JPH0264867A - Correspondence relation deciding system for character description result - Google Patents

Abstract

PURPOSE:To eliminate a trouble to prepare a dictionary by converting character described result into a phoneme row, and deciding the correspondence relation of the character described results by calculating the degree of difference at a phoneme level. CONSTITUTION:The title system is provided with a means to convert the objective character described result to be compared into the phoneme row, the means to calculate the degree of difference at the phoneme level and the means to decide the correspondence relation between the character described results themselves according to the degree of difference at the phoneme level. As for English language description, the means to convert English described result into the phoneme row consisting of the phonemes to be used in English language is provided, and in the case where the correspondence relation between KATAKANA (square form of Japanese syllabary) described result and the English described result is desired to decide, a tabling means to generate one or plural Japanese phonemes corresponding to each phoneme to be used in the English language is provided. Thus, a foreign language description dictionary need not be generated beforehand, and continuous renewal such as the registration of a newly-coined word, a derivative and a proper noun is unnecessary as well.

Description

DETAILED DESCRIPTION OF THE INVENTION A. Field of Industrial Application The present invention relates to a system for determining the correspondence between the results of notation using the same or different character notations, and more specifically, converts the results of character notation into a phoneme string, The present invention relates to a system that determines the correspondence of character transcription results by calculating the degree of dissimilarity at the phoneme level. The correspondence here may be determined in a binary manner such as correspondence or not (correspondence exists or no correspondence), or it may be determined in more detail depending on the degree of difference, for example, correspondence is strong, weak, or It may be determined that there is no such thing.

B. Prior art and its problems When writing foreign words using phonetic alphabets such as Japanese kana, there are often multiple ways to write them.An example of the original notation "interview" Accordingly, the notations include 1nterviey interview, interview, interview, interview, interview, interview, etc. These multiple notations all represent the same word "interview". However, 1, for example, "interbeen" does not represent this.

In situations like this, where it is required to determine whether or not the notation results represent the same word.

There are proofreading systems and information retrieval systems for Japanese word processors.

In the conventional proofreading system, Tanaka et al. "Analysis of Katakana sequences in scientific and technical literature abstracts" (Metric Japanese Linguistics Vol. 1.14,
No. 1, 1983), a dictionary system was used to register pairs of <original notation result - kana notation result>. Furthermore, in conventional information retrieval, a dictionary system is disclosed in Japanese Patent Application Laid-open No. 11932/1983.

However, the dictionary system has the following first-order problems.

・Since all newly coined words, derived words, and proper nouns must be registered in the dictionary, it takes time to create a dictionary.

Continuous updates are required.

- If the kana notation cannot be determined, all kana notations must be covered.

On the other hand, as a method for determining the correspondence between kana notation results, in addition to the method of storing all of the kana notation results that have a correspondence as suggested in the above publication into one entry in the dictionary, There is also a way to use uniform notation. A method for obtaining a unified notation is disclosed in a paper by Goto et al., ``Diversity of notation in technical terms that use katakana notation'' (Mita Library and Information Society Conference, 1985).

This applies exchange rules (for example, long sound, deletion of consonants, capitalization of consonants, rewriting f sound → h sound, V sound → b sound, etc.) sequentially from the first character string to the kana string to be processed. Apply it to get unified notation results. Then, it is determined whether or not there is a correspondence between kana strings based on whether the unified notation results match or do not match.

This method has the following problems.

In order to prevent groups of notations representing the same word from being judged as different words, it is necessary to set as many conversion rules as possible, but this increases the probability that different words will be regarded as the same.

For example, "alternative" and "alternative" both represent "alternative." In order to unify this, we add the conversion rule (゛ru' → long sound), which is combined with the long sound sign deletion rule to become (゛ru' → long sound → deletion). Therefore, this can be called "buckle". When applied to , it becomes ``back'' and it becomes impossible to distinguish between the two.

Means for Solving the C0 Problems The present invention has been made in view of the above problems, and includes means for converting character notation results to be compared into phoneme strings.

The present invention is characterized by comprising means for calculating the degree of dissimilarity at the phoneme level, and means for determining the correspondence between character notation results based on the degree of dissimilarity at the phoneme level.

For example, for katakana notation, means is provided for converting the katakana notation result into a phoneme string consisting of phonemes used in Japanese. Note that the phonemes used in Japanese in this specification are:
This is a reading that includes the concepts of consonants, long consonants, and consonants.

For English notation, a means is provided for converting the English notation result into a phoneme string consisting of phonemes used in English. When it is desired to determine the correspondence between the Katakana notation result and the English notation result, a table means is prepared that generates one or more Japanese phonemes corresponding to each phoneme used in English.

D. Embodiment D1: System for determining correspondence between English notation and kana notation The present invention will be described below with reference to a system for determining correspondence between English notation and kana notation. First, the flow of a series of processes performed by the system is shown in Figure 1, D2 to D.
8, details of each means shown in FIG. 1 will be explained.

D2. Japanese phoneme string generation means ■ Input the katakana notation result in the form of a character code from the keyboard, for example.

■ Check Table 1 for each letter to obtain the corresponding phoneme string.

(Example: "player" → phoneme sequence: pureiya) Consonants (tsu), long sounds (-), and persistent sounds (ya, yu, yo) are not used alone, so they are not considered phonemes, and the previous phoneme is (X l
Then, the consonant and long consonant have the length of X, and the consonant has the length of X.
It is treated in the form of a consonant element.

(See D4◎) The notation is X-+X, X-→X.

tsu X-+X,,X →X.

Ya Ja Ja Let X be -4X.

Yo J 0 Table 16 Japanese Kanah Phoneme Sequence Table ■ From left to right of the phoneme string obtained in ■, apply the conversion rules described below to obtain a unified phoneme string.

The main conversion rules are listed below.

1. Rule for deleting long sounds sandwiched between vowels Example: Kei gi-+ to Keui 2. Conversion rule for semi-vowels (y, w) w −'+ u other than the beginning of a word Example: sanldowici-+5 antdouici y −* i other than the beginning of a word Example: Pureiy-a → Pureiia 3. Conversion rules for diphthongs: Let the preceding vowel be V > + the following vowel v2.

