JP4103313B2 - Natural language processing apparatus and method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a natural language processing apparatus and method, and can be applied to, for example, a machine translation apparatus and method that performs machine translation using a translation pattern.
[0002]
[Prior art]
Machine translation devices based on translation patterns perform source language syntax analysis by applying source language patterns. Therefore, if there is no applicable source language pattern, the analysis fails, that is, the translation fails, and the translation result cannot be obtained.
[0003]
As a method for solving such inconvenience, Japanese Patent Laid-Open No. 5-290082 separately provides an analysis unit using grammar analysis in addition to an analysis unit using patterns. In addition, it is described that when an analysis that applies a pattern fails, a conventional method that does not use a pattern, that is, a source language is analyzed by a grammatical analysis.
[0004]
Also, an apparatus that takes a method of displaying an analysis failure and requesting the user to review the input sentence is also conceivable.
[0005]
[Problems to be solved by the invention]
However, the apparatus described in the above publication must be equipped with a plurality of analysis units, which can be said to be wasteful.
[0006]
Further, the device described in the above publication cannot enjoy the advantages of pattern-based translation processing, although it is a device that performs pattern-based translation processing. For example, pattern-based analysis can be performed even if it includes special expressions that cannot be obtained by grammatical analysis, even if the user can reflect their preferences for pattern analysis etc. However, it is impossible to enjoy such advantages.
[0007]
Furthermore, naturally, the method of causing the user to review the input sentence when the analysis fails causes a heavy burden on the user. Rather, the burden on the user is smaller when a reasonable translation result is presented.
[0008]
Therefore, even in the case where there is no applicable pattern in the translation processing based on the pattern, it is desirable to apply a pattern as much as possible and obtain a plausible translation result.
[0009]
The present invention has been made in consideration of the above points, and even when an analysis fails, the same analysis method can be applied to recover the failure without imposing a burden on the user. It is an object of the present invention to provide a processing apparatus and method.
[0010]
[Means for Solving the Problems]
  In order to solve such a problem, the first aspect of the present invention provides an input natural language sentence or a predetermined processing result for the natural language sentence.Analysis targetOn the other hand, in the natural language processing apparatus in which the analysis means performs analysis, (a) the analysis means uses at least the first analysis.Contains grammatically correct analysis informationBasic dictionary and (b) used by the above analysis means for analysis after analysis failureIt stores analysis information that can analyze the part of the analysis target that contains grammatical errors that humans are likely to fall intoOne or a plurality of analysis failure relief dictionaries, and (c) the basic dictionary as a dictionary used by the analysis meansOrThe above analysis failure relief dictionaryAt least one ofUse dictionary setting means to set(D) The use dictionary setting means sets the basic dictionary at the time of the initial analysis of the analysis means, sets the basic dictionary and the analysis failure relief dictionary at the time of analysis after the analysis failure, and (e ) When analyzing after analysis failure, the analysis means re-analyzes from the beginning using the basic dictionary and the analysis failure relief dictionary.It is characterized by that.
  According to a second aspect of the present invention, there is provided a natural language processing apparatus in which an analysis unit analyzes an input natural language sentence or an analysis target that is a predetermined processing result thereof. A basic dictionary storing grammatically correct analysis information for use in (b), and (b) an analysis target including grammatical errors that are likely to fall into humans, which is used in the analysis after the analysis failure. One or a plurality of analysis failure relief dictionaries storing analysis information capable of analyzing a part, and (c) at least one of the basic dictionary or the analysis failure relief dictionary as a dictionary used by the analysis means (D) the usage dictionary setting means sets the basic dictionary at the time of the first analysis of the analysis means, and at the time of analysis after the analysis failure, the basic dictionary and the analysis failure. Set the relief dictionary, the (e) above analyzing means, at the time of analysis after analysis fails, using the basic dictionary and the analysis fails relief dictionary, and executes the subsequent failed analyzed.
[0011]
  The second3The present invention provides an input natural language sentence or a predetermined processing result for the natural language sentence.Analysis targetWhereasPerformed by analysis meansIn the natural language processing method that performs analysis in the analysis process,(A)The above analysis process is used at least for the first analysisContains grammatically correct analysis informationThe basic dictionary and the above analysis process are used for analysis after analysis failureIt stores analysis information that can analyze the part of the analysis target that contains grammatical errors that humans are likely to fall intoOne or more analysis failure relief dictionaries, and(B)The basic dictionary as a dictionary used in the analysis stepOrThe above analysis failure relief dictionaryAt least one ofSet, Use dictionary setting meansA use dictionary setting process,(C) In the usage dictionary setting step, the usage dictionary setting means sets the basic dictionary at the time of the first analysis of the analysis step, and sets the basic dictionary and the analysis failure relief dictionary at the time of analysis after the analysis failure. (D) In the analysis step, the analysis means re-analyzes from the beginning using the basic dictionary and the analysis failure relief dictionary at the time of analysis after analysis failure.It is characterized by that.
