Electronic dictionary and method of codifying words therefor5297038Abstract Even-letter and odd-letter words are separately stored in an electronic dictionary and are codified by using predefined letter weight numbers and position weight numbers corresponding to different letters and letter positions within a word to calculate their hash values. These hash values are arranged sequentially in numerical order and divided into groups according to a predetermined number of higher digits. Numbers of hash values in these groups and the addresses of the first hash values in them are stored. When a letter array is inputted for spell-check, it is codified by the same formula and a corresponding group is identified from its high digits. A sequential comparison is made only with the hash values belonging to this selected group and until the stored hash value in the selected group exceeds the codified number. Claims What is claimed is: Description This invention relates to an electronic dictionary which may be made a part of a word processing apparatus such as a word processor, an electronic typewriter and an optical card reader. More particularly, the invention relates to a method of codifying words for such an electronic dictionary as well as a dictionary with words codified by such a method.
TABLE 1
__________________________________________________________________________
No. of
Letters
A B C D E F G H I J
__________________________________________________________________________
001 0001
0001
0001
0001
0001
0001
0001
0001
0001
0001
002 0022
0012
0020
0012
0009
0008
0008
0011
0012
0006
003 0060
0043
0051
0037
0040
0048
0041
0040
0037
0021
004 0114
0185
0168
0157
0070
0140
0132
0136
0048
0056
005 0263
0338
0425
0272
0119
0277
0245
0212
0078
0080
006 0400
0549
0719
0438
0257
0401
0359
0303
0167
0124
007 0515
0661
1006
0647
0383
0520
0443
0415
0251
0146
008 0583
0624
1154
0724
0470
0581
0377
0437
0322
0106
009 0597
0465
1092
0690
0522
0438
0322
0389
0430
0075
010 0493
0331
0941
0593
0442
0322
0223
0307
0466
0051
011 0355
0182
0722
0456
0316
0211
0121
0197
0432
0022
012 0237
0101
0483
0305
0234
0094
0071
0126
0328
0007
013 0132
0052
0264
0169
0108
0047
0041
0058
0222
0006
014 0054
0024
0132
0088
0045
0019
0016
0024
0095
0001
015 0027
0011
0057
0053
0021
0003
0003
0013
0071
0000
016 0017
0004
0035
0020
0013
0003
0002
0005
0039
0002
017 0007
0003
0029
0009
0003
0000
0000
0000
0025
0000
018 0001
0000
0006
0000
0003
0000
0000
0000
0005
0000
019 0000
0000
0001
0000
0001
0000
0000
0000
0004
0000
020 0000
0000
0001
0000
0002
0000
0000
0000
0004
0000
021 0000
0000
0000
0000
0002
0000
0000
0000
0000
0000
022 0000
0000
0001
0000
0000
0000
0000
0000
0000
0000
023 0000
0000
0000
0000
0001
0000
0000
0000
0000
0000
024 0001
0000
0000
0000
0000
0000
0000
0000
0000
0000
025 0000
0000
0000
0000
0000
0000
0000
0000
0000
0000
SUMMARY OF THE INVENTION It is therefore an object of the present invention in view of the above to provide an electronic dictionary which can be searched efficiently, say, for a spell-check purpose. It is another object of the present invention to provide a method of codifying words for storing them in an electronic dictionary. It is still another object of the present invention to provide an apparatus by which words can be codified in an efficient manner for storage in an electronic dictionary. The above and other objects of the present invention can be achieved by codifying words by (a) preliminarily defining letter weight numbers individually corresponding to letters and position weight numbers individually corresponding to letter positions in a word, (b) considering even-letter words and odd-letter words separately, (c) multiplying the letter and position weight numbers of each letter of a given word in a specified manner and dividing by a predefined divisor to obtain a multi-digit hash value, (d) splitting this multi-digit hash value into an index section and a data section, and (e) dividing these hash values into groups according to the index sections and arranging their data sections in ascending order within each group. BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are incorporated in and form a part of the specification, illustrate an embodiment of the present invention and serve to explain the principles of the invention. In the drawings: FIG. 1 is a block diagram of a word processing system for incorporating the present invention; FIG. 2 is a flow chart of a process according to the present invention for codifying words; and FIG. 3 is a block diagram of a control unit of the system of FIG. 1 for controlling the spell-check operation. DETAILED DESCRIPTION OF THE INVENTION The general construction of a word processing system to which an electronic dictionary of the present invention can be incorporated is broadly illustrated in FIG. 1 wherein numeral 1 indicates an input device for entering characters and word data into