The vowel string viv2 is converted according to Table 2.

Table 2 Conversion table for diphthongs (VIV2) - Mu (m u ) in front of the Pa line and B line → N (n example Rump → Rump 5, rules regarding vowels that are easily devoiced - kis-+kus Example Text → Chixt ( tekisuto) (tekusuto) Example "Player" Example here Blank: No conversion v : v Macon sign (): Optional a O→ a O e 1 → e Ua →ua, a, ua, a4, syllabic sound (n) Rules for deletion of the syllables before the lines S and M (disclosed in the above-mentioned paper by Goto et al.) Example Channel → Channel In the working area in memory, -
Then, store it.

D3. English phoneme string generation means■ Input the English notation result in the form of a character code from the keyboard, for example.

■ Obtain phoneme sequences from English spelling.

This algorithm already exists; for example, Elovitz et al.
Rules for Automatic Trans
lation of English Text
Phonetics” (ENGεEE Trans, Vo
ASSP-24, No. 6.1976), etc., so please refer to these documents for details.

In this specification, the English phonemes shown in Table 3 are used.

(■ Temporarily store the phoneme string obtained in ■ as the English phoneme string in the given English notation in the working area in memory.

Store.

Table 3 English phoneme symbol table D4. English phoneme, Japanese phoneme correspondence table means ■ Expression format of phonemes In order to compare English phonemes and Japanese phonemes, the expression format of phonemes is defined as follows.

Consonant phoneme (c) Mi Rc + F 1 + F 2 Vowel phoneme (v) ERv 1 + F 1 + F 3 + F 4 R (.) Corresponds to a phoneme that belongs to two consonants, Yomi R (V) in Japanese, corresponds to a phoneme that belongs to vowel. .

Reading in Japanese The J sounds of consonants, long sounds, and persistent sounds in Japanese are not included in the reading.Take ``Chara'' as an example.

Japanese phoneme sequence: kja Rc==, and Rv=a.

Fl: Priority order when there are multiple readings in Japanese corresponding to one phoneme F2: Presence or absence of a consonant (j sound) F3: Indicates whether it is a consonant or not.

F4: Taking "Chara" as an example, which indicates whether it is a long sound or not, the consonant phoneme = Rc (Yomi: k) + F1 (Priority: ■)
+F2 (Sound: j) Vowel phoneme = Rv (Yomi: a) 10F1 (Priority: ■) +
F3 (consonant = 6) 10F4 (20 long sounds) ■ Input the English phoneme string obtained in D3■.

■ Convert the English phoneme string from Table 2 to the expression format explained in ■.

In addition, the items in Table 4 @: Type of phoneme (c: consonant, V: mother) sound) 10■: If there is no vowel after a consonant, Represents the vowel added in Takana notation. vinegar. For example, /1/ in cat is day In Japanese, it is expressed as to. Therefore, +v=”o’.

F2 of vowel phoneme: represents adding a persistent consonant element to the preceding consonant. When comparing English and Japanese phoneme levels, it is used in addition to the preceding consonant F2.

Table 4 @: Phoneme BX c: PX BX c: BX BX c: TX c: TX BX c: DX c: DX BX c: KX c: KX c: KX BX c: GX English phoneme, Japanese phoneme correspondence table Sound: Fl: R (yomi): F2 (+j):
+v: Plosive ->/P/ :1 :p -:u Knee>/b/ :1 :b' :u Knee>
71. / :1 :t :2 :o ::2
:C° :u Knee＞/d/ :1 :ci'
:o ::2:Z ~>/ni/ : 1 : k 222g :3:C ->/g/ =1:g : u : : u : : u : : i : : u : c : GX c : GX FX c:MX c:MX FX c: NX c:NX NG c:NG c: NG FX c:LX FX c:FX c: FX x = 2: k: 3:z Nasal ->/m/: 1 2m :1:n! ->/n/ :1:nj :1:n ->10/ :1:n! :1 :n! gu consonant ->/ Q / :1:r fricative ->/f/ :1:h : 2 : b ->/v/ CH c : SH c : SH c:SH CH c : ZH FX c:HX c: HX FX c: RX CH c:CH c:CH CH c:JH c:JH c:JH c:JH ->/S/ :1:S :j :2:z :3:Shi->// :1 :Z ->/h / :1:h =220->/r/ :1:r Affricate->/15/ :1:c : 2 :k ->/d3 / =1:Z : 2:3:C:32g:3:0:u::u::U c:VX c:VX c:VX CH c:TH c:TH c:TH CH c:DH c: DH c:DH FX c: sX c: sX c: sX FX a : ZX c : ZX :1:b =3:0 :3:u ->10/ :1:S :2:z :2: ->/Convex/ :1:z :2:S :3: ->/s/ :1:S :2:z :3:C ->/z/ =1:z :2:S : u : : u : : u : : U : 0 : : u : Special symbol corresponding to plural c:Ts :1:c : :u :/*=T
XSX For example tea<ts) umbrella/c:DS:
1:z: :u:/$=I)XZX For example, c a r<d s)*/semi-vowel @:phoneme: Fl:R(yomi):F2(+
j):+v:JH->/j/a:JX:1:y+:u:
v:JX :1 :i ”WX ->/
W/ v:WX :1 :o ・v:WX
:1 :u'c:WX :3 :b
0: u: vowel @: 赫: Fl: R (youi): F2 (+j):
+v: F3(,): F40:IX->/i/ v:IX:1:i: : :2 :O:v:
IX:2:ai : : :O:O:v:IX:
3:e:::1:'7:v:IX
:3 :u :EE->/i:/v:EE :1 :i :v:EE
:3 :ia :v:EE :3 :i
e v:EE :2 :e :v:EE
:3 :ea :v:EE :3 :av:EE :3 :ai :v:EE :
3 :aie :E I ->/e/v:EI :1 :e v:EI :1 :av:EI :3 :ai v:EI :3 :1 EH->/ε/v:EH:1 :ev:EH:2 :iv:EH:3 :u v:EH:3 :a :1 :1 ::1 ::? : :O : :2 Salary j 9 :1: Nii? ::? :? v:AXIX:lai 20v:AXIX:2:i:3 AXUX ->/au/ v:AXUX:1:au: v:AXUX:2:a: v:AXUX:2:o: v:AXUX: 3:u: ○U->10/v:OU:l:o:v:OU
:2 :a :v:OU :3 :u
:V:○U:3:ui: v:OU :3 :oa :AW->/)/v:AW :1 :o :v:AW :
2:a:v:AW:3:ao
:v:AW :3 :au :AW
IX->10i/v:AWIX:1:oi: 20:0:3 20AE->/;e/v:AE:1:a:::2
:1 :v:AE:2:e :
: :') :3 :v:AE
:3:o : : :? :
? :UH->/A or lai v:tJH:1 :a : :
:2 :1 :v:UH:2:o
: : :O:1 :v:UH:2
:u :1 :1 :1 :
O:v:UH:3:e:::
? :':' :v:UH:3 :ia
: : :? :';'
:ER->/sa/ v:ER:1:a : : :O:
3:v:ER:3:e:;
:? :? ：Er-＞／＆／／
*ER+X (vowel) >Er+RX+X Liao/v:E
r:1:a:::?
:? :V:Er:1:ua:
: :? :Q :AA->/ / v:AA:1:o : : :2
:1 :v:AA:2:a :
: :1' :1 :AX I
X ->/a i / UX->/u/ v:UX :1 :u :UU->/u:
/v:UU:1:u:v:UU:
3:o:WX v:WX:1:u:v:WX:
3 :o :j xUU ＞/ j u
: / v:jxUU:1:u :2 v:jxUU:3:a :O v:jxUU:3:o :1 ->/:/ vknee:1:o: v: :2 :a ::3 :0 : :1 :3 ::
:'il :1 :: :3
:O: : :? :? :=2 :O 2020=1 :? : :O:3 : : :O:3 : ■ Each phoneme of the English phoneme string and the conversion result obtained in step (■) are paired and temporarily stored in the working area of the memory.