  According to a fourth aspect of the present invention, in the natural language processing method for analyzing the input natural language sentence or an analysis target that is a predetermined processing result thereof in an analysis step executed by the analysis unit, (a) the analysis step A basic dictionary that stores grammatically correct analysis information that is used in at least the first analysis, and an analysis that includes grammatical errors that humans are likely to fall into. One or a plurality of analysis failure relief dictionaries storing analysis information capable of analyzing the target portion, and (b) the basic dictionary or the analysis failure relief dictionary as the dictionary used by the analysis step A usage dictionary setting step executed by the usage dictionary setting means for setting at least one; (c) the usage dictionary setting means in the usage dictionary setting step for the first analysis of the analysis step; The basic dictionary and the analysis failure relief dictionary are set at the time of analysis after the failure of analysis, and (d) the analysis means at the time of analysis after the analysis failure in the analysis step. The analysis after the failure is executed using the basic dictionary and the analysis failure relief dictionary.
[0012]
DETAILED DESCRIPTION OF THE INVENTION
(A) First embodiment
Hereinafter, a first embodiment in which a natural language processing apparatus and method according to the present invention are applied to a machine translation apparatus and a machine translation method will be described in detail with reference to the drawings.
[0013]
(A-1) Configuration of the first embodiment
FIG. 1 is a block diagram illustrating a functional configuration of the machine translation apparatus according to the first embodiment. In practice, for example, a processing program or fixed data is loaded on an information processing apparatus such as a personal computer, and the machine translation apparatus of the first embodiment is constructed.
[0014]
In FIG. 1, the machine translation apparatus according to the first embodiment is mainly composed of an input / output unit 1.1, a translation processing unit 1.2, and a dictionary storage unit 1.3.
[0015]
The input / output unit 1.1 includes an input processing unit 1.11 that receives an input sentence to be translated from the input unit 1.01, and an output processing unit 1.12 that outputs the translated result to the output unit 1.02. Have.
[0016]
The translation processing unit 1.2 includes a morpheme analysis unit 1.21 that performs morphological analysis of an input sentence (source language sentence), a syntax analysis unit 1.22 that performs syntax analysis of the source language using a translation pattern, It has a syntax generation unit 1.23 that generates a syntax in the target language from the result of the syntax analysis, and a morpheme generation unit 1.24 that generates a morpheme in the target language from the result of the syntax generation.
[0017]
The syntax analysis unit 1.22 includes a dictionary lookup unit 1.222 that performs pattern dictionary lookup, and a syntax analyzer 1.221 that performs syntax analysis based on the dictionary lookup result.
[0018]
The dictionary storage unit 1.3 includes a source language morpheme dictionary 1.31 in which source language morpheme information is stored, a target language morpheme dictionary 1.32 in which target language morpheme information is stored, and source language and target language It has a translation pattern dictionary 1.33 in which translation patterns made up of pairs of patterns are stored, and a fail safe pattern dictionary 1.34.
[0019]
The fail-safe pattern dictionary 1.34 also stores translation patterns consisting of pairs of source language and target language patterns, but at least the source language pattern is erroneously expressed in the actual source language sentence ( A pattern is stored that takes into account the expression that a place that should be plural is singular or a place that should be third person singular is not. This will be exemplified in the operation section.
[0020]
The fail safe pattern dictionary 1.34 is used when the parsing unit 1.22 fails in the parsing by applying the translation pattern dictionary 1.33. The fail-safe pattern dictionary 1.34 includes patterns that are not grammatically correct but are apt to be input by the user (patterns that do not match the personalities of the main case and main verb), patterns that include abbreviations (patterns that omit a tat) A pattern for preventing analysis failure is registered in advance.
[0021]
The user inputs an input sentence using an input unit 1.01 such as a keyboard file reading device, and obtains a translation result by an output unit 1.02 on a display, a printer, a file storage device, or the like.
[0022]
(A-2) Operation of the first embodiment
Next, the translation operation (machine translation method of the first embodiment) of the machine translation device of the first embodiment will be described. Here, FIG. 2 is a flowchart showing the translation operation of the machine translation apparatus of the first embodiment, and FIG. 3 is a flowchart showing the syntax analysis processing of the syntax analysis unit 1.22.
