JPS62131372A - Natural language analyzing system - Google Patents

Abstract

PURPOSE:To perform quickly matching based on the patterns of verbs to easily perform plural context processings by reading sentences one by one to generate a control word which links preceding and succeeding sentences and listing independent words in sentences by preliminarily determined conceptual codes and linking independent words in the order of appearance. CONSTITUTION:Word signals of a morpheme analyzing part 1 which cuts out words from an input character string are stored in a queue 7 successively. A grammatical construction meaning analyzing part 5 and the morpheme analyzing part 1 perform asynchronous operations by pipeline connection. A grammatical construction meaning analyzer 12 takes out one word element from the queue 7; and if it indicates a pointer 52 to meaning information meaning dictionary information is stored in a buffer 11 from a declinable word meaning dictionary 3 or a substantives meaning dictionary 4 through a register 8 for dictionary input. If the word is a verb at this time, pattern matching between case patterns in the declinable word meaning dictionary 3 and nouns is performed. Thus, words are read out successively from the queue 7 to terminate the processing of one sentence.

Description

[Detailed description of the invention] (Industrial application field) The present invention relates to a natural language analysis method.

(Prior Art) Conventionally, as a (semantic) analysis method for natural language, one based on case grammar is known. This method is used, for example, in "Language Engineering," 1984, Makoto Nagao, PI) 163-P.
As described in P167 J, by defining a case pattern for each verb and pattern matching the rp and words in the sentence with the case pattern of the verb, we can formulate the action nominative case (e.g. human) and the object case (e.g. object case). ), the instrumental case (for example, a tool such as a key), etc. are determined. That is, this method searches for conceptual connections between words. Usually, several to several tens of such case types are predetermined. Case expressions are independent for each sentence, and there is no linking of information between multiple sentences. However, in order to see the anaphoric relationship of pronouns that span the span of 'Pi numbers, it is necessary to
As described in 00 to PP202, nouns are stored in a bushdown stack, and the first synonym that matches semantically in the stack is assumed to match that pronoun.

(Problems to be Solved by the Invention) However, in the conventional method described above, the analysis process is basically independent for each sentence, and the analysis process is performed only according to the case pattern matching results. Therefore, there was a problem in that it was unsuitable for performing context processing in particular in a plurality of sentences.

Therefore, the present invention aims to solve the above problems.

(Means for solving problems) The present invention is configured as follows.

(a) Read the text sentence by sentence.

(b) Create a control word to link the read sentence with the previously read sentence.

(c) List each independent word in the sentence by predetermined concept code.

(d) Link each independent word in the sentence in the order of appearance.

(e) Referring to a pattern match between a predetermined concept code corresponding to a verb case and a noun concept code,
Perform pattern matching between verbs and nouns in sentences.

Cf> Link the verb and noun that match the pattern.

(g) When the next sentence is read, a control word is used to link it to the facing sentence.

(Action) Creating the control JΔ in the above (b) for the sentence read in the above (a) has the effect of linking the sentences. In (c) above, by linking the analysis process with the concept of an independent word, in addition to speeding up the matching using each verb pattern, the meaning of the sentence can also be analyzed from the concept. (d), (e), (f) above
exhibits a matching effect depending on each pattern of verbs.

The above (g), together with the above (b), functions to link sentences and grasp the concept of independent words across a plurality of sentences.

(Example) First, the procedure of the present invention will be explained in detail with reference to FIGS. 1 and 2.

FIG. 1 is a flowchart showing the procedure of the present invention, and FIG.
The figure shows an analysis result (b) when the analysis process according to the present invention is applied to the example sentence (a). In FIG. 1, a sentence is first read (step 101), and a control word is created (step 102).

The control word is information for linking sentences, and is shown as SCW in FIG. Next, each independent word in the sentence is listed by concept code (step 103). This conceptual code is shown as C, -Cn in FIG. For example, if C1 is a person, C2 is a time, C3 is an object, and C0 is an action, when the example sentence in Figure 2 (a) is read, each independent word is listed as shown in Figure 2 (b). Ru. For example, in the concept of C1, independent words "A" and "B" are linked as shown by arrow 28 in the order of appearance. Next, each independent word is linked in the order of appearance (step 104). This link is indicated by arrow 27 in FIG. Next, pattern matching is performed between the verb and the noun in the sentence by referring to a predetermined pattern match between the concept code corresponding to the case of the verb and the concept code of the noun (step 105).

For example, regarding the verb “read”, it is assumed that “C2 is (is) C
If a concept code is specified such as "3", the independent word in C1 and the independent word in 03 are looked up, respectively.Then, the verb and noun that match the pattern are linked (step 106). In the case of ``read'', there is ``A'' in C1 and ``book'' in C3, so these are linked as shown by arrow 29.Additional words (particles, etc.) are linked directly to them. It is assumed that the control word is attached to the gender-independent word.When the processing of one sentence is completed and the next sentence is read (steps 107 and 101), a control word is created to link with the previous sentence. (Step 102).The sentence just read is then listed as shown in Figure 2(b) (
Step 103), pattern matching is performed (Steps 104 to 106).