Example 11 cuPI+English phoneme string: KX UHPX, and the conversion result of these phonemes is as follows.

Phoneme: Conversion result (1) Adjustment of the reading in Japanese corresponding to the phoneme Adjustment rules that take into account the phonetic environment before and after the phoneme, and the corresponding spelling, the priority of the reading (Fl), the possibility of a persistent sound (F
2) Change.

The main adjustment rules are described below.

Rules are expressed in the form of phoneme: condition → action to be taken when the condition is met.

MX: The following phoneme is I'X, BX, MX → Fl (reading: n ri = 0 In writing, it becomes "lamp".

NX: The subsequent phoneme is TX, DX, NX → Fl (reading: n ri = 0 PX: (Rc (P) 10 F1 (■) + F2 (20 sounds)
] D5. English phoneme string conversion result adjustment means ■ Adjust the English phoneme string conversion results obtained in D4■ according to the phonetic environment and spelling. The two adjustment items will be explained below.

In the book, it is called ``tent.''

EE: The corresponding spelling is e' → Fl (reading: e) = 0 Example: The kana notation for "meter" is "meter" The kana notation for "action" is "U"
Read H as 0'.

AE: The preceding phoneme is KX or GX, and the following phoneme is PX, BX, TX, DX, KX, GX, RX → F2=2 The preceding phoneme is KX or GX, and the following phoneme is (PX, BX, TX, DX, KX, GX, R
It is a consonant other than X) → F2 = I UH: The corresponding spelling is a' → Fl (read: a) = 0 Example The kana notation for "China" is "China" and UH is read as % aI.

AA: The following phoneme is the elongated sound ゛-'→Fl (reading:
a) = 0 Example The kana notation for "part" (PXro Prisoner 2 TX) is "bert" and AA is read as a'.

(2) Adjustment regarding the length of vowel phonemes located immediately before consonants or at the end of words In Japanese, the length of vowels is an important element for distinguishing words. Here, regarding the length of the vowel,
The vowel length adjustment value (denoted as Al) is calculated according to the adjustment rules described below. In D6 phoneme level difference calculation, this value is F3 (presence or absence of consonant), F4 (presence or absence of long consonant)
Use for comparison.

The corresponding spelling is -O/ → Fl (read: o) = 0 For explanation, the phoneme sequence of the target word is...v c
v2... (v: vowel phoneme to be adjusted, C: subsequent consonant phoneme, v2: vowel phoneme following c).

■ If the following consonant phoneme (c) is PX, TX, KX or above → A1 = 0 ■ If the subsequent consonant phoneme (c) is other than (PX, TX, KX) ■ Conversion adjusted with English phoneme and ■ The result and vowel length adjustment value (A1) are stored as a pair in a work area in memory.

D 6 , phoneme level difference calculation means ■ The Japanese phoneme string obtained in D2, ■ (hereinafter referred to as J phoneme string), and the adjusted English phoneme string conversion result obtained in D 5 tD (hereinafter referred to as E phoneme string) ) as input.

■ Calculate the degree of difference between the J phoneme string and the E phoneme string according to the procedure described below. First, the concept of a chunk will be explained (■-0 chunk) A chunk as used in this specification refers to a block created by dividing this phoneme string at the beginning of each consonant.

Example 1 “alternative” The English phoneme sequence is AW-LXTXERNX It becomes TJHTXI XVX.

From left to right, the consonants are LX, TX, NX, and TX.

Since it is VX, AW -I LX I TXERI NXU
HI TX I The consonants are t, n+ t, b from left to right, so it is alternative
lbu and divided into five chunks.

Example 3 In the case of “F Keerai”, there is one consonant (k), so
1 yank.

■-1 Chunk number matching is performed on the J phoneme string and E phoneme string of the phoneme level phase calculation, and the first matching (consonant part reading Re) and second matching are performed in order for those with the same chunk number, Penalty given to applicable items −
Let the sum total be the dissimilarity degree.

First, matching the number of chunks is performed.

(1) When the number of chunks in the rain phoneme sequence differs by 2 or more.