[0023]
The machine translation apparatus according to the first embodiment uses one sentence as an input unit. Further, in the following description, an example will be described in which the user inputs an English sentence “It shows a fact.” (Correctly “It shows a fact.”) Forgot to give s in the third person singular.
[0024]
When the user inputs the input sentence through the input unit 1.01, the input processing unit 1.11 receives it and passes it to the morpheme analysis unit 1.21 of the translation processing unit 1.2 (step S2.1). The morpheme analysis unit 1.21 performs morpheme analysis processing using information in the source language morpheme dictionary 1.31 (step S2.2).
[0025]
As shown in FIG. 4, the morphological analysis result is expressed by a tree structure with the root node as “root”. In the case of a morpheme that does not have a plurality of candidates, the standard form of each morpheme and morpheme information such as part of speech or change form are assigned immediately below the root node. On the other hand, in the case of a morpheme having a plurality of candidates (not shown in FIG. 4), information on each morpheme candidate is given as a child node of the or node.
[0026]
The syntax analysis unit 2.2 to which such a morpheme analysis result is given performs a syntax analysis process on the result using the translation pattern dictionary 33 (step S2.3). The syntax analysis unit 1.22 performs syntax analysis processing as shown in FIG.
[0027]
Before the main processing of syntax analysis is executed, a dictionary lookup unit 1.222 sets in advance which dictionary is used for syntax analysis. Initially, only the translation pattern dictionary 1.33 is set as a dictionary for analysis (step S3.1).
[0028]
FIG. 5 shows a part of the stored contents (translation pattern) of the translation pattern dictionary 1.33.
[0029]
In FIG. 5, the upper side represents an English pattern, the lower side represents a Japanese pattern, and the English pattern and the Japanese pattern are paired. Each language pattern is
[Language name: Pattern name Pattern component]
Consists of.
[0030]
The language name defines whether it is English (en) or Japanese (ja). As the pattern name, for example, a label according to a phrase structure rule such as VP or NP is applied. A pattern component consists of a word, a variable, or a sequence of words and variables. The variable is described by [arbitrary number: pattern name (may be accompanied by a pattern component)]. Arbitrary numeric parts indicate the correspondence between the paired patterns. In parsing, a pattern can have a nested structure by applying another pattern to a variable. Further, the word and the pattern name can have detailed information (feature information) such as semantic information. In addition, the word and pattern name can be referred to by converting the detailed information into a variable.
[0031]
FIG. 6 shows a part of the stored contents (translation pattern) of the fail safe pattern dictionary 1.34. The notation method of the stored translation pattern is the same as the translation pattern stored in the translation pattern dictionary 1.33.
[0032]
The syntax analysis unit 1.221 performs syntax analysis processing using each pattern registered in the translation pattern dictionary 1.33 as described above (step S3.2).
[0033]
Although the syntax analysis processing is omitted from the flowchart, the three processes of the pattern dictionary lookup processing, the pattern inspection processing, and the pattern application processing are repeatedly performed while confirming that the syntax analysis has not ended.
[0034]
The pattern dictionary lookup process, which is the first process, is a process of subtracting from the translation pattern dictionary 1.33 a translation pattern that may be applied next from the morphological analysis result and the result of the pattern application process so far. The pattern inspection process, which is the second process, is a process for inspecting each tree structure whether or not the translation pattern obtained as a result of dictionary lookup is applied to the tree structure currently being constructed. The pattern application process, which is the third process, is a process of actually applying the pattern to the tree structure based on the tree structure and the translation pattern that can be applied as a result of the inspection.
[0035]
FIG. 7 shows a tree structure in the middle of parsing for the input sentence. The nodes 7.1 to 7.5 in the tree structure of FIG. 7 correspond to the translation patterns (source language patterns) 5.1 to 5.5 shown in FIG. 5 (7.1: 5). .1, 7.2: 5.2, ...).
[0036]
The tree structure in the middle of parsing in FIG. 7 cannot continue parsing any further. This is because none of the translation patterns (source language patterns) 5.6 to 5.8 for the main case and verb phrase can be applied because the conditions for change (conditions for the third person singular) differ from the morphological analysis results. This is because the tree cannot be completed. That is, it is impossible to create a syntax tree using S (sentence) as a root node using all input words.
