JP2526670B2 - Word dictionary search device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial field of application) The present invention is a word for collating an input character string with a word dictionary and searching for a portion in the input character string where a word existing in the word dictionary appears. The present invention relates to a dictionary search device.
In particular, the present invention relates to a word dictionary search device applicable even when each character of an input character example has a plurality of candidates.

A word dictionary search device in the case where each character of an input character string is unique without a plurality of candidates is a part for performing a word dictionary search for a kana character string input from a keyboard in a kana-kanji conversion device, a machine translation device / sentence voice conversion device. -It is used in the lexical proofreading device, etc. for searching a word dictionary for a character string mixed with kanji and kana created by a word processor or the like.

The word dictionary search device when there are multiple candidates for each character of the input character string is a word dictionary search for selecting the most probable character from a plurality of candidate characters of the recognition result in a voice recognition device or a character recognition device. It is used for departments.

(Prior Art) The word dictionary search method conventionally used in the kana-kanji conversion device / sentence-speech conversion device is basically based on the document: “Kana-kanji conversion by computer” (Aizawa / Ehara, NHK).
Technical Research, Vol. 25, No. 5, pp. 23-60, 1973). That is, a process of extracting a partial character string from the input character string and searching a word in the word dictionary for which the partial character string has a notation (“kana notation” or “yomigana” in kana-kanji conversion) is matched. Perform (hereinafter, referred to as the first conventional technique.) For example, when performing a word dictionary search for an example of a mixture of kanji and kana such as "analyze a sentence", the first character
Partial characters such as "analyze sentence", "analyze chapter", "analyze", "analyze", "analyze", "do", "ru" in order to search for a word starting from each character position such as the second character Cut out the columns separately. Then, for each partial character string, a partial character string with the end trimmed is further generated, and each of them is searched from the word dictionary. That is, when searching for a word starting from the first letter, "analyze the sentence"
For substrings such as "analyze a sentence", "analyze a sentence", "solve a sentence", "sentence", "sentence", and "sentence", the search is repeated one after another to search for a word starting from the second character. , Analyze Chapter, Analyze Chapter, Analyze Chapter
The search is repeated one after another for substrings such as "solve chapter", "chapter", and "chapter". As a result, for the underlined sentences, sentences, chapters, and other substrings, words with matching notations can be found.

In the first conventional technique, the inspection of the word dictionary is repeated for a plurality of partial character strings. However, by devising the structure of the word dictionary, the time required for the search for one partial character string can be reduced. We are trying to shorten it. For example, words in the word dictionary are sorted in advance and a binary search is performed, or characters such as the first character and the second character are divided and common parts are collected to perform a tree-structured collation for each character. And so on. For the structure of such a dictionary and devising a search method, see the article: “The Art of Computer Prog.
ramming 3: Sorting and Searching "(DEKnuth, Addiso
n-Wusley, 1973).

However, this first conventional technique is supposed to be realized as a sequential program on a computer, and the device for shortening the required time is devised within the framework of sequential processing. Therefore, a process of comparing a partial character string of an input character string with a character string representing a word in a word dictionary is sequentially performed character by character. For example, to compare a substring "sentence" to the word "sentence" in the word dictionary, match "sentence" and "sentence", and when they match, match "chapter" and "chapter", Next, the collation process is repeated character by character such that "o" does not match and fails.

Further, in the first conventional technique, m is added to each character of the input character string.
In the case where there are candidates for each, it is necessary to generate in advance m ^L character strings by combining the candidates for the partial character string of length L. Then, for each of them, the partial character string with the end trimmed as described above is further generated, and the word dictionary is searched.

On the other hand, when comparing a character string having a certain length L with an input character string, it is considered that the collation of L sets at corresponding character positions is performed at the same time, instead of repeating the collation of each character. At that time, if the input character string is stored in the shift register, shifting the collation position shifts all the characters in the shift register one character at a time, rather than cutting out the partial character string again from the input character string. It can also be realized. This second conventional technique is described in Japanese Patent Application Laid-Open No. 63-261421, “Character String Processing Device” and Japanese Patent Application Laid-Open No. 63-261422, “Character String Matching Device”. In addition, this second
In the prior art of, there are no multiple candidates in the input string,
The number of character strings that can be searched is also limited to one (not as many as in a word dictionary).

