corpus 101, the content word definition table 103, the bilingual word dictionary 107, the morphological analysis unit 102, the part-of-speech unit 104, the phrase extraction unit 1
05, the phrase determination unit 106, the bilingual phrase rule table 108, and the bilingual phrase dictionary 109 are examples of the language conversion rule creation device of the present invention. Further, the bilingual phrase dictionary 109 of the present embodiment is an example of the phrase dictionary according to claim 6 of the present invention.

In the translator according to the present embodiment, first, before interpreting, the morphological analyzer 102 performs a morphological analysis of the source language sentence in the bilingual corpus 101, so that only the source language sentence is given a part-of-speech tag. Create For example, FIG.
In the example of the utterance sentence of “I would like to request a reservation for a room” of 120, a part of speech tag like 121 is given to the source language sentence. Next, the part-of-speech conversion unit 104 creates a part-of-speech bilingual corpus in which part of the word names are replaced with part-of-speech names in the source language sentence with the part of speech tag of the corpus. At this time, the word converted to the part of speech name satisfies the following conditions. (1) A word corresponding to the part of speech described in the content word table. (2) A word registered in the bilingual word dictionary and corresponding to the target language translation of the bilingual word dictionary exists in the corresponding target language bilingual sentence in the corpus.

In the example of the content word definition table 103 shown in FIG. 6, among the common nouns, the inflectional nouns, and the verbs described in the content word table, “room” registered in the bilingual word dictionary 107
And only “reservation” are made into parts of speech, and a corpus is created as shown in 122 by replacing these words with parts of speech. Further, the word name in the corresponding target language bilingual sentence is also replaced with the Japanese part of speech name as 123.

Next, with respect to the corpus in which some of the above content words have been replaced with part of speech names, the phrase extraction unit 105
Calculates the two-chain appearance frequency (hereinafter referred to as bi-gram) of each word or part of speech for the source language sentence and the target language sentence separately. The calculation formula is shown in (Equation 2).

[0057]

(Equation 2)

After calculating bi-grams for all source language sentences and target language sentences in the corpus, the phrase extracting unit 5 regards the two most frequently occurring words or parts of speech as one word and connects them. Then, the bi-gram is calculated again. As a result, for example, frequently-adjacent word pairs such as “O”, “Wish”, “Wish”, “Shi”, “Shi” and “Masu” are connected,
A phrase candidate "Please, please" is formed. In the target language, word pairs "I'd", "like", "like", and "to" are connected. The above connection and bi-gram calculation are repeated for all source language sentences and target language sentences separately until all the values of the bi-gram do not exceed a certain threshold. Then, individual words including the connected words are extracted as phrase candidates.

Next, the phrase determination unit 106 calculates the frequency at which each phrase appears simultaneously in the source language sentence and the target language sentence pair. The i-th source language phrase is J [i],
Assuming that the j-th target language phrase is E [j], the co-occurrence frequency K [i, j] of the phrases J [i] and E [j] is calculated by Expression (3).

[0060]

(Equation 3)

For example, in the example of FIG. 7, of the three bilingual sentences 130 described as a phrase string, the source language phrase “Please please” and the target language phrase “I'd like”
The co-occurrence frequency of "to" is 2 / (2 + 3), and the co-occurrence frequency of "I want to do it" is 1 / (1 + 3). A phrase pair whose frequency is equal to or more than a certain value is determined as a bilingual phrase, and a phrase number is added along with the frequency and registered in the bilingual phrase dictionary 109. Further, among the phrase candidates that have not been determined as a bilingual phrase, words that have already been converted to part of speech are registered in the bilingual phrase dictionary 109 as bilingual phrases by themselves. Other parts are registered in the phrase dictionary as pairs of each word string in the bilingual pair.

For example, in the example shown in FIG. 7, it is registered in the bilingual phrase dictionary 109 as 131.

After the phrase registration is performed in this way, the phrase numbers that co-occur in one sentence are recorded and registered in the bilingual phrase rule table 108 as a pair of phrase numbers. In the example of FIG.

The phrase bi-g of the above phrase number pair
The ram is obtained, and this is also recorded in the bilingual phrase rule table 108. That is, the source language corpus is represented by a phrase number string registered in the bilingual phrase dictionary, a phrase bi-gram is obtained using the corpus represented by the phrase number, and this is also recorded in the bilingual phrase rule table 8. The phrase bi-gram representing the probability of occurrence of the phrase j following the phrase i is represented by (Equation 4).