Representation results with these phoneme sequences are considered to be inconsistent. (Difference level = 100X difference in the number of chunks of both phoneme strings) (II) If the number of chunks of the rain phoneme strings differs by 1, then let the one with the larger number of chunks be the X phoneme string and the other - the X' phoneme string. Which chunk of the X phoneme sequence does not correspond to the X' phoneme sequence is determined by the following method. A chunk sequence created by sequentially removing one chunk from the first chunk of the X phoneme sequence,
A first matching is performed on the chunk sequence of the X′ phoneme sequence. When a chunk string is created by removing the i-th chunk from the The i-th chunk is regarded as a "non-corresponding chunk."

For example, in the case of "Keys" (first chunk: KxEE, second chunk: ZX) and "key" (first chunk: Ki), the second chunk (ZX) of "Keys" is the "uncorresponding chunk."

``If there is no "non-corresponding chunk", the two notations are considered to be inconsistent (difference = 100) "non-corresponding chunk"
If there is a consonant part, and the consonant part is listed in Table 5, a second matching is performed on the X' phoneme sequence by removing the "uncorresponding chunk" from the X phoneme sequence. The degree of dissimilarity is the penalty in the first matching - the sum of the penalties in Table 5 - the penalty in the matching of the two children - the sum.

11 If there is a chunk" that does not correspond and its consonant part is not listed in Table 5, both notations are considered to be inconsistent. (Difference = 100) Table 5 Consonant phonemes that do not need to match and penalties - Reading of the consonant part of the J phoneme sequence (JRc
) and reading candidates for the consonant part of the E phoneme sequence (ERci)
are compared in order starting from the first yank. E Rc i is shown in Table 4
It is given by R(yomi).

Example “cat” J phoneme sequence: kjat.

↓ Expressed by a combination of consonants and vowels.

(III) When the number of chunks match.

Perform the first matching. The second matching is performed only for chunks whose consonants are considered to match.

The degree of dissimilarity is the sum of the penalties in the first matching and the sum of the penalties in the second child matching.

The first matching and the second matching will be explained below.

■-21 Matching JRc ' k cat '' E phoneme string: KX AE ↓ TX Therefore, the reading candidates for the consonant part (KX) of the first yank of the E phoneme string are E Rc , = k , E Rc , = g +
E Rc , = c The reading candidates for the consonant part (TX) of the second chunk are ERc, = t, ERc2 = c.

According to the r5J language council report [Representation of foreign words].

Ti (t i) → Chi (ci), Dei (di) → Ji (z
It will be expressed as i). Therefore, when teeing, JRc=
When it is t, C or day, JRc is assumed to be d or 2 and compared.

(i) JRc and ERcl are singular-plural (t-c, d
-z), it is assumed that they match with a penalty of -(+2).

For example, the consonant part of the second yank of "cat" and "cats" is J Rc = t - E Rc L = c.

t and C match with a penalty of -2.

(ii) JRc and ERcl are voiced-unvoiced examples. The consonant part of the 33rd yank of "ladies" and "1adics" is J Rc = s, E Rc, :=
This is the relationship described in 1.

(a) In at least one of J Rc and E Rc 1, if this consonant phoneme is a voiced sound, the immediately preceding or immediately following consonant is an unvoiced sound, and if this consonant phoneme is an unvoiced sound and the immediately preceding or immediately following consonant is a voiced sound, the penalty -→+1 shall be.

This is because the voicedness or unvoicedness of the corresponding consonant may change depending on whether the consonant immediately before or after the consonant is voiced or unvoiced.

Example J phoneme example of "ladies": In redisu, d is a voiced sound and S is a voiceless sound, so this S and "'1adiCs"
The third yank of ERc = 2 is penalty -(+]2
) matches.

(b) For cases other than (a), the penalty is -→+3.

(ni) Those that do not fall under (i) and (ii) are considered to have different consonants. (Difference = 100) 2, J
If Rc is in ERc i, then J Rc = E Rc m, (i) ERcm
When the priority (Fl) of is 0 or l, the penalty -
→O (it) When the priority of ERcm is 2 (iii) ER
When the priority of cm is 3 or higher.

Mi (JRv) is compared with the reading candidate (ERvjk) of phoneme j that constitutes the vowel part of the E phoneme string.

For the sake of explanation, JRv = viv2-v (v is a Japanese phoneme) ERv = ERvIERv, -ERvm (ERvj is an English phoneme) ■ If the consonants of all chunks are considered to match in -2 1, ■ Phoneme string , E. Matching of three items of first order is performed for the phoneme sequence.

(a) Vowel part reading (Rv) (b) Presence of attenuation (F2) (c) Length of vowel-f phoneme at the end of each chunk (F3, F4
) (a) Matching of readings (Rv) of the vowel part In the vowel part of the J phoneme string, yon: the number of phonemes forming JRv, and m: the number of phonemes forming ERV, in order from the first chunk.

Create one selection arbitrarily from the reading candidates for each ERvj, and v
Starting from l, it is compared with the ev column in order. Those that can be omitted from the reading (phonemes enclosed in parentheses in JRv, evjk
20 things).

If the corresponding reading is not found in the other party's phoneme sequence, it is treated as missing. Furthermore, if the chunk of the E phoneme sequence ends with a consonant, the vowel of tv in Table 4 is supplemented for comparison.

Assume that for each ev column, a reading that matches up to the dxth phoneme of JRv is found, and the maximum dx is set as d.

Example "Key 4 (kTui) and 'key' (KXEE)
In the comparison, JRv=iui (v1=i+v,
=u, y, =i, n=3), ERv=EE Therefore, in comparing JRv and evl, v1=ev1□
Therefore, d1 = 1, and when comparing JRv and QV, v□≠evt2, d, = O1 Similarly, d n = 0 Therefore, d = maximum dx = d1 = 1 As above, it matches the reading of JRv Every time a reading candidate for ERvj (called cvjk) is found,
A penalty is obtained from the priority order of k and Table 6, and it is added to the degree of dissimilarity.

Table 6 Priority and penalty for the reading of the vowel part of the E phoneme string - However, if a priority of 0 exists as another candidate for the corresponding phoneme by adjusting D5-■-(1), the penalty will be - is (value in Table 6 + 1).