this system. This input device may include a keyboard of a known type, a tablet-like device, an optical card reader (OCR), or a magnetic tape device. Numeral 2 indicates a memory device connected to the input device 1. A core memory, an IC memory, a magnetic disk and the like may be used for this purpose. It is further connected to an output device 3 to be used for the output of data stored in the memory device 2. A printer, a display unit, a magnetic tape unit and a magnetic disk of all types can be used as the output device. The memory device 2 is also connected to an electronic dictionary 4 for checking the spelling of specified words comprising, for example, a core memory, an IC memory, a random-access memory (RAM), or a magnetic disk. As will be described below, this dictionary is provided with as processing means which exclusively serves to check the spelling of a given word, that is, the electronic dictionary 4 is adapted to determine in response to a request from the memory device 2 whether a given word is spelled correctly or not. Each of the aforementioned devices 1-4 is connected to a control unit 5 which may include a computer and serves to control the transmission of signals among the devices described above. Regarding a system as described generally above, the present invention teaches a method of codifying words such that they can be stored efficiently in a memory (that is, requiring only a small memory) and retrieved quickly from such a memory. In what follows, a method of codifying English words will be described first, followed by a method of storing such codified words and a spell-check method. These methods, which are to be described below, are in accordance with a preferred embodiment of the invention and intended to be illustrative, not to thereby limit the scope of the invention. Firstly, all English words to be registered are replaced by, or codified into 27-bit numerals. For this purpose, each letter of the alphabet is assigned a different binary "letter weight number". Since there are 26 capital letters and 26 lower case numbers, the letter weight numbers are binary numbers with 6 digits, or 6-bit numbers. Numerals (0-9) and certain symbols (such as the apostrophe) may be assigned distinctive letter weight numbess. Similarly, binary position weight numbers are assigned to different letter positions in a word. On the assumption that there is no word to be registered which contains 30 or more letters, only 29 of these position weight numbers are defined. On the further assumption that about a total of 100,000 words is going to be codified, position weight numbers are selected as binary numbers with 64 digits. With the letter and position weight numbers defined as above, each word to be registered is codified by multiplying the letter weight number of each letter of the word by the corresponding position weight number, adding these products for all letters of the word and obtaining a hash value. This is done according to a routine to be explained next by way of an example and the flow chart shown in FIG. 2. Consider an exemplary (fictitious) three-letter word "XYZ" as an example. After a working memory space is initialized (set equal to 0) (n1), the first letter in the given word ("X" in this example) is considered (n2) and the 6-bit letter weight number corresponding to this letter is retrieved (n3). Let this letter weight number be L, which is a 6-digit binary number. In the next step (n4), this 6-digit number is divided into two 3-digit numbers L1 and L2, L1 being the first 3-digits of L and L2 being the last 3-digits of L. In other words, if L=110010, L1 and L2 will respectively be L1=110 and L2=010 in binary. Next, since "X" is the first letter of the given word, the position weight number corresponding to the first position is retrieved (n5). Let this position weight number be P, which is a 64-digit binary number. In the next step (n6), this retrieved number P is similarly divided into two 32-digit numbers P1 and P2, respectively representing the first 32 digits and the last 32 digits of P. After these four numbers L1, L2, P1 and P2 are thus prepared, the value of (L1P1+L2P2) is calculated and then divided by the prime number closest to (2.sup.27 -1) to obtain a remainder (n7). This remainder would serve as the hash value for a single-letter word "X" and is temporarily stored at the aforementioned initialized work space (n8). Since "X38 , in this example, is not the final letter of the word under consideration (YES in n9), the next letter "Y" is considered in the next cycle with reference to the flow chart of FIG. 2. By proceeding similarly from Step n3 to Step n7, another remainder is obtained and this newly obtained remainder is added to the temporarily saved remainder obtained (n8). This sum of the remainders, which would serve as the hash value for a two-letter word "XY", is stored (temporarily) in