[0065]

(Equation 4)

For example, at 132 in FIG. 7, the phrase bi-gram of phrase 3 and phrase 1 is determined. In addition, regarding the rules between phrases 4, 5 and 2, the bi-grams of phrases 4, 5 and 5 and 2 are obtained and recorded in the bilingual rule table 108.

At the time of interpretation, first, the uttered source language voice is input to the voice recognition unit 110. The speech recognition unit 113 sequentially recognizes the recognized words in a chronological order using, for example, a network of words described as a phrase in the bilingual phrase dictionary 109 and a phrase bi-gram described in the bilingual rule table 108. Candidates are predicted.
The sum of the acoustic score based on the distance value between the acoustic model 113 and the input speech that has been learned in advance and the language score based on the phrase bi-gram is used as the recognition score, and a continuous word sequence that is a recognition candidate is determined by Nbest-search. Is done.

The recognized continuous word string is sent to the language conversion unit 11.
1 is input. The language conversion unit 111 converts the input continuous word string into a phrase string in the bilingual phrase dictionary 109, and searches for an inter-phrase rule corresponding to each phrase string. Then, the input source language recognition result sentence is converted into the target language sentence based on the target language phrase that is a bilingual translation of each phrase and the inter-phrase rule of the target language.

As described above, in the present embodiment, both the speech recognition unit 110 and the language conversion unit 111 use the bilingual phrase dictionary 109 and the bilingual phrase rule table 108.

The converted target language sentence is output from the output sentence generating unit 11.
2 to correct syntactic unnaturalness. For example, the use of definite or indefinite articles, pronouns, and verbs
Optimization such as personification, pluralization, and past tense are performed. The corrected target language translation result sentence is output, for example, as text.

In the above embodiment, rules are described in such a manner that the source language phrase and the target language phrase correspond to each other, and recognition is performed in units of this phrase, so that a part of the input sentence is an unknown partial sentence. Even if there is some or incorrect speech recognition, the correctly recognized and analyzed part allows the language converter to be properly processed and output. In addition, the bilingual phrase and inter-phrase rules are automatically determined by using the co-occurrence relationship between the frequency of adjacent words or parts of speech in the source language sentence and the target language sentence, and the frequently occurring word strings or part of speech strings in the parallel translation. By translating using a bilingual phrase rule, a language rule creating device that can automatically and efficiently generate a high-quality bilingual phrase dictionary with as little effort as possible is made possible.

In this embodiment, an interpreter has been described as an example of a language conversion apparatus. However, this is not the case with other language conversion apparatuses, for example, in which a spoken utterance is converted into a text sentence such as a written word. The same can be used in a language conversion device for converting to.

(Embodiment 4) In this embodiment, as in the third embodiment, an interpreter for performing conversion between different languages will be described as an example of a language converter. FIG. 8 is a block diagram of the interpreter of the present embodiment.

In this embodiment, the bilingual corpus 101, the content word definition table 103, the bilingual word dictionary 107, the morphological analysis unit 102, the part-of-speech unit 104, the phrase extraction unit 1
42, the phrase determination unit 143 includes a bilingual phrase inter-rule table 145, a bilingual phrase dictionary 144, and a phrase definition table 14.
1 is an example of a language conversion rule creation device of the present invention. Further, the bilingual phrase dictionary 144 of the present embodiment is an example of the phrase dictionary according to claim 6 of the present invention.

The translator according to this embodiment creates a bilingual corpus to which a part-of-speech tag is attached after morphological analysis, as in the third embodiment, before interpreting.

Next, the phrase extracting unit 142 connects words or parts of speech corresponding to the rules in accordance with a phrase definition table 141 in which words or parts of speech that are desired to be extracted as phrases are described in a regularized manner. For example, in the example 141 of FIG. 9, "+ (verb) + tai" is concatenated as a word by rules such as "verb + auxiliary verb" and "case particle + verb". In this way, for the corpus in which some of the content words described above are replaced with part-of-speech names and the words or part-of-speech strings are concatenated and regarded as one word, the source language sentence and the target language sentence are separately Calculate the frequency of two-chain occurrences of a word or part of speech (hereinafter referred to as bi-gram). The calculation formula is (Equation 2)
Is the same as

Further, the processing is repeated in the same manner as in the third embodiment until the values of all the bi-grams do not exceed the fixed threshold. Then, individual words including the connected words are extracted as phrase candidates, and the phrase determination unit creates a bilingual phrase dictionary 144 and a bilingual phrase rule table 145 as in the third embodiment. 9 is an example of a corpus in which words or parts of speech are linked according to the phrase definition table 141, and an example of a bilingual phrase dictionary 144 in which 152 is created.

