JPH05120331A - Homonym recognizing device - Google Patents

Abstract

PURPOSE:To improve the efficiency of recognition by providing a learning function to calculate a hit rate based on the original hit rate concerning the plural candidates of a homonym, to display those candidates in order of a large hit rate value and to transform them. CONSTITUTION:As a control means, a CPU 4 is equipped with a discriminating means, retrieving means and hit rate recalculating means. The discriminating means discriminates whether a predicate exists in a character string inputted from a keyboard 1 and containing a character string to be transformed into the string of Chinese characters (KANJI) or not. When the existence of the predicate is discriminated, in the dictionary of a fixed disk 3, a retrieving object is retrieved while being limited to a case frame concerning the KANJI string to be transformed in each character string according to the case frame decided by the predicate. Next, the hit rate is calculated again concerning the combination with the meaning element of meaning to be obtained by the retrieved predicate. Then, the KANJI strings are displayed at a CRT display device 5 successively from the KANJI string having the combination of the predicate and the meaning element with the higher hit rate.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a Japanese word processor, a computer device incorporating the Japanese word processor, a printing device, a Japanese voice recognition device, an automatic translation device for translating Japanese into another language, and other Japanese It is used as a recognition device. The present invention relates to an apparatus for recognizing Japanese homophones according to the context in the sentence.

[0002]

2. Description of the Related Art The applicant has found that when a kana-kanji conversion device includes a predicate in a character containing a character string to be converted into a kanji input for narrowing down a kanji string of a homonym,
Japanese Patent Application No. 1-313045 was filed for a technique that improves the homonym selection efficiency by narrowing down the homonyms to be converted according to the case frame concept.

This invention introduces the concept of "case frame" or "enlarged case frame" into a kana-kanji conversion dictionary, and is equipped with a predicate dictionary describing case frames and a term dictionary describing semantic features. By extracting predicates and matching which case position in the case frame of the noun phrase in the sentence, the homonyms are narrowed down.

[0004]

However, even in the invention of this earlier application, the recognition rate of the homonyms is reduced for unknown words not registered in the homonym analysis dictionary, and the expanded case frame is also used. There were cases where homonyms could not be recognized even according to the rules.

The present invention solves the above-mentioned problems.
An object of the present invention is to provide a homonym synonym recognition device in which the learning function is provided in the analysis dictionary by frequency calculation and the homonym synonym recognition rate is improved.

[0006]

According to the present invention, an input means for inputting a character string and an extremely large number of kanji strings (a single character or a plurality of characters corresponding to the character string) having a specific meaning and commonly used in Japanese ( A single character may be included including a string of one or more kanji, a kana character string, an English character string, or a combination thereof.
same as below. ) As a dictionary, and a control means for retrieving and reading from this dictionary a kanji string corresponding to a character string input from the input means and to be converted into a kanji string. The dictionary is based on case frames. Discriminating means designed to discriminate whether or not a predicate exists in a character string including a character string to be converted into a kanji character string input from the input means,
When it is determined by this determining means that a predicate is present, a search target is limited to the case frame in the dictionary for the kanji string to be converted of each character string according to the case frame determined by the predicate. In the homophone synonym recognition device having a search means, for a combination with a semantic meaning of a possible predicate searched by the search means, the hit rate is recalculated, and the predicate and the meaning with a high hit rate It is characterized in that it is provided with a means for displaying in sequence from a kanji string in combination with a prime.

[0007]

In the present invention, when the predicate is extracted from the character string, the term of the semantic element corresponding to the case frame of the predicate is selected, and a plurality of candidates for the homonyms are output and displayed, the hit ratio is calculated, Multiple candidates are output in descending order of the hit rate. By recalculating the hit rate each time and displaying the numerical values in descending order, the homonym recognition rate can be further improved.

[0008]

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

FIG. 1 is a block diagram of a homonym recognition device according to an embodiment of the present invention. In the figure, this device includes a keyboard 1 which is an input means for inputting a character string,
It consists of a read-only memory 2 for storing the first-level and second-level kanji sets defined in Japanese Industrial Standard C6226, and one or more characters corresponding to the character string input by referring to this read-only memory 2. Corresponds to a fixed disk control unit 3A that includes a fixed disk 3 that stores a very large number of Kanji strings that have a specific meaning and are commonly used in Japanese as a dictionary, and a character string that is input from the keyboard 1 and should be converted into a Kanji string. And a control means 4 for retrieving and reading the Kanji string to be read from this dictionary.