the place of the previously (temporarily) stored remainder value. The same cycle of steps is repeated for the last letter "Z" and the final remainder (the sum of three remainder values obtained for the three individual letters) is defined as the hash value of the given word. Since the prime number closest to (2.sup.27 -1) is used as the divisor throughout this routine according to this example, all hash values can be expressed as a binary number with 27 digits or less. If the divisor is changed to another prime number which is closest to a different power of 2, the number of digits of the hash values can be varied. Table 2 shows the exemplary hash values of words (written in octal) appearing at the beginning when arranged in alphabetical order corresponding to a certain particular choice of letter and position weight numbers. According to the present invention, these hash numbers are rearranged next in the ascending numerical order. Table 3 shows (for a certain exemplary dictionary) the lowest hash values (in octal) together with the corresponding words. According to a preferred embodiment of the present invention, all words to be registered are divided into those with an even number of letters ("even-letter words") and those with an odd number of letters ("odd-letter words") and the hash values corresponding to even-letter and odd-letter words are separately registered in a dictionary.
TABLE 2
______________________________________
Hash Value Word
______________________________________
627062512 aardvark
013662677 aardvarks
720357163 Aaron
411613330 abaci
201213330 aback
052615503 abocus
074557027 abacuses
614351261 avaft
305351462 avalone
261617146 abandon
005731615 abandoned
424072002 abandoner
503503327 abandoning
44542206 abandonment
303465146 abandons
421153261 abase
041154020 abased
475171636 abasement
621155434 abases
432753261 abash
573025663 abashed
573172665 abashes
117675467 abashing
506771636 abashment
374422475 abasing
421035261 abate
041036020 abated
475053636 abatement
155066547 abatements
621037434 abates
374304475 abating
614037503 abatis
636001027 abatises
561745260 abattoir
746545376 abattoirs
05727051 abbacies
435331130 abbacy
255457701 abbe
627457701 abbes
027462123 abbess
051423447 abbesses
466057701 abbey
666062054 abbeys
614733701 abbot
014736123 abbots
______________________________________
According to prior art routines, 8 bits (1 byte) are usually required to express a letter and hence 32 bits will be required to express a 4-letter word. Since Table 1 shows that the average number of the words to be registered is 5-6, the method of replacing each word by a 27-bit number would not be considered very effective. The present invention, therefore, incorporates an indexing method as follows. After each word to be registered has been converted into a 27-digit binary number and these hash values are rearranged in the ascending order (separately for even-letter and odd-letter words), each hash value is divided into the index portion (hereinafter simply referred to as the "index") consisting of the upper 11 digits when the hash value is expressed in binary and the data portion (hereinafter simply referred to as the "data") consisting of the remaining lower 16 digits. Words are then grouped together according to their index, that is, those words having the same index are grouped together. With reference to Table 3 (for both even-letter and odd-letter words), the 43 words from "nevus" to "accommodator", of which the hash values are less than 2.sup.16 (or 200000 in octal), all have the index "0" and they may be referred to as the zeroth group. Similarly, those words with hash values equal to o greater than 2.sup.16 and less than 2.sup.17, having the index "1", may be referred to as the first group, and so forth. In this manner, 2.sup.11 =2048 such groups (from the zeroth to the 2047th) are created separately for even-letter and odd-letter words. It now goes without saying in this connection that the number of groups can be increased or decreased if a different number of digits is given to the index portion. After 2048 groups of hash values are thus created (separately for even-letter and odd-letter words), the number of words belonging to each group is counted and stored. According to a particular embodiment of the present invention with a particular choice of letter and position weight numbers for codifying a typical dictionary containing about 74,000 words, the numbers of even-letter words in Groups 000, 001, 002, 003, . . . were 21, 14, 16, 13, . . . , respectively, in decimal. These numbers, as well as those for odd-letter words, are stored. The largest number of even-letter words in a group was 67. In other words, the spelling of any (even-letter) word can be checked by no more than 67 times of search. The 16-bit data portions of the hash values are sequentially stored in the dictionary from the zeroth address.