The operation at the time of interpretation is the same as that of the third embodiment.

In the above embodiment, after the words or parts of speech are linked according to the rules of the words or parts of speech that are to be regarded as predefined phrases, the adjacent frequencies of the words or parts of speech in the source language sentence and the target language sentence are determined. By automatically determining the bilingual phrases and inter-phrase rules using the co-occurrence relationship between frequent word strings or part-of-speech strings in bilingual translation, and performing language or stylistic conversion using these bilingual phrase rules, It is possible to provide a language conversion rule creation device that can more efficiently generate a high-quality bilingual phrase dictionary within a range that is minimized.

The bilingual phrases of the present embodiment are examples of the corresponding phrases of the present invention.

Further, in the present embodiment, an interpreter has been described as an example of the language conversion apparatus, but this is not the case with other language conversion apparatuses, for example, an unfoldable utterance sentence is written in a text sentence such as a written word. The same can be used in a language conversion device for converting to.

(Embodiment 5) In the third embodiment,
When constructing linguistic rules, some of the words in the corpus are converted to parts of speech to achieve more general and high-quality rules, but semantic coding instead of part of speech is also used. The effect can be expected. This embodiment will be described below with reference to FIG. Also in the present embodiment, description will be made using an interpreter that performs conversion between different languages.

In this embodiment, the bilingual corpus 201, the classified vocabulary table 216, the bilingual word dictionary 207, the morphological analysis unit 202, the semantic coding unit 215, the phrase extracting unit 205, and the phrase determining unit 206 The inter-rule table 208 and the bilingual phrase dictionary 209 are examples of the language conversion rule creation device of the present invention. Further, the bilingual phrase dictionary 209 of the present embodiment is an example of the phrase dictionary according to claim 6 of the present invention.

In the interpreter of this embodiment, as in the third embodiment, the morphological analysis unit 202 performs a morphological analysis of the source language sentence in the bilingual corpus 201 to give a part-of-speech tag to the source language sentence. Next, in the semantic encoding unit 215, in the morpheme sequence of the source language sentence, each morpheme and the classification vocabulary table 216
Is compared with the words written in the vocabulary table 216, and for the morphemes that match the words to which the semantic codes are given in the classification vocabulary table 216, the morpheme names are replaced with the semantic codes, so that some of the morphemes are Convert to semantically encoded morpheme sequence. At this time, the morphemes to be semantically encoded satisfy the following conditions. (Condition) A word registered in the bilingual word dictionary and corresponding to the target language translation of the bilingual word dictionary exists in the corresponding target language bilingual sentence in the corpus.

In the example of FIG. 11, only “room” and “reservation” which are registered in the bilingual word dictionary and are given codes in the classification vocabulary table are converted into meaning codes. A morpheme string replaced with the semantic code is created. Further, the word name in the corresponding target language bilingual sentence is also replaced with a semantic code like 2133.

Next, with respect to the corpus in which some of the content words have been replaced with meaning codes, the phrase extraction unit 20
In 5, the two-language appearance frequency of each word or meaning code is calculated separately for the source language sentence and the target language sentence. Equation (5)
Shown in

[0087]

(Equation 5)

After calculating bi-grams for all source language sentences and target language sentences in the corpus, the phrase extraction unit regards the two words or semantic code pairs that appear most frequently as one word and connects them. Then, the bi-gram is calculated again. As a result, for example, "O""Wish"
Word pairs such as "Wish", "Shi", "Shi", and "Masu" are concatenated to form a phrase candidate "Please". In the target language, word pairs "I'd", "like", "like", and "to" are connected.

The above connection and bi-gram calculation are repeated for all source language sentences and target language sentences separately until all of the bi-gram values do not exceed a certain threshold. Then, individual words including the connected words are extracted as phrase candidates.

