JPH02138662A - Natural language processing dictionary - Google Patents

Abstract

PURPOSE:To improve the grammatical processing accuracy and to reduce the process value by storing the functional meaning markers obtained by sorting the subject words/phrases based on a primary function structure of a subject. CONSTITUTION:A word dictionary memory includes an index memory of the fixed length which includes the reading ways and a pointer and a word information memory of the variable length which includes the descriptions, the grammar codes, the meaning markers, and the functional meaning markers. A case grammar is used to describe a grammar centering on the vocabularies. Then a meaning marker (conventional one or a functional meaning marker) and the auxiliary postpositional words are used as the means which designate the case frames of each vocabulary. Thus a grammar is easily described when the case grammar, a coupled value grammar, etc., are used in comparison with a case where only the meaning marker is used to process a language. Furthermore the process value is also reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a dictionary for natural language processing that is referred to when performing natural language processing such as kana-kanji conversion and natural language understanding.

[Prior Art] Conventionally, in natural language processing, dictionaries used when using methods such as case grammar and bond valence grammar have a notation that has a one-to-one correspondence with the target and a meaning marker attached to that notation. have.

The semantic markers are those that classify objects into several categories. For example, for the expressions (pigeon) and (chicken), the category of birds (birds) is expressed.
d) has a meaning marker. Word identification, semantic analysis, etc. have been performed by specifying the semantic markers using case grammars, valence grammars, and the like.

[Communication to be Solved by the Invention] However, it has been difficult to handle exceptions using only conventional semantic markers, making processing complicated. For example, suppose that the meaning marker (bird) is designated as the nominative of the verb "to fly."

Then, since the sentence "dove flies" has a semantic marker such that (pigeon) represents a bird (b t r d), consistency with the semantic marker is guaranteed, and it is actually a correct sentence. However, the sentence "A penguin flies" is actually an impossible sentence, even though (penguin) has the meaning marker (bird).

The present invention has been made in order to solve the above-mentioned problems, and when using case grammars, valence grammars, etc., it is possible to make it easier to describe the grammar, as well as to refine the grammar processing and reduce the amount of processing. The purpose of this paper is to provide a dictionary for natural language processing that can be used for natural language processing.

[Means for Solving the Problems] In order to achieve this object, the dictionary for natural language processing of the present invention uses, in addition to semantic markers, functional semantic markers that classify target words based on the main functional structure of the target. It is characterized by having.

[Operation] In the present invention having the above configuration, when processing a case grammar, a valence grammar, etc., a semantic marker and a functional semantic marker are read from the dictionary for natural language processing, and the semantic marker and the semantic marker specified in the grammar are read out from the dictionary for natural language processing. Processing such as matching is performed.

[Embodiment] An embodiment in which the present invention is embodied in a voice input word processor will be described below with reference to FIGS. 1 to 5.

The microphone 3 picks up information on the voice to be recognized,
The signal is configured to be converted into an electrical signal. The output terminal of the microphone 3 is connected to an amplifier 4 for amplifying the electrical signal, a low-pass filter 5 for limiting the frequency band of the electrical signal, and an amplifier 4 for converting the output of the low-pass filter 5 into a digital signal. D converter 6 is continuously connected.

An output terminal of the A/D converter 6 is connected to a first central processor unit (hereinafter referred to as a first CPU) 7 that performs voice recognition processing. This first CPU
7, a first read-only memory (hereinafter referred to as the first ROM) that stores a program that describes the processing;
8. A first random access memory (hereinafter referred to as a first RAM) 9 that serves as a work area for processing by the first CPU 7 is connected. This first CPU 7 recognizes the voice by processing the digital signal converted by the A/D converter according to a program stored in the first ROM 8, and outputs a plurality of kana character candidates. It is configured.

A second central processor unit (hereinafter referred to as second CPU) 11 is connected to the first CPIJ 7 via a shared memory 10 that can store voice recognition results. This second CPU 11 has a second read-only memory (
A second random access memory (hereinafter referred to as a second RAM) 15 serving as a work area for processing of the second CPUI 1 (hereinafter referred to as a second ROM) 14, a second random access memory (hereinafter referred to as a second RAM) 15, a dictionary for natural language processing of the present invention. A corresponding word dictionary memory 12 and a grammar memory 13 containing grammar necessary for processing by the second CPU II are connected. This second CPUI
1 is configured to convert the voice recognition result stored in the shared memory 10 into kana-kanji according to a program stored in the second ROM 14, and to order candidates for sentences containing kana-kanji after the kana-kanji conversion.

Furthermore, a display device 16 for outputting ordered kana-kanji mixed sentences is connected to the second CPU II.

FIG. 1 shows an example of a semantic marker used in this embodiment.

For example, (bird) is a bird, (conveyance)
represents a vehicle, (fly) represents something that flies, (walk) represents something that walks, and (run) represents something that runs.

FIG. 2 shows the structure of the word dictionary memory 12.

The word dictionary is composed of a fixed-length index memory 1 containing pronunciations and pointers, and a variable-length word information memory 2 containing notations, grammar codes, meaning markers, and functional meaning markers.

The operation of this embodiment configured as above will be explained below with reference to the drawings.

