JPH0628392A - Part of speech selection system - Google Patents

Abstract

PURPOSE:To provide the part of speech selection system capable of reducing a memory capacity on the part of speech selection system automatically selecting the kind of a part of speech of each word in a sentence (text). CONSTITUTION:The system consists of a vocabulary possibility retrieval section 1 receiving an input text and retrieving the vocabulary possibility for each word, vocabulary possibility dictionary 2 being the retrieval object of the section 1, context possibility detection section 10 detecting the possibility of the context arranged with plural parts of speech based on the vocabulary possibility retrieved from the section 1, context possibility calculation section 20 using a neural network inputting the input pattern given from the section 10 and calculating the context possibility according to the kind of a part of speech of the next word, and an evaluation value calculation section 6 receiving the output from the context possibility detection section 10 and calculating the evaluation value according to a prescribed procedure based on the possibility of the context arranged with plural parts of speech and the vocabulary possibility of words of head or end arranged with a part of speech.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language processing technique, and more particularly to a part-of-speech selection system for automatically selecting the type of part-of-speech of each word in a text.

[0002] In recent years, research on voice synthesis has advanced, and it has begun to be used in various fields such as reading systems and response systems. In order to improve the naturalness of synthesized speech, it is necessary to extract the linguistic information of the output sentence and generate a pattern for generating accents and intonations corresponding to it. The part of speech that forms the basis of such language information. Part-of-speech refers to a type of word such as a noun, a verb, or an adverb. Since the voice synthesis system is required to be portable, a system that can easily extract the part of speech of a word in a sentence is strongly desired.

[0003]

2. Description of the Related Art Recently, a method of determining the part of speech of a word in a sentence by using statistical information has been used. This is, for example, “A Stochastic Parts”
Program and Noun Phase
Paser for Unrestricted Te
xt ”(ACL processes 2nd A
pplied Natural Language P
processing, the vocabulary probability, which is the probability that a given word takes a part of speech, and the context probability that indicates the frequency of appearance of the part-of-speech sequence are used to determine the part of speech of each word in the sentence. .

FIG. 19 is a block diagram showing a configuration example of a conventional device. In the figure, 1 is a vocabulary probability search unit that receives an input text and searches for a vocabulary probability for each word, and 2 is a vocabulary probability dictionary that is a target of search by the vocabulary search unit 1, in which the vocabulary and the probability of the vocabulary are stored. Reference numeral 3 is a part-of-speech sequence creation unit that creates a part-of-speech sequence of each word based on the output of the vocabulary probability search unit 1.

Reference numeral 4 denotes a context probability detecting unit for detecting context probabilities of a plurality of parts of speech based on the vocabulary probabilities searched by the vocabulary probability searching unit 1, and 5 a context for storing context probabilities of the parts of speech array. A probability table, 6 is an evaluation value calculation unit that receives the output of the context probability detection unit 4 and calculates an evaluation value from the context probabilities of a plurality of part-of-speech sequences and the vocabulary probabilities of the first and last words of the part-of-speech sequence. is there. The operation of the apparatus configured as described above will be described below.

First, the vocabulary probability search unit 1 inputs an input text (sentence) and reads the vocabulary probability for each word from the vocabulary probability dictionary 2. For example, if the input text is I love her very much. It was assumed that the sentence was. Here, a three-word part-of-speech sequence is considered as a category of a part-of-speech sequence of words. The vocabulary probability search unit 1 extracts the vocabulary probability of each word forming this sentence from the vocabulary probability dictionary 2.

The vocabulary probability dictionary 2 has a structure as shown in FIG. 20, for example. The content shown in the figure is obtained by extracting only the words in the input sentence, and actually, all the words are stored in alphabetical order as the appearance probabilities for each type of part of speech.

Here, consider the case where processing is performed after a sentence.

The part-of-speech sequence creating unit 3 first creates a part-of-speech sequence of (Z ##) for the period ". ##" of the sentence, and supplies it to the context probability detecting unit 4. Here, # is a symbol indicating the end or beginning of a sentence. The context probability detecting unit 4 extracts the context probability of the part-of-speech arrangement of (Z ##) from the context probability table 5. The context probability table 5 stores, as context probabilities, appearance probabilities for all patterns that form a combination of three parts of speech.