Hereinafter, similarly to the third embodiment, the translated phrase is determined by the phrase determining unit 206 and registered in the translated phrase dictionary 209. Further, an inter-phrase language rule and a phrase bi-gram are created and registered in the bilingual inter-phrase rule table 208 as in the third embodiment.

At the time of interpretation, the operation is the same as in the third embodiment.

In the above embodiment, rules are described in such a manner that the source language phrase and the target language phrase correspond to each other.
By performing recognition in units of this phrase, even if a part of the input sentence is an unknown sub-sentence or a part of the speech recognition is incorrect, the correctly recognized and analyzed part is appropriately processed and output. Enables a language translator. In addition, the bilingual phrase and inter-phrase rules are automatically determined using the co-occurrence relationship between the frequency of adjacent words or meaning codes in the source language sentence and the target language sentence and the frequently occurring word strings or meaning code strings in the bilingual translation. By performing translation using this bilingual phrase rule, a language rule creating device capable of automatically and efficiently generating a high-quality bilingual phrase dictionary with minimal human intervention becomes possible.

In the present embodiment, an interpreter has been described as an example of the language conversion apparatus. However, this is not the case with other language conversion apparatuses, for example, in which a spoken utterance sentence is converted into a text sentence such as a written word. The same can be used in a language conversion device for converting to.

(Embodiment 6) In the fifth embodiment,
When building linguistic rules, phrases were created by linking frequently adjacent words or parts of speech and semantic codes.
After you create a phrase, you can evaluate sentence complexity,
It is possible to form a phrase that has higher quality and can guarantee the recognition rate.

An embodiment of the language conversion rule creating device will be described below with reference to FIG.

The bilingual phrase dictionary in the present embodiment is an example of the phrase dictionary according to claim 6 of the present invention.

As in the previous embodiment, after morphological analysis, a semantic encoding unit 213 creates a bilingual corpus in which some morphemes are converted into semantic codes. Further, the phrase extraction unit calculates a bi-gram of each word or meaning code separately for the source language sentence and the target language sentence. The calculation formula is the same as (Equation 5).

Further, the processing is repeated in the same manner as in the previous embodiment until all the values of the bi-gram do not exceed the fixed threshold. Then, individual words including the connected words are extracted as phrase candidates.

When performing the above processing, the sentence complexity calculation unit 2
At 18, a bi-gram of each word or meaning code is calculated, and b
When performing the connection process based on the value of the i-gram, the sentence complexity between the case where each word pair is connected and the case where each word pair is not connected is calculated and compared. The sentence complexity is calculated by (Equation 6).

[0100]

(Equation 6)

As a result of the comparison, the phrase extraction unit 217 that removes from the phrase candidates those sentences which increase the sentence complexity by linking each word or meaning code.

For the phrases remaining as the phrase candidates in the above processing, the phrases are determined under the same conditions as in the previous embodiment, and the bilingual phrase dictionary 209 and the inter-phrase rule table 208 are determined.
To determine.

In the above-described embodiment, when a bilingual phrase is determined, the bilingual phrase is automatically extracted from the corpus by using the sentence complexity of the bilingual corpus that has been word-classified by the semantic code. This makes it possible to efficiently generate a high-quality bilingual phrase dictionary without using human resources as much as possible. Also, the measure of sentence complexity is
Since it is closely related to a measure of whether a phrase is appropriate for speech recognition, it is possible to automatically extract a phrase while guaranteeing recognition accuracy.

In the present embodiment, an example has been described in which phrases are extracted by using a corpus in which some words are converted into meaning codes. However, similar effects are expected when phrases are extracted by using a corpus in which parts of speech are converted. it can.

Further, in the fourth embodiment, an example has been described in which a phrase is extracted from a phrase definition table by using a bilingual corpus to which a part-of-speech tag is attached. A similar effect can be expected even when a phrase is extracted from a phrase definition table by using a corpus in which a part of the word is converted into a meaning code.

Further, in the first to fifth embodiments, the language conversion device has been described as including the speech recognition unit, the language conversion unit, and the output sentence generation unit, but the present invention is not limited to this. FIG.
As shown in (1), a speech synthesis unit that performs speech synthesis on the translation result sentence output by the output sentence generation unit 212 may be provided. Then, the speech synthesizing unit performs the speech recognition
0, speech synthesis is performed using the same bilingual phrase rule table 208 and bilingual phrase dictionary 209 used by the language conversion unit 211. This solves the problem that even if there is an unlearned part in the input speech sentence or a part of speech recognition causes an error, the speech synthesis result for the whole sentence is not output at all. It can be expected that an appropriate sound can be output.