[0037]
Therefore, in this case, the result of the parsing failure is obtained in the process of determining whether the parsing has succeeded (step S3.3). If the syntax analysis is successful, the syntax analysis process ends and the syntax analysis result is given to the syntax generation unit 1.23.
[0038]
If the syntax analysis in step S3.2 fails, it is determined whether or not the fail-safe pattern dictionary 1.34 was used in the syntax analysis (step S3.4). At this time, since the fail-safe pattern dictionary 1.34 is not yet used, both the translation pattern dictionary 1.33 and the fail-safe pattern dictionary 1.34 are set as pattern dictionaries used for syntax analysis (step S3. 5) Returning to the main parsing process (step S3.3), the parsing process is performed from the middle of the parsing that failed last time.
[0039]
As described above, FIG. 6 shows a storage example of the fail safe pattern dictionary 1.34. In the fail-safe pattern dictionary 1.34, a pattern with relaxed conditions such as a pattern 6.3 composed of pattern constituent elements having no conditions related to changes as shown by reference numerals 6.1 and 6.2 is prepared. Keep it. In other words, a pattern that can handle an input sentence whose change is not correct is prepared.
[0040]
Accordingly, pattern 6.3 is applied to the input sentence in the main parsing process (step S3.3) with both translation pattern dictionary 1.33 and fail-safe pattern dictionary 1.34 set. Then, a tree structure having S (sentence) as a root node as shown in FIG. 8 can be created, the syntax analysis is successful, and the syntax analysis processing is completed.
[0041]
8 correspond to the constituent elements 6.1 and 6.2 of the pattern 6.3 in FIG.
[0042]
If the main parsing process fails even if the main parsing process is performed with both the translation pattern dictionary 1.33 and the fail safe pattern dictionary 1.34 set, the parsing process ends and the parsing process fails. Is displayed and output, for example.
[0043]
The obtained syntax analysis result is given to the syntax generation unit 1.23, and the syntax generation unit 1.23 executes a syntax generation process (step S2.4 in FIG. 2). The tree structure generated by the syntax generation unit 1.23 is given to the morpheme generation unit 1.24, and the morpheme generation unit 1.24 performs morpheme generation processing (step S2.5). In the morpheme generation process, the syntax generation results are arranged in order from the left, Japanese words located at the end node, and each word is arranged using the target language morpheme dictionary 1.32, and the verb usage form is adjusted. Shape it. The output processing unit 1.12 receives the morpheme generation result (translated sentence), and the output unit 1.02 displays and outputs it or stores it in a file.
[0044]
For the above-mentioned input sentence “It show a fact.”, The translation result “It indicates fact” is obtained.
[0045]
(A-3) Effects of the first embodiment
According to the machine translation apparatus and method of the first embodiment, the following effects can be obtained.
[0046]
(A) In Japanese Patent Application Laid-Open No. 5-290082 referred to in the section of the prior art, since the translation failure is remedied by grammatical analysis, the advantage of pattern translation cannot be enjoyed. Since it is a relief to use the safe pattern dictionary 1.34, the advantage of pattern translation can be enjoyed. For example, when there are patterns registered by the user or patterns related to special expressions in the translation pattern dictionary 1.33, these patterns can be appropriately reflected in the analysis even when the analysis failure is remedied.
[0047]
(B) In the method of the above publication, since both pattern analysis and grammatical analysis are used, it is necessary to develop and manage a dictionary and an analysis engine for both different specifications. In the embodiment, since everything is performed on the pattern analysis, the dictionary and the analysis engine have exactly the same specifications, and development, management, and improvement of efficiency are achieved. Also, the analysis unit configuration is simple.
[0048]
(C) Since most of the translation results are obtained by using the fail-safe pattern dictionary 1.34, it is almost unnecessary to ask the user to review the input sentence, and the user is not burdened. I'm going to live.
[0049]
(D) Although both the translation pattern dictionary 1.33 and the fail safe pattern dictionary 1.34 may be used from the beginning, the syntax analysis using only the translation pattern dictionary 1.33 as in the first embodiment If the fail-safe pattern dictionary 1.34 is used when it fails, the parsing processing time for a correct input sentence can be shortened. That is, when both the translation pattern dictionary 1.33 and the fail safe pattern dictionary 1.34 are used from the beginning, the number of patterns to be searched, the number of patterns to be judged for conformance, etc. increase, and the processing time for parsing the correct input sentence Will also be long.
[0050]
(A-4) Modified embodiment of the first embodiment
The translation pattern dictionary 1.33 can be used not only in the dictionary that the system has from the beginning but also in a dictionary created by the user.