FIG. 21 is a diagram showing a configuration example of the context probability table. In the figure, only a part thereof is shown. In addition, in the case of English as a part-of-speech sequence, it is the easiest to classify 3
The word part-of-speech sequence is used. The probability that each combination of part-of-speech sequences can take is obtained by statistically obtaining the possible probability with reference to a large amount of sentences.

The evaluation value calculation unit 6 multiplies the context probability of the part-of-speech sequence of (Z ##) by the vocabulary probability of (.Z) to obtain an evaluation value. For example, the context probability of (Z ##) is 0.983,
If the vocabulary probability of (.Z) is 1.00, the evaluation value will be 0.983.

Next, the part-of-speech sequence creating section 3 uses the word sequence "m".
uch. # "(ADJ Z #), (NOUN Z #), (ADV
Z #). The context probability detecting unit 4 uses these parts-of-speech sequences (ADJ Z #), (NOUN Z #), (A
DV Z #) context probability table 5
Extract from. Here, ADJ is an adjective, NOUN is a noun, and ADV is an adverb. FIG. 24 shows classification of parts of speech.
In the case of English, as shown in the figure, there are 20 kinds of parts of speech, and here, for each part of speech, the part of speech symbol as shown in the figure is used.

The evaluation value calculation unit 6 returns NO when "much" is ADJ in each part of speech sequence.
The evaluation value is calculated by multiplying the context probability, the vocabulary probability, and the corresponding cumulative evaluation value up to that time when UN is taken and when ADV is taken.

Next, the part-of-speech sequence creating section 3 uses the word sequence "ve".
ry much. Part of speech for "(ADV ADJ
Z), (ADV NOUNZ Z), (ADV AD
VZ), (ADJ ADJ Z), (ADJ NOUN
ZZ), (ADJ ADV Z) are created.

The context probability detecting section 4 uses the sequence of parts of speech (A
DV ADJ Z), (ADV NOUNZ Z),
(ADV ADV Z), (ADJ ADJ Z),
The context probabilities for (ADJNOUNZ Z) and (ADJ ADV Z) are extracted from the context probability table 5. The evaluation value calculation unit 6 calculates
The evaluation value of each part-of-speech sequence is calculated by multiplying the context probability, the vocabulary probability, and the cumulative evaluation value for each of the cases where "very" takes ADV and ADJ.

When the same operation is repeated until "## I", five parts-of-speech sequences I love her very much. The evaluation values for all possible part-of-speech sequences can be obtained. Then, the one with the highest evaluation value is selected as the part-of-speech sequence. In this case, the selected part-of-speech sequence is ## PPRON V PRON ADV ADV Z ##.

[0017]

In the case of the above-described conventional apparatus, since the context probabilities of the part-of-speech arrangement are stored in the table (context probability table 5), as the number of categories of part-of-speech classification and the number of part-of-speech arrangements increase. There was a problem that a large amount of memory was required.

The present invention has been made in view of the above problems, and an object thereof is to provide a part-of-speech selection system capable of reducing the memory capacity.

[0019]

FIG. 1 is a block diagram showing the principle of the present invention. The same parts as those in FIG. 19 are designated by the same reference numerals. In the figure, 1 is a vocabulary probability search unit that receives an input text and searches for a vocabulary probability for each word, and 2 is a vocabulary probability dictionary that is a target of search by the vocabulary search unit 1, in which the vocabulary and the probability of the vocabulary are stored. Is.

Reference numeral 10 is a context probability detecting unit for detecting context probabilities of a plurality of parts of speech based on the vocabulary probability retrieved by the vocabulary probability retrieving unit 1, and 20 is an input pattern given from the context probability detecting unit 10. To calculate the context probability according to the type of part of speech of the next word, 6 is a context probability calculating unit using a neural network, 6 is the context probability detecting unit 1
The evaluation value calculation unit receives an output of 0 and calculates an evaluation value according to a predetermined procedure from the context probabilities of a plurality of part-of-speech sequences and the vocabulary probability of the first or last word of the part-of-speech sequence.

[0021]

In advance, the context probability calculation unit 20 is made to learn the probability of the part-of-speech arrangement. For example, in the case of a 3-word part-of-speech sequence,
The neural network is made to input two part-of-speech input patterns from the back, and the probability according to the kind of part-of-speech of the next word (first word) at that time is given in advance as a teacher pattern, and the teacher pattern and the output of the neural network are Make them learn to be equal.