A CRT display device 5 for optically optically displaying a character string input from the keyboard 1 and a Chinese character string read by the control means 4 is a display character data memory 5A and a dot image display data memory 5B. , A display control data memory 5C and a display section 5D of the CRT system. Floppy disk 7 for storing conversion results
And a printer 8 for printing out. Reference numeral 9 is a write / read memory that can be expanded, and is usually replaced by the fixed disk 3. Reference numeral 10 is a line interface compliant with the RS-232C standard.

Here, in the dictionary stored in the fixed disk 3, dictionary items to which case frames are added (called predicates) and dictionary items to which semantic features are added (terms).
The semantic feature is a marker that displays the approximate meaning of a term such as a noun, and corresponds to the semantic feature in the case frame that is the attribute of the predicate. First discriminating means for discriminating whether or not a predicate is present in a character string including a character string to be converted, and when it is discriminated by the first discriminating means that the predicate is present, the fixed disk 3 is stored. A second discriminating means for discriminating whether or not a homonym is present in the stored kanji string corresponding to this character string, and when it is discriminated by the second discriminating means that a homonym is present. In addition, according to the case frame determined by the predicate, a kanji string to be converted of each character string in which homonyms exist is searched as a search target in the dictionary in which the term with the semantic composition corresponding to this case frame is registered. You In the control means 4 including the search means, the control means 4 includes the recalculation means for the hit rate, and it is determined in the first determination means and the second determination means that there is a homonym, and this determination is made. For multiple homonyms searched based on the result, the hit ratio between the case frame of the predicate and the corresponding term is recalculated, and based on this recalculation result, a plurality of homonyms are ordered in descending order of hit ratio. It is characterized by displaying candidates, improving kana-kanji conversion efficiency, and increasing homonym recognition rate.

Here, the semantic features will be described. The semantic feature corresponds to the semantic feature in the case frame that is the attribute of the predicate by the meaning indicator attached to the “term” such as a noun in the dictionary. HUMN (human), ANIM (animal), PL
AC (location), ACTN (motion), CONC (concrete), ABST (abstract), CHAR (character information), INST (tool means), PLNT
(Plant), TIME (time), PHEN (phenomenon, involuntary movement),
MENT (thinking), DIRC (direction), DEGR (numerical expression, degree), VEHC (vehicle), ROLE (role), ATRB (attribute),
DIVR (no restrictions on the types of terms that can appear)
And the like, and corresponds to the semantic feature in the case frame that is the attribute of the predicate, and the predicate is the word to which the case frame is added in the dictionary.

Hereinafter, the recalculation processing of the hit rate in this embodiment will be described by way of example.

FIG. 2 shows a control flow chart in the homonym recognition device of this embodiment. In the present embodiment, in the conversion result confirmation process, as shown in FIG. 3A, after a selection instruction is given from the candidates and confirmed, a recalculation process of the hit ratio of the semantics is performed, and a plurality of candidates are selected. When displaying, the feature is that the numerical values of the recalculated hit rate are displayed in descending order. By this recalculation processing of the hit rate, learning is performed based on the appearance frequency of a plurality of candidates for homonyms and the recognition rate is improved.

FIG. 4 is a flow chart for explaining the recalculation processing of the semantic elements shown in FIG. 3A. This recalculation processing is divided into recalculation of the hit ratio of the semantic characteristic of the predicate and the semantic characteristic of the term. Be done. The hit rate is represented by HR in the drawing.

The recalculation of the hit rate of the semantic feature of the predicate adds the hit rate of the term having the same meaning as the self to the hit rate of itself between the determined predicate and the term. Further, the recalculation of the hit rate of the semantic features of the term receives and recalculates all the semantic features of the predicate and the hit rate between the determined predicate and the term.

Next, the recalculation of the hit rate of the semantic features of the predicate will be described with reference to its calculation formula.