Japanese Unexamined Patent Publication No. Sho 62-67636 "Reference system" and references:
"Dictionary Matching Algorithm for High-speed Language Processing in Spoken Japanese Input System" (Hamaguchi and Suzuki, Transactions of the Institute of Electronics, Information and Communication Engineers, Vol. J70-D, No. 8, pp. 1589-1596 198)
7 years), the third prior art is shown.

The third conventional technique assumes that the input character string has a plurality of character candidates. First, the character type is M (for example, in the JIS character code table, M = 8 for hiragana).
3), at each character position such as the first and second characters,
One M-bit memory is prepared and each bit corresponding to a plurality of candidate characters is set to 1. And at the time of matching,
For each word in the word dictionary, the first letter of the word notation
For each character position such as the second character, the bit content (1 or 0) of the corresponding character is simultaneously read from the corresponding M-bit memory. When 1 is read from all the M-bit memories, it means that a word in the word dictionary appears in the input character string.

(Problem to be Solved by the Invention) First, as described above, the first conventional technique has a drawback that the word dictionary search for a large number of partial character strings must be repeated. This drawback is especially noticeable when there are multiple candidates for each character in the input string. If there are m candidates for each character in the input string of length K, and the length of the longest substring is L (usually, the length of the longest word in the word dictionary is L), Maximum ( ^mL x L x K)
It becomes necessary to repeat the search for the respective partial character strings.
Usually, since conditions are set in advance to avoid searching for unnecessary sub-character strings, (m ^L × L × K) is the worst case, but the number of searches is still large.

A second drawback of the first conventional technique is that a certain character string is compared with a character string representing a certain word in the word dictionary one by one in order, and therefore the comparison takes time.

The second prior art addresses the second shortcoming of the first prior art. However, the second conventional technique cannot be applied when there are a plurality of candidates in the input character string. Also, the number of character strings to be searched is limited to one, and considering the comparison with many character strings as in a word dictionary, the difference in the notation length of each word in the word dictionary becomes a problem. Not applicable.

The third prior art addresses two drawbacks of the first prior art. However, in order to store the input character string, it is necessary to prepare the memory having the number of bits corresponding to the character type only for the character string length, so the memory size becomes considerably large when there are many character types. There is a problem that ends up.
In the case of kana-kanji conversion or voice recognition, the input character string is limited to about 100 types of hiragana (or phoneme characters), so it does not pose a problem, but sentence-to-speech conversion or machine translation for kanji-kana mixed character strings. In such cases, there are 3000 to 4000 types of characters, including kanji, which causes a large memory problem.

Further, in the third conventional technique, only a word starting from the beginning of the input character string is searched. Therefore, when searching a word starting from the second character or a word starting from the third character, the input character There is also a problem that it is necessary to re-register the column in the memory.

Further, in the third conventional technique, it can be seen that a matched word has appeared from the beginning of the input character string to a certain length,
It was necessary to memorize the address in the word dictionary of the word, determine the length of the word again, and check the notation of the word in the word dictionary again to obtain the length.

An object of the present invention is to eliminate the above-mentioned drawbacks of the prior art, and provide a word dictionary search device capable of high-speed matching with a word dictionary even when there are a plurality of candidates for each character of an input character string. That is. In addition, a word dictionary search device provided with a mechanism for easily calculating the length of the detected word is also provided.