Thereafter, the context probability calculating unit 20 inputs the pattern of the part-of-speech arrangement given from the context-probability detecting unit 10, and calculates the context probability according to the type of the part-of-speech of the next word. For example, in the case of a three-word part-of-speech sequence, two part-of-speech sequences are input as an input pattern, and the probability of taking the part-of-speech sequence of the next word (context probability) is calculated for each type of part-of-speech. Based on the calculated context probabilities, the evaluation value calculation unit 6 calculates the evaluation value for each pattern of the 3-word part-of-speech arrangement.

As described above, according to the present invention, the neural network forming the context probability calculating unit 20 is made to learn the probability of the part-of-speech arrangement, so that, for example, in the case of three-part-of-speech arrangement, it corresponds to two parts of speech from the rear. If a pattern is input as an input pattern, the probability (context probability) of the next third part of speech (part of speech of the first word) is output for each type of part of speech, so the context probability information is also stored as a context probability table. It is possible to provide a part-of-speech selection system that eliminates the need and can reduce the memory capacity.

[0024]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The operation of the present invention will be described in detail below with reference to the drawings.

FIG. 2 is a diagram showing a configuration example of the context probability calculation unit 20. 21 is a neural network, which is an input layer L
1, an intermediate layer L2 and an output layer L3. The input layer L1 is provided with (combination of part-of-speech numbers-1) combinations of the number of part-of-speech categories. For example, the part-of-speech sequence described above is 3
In the case of the number of pieces, the input layer L1 is composed of two combinations of the number of parts of speech categories. The output device L3 has a combination of one category number. The unit of combination of the number of parts of speech categories is, for example, the number of parts of speech in English, which is 20.
Individual pieces (20 units) are used.

Reference numeral 22 is a comparison unit for comparing the output value of the output layer L3 with the teacher pattern. The comparing unit 22 outputs the output layer L
The synaptic weight between the input layer L1 and the intermediate layer L2 is changed and the weight between the intermediate layer L2 and the output layer L3 is changed so that the output (probability) of 3 becomes equal to the teacher pattern. Finally, learning ends when the output value of the output layer L3 and the teacher pattern match. At this time, the weight of the synapse between the input layer L1 and the intermediate layer L2 and the intermediate layer L2 and the output layer L
The weight between 3 is fixed to a certain fixed value.

FIG. 3 is a block diagram showing the configuration of an embodiment of the present invention. The same parts as those in FIG. 1 are shown with the same reference numerals omitted. In the figure, the context probability detecting unit 10 inputs the words from the vocabulary probability searching unit 1 and creates a predetermined number (for example, three) of a part-of-speech sequence creating unit 11 and an output of the part-of-speech sequence creating unit 11. Input pattern creating unit 12 which receives the input pattern and outputs the output of the input pattern creating unit 12 to the context probability calculating unit 20 to receive the context probabilities for each part of speech obtained from the neural network and output values. The output value selection unit 13 for selecting

The evaluation value calculation unit 6 includes an evaluation value calculation unit 31 for calculating an evaluation value from the context probability and the vocabulary probability given by the output value selection unit 13, and a plurality of outputs from the evaluation value calculation unit 31. It is composed of a maximum value selection unit 32 that selects the one with the maximum evaluation value from among them. Then, the output of the maximum value selection unit 32 becomes the desired part-of-speech sequence. The operation of the apparatus configured as described above will be described below.

The neural network 21 in this embodiment selects a part-of-speech in the reverse direction from the last word to the first word in the input text based on the frequency of the part-of-speech arrangement extracted from a large amount of data in advance and learns it. Any of the methods of learning by selecting a part of speech positively from the first word toward the last word can be used. (When selecting a part of speech from the last word) FIG. 4 is a diagram showing an example of creating a learning pattern when selecting a part of speech from the last word. Here, a three-word part-of-speech sequence is used. For example, consider a case where three adverbs are lined up like ADV ADV ADV. The frequency of such a part-of-speech arrangement is 307 when a large amount of data is searched, and its context probability is 0.07. This 0.07 is
It is calculated as a value obtained by dividing the frequency 307 by the sum of the frequencies of the two last two parts of speech in the three-word part-of-speech sequence. Thus, the frequency of the arrangement of NOT and two adverbs such as the next part-of-speech sequence NOT ADV ADV is 78 and its context probability is 0.02. And so on
The last two are all ADV and the first one is 20 categories
The probability of taking any part of speech of is shown as the context probability in FIG. This context probability is used as the neural network 21
I will explain how to learn.