Furthermore, if there are multiple ev's where d=n, the minimum penalty is taken as the penalty here.

At the end of matching, if the following conditions are met, the corresponding penalty will be added.

(i) When d<n ・When d=o, add the penalty of (nX3) ・When d>O, add the penalty of ((nd-d)X2) (ii) When d=n , penalty when there is an E phoneme that was not used in matching (number of remaining E phonemes x 2) - In the above example, "kiui" and "ke
y ” (K X E E ), v1=ev,,
= i and CVll priority = 1, so the penalty is -0, but since d = 1 and n = 3, (i) satisfies d < n and d > O, and the penalty of (3-1) x 2 = 4 - is added. Therefore, the total penalty is 4.

In the corresponding chunks of matching J and E phoneme sequences of 41 F2 of b.

Penalties are determined according to the presence or absence of persistent sounds (F2) as shown in Table 7.

Table 7 Penalty due to F2 (presence or absence of persistent consonant) of J phoneme string and E phoneme string - Definition of vowel phoneme length (denoted as K) of J phoneme string KEF4
-F3 (c) Calculate the penalty using the penalty calculation table 9.

Table 9 Pena (2) E in vowel phoneme awe matching
Definition of the vowel phoneme length (A) of a phoneme string AEA +
Go to A2〇: Adjustment value due to phonological environment and spelling (D5-■-(2
)) Values given by Table 8 from F3 and F4 in Table 4 of A-Pa Table 8 Length of the vowel phoneme of the E phoneme string (A2) However, in Japanese kana notation, Tee is sometimes written as Chee-day and Day. Therefore, the combinations of ti and te, cli and de,
The difference obtained by ■■ without giving a penalty is stored in the memory work area.

D7 Correspondence determination method at character notation level■ D6■
The dissimilarity obtained in is input.

■ Determine an appropriate threshold depending on the application case.

For example, if this is 3 If the degree of difference is 3, both notation results are considered to correspond.

If the degree of difference is 3, it is considered that there is a possibility that the two notation results correspond.

If the degree of difference>3, it is assumed that the two notation results do not correspond.

The correspondence relationship is determined as follows.

■ Output is performed, for example, in the form of a display to the user.

As a specific example, a proofreading system may display a message saying ``Multiple transcription results exist for the same word.'' For application examples, please refer to D17 and D18.

D8. Specific example of judgment (comparison of English notation results and kana notation results) Example 1: Kana notation result “Fuzzy” and English notation result” f
uz z y ” is determined.

1. “Fuzzy” D2. Day is converted into J phoneme sequence=:hazi.

2. Convert "hazi" into the D4-■ expression format.

Phoneme reading (RC/v): Fl: F2: F3: F4 (persistent sound) (long sound) (continent sound) h Rc=h: ■: O:a
Rv=a: ■ ::O:Oz R
c=z : ■ : 0 : i R
v=i: ■::0:63 e”
f u z z y” is converted into E phoneme sequence = FXUH2XEE by means of sacrificial composition.

4. D4 English sound, Japanese phoneme pair, by tape small hand,
Using Table 4, convert to the expression format of D4-■.

Phoneme reading (Rcv): Fl: F2: F3
: F4 (persistent sound) (continuous sound) (long sound) FX Rc=h :■ : 0 :Rc=b :
■: 0: UHRv=a:■::2:I Rv=o:■::O:I Rv=u:■::1:0 Rv==e:■::? :? Rv=ia :■ : :? Near ZX Rc
=z :■ : 0 :Rc=s :■ : 0
: EE Rv=i :■ ::1:3Rv=ia:■
: :0 Near Rv=ie:■ ::O:
2 Rv=e :■ ::2:1 5. O5'' (7)
Changes are made as follows depending on P Disaster 1 Kedan.

D5-■-(1) No applicable items The corresponding vowels in D5-■-(2) are UH and EE.

UH: A1 from D5-■-(2)-■(1) (if)
=O EE: D5-■-(2)-■(1) From (it), A1=0 The expression of "fuzzy" obtained in 6.2 and the "f" obtained in 4 and 5
Using the expression "Uzzy", the degree of dissimilarity is calculated as follows based on the degree of hair growth of the skin.

(i) Division into chunks "Fuzzy"→(ha) (zi) "fuzzy"-+ (FXUH) (ZXEE) Since the number of chunks is 2 and matches, the first matching is performed.

(n) First matching ・Chunk: h and FX It is.

(JRc=ERc1) and (E Rcl priority=1
), so the penalty is O.

・Chunk 2: Z and Zx It is.

Similarly, the penalty becomes O.

It was found that the consonant parts matched with the degree of difference = 0 + 0 = 0.

Next, a second matching is performed.

(City) Second matching chunk 1: a and UH (a) Matching of reading (Rv) J Rv = a , (b) Matching of presence/absence of persistent sound (F2) F2 (ha) =
Since O1F2 (FXUH)=O, the penalty becomes O.

(c) Matching K of vowel phoneme lengths (F3, F4) (a) =O A (UH: a) = Al + A2 = O + (1-2) = -
1 Since the conditions of =O and IAI≦1 in Table 9 are met, the penalty is 0.

・Chunk 2: i and EE (a) Matching readings (Rv) J　Rv　zi　.

Since (J Rv = E RVIL) and (E RV, priority order = 1), the penalty is 0.

It is.

Since (J Rv = E Rv, ,) and the priority of (E Rvl, = 1), the penalty is 0.

(b) Matching F2 (zi
) = Q, F2 (ZXEE) = O, the penalty is O
becomes.

(c) Matching of vowel phoneme lengths (F3, F4) K (i) = 6 A (EE: 1) = AI + A2 = O + (3-1) =
2 Since the conditions in Table 9 of ≠0 and IK-AI≦4 are met, the penalty is 0.

From the above, the degree of difference is 20.

7. D7 Sentence level relationship “
It is determined that "fuzzy" and "fuzzy" correspond.

Example 2: For the English notation result “Q u p”, the kana notation results 1 “cup”, 2 “kop”, 3 “curve J”
A judgment is made when three of the above are input.