(Means for Solving the Problem) In the present invention, the first to mth candidates (m is m ≧
An input device for a character string having m types of candidates up to 2) and one word notation at each address having a data width of n characters (n is an integer of n ≧ 1) A word dictionary memory (number of registered words ≧ 2) in which predetermined residual symbols are filled in a portion less than n characters, and one shift clock each time m types of candidates for one character are input by the input device. A controller that generates a determination clock and a counter clock the number of times corresponding to the total number of words in the word dictionary memory, and a word dictionary memory that performs reset in synchronization with the shift clock and count-up in synchronization with the counter clock. An address counter and a first character string input by the input device
2nd ... The 1st for n characters each of which is forwarded character by character in synchronization with the shift clock corresponding to the mth candidate
Second ... The mth candidate shift register and the first character of the data for n characters read from the word dictionary memory.
When the character at the corresponding position corresponding to the character, ..., Nth character matches the character at the same position in any of the first, second, ..., mth candidate shift register, a match signal is output and A first-character / second-character ... N-character comparison circuit that outputs a coincidence signal and a residual detection signal when the residual symbols match.
In synchronization with the judgment clock, the first character, the second character,
··· A determination circuit for determining that a word existing in the word dictionary memory appears in the character string input by the input device when a match signal is detected from all the n-th character comparison circuits; A word dictionary search device comprising: a word length calculation circuit that calculates a word length based on a detection signal.

(Example) The structure and operation of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of the first embodiment of the word dictionary search device of the present invention. Hereinafter, each component will be described first.

The input device 5 is a device for inputting a character string in which m types of candidates from the first candidate to the m-th candidate (m is an integer of m ≧ 2) exist for each character. For example, a character recognition device. The input device 5 simultaneously outputs m candidates for each character, and informs the controller 7 of the output timing by the input clock 50.

The word dictionary memory 1 stores one notation of a word at each address having a data width of n characters (n is an integer of n ≧ 1), and a portion less than n characters is predetermined. It is packed with residual symbols. FIG. 2 is a diagram showing an example of contents of the word dictionary memory 1 (n = n in FIG. 2).
4). In FIG. 2, Δ represents the residual symbol. Assuming that the address of the word dictionary memory 1 in FIG. 2 is a, n (= 4) characters of "Japan ΔΔ" are read out at the same time. The word dictionary memory 1 can be realized by using an IC memory or the like. Normally, the kanji code is represented by 16 bits, so the data width of the word dictionary memory 1 in FIG.
4 = 64 bits. In current IC memory, the data width that can be read at the same time is about 8 bits.
It can be realized by arranging 8 IC memories in parallel.

Whenever the input device 5 inputs m types of candidates for one character, the controller 7 outputs the shift clock 1
And the determination clock and the counter clock are generated a number of times corresponding to the total number of words in the word dictionary memory 1. FIG. 3 is an example of a time chart of input / output signals of the controller 7. In the time chart of FIG. 3, every time the input clock 50 is input from the input device 5, first, the shift clock is input.
70 is output once (the input clock 50 is output as it is as the shift clock 70), and subsequently, the determination clock 72 and the counter clock 71 are alternately output N times. However,
The counter clock 71 may be (N-1) times. here,
N is the total number of words in the word dictionary memory 1. The controller 7 that operates according to such a time chart can be easily realized by those skilled in the art.

The address counter 4 is a counter that resets in synchronization with the shift clock 70 and counts up in synchronization with the counter clock 71, and outputs the counter value as the address value of the word dictionary memory 1. It can be realized with a conventional counter IC.

The i-th candidate shift register 2 is the i-th candidate of the character string input by the input device 5 (i is an integer 1 ≦ i ≦ m)
Is a shift register that stores n characters while sequentially feeding each character in synchronization with the shift clock 70. The shift register 2 is provided for each of m candidates for each character of the input character string, and the m candidates of the first candidate shift register, the second candidate shift register, ... is there. FIG. 4 is a diagram showing a configuration example of each shift register 2. When one character is represented by d bits, the i-th candidate shift register 2 can be realized by (d × n) D flip-flops synchronized with the shift clock 70 as shown in FIG. The ones lined up in
n units are connected in series). The d pieces connected in parallel correspond to one character, and the outputs thereof are sent to the comparison circuit 3 collectively.