Voices uttered at appropriate intervals (eg, sentence units) are collected by the microphone 3 and converted into electrical signals. This electrical signal is amplified to an appropriate level by an amplifier 4, and its frequency band is limited by a low-pass filter 5. The output of the low-pass filter 5 is converted into a digital signal by the A/D converter 6 and input to the first CPU 7.

The first CPU 7 processes the input digital signal based on the program stored in the first ROM 8 to recognize the voice, outputs a plurality of kana character candidates for each syllable, and stores them in the shared memory 10. .

For example, FIG. 4 shows an example of kana character candidates for each syllable when uttering "Hato ga Tobi°."

The second CPU II converts the kana character candidates stored in the shared memory 10 into kana-kanji characters using the word dictionary memory 12 to create a sentence containing a plurality of kana-kanji characters. Furthermore, the second
The CPU II orders a plurality of kana-kanji mixed sentences using the grammar memory 13 in a manner described later, and displays them on the display device 16.
Output to.

At the time of kana-kanji conversion, the combination of kana character candidates stored in the shared memory 10 is compared with the readings in the index memory 1 in the word dictionary memory 12 to find matching readings. Then, the word information memory 8 is accessed according to the pointer corresponding to the pronunciation, the notation and grammar code are read out, and kana-kanji conversion is performed using them. Further, the pointer is attached to a candidate sentence containing kana, kanji, and kanji and stored, so that word information such as a meaning marker can be read out as needed.

Next, we will explain how to order multiple candidates for sentences containing kana and kanji.

As a grammar, we use case grammar, which describes grammar mainly using predicates. Semantic markers (conventional semantic markers or functional semantic markers) and particles are used as means for specifying the case frame of each predicate. For example, the conventional semantic marker (human
), the particle ``ga, wa'' is specified, the functional meaning marker (read-and-write) is specified as the objective case, and the particle ``wo'' is specified. Note that (hurnan) represents a human being, and (read-and-write) represents something that can read and write.

Suppose that the kana character candidates in Figure 4 are converted to kana-kanji, and the following candidates for sentences containing a plurality of kana-kanji characters are created.

1 “The back flies 0 2 “The doves fly.

3# Dove Chases 2 The second CPU 11 performs the following processing according to the flowchart in FIG. First, it is checked in Sl whether there are candidates for sentences containing kana and kanji, and if there are, one of the candidates is extracted in S2. In the example, "the back flies" is extracted. Next, in S3, it is checked whether there are any words with case frames. In the example, “fly”
has a case frame and a functional semantic marker (f
ly), the particle "ga" is specified. Case frames are matched in S4, but clause 4 before “fly”
For ``goga'', the particle matches the case frame specification, but ``goo does not have a functional semantic marker (fly).

Therefore, matching of functional semantic markers fails and S
5 priority calculation does not give a very high priority. Next, in S6, it is checked whether there are any candidates for remaining kana-kanji mixed sentences.
If a candidate exists, the process returns to S2 and the above steps are repeated.

In the example, next, “dove flies” is extracted, and since “dove” has a functional meaning marker (fly), a high priority is obtained in the priority calculation of S5, and the output order is also “ "Doves fly" will be in - position.

As another example sentence, suppose you utter the sentence "The penguin chases it." Candidates for sentences containing kana and kanji are: 1. The penguin bears it. 2. The penguin chases it. 3. The penguin flies in the sky.

Suppose that we obtain This time, since "Penguin 2 does not have a functional meaning marker (fly)," the priority of "The penguin flies in the sky" will be low, and the priority of "The penguin chases it" will be high.

(“chasing” and “obliging” each have case frames, but detailed explanations are omitted.) If we try to do what we have explained above using only semantic markers,
It becomes difficult to specify the meaning marker for the nominative of “fly”. If we were to specify (bird), we would also have to specify exceptions such as "penguin" or "chicken".Also, when it comes to "airplane", there is nothing we can do with the meaning marker "(conveyance)", so we can use (, which represents an aircraft). air
craft), and as a nominative meaning marker for “fly”, (birdSaircraft).
It will be specified like this. As mentioned above, if you try to process case grammar using only semantic markers,
This requires measures such as specifying exceptions and subdividing semantic markers, making it difficult to create a grammar. Furthermore, since exception processing and the like are performed, the amount of processing increases.

[Effects of the Invention] As is clear from the detailed description above, according to the present invention, when using a case grammar, an associative value grammar, etc., the grammar can be written more easily than when processing is performed using only semantic markers. , it is also possible to reduce the amount of processing.

[Brief explanation of the drawing]

1 to 5 show embodiments embodying the present invention. FIG. 1 is a diagram showing an example of a word marker, and FIG. 2 is a diagram showing an example of stored contents of a word dictionary memory. , FIG. 3 is a diagram showing the overall configuration of the apparatus of this embodiment, FIG. 4 is a diagram showing an example of a speech recognition result, and FIG. 5 is a diagram showing a flowchart for ordering candidates for sentences containing kana and kanji. In the figure, 1 is an index memory that stores target words, 2
12 is a word information memory that stores semantic markers and functional semantic markers, and 12 is a word dictionary memory that corresponds to the natural language processing dictionary of the present invention.

Claims (1)

[Scope of Claims] 1. A dictionary for natural language processing that stores target words and phrases in association with semantic markers that classify the target words based on relatively superficial criteria, comprising: A dictionary for natural language processing, characterized in that functional and semantic markers classified based on structure are further associated and stored.