Furthermore, all or some of the functions of each component of the language conversion device or language conversion rule creation device of the present invention may be realized using dedicated hardware.
Further, it may be realized by software by a computer program.

Further, there is also provided a program recording medium storing a program for causing a computer to execute all or a part of the functions of each component of the language conversion device or language conversion rule creation device of the present invention. It belongs to the present invention.

[0109]

As is clear from the above description, the present invention can always output a recognition result which can be converted into a target language sentence, and therefore, a part of an input sentence is an unknown partial sentence or a speech recognition Even if a part is incorrect, it is possible to provide a language conversion rule creating device and a language conversion device that enable a correctly recognized and analyzed part to be appropriately processed and output.

Further, according to the present invention, even if there is an unlearned part in an input speech sentence or a part of speech recognition causes an error, it is possible to convert only a part which is correctly recognized and to which an appropriate analysis rule is applied. It is possible to provide a language conversion rule creation device and a language conversion device that can always output a partial conversion result.

Further, the present invention can provide a language conversion rule creating apparatus which can automatically create a language rule with minimum effort.

Further, the present invention can provide a language conversion rule creating apparatus which can create a high-quality language rule automatically and more efficiently with as little labor as possible.

Further, according to the present invention, it is possible to provide a language conversion rule creating apparatus which enables automatic and efficient creation of high quality language rules.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a language conversion device according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a language conversion device according to a second embodiment of the present invention.

FIG. 3 is a view for explaining creation of a language rule according to the first embodiment of the present invention;

FIG. 4 is a view for explaining creation of an optimal language rule according to the second embodiment of the present invention;

FIG. 5 is a block diagram showing a configuration of a language conversion device and a language rule creation device according to a third embodiment of the present invention.

FIG. 6 is a view for explaining creation of a language conversion rule according to the third embodiment of the present invention;

FIG. 7 is a diagram showing an example of a bilingual phrase rule table and a bilingual phrase dictionary according to the third embodiment of the present invention.

FIG. 8 is a block diagram illustrating a configuration of a language conversion device and a language rule creation device according to a fourth embodiment of the present invention.

FIG. 9 is a diagram illustrating an example of a phrase definition table according to the fourth embodiment of the present invention.

FIG. 10 is a block diagram showing a configuration of a language conversion device and a language rule creation device according to a fifth embodiment of the present invention.

FIG. 11 is a diagram illustrating creation of a language rule according to a fifth embodiment of the present invention.

FIG. 12 is a block diagram illustrating a configuration of a language conversion rule creation device according to a sixth embodiment of the present invention.

FIG. 13 is a block diagram illustrating a configuration example of a language conversion device having a speech synthesis unit.

FIG. 14 is a diagram showing an example of a language rule used in a conventional language conversion device.

FIG. 15 is a block diagram showing a configuration of a conventional language conversion device.

[Explanation of symbols]

 Reference Signs List 1 bilingual corpus 2 language rule reproducing unit 3 language rule in phrase 4 language rule between phrases 5 sentence generation rule 6 microphone 7 speech recognition unit 8 acoustic model 9 language conversion unit 10 output sentence generation unit 101 bilingual corpus 102 morphological analysis unit 103 content word Definition table 104 Part-of-speech generation unit 105 Phrase extraction unit 106 Phrase determination unit 107 Bilingual word dictionary 108 Rules table between bilingual phrases 109 Bilingual phrase dictionary 110 Speech recognition 111 Language conversion 112 Output sentence generation 113 Sound model 114 Sentence generation rules

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G06F 15/38 Z

Claims (14)