The j-th character comparison circuit 3 has the value 1 indicated by the address counter 4.
The j-th character (j is an integer 1 ≦ j ≦ n) of the data of n characters read from the word dictionary memory 1 for one address is j of any one of the m shift registers 2.
It is a circuit that outputs the match signal 30 when the second character is matched, and outputs the match signal 30 and the residual detection signal 31 when the residual symbol is matched. The comparison circuit 3 is provided for each of the data width n characters of the word dictionary memory 1,
There are n first character comparison circuits, second character comparison circuits, ..., Nth character comparison circuits. FIG. 5 is a diagram showing a configuration example of the j-th character comparison circuit 3. When one character is represented by d bits and m shift registers 2 are provided, the j-th character comparison circuit 3 operates as shown in FIG.
It can be configured with a bit comparator and one OR gate. The m d-bit comparators compare the j-th character of the n-character data read from the word dictionary memory 1 with the j-th character of each shift register 2, and the remaining one d-bit comparison The instrument checks whether the j-th character of the n-character data read from the word dictionary memory 1 is a residual symbol.
At the final OR date output, when a match is detected in any of the (m + 1) comparators, the match signal 30
Is output. The output of the d-bit comparator for checking whether the j-th character of the n-character data read from the word dictionary memory 1 is also a residual symbol is used as the residual detection signal 31.

The determination circuit 6 synchronizes with the determination clock 72, and when a match signal is detected from all of the n comparison circuits, the determination circuit 6 includes the word dictionary memory 1 in the character string input by the input device 5.
It is a circuit that determines that a word existing in the word has appeared. FIG. 6 is a diagram showing a configuration example of the determination circuit 6. The decision circuit 6 can be realized by one AND gate and one D flip-flop as shown in FIG.

The word length calculation circuit 8 is a circuit that calculates the word length 80 based on the residual detection signals 31 output from the n comparison circuits 3. FIG. 9 is a diagram showing a configuration example of the word length calculation circuit 8. In the circuit of FIG. 9, the j-th character comparison circuit (here, 2
When the residual detection signal 31 has arrived from ≤j≤n) and the residual detection signal 31 has not arrived from the (j-1) th character comparison circuit, (j-1) is output as the word length 80. When the residual detection signal 31 does not reach from any of the first character comparison circuit to the nth character comparison circuit, the value of n is output as the word length 80. When the residual detection signals 31 arrive from all the nth character comparison circuits from the first character comparison circuit, 0 is set to the word length 80.
(However, this is a case where the word dictionary memory 1 contains a word with a word length of 0, which is usually unthinkable).

 Next, the operation of this embodiment will be described using an example.

FIG. 7 shows an example of changes in the contents of the first candidate shift register, the second candidate shift register, and the third candidate shift register when n = 4, m = 3 and the length of the input character string is 4. It is a figure. In the 3 × 4 matrix of FIG. 7, one row corresponds to the contents of each shift register 2, the row direction represents the character position (1 to 4), and the column direction represents the candidate level (1 to 3). There is. Regarding the character strings input to these shift registers 2, the candidates for the first character are “day” in order from the first candidate.
"White", "eyes", second character candidates are "tree""book" in order
"Large", the third character is "den", "thunder""fog", and the fourth character is "ki""island""kai" in order. The shaded area indicates that no character is stored.

In FIGS. 7A to 7H, (a) → (b) →
The change of (c) → (d) → (e) → (f) → (g) → (h) indicates a change that occurs each time the shift clock 70 is generated once. Then, in each of the states (a) to (g), the counter clock 71 and the determination clock 72 are generated N times. The address counter 4 has a shift clock 70
Reset by the counter clock 71
Since it is counted up times, in each state, the notation of all words (N pieces) is read from the word dictionary memory 1 in order from the first word to the last word, and n words (in this example, 4 words) are read.
7), the contents of the shift register shown in FIG. 7 are compared with each other.

As a result, in each state, the determination circuit 6 detects the appearance of the following words in the word dictionary memory 1, for example.
Then, for these, the word length output from the word length calculation circuit is a value within <>.