Here, the neural network 21 of FIG. 2 has two sets of inputs, and the types of parts of speech are both AD.
Let V be. This ADV is shown in a state where "1" is set in the first unit of the 20 units. Here, the use of 20 units as the number of parts-of-speech categories corresponds to that in the case of English, there are 20 kinds of parts-of-speech as shown in FIG.

FIG. 5A is a diagram showing an input pattern example at this time. Now, the input pattern creating unit 12 causes the neural network 21 to input a pattern as shown in FIG. Further, for example, when the last two words in the three-word part-of-speech sequence are NOUN ADVs, the input pattern is as shown in FIG. 5 (b). In the case of NOUN, the number is "4" from FIG. In FIG. 24, since the number starts from 0, the fact that it is "4" indicates the fifth. Therefore, in the case of NOUN, "1" stands fifth from the front.

FIG. 6 is an explanatory diagram of learning at this time. The input layer L1 receives a total of 40 units in two sets of 20 units each. A "1" in the first unit of each set is an adverb. The output is 20 units. On the other hand, the teacher pattern given to the comparison unit 22 of the neural network 21 is as shown in FIG. This teacher pattern uses the context probability shown in FIG. 4 as it is.

That is, since the probability of the part-of-speech sequence that becomes ADV ADV ADV is 0.07, this value is given to the first unit of 20 units of the output layer L3 as a teacher signal. The next unit is given a probability of part-of-speech sequence of 0.02, which is NOT ADV ADV. Less than,
The values in FIG. 7 are given to the comparison unit 22 as a teacher pattern. The last unit is given a part-of-speech sequence probability of 0.08, which is # ADV ADV. In the configuration of FIG. 6, the output of each unit of the output layer L3 is 0.0
The same input pattern is given many times until the value of 7, 0.02, ..., 0.08 is obtained and the neural network 21 is trained.

In the above, the case of obtaining the context probability according to the type of part of speech taken by the first word when the latter two are ADV, ADV has been taken as an example, but the part of speech sequence is not limited to this. In addition to the pattern shown in FIG.
There are 400 learning patterns. The neural network 21 of FIG. 6 is also made to perform learning for each combination of patterns.

In this way, when the input text is input to the device shown in FIG. 3 in the state where the learning for all the three-word part-of-speech sequences is completed, the context probability calculation unit 20 selects three input words according to the input pattern. The context probability of the part-of-speech sequence is output. The output value selecting unit 13 selects an output value of the probability of the target part-of-speech arrangement from the output of the context probability calculating unit, and gives it to the evaluation value calculating unit 31.

The evaluation value calculation means 31 calculates the evaluation value by multiplying the context probability, the vocabulary probability of the leading word and the cumulative evaluation value of the corresponding part-of-speech string up to that point. Evaluation value calculation means 31
Since the evaluation values are calculated for all possible three-part-of-speech sequences, the maximum value selection unit 32 selects the largest part-of-speech sequence among these evaluation values and outputs it as the part-of-speech sequence to be obtained. (When selecting a part of speech from the first word) When selecting a part of speech from the first word, it can be considered in the same manner as when selecting a part of speech from the last word.

FIG. 8 is a diagram showing an example of creating a learning pattern when selecting a part of speech from the last word. Here, a three-word part-of-speech sequence is used. For example, consider a case where three adverbs are lined up like ADV ADV ADV. The frequency of such a part-of-speech arrangement is 307 when a large amount of data is searched, and its context probability is 0.07. This 0.07 is
The frequency of 307 is ADV, AD before the last two parts of speech in the three-word part-of-speech sequence
It is obtained as a value divided by the sum of the frequencies of the three part-of-speech sequences, which is V. In this way, the frequency of the arrangement of NOT and two adverbs such as the following part-of-speech sequence ADV ADV NOT is 6,
Its context probability is 0.00. Similarly, the probability that all of the preceding two are ADVs and the last one is any part-of-speech with a category number of 20 is shown as a context probability in FIG. A method for causing the neural network 21 to learn this context probability will be described.