The Ef prime sequence of l, II cuPII is (KXUH)
(PX) (() represents a chunk). From Table 4, the conversion results are as follows.
: F4 (continuous sound) (continental sound) (long sound) KXRc=ni:■ : 0 : Rc=g:■ : 0 : Rc=c:■ : 0 : UHRv=a:■ : :2:I Rv=o:■ : :O:I Rv=u :■ :1: 1 :0Rv=e
:■ : :? :? Rv=ia: ■ : :? :? PX Rc=P : ■ : 0 : is obtained.

2. Modify the conversion result of 1 using adjustment rules.

D5-■-(1): No applicable item D5-'s-(2): Target vowel phoneme is UH■(1)
(City), A1=-1 3, kana notation result 1 Find the degree of difference between "cup" and "cup".

The J phoneme sequence is (k) (pu) (() represents chunks) (i) Since the number of chunks is equal to 2, the first matching is performed.

(ii) Penalty due to matching of KX to the first matching chunk.
0 - Chunk 2: Penalty for matching p and PX - 〇 (iii) Second matching chunk 1: a and UH (a) Matching reading (Rv) JRv = ERv,, (UH: Penalty - 0 for a) (b) Matching of presence/absence of persistent sounds (F2) F2 (ka)
= F2 (KXUH) = Penalty -0 for = O (c) Matching vowel phoneme length (F3, F4) K (a) = -6 A (UH: a) = A1 + A2 = -1 + (1-2
)=-2 In Table 9, the conditions of ≠O and lK-A I≦4 are met,
Penalty -0 ° Chunk 2: Since the second chunk (px) of the E phoneme sequence ends with a consonant, the +V vowel is captured from D4■ and compared.

Therefore, J vowel phoneme = U and E supplementary vowel phoneme (+v) = u
We will compare (a).

Both (b) and (c) match, resulting in a penalty of -0.

From the above, the degree of difference (“cup r cup”)=0.

4. Calculate the degree of difference between kana notation result 2 “kotsupu” and “a u p”.

The J phoneme sequence is (k) (pu) (() represents a chunk) (i) The number of chunks is 2 and they match.

(it) As in the case of the first matching 3, the penalty becomes -〇. (iii) Compared to the case of the second matching 3, only the vowel part (0) of chunk 1 is different, so "mi" and xi A comparison of U Hpp will be described.

(a) In reading (Rv) matching J Rv -E Rv x x (U H: o), the priority of ERVtz is 2, so the penalty is +1.

(b) Matching of presence/absence of persistent sounds (F2) F2
Since (k o) = F 2 (KXUH) = 0, the penalty is O.

(c) Matching of vowel phoneme lengths (F3, F4) K(ru)=-6 A (UH: o)=AI+A2=-1+(1-0)
=0 The conditions of Kf-O and l K-A I =6 in Table 9 are met, and the penalty is +1.

From the above, the degree of dissimilarity (“c u p 11 - “kotup”) =
+1+1=2.

5. Find the degree of difference between kana notation result 3 “curve” and “Q u p ”.

The J phoneme sequence is (ka) (pu) (() represents a chunk) (i) The number of chunks is 2 and they match.

(n) First matching Similarly to case 3, the penalty becomes -0.

(nu) Compared to the case of second matching 3, the only difference is the vowel part of chunk 1), so a comparison between ``a'' and ``UH'' will be described.

(a) Matching of reading (Rv) JRv=ERv,, (UH: a), penalty -0 (b) Matching of presence/absence of persistent sound (F2) F2 (ka)
= F2 (KXUH) = O, penalty -0 (c) Matching K of vowel phoneme length (F3, F4) (a) = +6 A (UH: a) = A1 + A2 = -1 + (1-2) =
-2 In Table 9, ≠0 and 1K-AI>7 (...) (a) apply, and the penalty - is +3.

From the above, the degree of difference (“cupH−r cup”)=3.

6. When determining the correspondence between "Qup" and the above three kana notations using the threshold value of D7, it is found that "r cup" and "kotsupu" correspond to 14. "Q u p" and "curve" can correspond. There is a gender (similarity).

D9. Used in combination with a dictionary Foreign words that have been written to sound far apart from the original sound, such as "s a 11 a d" and "salad"
, "doQfiar" and "dollar", accuracy can be improved by using a dictionary together.

Such foreign words have a strong sense of being Japanese and are limited in number, so it is easy to prepare such a dictionary.

Dlo, a system for determining the correspondence between katakana and hiragana notations, and a system for determining the correspondence between katakana and hiragana, and katakana and katakana.

Since hiragana and katakana have a one-to-one correspondence, only the determination of the correspondence between katakana notations will be explained using FIGS. 2 and 3.

The difference between FIG. 2 and FIG. 3 is the presence or absence of related reading handling means. When there are rain phoneme strings that are subject to correspondence determination that do not match in the presence or absence of persistent consonants and vowel phoneme readings (ignoring differences in consonants and long consonants).

Use this.

Hereinafter, the details of each means shown in FIG. 2 will be explained in D11 to D015.

D110 Japanese phoneme string generation means Do the same as D2.

D12. Related Reading Measures ■ Input either one of the Japanese phoneme strings obtained by Dll.

■ If there is a phoneme string in the original language (from
Make X2) correspond.

If X represents a Japanese phoneme (sequence) other than j, then D4
The relationship between R and Fi of the expression format described in -■ is as follows.

X-) Reading (R) = x, F2 (persistent sound) = 0, F3 (consonant sound) = 0, F4 (long sound) = O
F3 (consonant) = 0, F4 (long consonant) = 6 e' is composed of e-+1 (F1 = 2) < 3 chunks or more in Table 10, and the consonants following e are t, P.

s, h. 〉, so i' is regarded as a related reading, and the conversion result is as follows.

■ Store the results obtained in ■ in the memory work area.