TABLE 3
______________________________________
Hash Value Word
______________________________________
000000745 nevus
000004213 entrenches
000005075 Eurodollar
000011631 billhooks
000015105 nexus
000016265 negus
000017145 breakfronts
000017320 pieta
000020272 correspondents
000021333 sentinels
000033540 starboard
000037417 interchanges
000037552 pasta
000040713 indolent
000042775 simian
000044535 adsorbate
000045371 recedes
000052231 fabricated
000057151 mansards
000060010 dredge
000065030 ciliate
000065473 insolent
000070742 varnish
000076105 dentists
000103410 drudge
000106374 ilex
000112020 overcapitalize
000113447 lifestyle
000114605 tangiest
000121715 desiccated
000131466 dredged
000136045 pageants
000136653 silliest
000142357 repairers
000152333 sorghum
000155066 drudged
000157430 queenliest
000157607 medically
000163073 cattily
000166277 wayfarer
000172567 sprains
000172636 strains
000173734 accommodator
000202554 surliest
000204623 musically
______________________________________
The address in the dictionary at which the first data portion of each group (for both even-letter and odd-letter words) is calculated next. For the purpose of description, let us now consider only the part corresponding to even-letter words. For the zeroth group, the first data portion is stored at Address 0. The address of the first data portion of the first group is equal to the number of hash numbers (or words) in the zeroth group. (The real address in bytes is twice this number because each 16-bit data portion requires 2 bytes to store. For the convenience of explanation, this factor of 2 will be ignored in the description which follows.) In the case of the example discussed above, there are 21 even-letter words (or data portions thereof) in Group 000. Thus, the address of the first word in Group 000 is 0 and that of the first word in Group 001 is 21. Similarly, since the numbers of (even-letter) words in Groups 001 and 002 are 14 and 16, respectively, the addresses of the first words in Groups 002 and 003 are 21+14= 35 and 35+16= 51, respectively. In other words, the addresses of the first words of the groups can be readily calculated by sequentially adding the numbers of words in the groups in the ascending order and the addresses thus calculated for the first words in the groups corresponding to both even-letter and odd-letter words are also stored in a memory device for reference, as will be explained below. For the purpose of reference, the address of the first data portion of the n.sup.th group (n=0 -2047) will be denoted by A.sub.n. Next, a method of using this dictionary for a spell-check operation (or for retrieving a word) is described. First, it is determined whether the word to be retrieved is an even-letter word or an odd-letter word and the corresponding dictionary is selected. Next, the word to be retrieved is converted into a 27-bit number in the same way described above and this 27-bit number is divided into the 11-bit index portion representing the higher 11 digits and the 16-bit (2-byte) data portion. From this index portion, it is immediately ascertained which group (for even-letter or odd-letter words) this word belongs to and, if this word belongs to the m.sup.th group, it can be ascertained that there are (A.sub.m+1 -A.sub.m)= N.sub.m words in this group. In other words, the number of data to be checked is at most N.sub.m. Since the N.sub.m data in this group (from A.sub.m to A.sub.m+1 -1) are arranged in the ascending order, however, it is not necessary to check all of these data. The comparison starts from the first word in the group and as soon as the data value exceeds the 16-bit data portion of the hash value for the word to be retrieved, it can be concluded that the retrieval failed, or that the input word is not in the dictionary. In summary, even after the group to be checked has been determined, it is not always necessary to make a comparison with every one of the words in that group and hence the time required for retrieval (spell-check) is substantially reduced. An example of a circuit structure for the control unit 5 (of FIG. 1) with which words can be codified and a given word (or a letter array) can be retrieved for the purpose of spell-check as described above is illustrated in FIG. 3, and its function will be explained below with reference to the flow chart of FIG. 2. A word to be retrieved (represented symbolically, for example, as XYZ as in the previous example) is entered as a letter array through the input device 1, ending with a punctuation signal indicating the end of the letter array. Evenness or oddness of the inputted word is immediately determined thereupon. When the first letter (X in this example) is taken in (n2 in FIG. 2), a letter weight table 12 is referenced and a 6-bit letter weight number L corresponding to the inputted first letter is retrieved (at 13) and the first three digits L1 and the second three digits L2 are separately stored (at 14). Since this is a first letter in the inputted word, a 64-bit position weight number P for the first position is retrieved from a position weight table 16 (to 16). As done with the letter weight number L, the retrieved position weight number P is