[0051]
In the above description, the case where syntax analysis has failed and the re-analysis using the fail-safe pattern dictionary has been started from the middle of the previous failed syntax analysis. You can start parsing again.
[0052]
As described above, although the effect is somewhat lower than in the first embodiment, both the translation pattern dictionary 1.33 and the fail safe pattern dictionary 1.34 may be used from the beginning to perform syntax analysis. Further, the user may be able to specify whether to use the fail-safe pattern dictionary 1.34 from the beginning or after the failure of syntax analysis.
[0053]
(B) Second embodiment
Next, a second embodiment in which the natural language processing apparatus and method according to the present invention are applied to a machine translation apparatus and a machine translation method will be described in detail with reference to the drawings.
[0054]
(B-1) Configuration of the second embodiment
FIG. 9 is a block diagram showing a functional configuration of the machine translation apparatus according to the second embodiment. The same or corresponding parts as those in FIG. 1 according to the first embodiment described above are assigned the same or corresponding reference numerals. It shows.
[0055]
The machine translation apparatus according to the second embodiment is different from the first embodiment in that a plurality (X) of failsafe pattern dictionaries 1.341 to 1.34X are provided as failsafe pattern dictionaries. ing.
[0056]
Each fail-safe pattern dictionary 1.341,..., 1.34X is similar in that it is used when analysis fails.
[0057]
However, for example, a fail-safe translation pattern corresponding to a mistake with a high occurrence rate in an input sentence is stored in a fail-safe pattern dictionary (for example, 1.341) with a small number, and a mistake with a low occurrence rate in an input sentence is handled. Thus, the fail-safe translation pattern is distinguished from the application efficiency so that it is stored in a large-number fail-safe pattern dictionary (for example, 1.34X). In addition, for example, a translation pattern corresponding to a small number of mistakes in one sentence is stored in a fail-safe pattern dictionary with a small number (for example, 1.341), and a translation pattern corresponding to a large number of mistakes in one sentence is As stored in a large-numbered fail-safe pattern dictionary (for example, 1.34X), it is distinguished from the number of mistakes. A distinction (priority order) may be given from another viewpoint. The fail-safe pattern may be applied not only to the grammatical mistake but also to the omission of the relative pronoun tat (grammatically correct). You just store in.
[0058]
As described above, since the second embodiment includes the plurality of fail safe pattern dictionaries 1.341 to 1.34X, the functions and processes of the syntax analysis unit 1.22 are also different from those of the first embodiment. Is different.
[0059]
(B-2) Operation of the second embodiment
The machine translation apparatus according to the second embodiment is different from the first embodiment only in the operation of the parsing process, and the parsing process will be described below. Here, FIG. 10 is a flowchart showing the operation of the syntax analysis process of the machine translation apparatus of the second embodiment.
[0060]
The syntax analysis unit 1.22 sets the translation pattern dictionary 1.33 as a pattern dictionary to be used (step S10.1), and sets a parameter n that defines the failsafe pattern dictionary to an initial value 0 (step S10. 2) The main parsing process is executed (step S10.3). Then, after incrementing the parameter n by 1 (step S10.4), it is determined whether or not the analysis is successful (step S10.5).
[0061]
If successful, the series of parsing processing is terminated. On the other hand, if it fails, it is determined whether or not the value of the parameter n is equal to or less than the number X of the fail safe pattern dictionaries 1.341 to 1.34X (step S10.6). Here, the value of the parameter n being larger than the number X of failsafe pattern dictionaries means that the analysis failed even when all the failsafe pattern dictionaries 1.341 to 1.34X are applied. End the analysis process. On the other hand, if the value of the parameter n is equal to or less than the number X of failsafe pattern dictionaries, the failsafe pattern dictionary 1.34n defined by the parameter n is additionally set as a pattern dictionary used for syntax analysis (step S10. 7) Returning to the main syntax analysis processing in step S10.3 described above.
[0062]
Hereinafter, the syntax analysis process of the second embodiment will be described in detail using the following three input sentences (example sentence 1 to example sentence 3) as appropriate.
[0063]
“She found the ready crying” (example sentence 1)
“She is the ready living here.” (Example sentence 2)
“It wanted tending.” (Example sentence 3)
Here, it is assumed that the stored contents (translation pattern) of the translation pattern dictionary 1.33 are as shown in FIG. Further, the number of fail safe pattern dictionaries X is 2, the stored contents (translation pattern) of the fail safe pattern dictionary 1.341 are as shown in FIG. 12, and the stored contents (translation pattern) of the fail safe pattern dictionary 1.342. Is as shown in FIG.