Next, a configuration example for implementing the present invention will be described.

FIG. 3 is a block diagram showing the hardware of this configuration example. In the figure, 1 is a morphological analysis unit, 2 is a morphological dictionary (MD), 3 is a word-semantic dictionary (VO), 4 is a nominal-semantic dictionary (SO), 5 is a syntactic-semantic analysis unit, and 6 is a morphological analysis unit 1 A signal line 7 connects the output of
is a queue (FIFO) that stores data in first-in, first-out format, and 8 is a register for dictionary manual operation (D
! 1) and 9 are control register groups (
cR5), 9-1 is the current SCW address (c5CW), and 9-2 is the current embedded statement nest number (
GNNO), 10 is a porter (cT) to a case table (cT) that associates particles with independent words connected to them.
PS), only the embedded statement level exists. For example, in the example shown in FIG. 2(a), there is one embedded statement, so there are two pointers at level 0 and level 1. 11 is a buffer (TB) that sequentially stores the results of analyzing sentences.
), and the analysis result of FIG. 2(b) is stored here. 1
2 is a syntactic and semantic analyzer (SSA), 13 is a morpheme dictionary 2,
Pragmatic meaning dictionary 3, nominal meaning dictionary 4, and syntax-semantic analysis unit 5
A signal line 14 is an internal bus used for syntactic and semantic analysis.

FIG. 4 is a diagram showing the internal structure of various control words that appear in the meaning expression of a sentence. 30 is a control word (SCW);
1 is a forward pointer (FPTR), which corresponds to the SCW of the next statement. 32 is a backward pointer (BPTR), which corresponds to the SCW of the previous sentence. 33 is the word pointer (WPTI()
This corresponds to arrow 27 in FIG. 2(b). 34 is the sentence number (STNO) 1, and 35 is the internal representation of the independent word. 36 is a forward pointer (FPTR), 37 is a word pointer (WPTR), 38 is an embedded sentence nesting level number, 39 is attached word information connected to this word,
40 is morphological dictionary information, 41 is a semantic dictionary pointer, 42
is a part that differs depending on the part of speech and is used for various types of control. The independent words "Aoo", "yesterday", etc. in FIG. 2(b) are expressed by elements with reference numbers 36-42.

FIG. 5 is a diagram showing the format of data flowing through the signal line 6 in FIG. In the figure, 47 is a pointer to the headword, 48 is part of speech information, 49 is a conjugation form, 5o is an elemental concept that the word has,
Reference numeral 51 indicates connection information with preceding and succeeding words, and reference numeral 52 indicates a pointer to the lexical meaning dictionary 3 or the nominal meaning dictionary 4 in FIG.

FIG. 6 is a diagram showing the correspondence between particles and verbs connected thereto. 55 is each table (cT) generated one for each sentence level, 56 is the entry number of the first occurrence of a part of speech, and 57 is the entry number of the last particle occurrence. 58
．． 59 are paired for each particle code entry, and 58
59 indicates a pointer to an independent word connected to the particle, and 59 indicates a portion containing a pointer for linking the particles in the order in which they appear.

Next, the operation will be explained.

First, an input sentence is input to the morphological analysis unit 1 one character at a time. The morphological analysis unit 1 performs the work of cutting out words from an input character string. This is usually done using the morpheme dictionary 2 using the known longest match method or the least clause method. Cut out middle 0
The words are sequentially stored in queue 7 through signal line 6 in accordance with the data format shown in FIG. Therefore, morphological analysis section 1
and the syntax analysis unit 5 operate asynchronously (pipeline connection). The syntax 7α taste analyzer 12 generates in the buffer 11 an expression of the analysis result shown in FIG. 2(b). At the beginning of the sentence @ (before the sentence begins), 5CW16 is placed in the syntactic and semantic analyzer 12, and pointers for each element concept 01~
Cn is secured at consecutive memory addresses, and its interior is initialized to zero clear. At this time, CN N O9-2
Also set it to zero (the top level of the sentence is 0). Here, the start address of the direct SCW is control register group 9
Since it is set in C3CW9-1 in C3CW9-1, read it and write the address of the newly secured SCW to the first FPTR31 pointed to by that address, and write it to the BPTR32 of the newly secured SCW to write the new SCW address. Load into C3CW9-1. This allows the Tatego text to be linked. After this work, syntactic and semantic analyzer 1
Step 2 begins syntactic and semantic analysis of the sentence.

First, one element is taken from the cue 7, and if it points to the pointer 52 to semantic information, follow that pointer to the pragmatic semantic dictionary 3 or the nominal semantic dictionary L! From F 4, α
The taste dictionary information is stored in the buffer 11 via the dictionary input register 8. Then, the value of the pointer 52 is rewritten to the start address of the semantic dictionary stored in the buffer 11. Then, the syntactic and semantic analyzer 12 secures the area outside the area indicated by the internal representation 35 after independence in the buffer 11. Then, this start address is written to WPTR:13 of the SCW. In addition, since the element concept to which this part belongs is obtained from the area with reference number 50 in the data format in FIG.
Change δ to the address secured within. With the above operation,
This means that part 21 of A in the example sentence of FIG. 2(a) has been generated. Note that the nest 38 stores the value of CN N 09-2.