Here, it is assumed that there are two sets of inputs and that the type of speech is ADV. This ADV is indicated by "1" in the first unit of the 20 units.
Is shown in an upright position.

FIG. 9 is a diagram showing an input pattern example at this time. Now, the input pattern creating unit 12 causes the neural network 21 to input a pattern as shown in FIG.
The configuration of the neural network at this time is the same as in FIG.

The input layer L1 receives a total of 40 units in two sets of 20 units each. A "1" in the first unit of each set is an adverb.
The output is 20 units. On the other hand, the teacher pattern given to the comparison unit 22 of the neural network 21 is as shown in FIG. This teacher pattern is shown in Figure 8.
The context probability shown in is used as it is.

That is, since the probability of the part-of-speech sequence that becomes ADV ADV ADV is 0.07, this value is given to the first unit of 20 units of the output layer L3 as a teacher signal. The next unit is given a probability of 0.00 of the part-of-speech sequence that becomes ADV ADV NOT. Less than,
The values of FIG. 10 are given to the comparison unit 22 as a teacher pattern. The last unit is given a probability of 0.00 of the part-of-speech sequence that becomes ADV ADV #. In the configuration of FIG. 6, the output of each unit of the output layer L3 is 0.0
The neural network 21 is trained by giving the same input pattern many times until it takes 7, 0.00, ..., 0.00.

In the above, the case where the context probability is obtained according to the type of part of speech taken by the last word when the last two words are ADV, ADV has been taken as an example, but the part of speech sequence is not limited to this, and is extremely limited. There are many patterns. The neural network 21 of FIG. 6 is also made to perform learning for each combination of patterns.

In this way, when the input text is input to the device shown in FIG. 3 in the state where the learning for all the three-word part-of-speech sequences is completed, the context probability calculation unit 20 selects three input words according to the input pattern. The context probability of the part-of-speech sequence is output. The output value selecting unit 13 selects an output value of the probability of the target part-of-speech arrangement from the output of the context probability calculating unit, and gives it to the evaluation value calculating unit 31.

The evaluation value calculation means 31 calculates the evaluation value by multiplying the context probability and the vocabulary probability of the last word by the cumulative evaluation value of the corresponding part-of-speech string up to that point. Evaluation value calculation means 31
Since the evaluation values are calculated for all possible three-part-of-speech sequences, the maximum value selection unit 32 selects the largest part-of-speech sequence among these evaluation values and outputs it as the part-of-speech sequence to be obtained.

Next, the operation of the device of the present invention will be described using a specific example. Here, a case is considered in which the part-of-speech of each word of the input text is selected from the end to the beginning by using the part-of-speech classified into 20 categories shown in FIG. . Here, the part of speech is selected for the sentence I appealed to the children to make less noise. The vocabulary probability search unit 1 searches the vocabulary probability dictionary 2 for each word in the input text, and extracts the part-of-speech category and its probability for each word. FIG. 11 is a diagram showing the part-of-speech categories thus extracted and their probabilities. Corresponding to the sentence, the part-of-speech category of each word and its vocabulary probability are shown.

The part-of-speech sequence creating section 11 creates a part-of-speech sequence of three words for this search result. FIG. 12 is a diagram showing a three-word part-of-speech sequence created in this way. All part-of-speech sequences that may be realized later in the sentence have been created.

The input pattern creating unit 12 creates an input pattern as shown in FIG. 5 from the two parts of speech after these parts of speech alignment and gives it to the context probability calculating unit 20 using a neural network. The context probability calculating unit 20 outputs the probability of the first word from the latter two parts of speech as the context probability for each part of speech category.

The output value selection unit 13 selects a value for the first part of speech of the part-of-speech sequence for the output of the context-probability calculation unit 20 for the given input pattern, and sets it as the context probability of the part-of-speech sequence.

The evaluation value calculation means 31 includes an output value selection unit 13
A value e obtained by multiplying the context probability Pb obtained from the vocabulary probability Pb and the vocabulary probability Pv for the first part-of-speech of the part-of-speech sequence searched by the vocabulary probability search unit 1 by the cumulative evaluation value e ′ for the part-of-speech sequence created from the last word of the input text. Is a new evaluation value. The new evaluation value e is expressed by the following equation.