[0064]
Initially, a translation pattern dictionary 1.33 shown in FIG. 10 is set as a dictionary used for syntax analysis (step S10.1). Next, 0 is set to the parameter n (step S10.2), the main processing of syntax analysis is performed (step S10.3), and 1 is added to the parameter n after completion of the syntax analysis (step S10.4).
[0065]
When the example sentence 1 is an input sentence, the nodes 14.1 and 14.4 to 14.9 of the parsing result (tree structure) shown in FIG. 14 are respectively the source language patterns 11.1, 11 of FIG. As apparent from the correspondence to .4 to 11.9, the syntax analysis using only the translation pattern dictionary 1.33 succeeds, and the syntax analysis processing is terminated.
[0066]
On the other hand, when the example sentence 2 or the example sentence 3 is an input sentence, as shown in FIGS. 15 and 16, respectively, the syntax analysis cannot be performed only by applying the translation pattern dictionary 1.33. Therefore, the parameter n (= 1) Is less than the number of fail-safe dictionaries X (= 2), and is less than the number X (step S10.6), it is added to the translation pattern dictionary 1.33 and the fail-safe pattern dictionary 1.341 (the end of the code) 1 is the value of parameter n) (step S10.7), and the main parsing process is executed again (step S10.3).
[0067]
When the example sentence 2 is an input sentence, the pattern shown in FIG. 12 is applied to a portion where the pattern cannot be applied in a state where only the translation pattern dictionary 1.33 is applied by the additional set of the fail safe pattern dictionary 1.341. 12 A is applied, and the tree structure 17 shown in FIG. A can be constructed, the parsing is successful, and finally the parsing result shown in FIG. 17 is obtained.
[0068]
On the other hand, when the example sentence 3 is an input sentence, even if the fail-safe pattern dictionary 1.341 is additionally set, the syntax analysis still continues to fail, so the fail-safe pattern dictionary 1.342 is further added and again. Parse.
[0069]
An additional set of fail-safe pattern dictionaries 1.342 allows the pattern 13. As shown in FIG. 18, a tree structure 19. B can be constructed, the parsing is successful, and finally the parsing result shown in FIG. 18 can be obtained.
[0070]
Thereafter, the syntax generation process and the morpheme generation process are sequentially executed, and for each of the example sentence 1, the example sentence 2 and the example sentence 3, "she understood that a woman is crying", "she lives here. "I am a woman.", "It wants to be repaired."
[0071]
(B-3) Effects of the second embodiment
According to the machine translation apparatus and method of the second embodiment, in addition to the effects of the first embodiment, the following effects can also be achieved.
[0072]
(A) By providing a plurality of fail-safe pattern dictionaries, a priority order can be set for a pattern to be relieved. For example, in the case of the second embodiment, regarding the present participle verb, the pattern having the property of postfix noun modification of the present participle has priority over the pattern having the property of noun derivation in the present participle form alone. The ranking is high.
[0073]
(B) In addition, since a plurality of fail-safe pattern dictionaries are gradually increased and set, a parsing result can be obtained quickly for those having a high priority of the cause of the parsing failure.
[0074]
(B-4) Modified embodiment of the second embodiment
In the above description, the fail-safe pattern dictionary is used as a dictionary for relieving failure. However, it can be used not for the purpose of relieving failure but also as a means for setting the priority of the dictionary to be applied. For example, a dictionary composed of patterns registered by the user is the translation pattern dictionary 1.33 in the second embodiment, a grammatical pattern among the translation pattern dictionaries of the system is a fail safe pattern dictionary 1.341, and a translation of each word If the pattern related to is a fail-safe pattern dictionary 1.342, the pattern registered by the user is parsed with the highest priority, and if the analysis is successful only with the pattern registered by the user, the result is output as a translation result. Next, add a grammar pattern from the translation pattern dictionary of the system and analyze it. If the analysis is successful, output the result as a translation result, giving priority to the pattern registered by the user as much as possible. Can be translated.
[0075]
(C) Other embodiments
In each of the above embodiments, the present invention is applied to a machine translation device and method using a translation pattern. However, each of the above embodiments is characterized in processing until a syntax analysis result is obtained. The technical idea of each embodiment can be applied to other natural language processing apparatuses and methods. For example, the technical idea of the present invention can be applied to the part until the answer sentence is parsed by using a pattern (natural language pattern) in the question and answer apparatus.