[Claims] 1. A sentence to be subjected to language conversion input by voice or text (hereinafter, referred to as a source language sentence, a sentence correspondingly converted to a target language sentence is referred to as a target language sentence) and a target language sentence Storage means for storing language rules obtained by learning grammatical or semantic constraint rules for words or word strings from a learning database (hereinafter referred to as a bilingual corpus) paired with A speech recognition unit that performs speech recognition of the input speech using a language rule and outputs a recognition result in a sentence to be subjected to language conversion; and a language conversion using the same language rule as used in the speech recognition unit. A language conversion unit for converting a sentence to be converted into a language-converted sentence. 2. The language rule divides a sentence to be subjected to language conversion and a converted sentence into a part (referred to as a form-independent phrase) which forms a semantic unit together with the form-independent phrase. 2. The language conversion apparatus according to claim 1, wherein the language conversion apparatus is formed by dividing a language rule in a phrase and a language rule between the body-type independent phrases into rules. 3. The linguistic rule is created by regularizing co-occurrence or connection between a grammatical or semantic rule in the type-independent phrase and the type-independent phrase. The language conversion device according to claim 2, wherein 4. The language conversion apparatus according to claim 1, further comprising a speech synthesis section that performs speech synthesis on the language-converted sentence using the same language rule as that used in the language conversion section. apparatus. 5. A linguistic rule group in which linguistic rules having the same target language sentence are grouped into the same category among the linguistic rules is subjected to linguistic conversion of the linguistic rules included in the linguistic rule group. An inter-rule distance calculation unit that calculates an acoustic inter-rule distance of a sentence; and optimizes the rule group by merging language rules whose calculated distances are close to each other in order to increase the recognition level of speech recognition. The language conversion device according to any one of claims 1 to 4, further comprising an optimum rule creation unit. 6. A part that calculates the frequency of adjacent words or parts of speech in a bilingual corpus and a source language sentence and a target language sentence in the bilingual corpus, and connects the frequent words and parts of speech to form a semantic unit. A phrase extraction unit that extracts a sentence (hereinafter, referred to as a phrase); and a phrase determination unit that determines a corresponding phrase by examining a relationship between phrases in a source language and a target language in the phrases extracted by the phrase extraction unit. And a phrase dictionary that stores the determined corresponding phrase. The phrase dictionary is used when performing language conversion, and the language conversion is performed when a source language sentence is input. Language conversion characterized by performing language or style conversion by comparing a sentence with the corresponding phrase stored in the phrase dictionary Law creating apparatus. 7. The language conversion rule creating device according to claim 6, wherein the phrase determining unit determines a corresponding phrase by examining a co-occurrence relationship between phrases in a source language and a target language. 8. A morphological analysis unit for converting a source language sentence of the bilingual corpus into a word string, and using a result of the morphological analysis unit, a part or all of the words of the source language sentence and the target language sentence are part-of-speech names 7. The language conversion according to claim 6, further comprising a part-of-speech conversion unit that creates a bilingual corpus replaced by: a. Rule making device. 9. A part-of-speech word dictionary comprising a source language and a target language bilingual word dictionary, wherein the part-of-speech unit converts words that are associated with the bilingual word dictionary and whose source language is a content word. 9. The language conversion rule creating device according to claim 8, wherein: 10. A morphological analysis unit for converting a source language sentence of the bilingual corpus into a word string, and using the result of the morphological analysis unit to classify words by regarding semantically similar words as the same class, A bilingual corpus in which some or all of the words in the source language sentence and the target language sentence are replaced with the codes in the classification vocabulary table, based on a table in which the same codes are assigned to the words in the class (hereinafter referred to as the classification vocabulary table) 7. The language conversion rule creation according to claim 6, further comprising: a semantic encoding unit that creates a phrase; and wherein the phrase extracting unit extracts a phrase from the bilingual corpus replaced with a code by the semantic encoding unit. apparatus. 11. A bilingual word dictionary for a source language and a target language, wherein the semantic coding unit converts only words associated with the bilingual word dictionary into semantic codes. Language conversion rule creation device described. 12. The phrase extracting unit extracts a phrase by also using a phrase definition table in which a word or a part-of-speech sequence to be preferentially regarded as a phrase is stored in a pair of a source language and a target language in advance. 7. The language conversion rule creating device according to claim 6, wherein: 13. A sentence complexity calculation unit for calculating perplexity (sentence complexity) of a corpus, wherein the phrase extraction unit extracts a phrase using the adjacent frequency of a word or a word class and the sentence complexity. The language conversion rule creating device according to any one of claims 6 to 13, wherein the language conversion rule creating device performs extraction. 14. A program for causing a computer to execute all or a part of the functions of each component of the language conversion device or language conversion rule creation device according to any one of claims 1 to 13. A program recording medium characterized by the above-mentioned.