E = Pb × Pv × e ′ (1) FIGS. 13 and 14 are views showing the evaluation value calculation results obtained in this way. 13 and 14 are continuous, and FIG. 14 comes after FIG. In the figure, A is the context probability obtained by the above-mentioned sequence in the context probability calculating unit 20, B is the vocabulary probability according to the type of part of speech of the first word, and C is the cumulative evaluation value obtained in the process up to that point. Reference characters e and D respectively represent a part-of-speech arrangement corresponding to the cumulative evaluation value.

From the word at the end of the input text, three part-of-speech sequences are created for all of the part-of-speech categories created by the word, and for all of the part-of-speech sequences created up to that time, after the part-of-speech sequence, the three part-of-speech sequence is created. Search probability Pb. The product of this context probability Pb, the vocabulary probability Pv of the word in the part-of-speech category, and the evaluation value e ′ of the part-of-speech sequence created up to that point is set as the evaluation value e of the new part-of-speech sequence.
As shown in the figure, after the evaluation values of all the parts-of-speech sequences up to the first word of the input text can be calculated, the maximum value selection unit 32
Selects the one with the largest evaluation value from these part-of-speech sequences and determines the part-of-speech of each word in the input text.

FIG. 15 is a diagram showing the part-of-speech selection result thus obtained. This matches the part-of-speech arrangement at the final stage of FIG.

FIG. 16 is a block diagram showing the configuration of another embodiment of the present invention. The same parts as those in FIG. 3 are designated by the same reference numerals. In the figure, 1 is a vocabulary probability search unit that receives an input text and searches for a vocabulary probability for each word, and 2 is a vocabulary probability dictionary that is a target of search by the vocabulary search unit 1, in which the vocabulary and the probability of the vocabulary are stored. Reference numeral 10 is an input pattern creation unit as a context probability detection unit that receives the output of the vocabulary probability search unit 1 and creates an input pattern.

Numeral 20 is a context probability calculator using a neural network for inputting the input pattern given from the input pattern generator 10 to calculate the context probability according to the type of part of speech of the next word. Reference numeral 31 is the context probability calculator. An evaluation value calculation means for receiving an output of the unit 20 and calculating an evaluation value from the given context probability and vocabulary probability, and 32 is the evaluation value calculation means 3
The maximum value selection unit 32 selects the one having the maximum evaluation value from the plurality of outputs from 1. The evaluation value calculation unit 31 and the maximum value selection unit 32 constitute the evaluation value calculation unit 6. The operation of the apparatus configured as described above will be described below.

In this embodiment, instead of inputting the input pattern of the latter two part-of-speech sequences as "1" or "0" as shown in FIG. The evaluation value indicating the probability of taking is input.

As in the embodiment of FIG. 3, the vocabulary probability search unit 1
Searches the lexical probability dictionary 2 for each word in the input text, and extracts the part-of-speech category and its probability for each word as shown in FIG. Here, the evaluation value at the time of selecting a part of speech is expressed for each word by the vocabulary probability of the part of speech category, and the next two words or the previous two words of the part of speech category adjacent to the context probability calculating unit 20 using the neural network are displayed. Enter the probability.

17 and 18 are views showing examples of evaluation value calculation. 17 and 18 show a series of operations, and FIG. 18 is followed by FIG. When selecting a part-of-speech from the last word of the input text, the input pattern creating unit 10 selects the last two words of the three-word part-of-speech sequence.
The evaluation value E of the word is input to the neural network 21, and the obtained output becomes the context probability Pb.

The probability of each part-of-speech category of the context probability Pb is multiplied by the probability of each part-of-speech category of the word obtained by the vocabulary probability retrieving unit 1 to obtain an evaluation value E of the word.

The maximum value selection unit 32 selects, for each word of the input text, the part-of-speech category having the largest evaluation value E as the part-of-speech of the word. For example,
In the case of FIG. 17, when Wn = 6, “make” is evaluated as a word. The evaluation value E at this time is the maximum value of 0.34 for the tenth unit from the front. The tenth part of speech from the front is a verb (V) from FIG. The result of selecting the part of speech in this way is also shown in FIG.
Will be the same as

According to this embodiment, by inputting the evaluation value indicating the probability of taking the part-of-speech category to the neural network 21, it is possible to reduce the number of times of searching the context probability, and to perform the part-of-speech selection at high speed. You can

In the above embodiment, the case where the present invention is applied to English has been described, but the present invention is not limited to this. The present invention can be applied to Japanese and other sentences.