The recalculation of the hit rate of the semantic features of the predicate is performed by calculating the number of the determined semantic features of the predicate by n, and the original hit rates are vj1, vj2, ... Vjn The sum of the hit rates vj1 + vj2 + ... The hit rates of the semantics of the terms corresponding to them are vk1, vk2, ... vk
n (when there is no corresponding semantic element, the value is 0) The sum of the hit ratios of terms is vk1 + vk2 + ... + vkn = Svk If the new hit ratio of the semantic features of the predicate to be obtained is j1, j2, ... jn, j1 = {(vj1 / Svj) + vk1} / (1 + Svk) * Svj j2 = {(vj2 / Svj) + vk2} / (1 + Svk) * Svj ... jn = {(vjn / Svj) + vkn} / (1 + Svk) * Svj.

In the case of the term of unknown word, the sum Svk of vk1, vk2, ... Vkn and the hit rate is “0”, so jn = {(vjn / Svj) +0} / (1 + 0) * Svj = (Vjn / Svj) Svj = vjn, and the hit rate of the predicate does not change. If there is only one predicate element, then Svj = vj1, Svk = vk1, so j1 = {(vj1 / vj1) + vk1} / (1 + vk1) * vj1 = vj1 and the predicate hit rate is the term It doesn't change regardless of the hit rate.

The recalculation of the hit rate of the semantic features of the term is performed by calculating the number of the semantic features of the determined predicate as m, the original hit rates of ak1, ak2, ... Is 0) Sum of hit rates ak1 ＋ ak2 ＋… ＋ akm ＝ Sak Semantic hit rates of terms corresponding to them are aj1, aj2,…
ajm (If there is no corresponding semantic feature, the value is 0.) The sum of the hit ratios of predicates is aj1 + aj2 + ... + ajm = Saj. = {(Ak1 + aj1 / Saj) / (Sak + Saj / Saj) = (ak1 + aj1 / Saj) / (Sak + 1) k2 = {(ak2 + aj2 / Saj) / (Sak + Saj / Saj) = (ak2 + aj2 / Saj) / (Sak + 1)… km = {(akm + ajm / Saj) / (Sak + Saj / Saj) = (akm + ajm / Saj) / (Sak + 1).

In the case of an unknown word, ak1, ak2, ... akm, and
Since Sak is “0”, k1 ＝ aj1 / Saj k2 ＝ aj2 / Saj… km ＝ ajm / Saj

In the case of the term of the known word, the sum Sak of hit rates is usually "1", so k1 = (ak1 + aj1 / Saj) / 2 k2 = (ak2 + aj2 / Saj) 2 ... km = (akm + ajm / Saj) / 2.

Next, a concrete example of the recalculation processing of the hit rate when a character string is input will be described.

There are dictionaries in the dictionary which are given numerical values of the possible semantic features, surface layer, and hit rate for the following predicates "grow" and "plant", and for terms, "Poinsettia" and "teeth". , "Leaves" have dictionaries, and the semantic features of "growing" and "planting", surface layer, and hit rate are as follows.

[0025]

[0026]

The initial value of the hit rate of the predicate is 1.0 per deep case, and the hit rate of terms is always 1.0 per term.

Here, between the determined predicate and the term, the predicate adds the hit rate of the term having the same semantics as its own to its own hit rate, and the term receives all the semantic features of the predicate and its hit rate. , Recalculate the hit rate. The total hit rate of the predicate candidate group does not change before and after the calculation. By recalculating the hit rate when "Poinsettia grows" as the input string, "Growth" A1 + SEGM is {(1.0 / 2 + 0) / 1} * 2 = 1 B1 + PLAT is {(0.5 / 2 + 0) / 1} * 2 = 0.5 + PART is {(0.5 / 2 + 0) / 1} * 2 = 0.5 "Boinsettia" SEGM (0 + 1.0 / 2) / 1 = 0.5 PLAT (0 + 0.5 / 2) / 1 = 0.25 PART (0 + 0.5 / 2) / 1 = 0.25. In addition, the hit rate in the case of "plant poinsettia" as the input character string is recalculated. Regarding "planting" at this time, jn = {(vjn / Svj) + vkn} / (1 + Svk) * S described above
If you calculate the hit rate with the formula of vj, "plant" A2 + PLAT is (1/2 + 0.25) /1.25*2 = 1.2 B2 + IDEA is (1/2 + 0) /1.25*2=0.8 As for "Poinsettia", the above km = (akm + a
jm / Saj) / 2, To get