Table 100 Related reading table for Japanese phoneme sequences z-+ reading (
R) = x, F2 (continent) = 0, F3 (consonant) = 6, F4
(Long sound) = O If X represents x, x, or delusion, then
X), F4=F4 (X) Example “Report J ((
r eNp oHt o)) D13. Phoneme level difference calculation means■ Japanese phoneme string obtained with Dll (referred to as J phoneme string) and D
The Japanese phoneme string (referred to as the J' phoneme string) obtained in 12■ is input.

■ Calculate the degree of difference between the J phoneme string and the J' phoneme string according to the following procedure.

First, match the number of chunks (1) If the number of chunks differs by 2 or more, (■) If the number of chunks differs by 1, the process is D6-■-1 (1)
Same as (U) 6 (m) When the number of chunks matches.

(1) First matching Proceed in the same manner as D6-■-2.

(2) Second matching If the consonant parts of all chunks are considered to match in the first matching of (1), the following three items are matched for the J phoneme string and J' phoneme string, and the degree of dissimilarity is determined. is the sum of penalties for the applicable items.

(a) Matching of vowel reading (Rv) Matching method is D6-■-3- Same as (a) - (J'Rv corresponds to ERv) Penalty - Table 11 instead of Table 6 use.

Table 11. (b) Matching of vowel part reading of J' phoneme string (b) Presence or absence of persistent sound (F2) D6-■-3-
Same as (b) (c) Length of the last vowel phoneme of each chunk (F3, F4)
Matching When the length of the last vowel phoneme of each chunk of the J phoneme string is , and the length of the last vowel phoneme of each chunk of the J' phoneme string is , then the definition of K, ni' is D6-■ −
3-(c) Same as (F4-F3)), the penalty due to the combination of K and ni' is determined as shown in Table 12.

Table 12. For length matching of vowel phonemes Store the dissimilarity obtained in ■■ in the memory work area.

D140 Correspondence determining means at character notation level This is performed in the same manner as D7.

D15. Specific example of correspondence relationship determination (comparison of kana notation results) The correspondence relationship between the kana notation results “report” and “report” is determined.

1. Convert the above kana notation result into a phoneme string using Dll day rawhide 1 roll.

J phoneme sequence of “report” = r e p o t.

J′ phoneme sequence of “report” = rip 100t.

2. Convert the J phoneme string (repoto) into the D4-■ expression format.

Phoneme reading (Rc/v): Fl: F2: F3
: F4 (continuous sound) (continuous sound) (long sound) r Rc=r: ■ : O:e R
v=e: ■: :O: .

p Rc=p: ■ : O:o Rv=
o: ■ : :O: 6t R
c=t: ■ : 0 :o Rv=o:
■ : :0: Similarly to 03.2, convert the J' phoneme sequence (ripoto) into the D4-@ expression format.

4. When applying D12 Yomi... to the JT phoneme string, the only corresponding phoneme is "i°'. Therefore, the following result is obtained.

Phoneme reading (RC/v): Fl: F2: F3
: F4 (consonant) (persistent) (long) i i : ■ :: O: O l e : ■ :: O: 0 Other phonemes are the same as the J phoneme sequence.

The degree of difference is calculated for the results obtained in 5°2 and 4 using the D13 phoneme level difference calculation means.

The chunks of the J phoneme sequence are (r e) (p o) (t o)
J′ phoneme sequence tr (r imp o) (t o
), and the vowel phoneme 11 of the first yank eI7 and i''
Only the matching of these two will be described.

Since the J phoneme C matches the related reading C (priority = 2) of the J' phoneme i, the penalty is +2 from Table 11.

From the above, the degree of difference (“report”−“report”)=2.

6. Sentence D14 Based on the relationship at the notation level, it is determined that "report" and "report" are "corresponding."

D16. Expansion to other character notations Although the present invention has been described above as a system for determining the correspondence between kana notations and between English notation and kana notation, the present invention can also be applied to a similar system for other notations, such as between French notation and kana notation. It is also possible to apply In that case, it is necessary to prepare a French phoneme string generation means in place of the above-mentioned English phoneme string generation means, but since algorithms for generating French phoneme strings from French spellings are known, the preparation is easy.

Furthermore, like Kana, there are Hangul characters that are created based on pronunciation.1 If the idea of the present invention is applied as is, it is also possible to create a system for determining the correspondence relationship between English and Hangul notations.D17 ．． Application Example 2: A function of a proofreading system in a Japanese word processor An example in which the present invention is applied to a proofreading system in a Japanese word processor to detect variations in foreign word notation will be described with reference to FIG.

Symbols in Figure 4: Explanation 1: The user uses the keyboard to control Enter Japanese text into your computer. Strengthen.

The second system is the input text From katakana string (katakana notation) result), Alphabella 1-column (here Let's take the English notation result) put out.

3: Any two of the notation results obtained in 2 In combination, alphabet string combination of one alphabet string For all pairs except .

Perform dissimilarity calculation according to steps 4 to 6. cormorant.

4: Perform chunk number matching.

The total sum of penalties is the degree of dissimilarity. Ru.

5:4 to be considered a match Only then, perform the first matching. Now. Add penalty to dissimilarity Calculate.

In a conflict that is considered to be a 6:5 agreement. Perform the second matching only when Now. Add penalty to dissimilarity Calculate.

7: The degree of difference is as determined in advance If the value is less than or equal to the value, these notation results is the variation in spelling of the same word. It is determined that

It was determined that there was a discrepancy in the notation at 8 near. For notation result pairs, display e.g. character display through a digital device. By changing the color of Warn the.

9: The user must follow the warning and If possible, standardize the notation results.

10 system writes and saves the modified text, for example on disk D1.

D18. Application example (1) Information retrieval system An example of application of the present invention to a document retrieval system will be described with reference to FIG. However, the effect of applying the present invention can be obtained only when the input keyword is written in Katakana or English.
Assume that the keyword is written in one of these notations.

Reference numerals in FIG. 5: Explanation 11: The user inputs the keyword of the document to be searched into the computer from the keyboard. (Let it be 1-KWD) 12 System reads the keyword of each document from the literature database D2 (Let this be P K W D I J ; i: document number, j: keyword number) 13: P KWDij and I - Calculate the KWD dissimilarity.

14: If the degree of dissimilarity is less than or equal to a predetermined threshold, the pair is considered to correspond.