Next, when one element is extracted from the queue 7, it is an attached word <is>, and this information is written in the area 39. Furthermore, in particle processing, the case table ([';
In T), write the pointer section to the independent word of the corresponding particle entry section and the appearance link section. If it is found that the sentences are nested, the syntax-semantic analyzer 12 converts GNNO9-2 into r
Increment and lower the level by one), buffer ll
An area for a new case table 55 is secured within the area, and its start address is written to the corresponding portion of CTPSIO. When returning from the nest, set the corresponding part of (: TPSlo to zero, and (:
Decrement NN09-2.

When a verb is retrieved from the queue 7, an area 35 is reserved in the buffer 11 and a value is set, similar to other independent word processing. Verb processing involves pattern matching between case patterns existing in the word-semantic dictionary 3 and nouns. If Kowa uses the links 17 to 20 by element concept and the element concept in the case pattern, and if the particle code taken by the case pattern is specified in the semantic dictionary, priority is given to case table 55 corresponding to CNN09-2. to search for the specified particle code entry. In particular, when the verb is in the adnominal form, processing is performed after the next noun is processed. If it does not exist in this case table, it is searched for using the case table at the next higher level. In the case of the example shown in FIG. 2(a), pattern matching in <read> 23 is performed at level 1 since it is an embedded sentence, and pattern matching in <return> 26 is performed at level 0. For those whose patterns match, a pointer to the case pattern-compatible word in the word-semantic dictionary 3, which is transferred to the buffer 11, is written. Note that in the example of FIG.
is not marked. This is because these are not required paths.

When the processing of one sentence is completed as described above, the syntactic and semantic analyzer 12 clears CNN09-2 to zero and repeats the processing of the next sentence.

As described above, in this configuration example, the matching results based on case patterns and the links of words by element concept are integrated and managed in one area, so the meaning of the sentence is not only verb-centered, but also linked by element concept. It can be easily extracted from word strings that are In addition, since sentences are linked and managed, it is possible to trace back the same elemental concept, making it easier to search for abbreviations and pronoun anaphors. Furthermore,
By using a case table and specifying the level numbers of embeddings, matching case patterns becomes fast and easy. Therefore, natural language analysis that performs efficient context processing becomes possible.

(Effects of the Invention) As described below, according to the present invention, it is possible to easily and reliably perform context processing over a plurality of sentences. The present invention is suitable for application to computer interfaces, document processing, machine translation, and the like.

[Brief explanation of drawings]

FIG. 1 is a flowchart showing the procedure of the present invention, FIG. 2 is a diagram showing an example of an analysis result according to the analysis process of the present invention, FIG. 3 is a block diagram of a configuration example for implementing the present invention, and FIG. 5 is a diagram showing the internal structure of various control words, FIG. 5 is a diagram showing the format of data flowing through the signal line 6, and FIG. 6 is a diagram showing the correspondence between particles and verbs connected thereto. 1... Morphological analysis unit, 2... Morphological dictionary (M
D), 3... Vocabulary semantic dictionary (VO), 4... Denotative semantic dictionary (SD), 5... Syntactic and semantic analysis unit, 6...
・Signal line, 7... Queue (FIFO), 8... Dictionary input register (DR), 9... Control register group ((:R5), 10... Pointer ((:TP
S), 11...-Buffer (TB), 12... Syntax and semantic analyzer (SS8), 13... Signal line,
14-...Internal bus,) 5...Example sentence,
16...Le control word (SCW), +7~20-...
Concept code (c+~C,) 27...Independent words connected in order of appearance, 28...Independent words connected by concept,
29...Verb case pattern and its equivalent word connected, 30...Control word (SCW), 31...Forward pointer (FPTR), 32...Backward pointer (BPTR), 33...・Word pointer (WPTR), 34...Sentence number (STNO)
, 35... Internal representation of independent word, 36... Face pointer (FPTR), 37... m-word pointer (WPTR), 38... Nesting level number of embedded sentence, 39... Attachment Word information, 40...Morphological dictionary +? I report, 41...Semantic dictionary pointer, 47...Pointer to headword, 48...Part of speech information, 49...Conjugation form, 50...Elemental concepts possessed by words, 51...Interview Connection information with the word, 52...Pointer, 55...Case table (cT)
, 56... Entry number of the first part of speech, 57... Entry number of the last part of speech, 58... Pointer to the independent word connected to the particle. 59...Pointer for linking particles in order of appearance.

Claims (1)

[Claims] (a) Read a sentence sentence by sentence, (b) Create a control word to link the read sentence with a previously read sentence, (c) Each independent word in the sentence (d) link each independent word in the sentence in the order of appearance; (e) pattern match the concept code corresponding to the predetermined verb case with the noun concept code. , perform a pattern match between the verb and noun in the sentence, (f) link the pattern-matched verb and noun, and (g) use the control word when the next sentence is read. A natural language analysis method that is characterized by linking with the previous sentence.