[0062]

As described above in detail, according to the present invention, by using the neural network to obtain the context probability, it is not necessary to have the context probability table, and the memory capacity can be reduced. It is possible to provide a part-of-speech selection system capable of performing.

[Brief description of drawings]

FIG. 1 is a principle block diagram of the present invention.

FIG. 2 is a diagram showing a configuration example of a context probability calculation unit.

FIG. 3 is a configuration block diagram showing an embodiment of the present invention.

FIG. 4 is a diagram showing an example of creating a learning pattern when selecting a part of speech from the last word.

FIG. 5 is a diagram showing an example of an input pattern.

FIG. 6 is an explanatory diagram of learning.

FIG. 7 is a diagram showing an example of a teacher pattern.

FIG. 8 is a diagram showing an example of creating a learning pattern when selecting a part of speech from the first word.

FIG. 9 is a diagram showing an example of an input pattern.

FIG. 10 is a diagram showing an example of a teacher pattern.

FIG. 11 is a diagram showing extracted part-of-speech categories and their probabilities.

FIG. 12 is a diagram showing a created part-of-speech sequence of three words.

FIG. 13 is a diagram showing an evaluation value calculation result obtained by the present invention.

FIG. 14 is a diagram showing an evaluation value calculation result obtained by the present invention.

FIG. 15 is a diagram showing a part of speech selection result according to the present invention.

FIG. 16 is a configuration block diagram showing another embodiment of the present invention.

FIG. 17 is a diagram showing an example of evaluation value calculation.

FIG. 18 is a diagram showing an example of evaluation value calculation.

FIG. 19 is a block diagram showing a configuration example of a conventional device.

FIG. 20 is a diagram showing an internal configuration example of a vocabulary probability dictionary.

FIG. 21 is a diagram showing a configuration example of a context probability table.

FIG. 22 is a diagram showing classification of parts of speech.

[Explanation of symbols]

 1 vocabulary probability search unit 2 vocabulary probability dictionary 6 evaluation value calculation unit 10 context probability detection unit 20 context probability calculation unit

Claims (4)

[Claims] 1. A vocabulary probability search unit (1) for receiving an input text and searching for a vocabulary probability for each word, a vocabulary and a probability of the vocabulary are stored, and the vocabulary search unit (1) is stored.
A vocabulary probability dictionary (2) to be searched for and a context probability detecting unit (10) for detecting context probabilities of a plurality of parts of speech based on the vocabulary probability searched by the vocabulary probability searching unit (1). And a context probability calculating unit (20) using a neural network for inputting an input pattern given from the context probability detecting unit (10) to calculate a context probability according to the type of part of speech of the next word, An evaluation value calculation unit (6) that receives an output from the context probability detection unit (10) and calculates an evaluation value according to a predetermined procedure from context probabilities of a plurality of part-of-speech sequences and vocabulary probabilities of words at the beginning or end of the part-of-speech sequence. ) Part-of-speech selection system. 2. When the context probability calculating unit (20) calculates the context probability of an n-word part-of-speech sequence, the part-of-speech pattern of the (n−1) th word from the beginning of the part-of-speech sequence is input as (n -1) The part-of-speech selection system according to claim 1, wherein the part-of-speech selection system is configured to output a probability that each part-of-speech category of an n-th word with respect to a part-of-speech arrangement of words. 3. When the context probability calculating unit (20) calculates the context probability of an n-word part-of-speech sequence, the part-of-speech pattern of the second to the end n of the part-of-speech sequence is input (n-1). 3. The part-of-speech selection system according to claim 1, wherein the part-of-speech selection system is configured to output a probability that each part-of-speech category of a leading word with respect to a word part-of-speech sequence is output. 4. When the context probability calculating unit (20) calculates the context probability of an n-word part-of-speech arrangement, an evaluation value indicating the probability of taking each part-of-speech is input to (n-1) part-of-speech arrangement patterns. 4. The part-of-speech selection system according to claim 2, wherein the part-of-speech selection system outputs the probability of each part-of-speech category of the n-th or the first word.