J, 5: Keywords considered to correspond (p-Kwo
Information on document (i) with i, j) is read from the 1-document database.

16: Display the information obtained in 15 on the display terminal.

In the conventional system, P-KWD completely matches I-KWD.
Because KWDijL was not allowed, if there were variations in the notation of keywords, the only option was to include them all in the literature keywords or request the user to input them in a uniform notation. but.

This system can calculate the degree of dissimilarity between kana and English spellings, so users can, for example, use the input keyword ``fan'' to obtain documents with document keywords ``fuzzy J, r hashi'', ``fuzzy'', etc. .

E. Effects According to the present invention, compared to a conventional character notation result correspondence determination system using a dictionary.

・No need to create a foreign word notation dictionary in advance ・Newly coined word 1
It has the advantage that there is no need for continuous updates such as registration of derived words and proper nouns in the dictionary.

Furthermore, even when the present invention is applied only to a system for determining the results of kana notation, an appropriate threshold value can be set compared to the conventional system for determining the correspondence relationship (match/discrepancy'9X) between unified notations. By setting, it is possible to change the range of what is determined to be a variation in one notation.

・Also, you can display items in descending order of correspondence. There is an advantage.

[BRIEF DESCRIPTION OF THE DRAWINGS] Fig. 1 is a diagram showing an embodiment of a correspondence determination system between English notation and kana notation to which the present invention is applied; Fig. 2 and 3 are to which the present invention is applied. FIG. 2 is a schematic diagram showing an embodiment of a correspondence relationship determination system between kana notations. FIG. 4 is a diagram for explaining an example of application of the present invention to a proofreading system in a Japanese word processor. FIG. 5 is a diagram for explaining an example of application of the present invention to an information search system. Applicant IBM Japan Co., Ltd. Agent Patent Attorney Takashi Tonmiya (1 other person) Kana notation Houya Katan 1 illustration Eigo notation result Kakakana SL self result. ``Reho-←'' Himokat/Katakana notation combination ``Riport''/``Risai S-to'' Japanese tube phoneme sequence repδt0 Nihon Yoko 1 ni electric furnace'' rip5t. Ainobacteria = 2 Aien Ryo 20 Flute 2 Kafukana notation Thread confession "F7: / - J Hiraka - / Katakana notation result "13, Oshi" / "Fu7shi" Japanese sound # White 1 aZ Japan A phoneme sequence az Dissimilarity: 0 Figure 3

Claims (5)

[Claims] (1) A system for determining the correspondence between a notation result by a first character notation method and a notation result by a second character notation method, the system comprising: (a) converting the notation result by the first character notation method into a first phoneme; (b) means for converting the notation result by the second writing system into a phoneme string consisting of phonemes selected from the second phoneme group; (c) (d) table means for generating one or more phonemes in the first phoneme group corresponding to each phoneme in the second phoneme group; (d) using the means in (a) above to generate the first phoneme; The phoneme string obtained by converting the notation result according to the character notation method and the above (b
) and one or more phoneme strings obtained by converting the notation result according to the second character notation method using the means of (c), and determine the degree of difference at the first phoneme group level. (e) Based on the degree of difference at the first phoneme group level,
A system for determining a correspondence relationship between character notation results, comprising means for determining a correspondence relationship between a notation result according to the first character notation method and a notation result according to the second character notation method. (2) A system for determining the correspondence between a notation result according to a first character notation method and a notation result according to a second character notation method, comprising: (a) a notation result according to a first character notation method from a group of phonemes; (b) means for converting the notation result by the second character notation method into a phoneme string consisting of the phonemes selected from the above group of phonemes; Table means for generating, for each phoneme of a sub-group that occupies at least a portion, one or more phonemes from the group of phonemes that are related to the phoneme; (d) the means of (a) above; The phoneme string obtained by converting the notation result according to the first character notation method using
) and (c) to compare the notation result with one or more phoneme strings obtained by converting the notation result to the second character notation method, and determine the degree of difference at the phoneme level of the group. (e) means for determining the correspondence between the notation result according to the first character notation method and the notation result according to the second character notation method based on the degree of difference at the phoneme level of the group; A correspondence determination system for character notation results, characterized by comprising: (3) A system for determining the correspondence between a first notation result and a second notation result based on the same character notation method, which (a) determines the notation result by the above character notation method from a phoneme selected from a group of phonemes; (b) for each sub-group phoneme that occupies at least a portion of the group, one or more phonemes from the group of phonemes that are related to the phoneme are generated; (c) a phoneme string obtained by converting the first transcription result using the means (a) above; (d) means for calculating the degree of dissimilarity at the phoneme level of the group by comparing one or more phoneme strings obtained by converting the notation results of (d) the degree of dissimilarity at the phoneme level of the group; A correspondence relationship determination system for character notation results, comprising means for determining a correspondence relationship between a notation result by the first character notation method and a notation result by the second character notation method, based on the above. (4) A system for determining the correspondence between a notation result according to a first character notation method and a notation result according to a second character notation method, the system comprising: (b) means for converting the notation result by the second character notation method into a phoneme string consisting of the phonemes selected from the group of phonemes; (c) the above (a) ), the phoneme string obtained by converting the notation result according to the first character notation method, and the above (b
(d) means for calculating the degree of difference at the phoneme level of the group of phonemes by comparing the phoneme sequence obtained by converting the notation result according to the second character notation method using the means of (d) the above-mentioned method; A character characterized by comprising means for determining the correspondence between the notation result according to the first character notation method and the notation result according to the second character notation method based on the degree of difference at the level of a group of phonemes. Correspondence determination system for notation results. (5) A system for determining the correspondence between notation results based on the same character notation system, which (a) converts the notation results based on the above character notation system into a phoneme string consisting of phonemes selected from a group of phonemes; (b) means for comparing phoneme strings obtained by converting orthographic results using the means in (a) above to calculate the degree of difference at the phoneme level of the group; (c) A system for determining the correspondence of character notation results, comprising: means for determining the correspondence of the notation results according to the character notation method based on the degree of difference at the phoneme level of the group.