[0076]
Moreover, although each said embodiment showed what applied this invention to the machine translation apparatus and method using a translation pattern, the technical idea of each said embodiment is also applied to the machine translation apparatus and method which does not use a translation pattern. Can be applied. For example, in the case of a device that parses a grammar rule by applying a grammar rule, the grammar rule dictionary may be separated into one that is applied from the beginning and one that is applied when analysis fails.
[0077]
Furthermore, the technical idea of the present invention can be applied not only to syntax analysis but also to morphological analysis. That is, the morphological analysis dictionary may be separated into one that is applied from the beginning and one that is applied when analysis fails. For example, an English verb that changes irregularly may be prepared in a morphological analysis dictionary to which “ed” is added to the original form of the verb when the analysis fails.
[0078]
【The invention's effect】
As described above, according to the natural language processing apparatus and method of the present invention, at least the basic dictionary used in the first analysis and one or more analysis failure relief dictionaries used in the analysis after the analysis failure are provided. Since a dictionary and / or analysis failure remedy dictionary is set appropriately and analysis after analysis failure is performed, even if analysis fails, the same analysis method can be applied to recover the failure without imposing a burden on the user it can.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a functional configuration of a machine translation apparatus according to a first embodiment.
FIG. 2 is a flowchart showing a translation operation of the machine translation apparatus of the first embodiment.
FIG. 3 is a flowchart showing details of syntax analysis processing according to the first embodiment;
FIG. 4 is an explanatory diagram illustrating an example of a morpheme analysis result according to the first embodiment.
FIG. 5 is an explanatory diagram showing an example of stored contents of a translation pattern dictionary 1.33 of the first embodiment.
FIG. 6 is an explanatory diagram illustrating an example of contents stored in a fail-safe pattern dictionary 1.34 according to the first embodiment;
FIG. 7 is an explanatory diagram illustrating a tree structure in the middle of syntax analysis according to the first embodiment;
FIG. 8 is an explanatory diagram illustrating a syntax analysis result that also uses the fail-safe pattern dictionary of 1.34 according to the first embodiment;
FIG. 9 is a block diagram illustrating a functional configuration of a machine translation apparatus according to the second embodiment.
FIG. 10 is a flowchart showing details of syntax analysis processing according to the second embodiment;
FIG. 11 is an explanatory diagram showing an example of stored contents of a translation pattern dictionary 1.33 of the second embodiment.
FIG. 12 is an explanatory diagram illustrating an example of contents stored in a fail-safe pattern dictionary 1.341 according to the second embodiment;
FIG. 13 is an explanatory diagram illustrating an example of contents stored in a fail-safe pattern dictionary 1.342 according to the second embodiment.
FIG. 14 is an explanatory diagram illustrating a syntax analysis result of Example sentence 1 using the translation pattern dictionary 1.33 of the second embodiment.
FIG. 15 is an explanatory diagram showing a tree structure in the middle of syntax analysis of example sentence 2 using the translation pattern dictionary 1.33 of the second embodiment;
FIG. 16 is an explanatory diagram showing a tree structure in the middle of syntax analysis of example sentence 3 using translation pattern dictionary 1.33 of the second embodiment;
FIG. 17 is an explanatory diagram illustrating a syntax analysis result of Example sentence 2 that also uses the fail-safe pattern dictionary 1.341 according to the second embodiment;
FIG. 18 is an explanatory diagram illustrating a syntax analysis result of Example sentence 2 that also uses the fail-safe pattern dictionary 1.342 according to the second embodiment;
[Explanation of symbols]
1.1 ... I / O unit,
1.2 ... Translation processing part,
1.21 ... Morphological analyzer
1.22 ... syntax analysis part,
1.3 ... dictionary storage unit,
1.33 ... Translation pattern dictionary,
1.34, 1.341 to 1.34X: Fail Safe pattern dictionary.