Next, regarding "tooth growth", if "growth" is calculated by the formula of j1 = vj1, "growth" A1 + SEGM is 1.0 "teeth" is km = (akm + aim / Saj) / 2 From the formula, "teeth" SEGM (1.0 + 1.0 / 1) /2=1.0

Next, regarding “leaves grow”, jn = {(v
jn / Svj) + vkn} / (1 + Svk) * Svj "grown up" B1 + PLAT is (0.5 / 1 + 0) / 2 * 1 = 0.125 1 + PART is (0.5 / 1 + 1.0) / 2 * 1 = 0.625 To get

Next, regarding "planting Poinsettia", jn = {(vjn / Svj) + vkn} / (1 + Svk) * Svj, jn = vjn
Than Also for "Poinsettia", km = ajm / Sa mentioned above.
When calculated by j formula, Becomes For "planting flowers", jn = {(vjn / Svj) +
"Planting" from vkn} / (1 + Svk) * Svj A2 + PLATA {(1/2) +0.5} /1.5*2=1.333 B2 + IDEA {(1/2) +0.0} /1.5*2=0.667 If you change the order of conversion, "planting flowers" is jn = {(vjn / Svj) + vkn} / (1 + Svk) * Svj jn = vjn, and "planting" A2 + PLATA is {(1/2) +0.5} /1.5*2=1.333 B2 + IDEA = {(1/2) +0.0} /1.5*2=0.667 The result changes depending on the conversion order, and the previous information is weighted.

[0032]

As described above, by providing a learning function for recalculating the hit rate of a plurality of homonymous synonyms based on the original hit rate and displaying the values in descending order of the hit rate, conversion is performed. The recognition efficiency can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram of a homonym recognition device according to an embodiment of the present invention.

2A and 2B are control flow charts of the embodiment.

3 (A), (B), and (C) are the same control flow charts.

FIG. 4 is a control flow chart of the hit rate recalculation process of FIG.

[Explanation of reference symbols] 1 keyboard 1A, 8A interface 2 read-only memory 3 fixed disk 3A fixed disk control unit 4 control means 5 CRT display device 5A display character data memory for storing text character information displayed on display unit 5D 5B Dot image display data memory that stores image data in dot units 5C Display control data memory that stores mixed control data of text character information and image data 5D Display section by CRT system 6 Write / read memory (640K bytes capacity) 7 Floppy disk (capacity 1M bytes and 640K bytes) 7A Floppy disk control unit 8 Printer 9 Expandable write / read memory 10 Line interface

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] October 7, 1992

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] Brief description of the drawing

[Correction method] Change

[Correction content]

[Brief description of drawings]

FIG. 1 is a block diagram of a homonym recognition device according to an embodiment of the present invention.

FIG. 2 is a control flow chart of the embodiment of the present invention.

FIG. 3 is a control flow chart of the embodiment of the present invention.

FIG. 4 is a control flow chart of the embodiment of the present invention.

FIG. 5 is a control flow chart of the embodiment of the present invention.

FIG. 6 is a control flow chart of the embodiment of the present invention.

7 is a control flow chart of the hit rate recalculation process of FIG. 4;

[Procedure Amendment 2]

[Document name to be corrected] Drawing

[Correction target item name] All drawings

[Correction method] Change

[Correction content]

[Figure 1]

[Fig. 2]

[Figure 3]

[Figure 4]

[Figure 5]

[Figure 7]

[Figure 6]

Claims (1)

[Claims] 1. An extremely large number of kanji strings (one or more kanji strings, which consist of one or more characters corresponding to the character string and have a specific meaning and which are commonly used in Japanese, A kana character string, an English character string, and a combination thereof, and may be a single character. The same applies hereinafter), and a storage means for storing as a dictionary, and a character string input from the input means and to be converted into a kanji character string. The dictionary is designed based on the case frame, and the predicate exists in the character string including the character string to be converted into the Kanji string input from the input unit. Determining means for determining whether or not to perform, and when the determining means determines that a predicate exists, the dictionary for the kanji character string to be converted of each character string according to the case frame determined by the predicate In the homonym recognition device provided with a search means for limiting the search target to the case frame in the above, in the combination with the semantics of the possible meaning of the predicate searched by the search means, hit A homonym synonym recognition device comprising means for recalculating the rate and displaying in order from a kanji string of combinations of predicates and semantics with a high hit rate.