also divided into the first 32 digits P1 and the second 32 digits P2 and they are separately stored (at 17). Products L1.times.P1 and L2.times.P2 are calculated (at 18 and 19) and the sum of these two products are obtained (at 20). The sum thus obtained is divided by the prime number closest to (2.sup.27 -1) to obtain a remainder (at 21) and the hash value thus obtained is stored (at 22). This is repeated for the second and subsequent letters, each time adding the residue obtained at 21 to the previous hash value stored at 22. When all letters of the inputted word is thus processed, the value then stored at 22 is the hash value of the inputted word. The dictionary 4 is divided into an even-letter portion 41 and an odd-letter portion 42. For the purpose of spell-check, evenness or oddness of the inputted word, for which the hash value is now stored at 22, is checked. In the present example, XYZ is an odd-letter word and spell-check is carried out with the odd-letter portion 42 of the dictionary 4. For the comparison (at 23), the index portion of the hash value at 22 is checked to thereby determine which of the 2048 groups in the odd-letter portion 42 should be referenced. When the group is ascertained, an address table 24 listing the address of the first word in each group (of odd-letter words) is consulted to determine from what address of the dictionary 4 (or the odd-letter portion thereof) a search (or comparison) should be carried out. As mentioned above, since words within the same group are arranged in ascending order of their hash values (or the data portions thereof) in the dictionary 4, the search within the identified group need not be carried out by comparing the hash value at 22 with every stored hash value in the group. The search may be stopped when the data portion of the hash value stored at 22 exceeds that of a word in the group. In this manner, the time required for a search can be significantly reduced. The result of the search is outputted through an output device 3. The foregoing description of a preferred embodiment of the invention has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed, and obviously many modifications and variations are possible in light of the above teaching. For example, the present invention has been described above by way of a spell-check routine. The method of and apparatus for codifying words according to the present invention in connection with an electronic dictionary can be also used for adding a new word to the dictionary or deleting a selected word from the dictionary. The dictionary 4 may further include both a main dictionary provided by the system itself and a user dictionary serving to store only words selected by the user. The user dictionary may similarly have an index section, a first data section for storing user-selected words as numerical data by the aforementioned hash method, and a second data section with areas for storing these words as character data. When the user wishes to check the spelling of a word, its hash value is calculated in the same manner as described above and a high speed search is carried out from the index section. The second data section may be used principally for deleting a word from the user dictionary. In summary, electronic dictionaries of the present invention can be processed efficiently and speedily because words to be stored are grouped together in such a way that the number of groups and hence the numbers of words in individual groups can be adjusted by increasing or decreasing the number of digits in the index portion of the aforementioned hash values. If the aforementioned number of digits of the index is made sufficiently large such that the number of groups is greater than if the words are grouped by the initial letter or the number of letters in the word, as previously attempted, the number of words in each group can be reduced and hence the time required for the search can also be reduced. Moreover, codification is effected by the present invention in units of words rather than in units of letters as done previously such that the memory space for each word can be made not only uniform but also small. In other words, the total memory capacity required of the whole dictionary can also be reduced. In addition, since the data belonging to each group are arranged in ascending order, the data of a test word need not be compared with all data in the selected group. Since a conclusion of an error can be reached before comparison is made with all data in the group, the processing time can be further reduced. Still another advantage according to the present invention is achieved by dividing words into even-letter and odd-letter words. This method of grouping words, for example, by means of alternately emitted ON and OFF signals from the input device 1, is advantageous because evenness and oddness of a word can be more easily ascertained in this manner than by an ordinary method of determining a group from the content of the data. In summary, such modifications and variations that may be apparent to a person skilled in the art are intended to be included within the scope of this invention.
|
Same subclass Same class Consider this |
||||||||||