Claims (12)

In the natural language processing apparatus in which the analysis means analyzes the input natural language sentence or the analysis target which is a predetermined processing result for the natural language sentence,
A basic dictionary storing grammatically correct analysis information used by the analysis means at least in the first analysis;
One or a plurality of analysis failure relief dictionaries storing analysis information that can be used for analysis after analysis failure, and the analysis means can analyze a part to be analyzed that includes a grammatical error that is likely to fall ;
As a dictionary in which the analysis means is utilized, it possesses a usage dictionary setting means for setting at least one of the basic dictionary or the analysis fails relief dictionary,
The use dictionary setting means sets the basic dictionary at the time of the initial analysis of the analysis means, sets the basic dictionary and the analysis failure relief dictionary at the time of analysis after the analysis failure,
The natural language processing apparatus characterized in that the analysis means re-analyzes from the beginning using the basic dictionary and the analysis failure relief dictionary at the time of analysis after analysis failure . In the natural language processing apparatus in which the analysis means analyzes the input natural language sentence or the analysis target which is a predetermined processing result for the natural language sentence,
A basic dictionary storing grammatically correct analysis information used by the analysis means at least in the first analysis;
One or a plurality of analysis failure relief dictionaries storing analysis information that can be used for analysis after analysis failure, and the analysis means can analyze a part to be analyzed that includes a grammatical error that is likely to fall ;
As a dictionary in which the analysis means is utilized, it possesses a usage dictionary setting means for setting at least one of the basic dictionary or the analysis fails relief dictionary,
The use dictionary setting means sets the basic dictionary at the time of the initial analysis of the analysis means, sets the basic dictionary and the analysis failure relief dictionary at the time of analysis after the analysis failure,
The natural language processing apparatus characterized in that the analysis means executes analysis after failure using the basic dictionary and the analysis failure relief dictionary at the time of analysis after failure of analysis . A priority is assigned to the plurality of analysis failure relief dictionaries, and the use dictionary setting means additionally sets an analysis failure relief dictionary having a low priority as a dictionary to be used in the next analysis for each analysis failure. The natural language processing apparatus according to claim 1 or 2 . At least two of the basic dictionary and the analysis fails remedy dictionary, natural language processing apparatus according to any one of claims 1 to 3, characterized in that which was distributed normal dictionary contents. Natural language processing apparatus according to any one of claims 1 to 4, characterized in that analysis the analyzing means performs is parsed. 6. The natural language processing according to claim 5 , wherein the syntax analysis performed by the analysis means uses a collation with a natural language pattern prepared in advance in the basic dictionary and / or the analysis failure relief dictionary. apparatus. In a natural language processing method for analyzing an input natural language sentence or an analysis target that is a predetermined processing result thereof in an analysis step executed by an analysis unit ,
A basic dictionary that stores grammatically correct analysis information used by the analysis process at least for the first analysis, and a grammatical error that the analysis process uses for analysis after an analysis failure is likely to occur. together and a one or more analysis failure relief dictionary stores analysis information that may analyze portion of the analysis target including,
As a dictionary used by the analysis step, there is a use dictionary setting step executed by a use dictionary setting unit for setting at least one of the basic dictionary or the analysis failure relief dictionary,
In the usage dictionary setting step, the usage dictionary setting means sets the basic dictionary at the time of the first analysis of the analysis step, sets the basic dictionary and the analysis failure relief dictionary at the time of analysis after the analysis failure,
The natural language processing method characterized in that the analysis means re-analyzes from the beginning using the basic dictionary and the analysis failure relief dictionary at the time of analysis after analysis failure in the analysis step . In a natural language processing method for analyzing an input natural language sentence or an analysis target that is a predetermined processing result thereof in an analysis step executed by an analysis unit ,
A basic dictionary that stores grammatically correct analysis information used by the analysis process at least for the first analysis, and a grammatical error that the analysis process uses for analysis after an analysis failure is likely to occur. together and a one or more analysis failure relief dictionary stores analysis information that may analyze portion of the analysis target including,
As a dictionary used by the analysis step, there is a use dictionary setting step executed by a use dictionary setting unit for setting at least one of the basic dictionary or the analysis failure relief dictionary,
In the usage dictionary setting step, the usage dictionary setting means sets the basic dictionary at the time of the first analysis of the analysis step, sets the basic dictionary and the analysis failure relief dictionary at the time of analysis after the analysis failure,
The natural language processing method, wherein the analysis means performs analysis after the failure using the basic dictionary and the analysis failure relief dictionary at the time of analysis after failure in the analysis step . Priorities are assigned to the plurality of analysis failure relief dictionaries, and the use dictionary setting means uses the analysis failure relief dictionary having a low priority in the next analysis for each analysis failure in the use dictionary setting step . The natural language processing method according to claim 7, wherein the natural language processing method is additionally set. 10. The natural language processing method according to claim 7 , wherein at least two of the basic dictionary and the analysis failure repair dictionary are obtained by sorting normal dictionary contents. The natural language processing method according to claim 7 , wherein the analysis performed in the analysis step is a syntax analysis. The natural language processing according to claim 11 , wherein the syntax analysis performed in the analysis step uses collation with a natural language pattern prepared in advance in the basic dictionary and / or the analysis failure relief dictionary. Method.

Images