(A) Not applicable (b) Not applicable (c) Not applicable (d) "Sun"<1>"Japan"<2>"NEC"<4>"Nihondai"<2>"White"<1>"Shiraki"<2>"Eyes"<1> (e) "Tree"<1>"Book"<1>"Large"<1> (f) "Den"<1>"Electricity"<2>"Mist<1> “Kirishima” <2> “Thunder” <1> (g) “Qi” <1> “Island” <1> Of these, the contents of the shift register in FIG. 7 in the state (d) and the word dictionary Word "Japan" (= "Japan △△")
The operation of each comparison circuit 3 when comparing and will be described. 1
The character comparison circuit generates a coincidence signal 30 when "Japan" and "Day" coincide with the first character "Day" of the first candidate shift register. The second character comparison circuit generates a coincidence signal 30 when the “book” of “Japan ΔΔ” and the second character “book” of the second candidate shift register match. Both the third character comparison circuit and the fourth character comparison circuit are "△" of "Japan △△".
(Residual symbol) is detected to generate a coincidence signal 30 and a residual detection signal 31. As a result, the decision circuit 6 receives the coincidence signals 30 of all the comparison circuits 3 and detects the appearance of the word. At that time, the word length detection circuit 8 receives the residual detection signal 31 from the third character comparison circuit and the fourth character comparison circuit and outputs 2 as the value of the word length 80.

When the input character string length is K, the sequential feeding in the shift register 2 needs to be performed at least (K + n-1) times. Therefore, after inputting the input character string having the length K, the input device 5 needs to further perform (n-1) dummy character string inputs. Alternatively, the controller 7 detects the end of the input character string, and further performs the shift clock 70 once, the counter clock 71, and the determination clock.
The cycle of 72 times N times may be repeated (n-1) times.

Further, during the first (n-1) times of sequential feeding (from (a) to (c) in FIG. 7), since the input character string has not reached the beginning of the shift register 2, the word dictionary memory 1
There is no point in matching with. So during that time,
The controller 7 may generate the shift clock 70 continuously without generating the counter clock 71 and the determination clock 2.

In the above, an example of a general case in which there are m candidates for each character of the input character string has been shown. However, a word dictionary for word dictionary search in kana-kanji conversion or analysis of a sentence created by a word processor. In search or the like, each character in the input character string is one type (m = 1). FIG. 8 is a block diagram showing an example of a word dictionary search device for such a case. In this case, one shift register is sufficient as in the second prior art, and is excluded from the scope of the claims of the present invention. The constituent elements / operations may be m = 1 in the embodiment shown in FIG.

(Effects of the Invention) As described above, according to the present invention, a character string composed of many kinds of characters such as Chinese characters can be collated with a word dictionary at high speed even if each character has a plurality of candidates. A word dictionary search device capable of In particular, even for the number of candidates for each character in the input character string, the input character string can be matched with one word in the word dictionary within about two clocks, regardless of the notation length of the word in the word dictionary. Is very effective.

Furthermore, as shown in the embodiments, each component of the present invention can be realized by combining a small number of logic ICs. Therefore, the advantage of using LSI technology is that it can be realized as a very small device. In addition, unlike the first conventional technology, it can be realized not as software on a general-purpose computer but as dedicated hardware / special LSI, so the clock frequency itself is considerably higher than that of a general-purpose computer. It is possible to set it, and the high speed is also excellent in this respect.

Due to the high speed as described above, the present invention does not limit the number of words to be matched in the word dictionary as in the first conventional art described above, and even if matching is performed with all the words in the word dictionary, Although it is considered that a sufficiently high speed word dictionary search device can be obtained, as a result, there is an advantage that the word dictionary does not have to be sorted in the code order of the notation. Therefore,
Even if words are added or deleted, it is not necessary to reorganize the word dictionary, and maintenance of the word dictionary is extremely easy.

Further, in the word dictionary search device of the present invention, not only can the word in the word dictionary appear in the input character string but also the length of the word can be obtained at the same time, so the word length is read again from the word dictionary. There is no need, and the processing efficiency is high.

[Brief description of drawings]

1 is a block diagram showing the configuration of the first embodiment of the present invention, FIG. 2 is a diagram showing an example of the contents of the word dictionary memory 1, and FIG. 3 is an example of a time chart of input / output signals of the controller 7. , FIG. 4 is a diagram showing a configuration example of the shift register 2, and FIG.
FIG. 6 is a diagram showing a configuration example of the comparison circuit 3, and FIG. 6 is a determination circuit 6
7A to 7H are diagrams showing examples of changes in the contents of the shift register 2, and FIG. 8 is a block diagram showing an example of a word dictionary search device with m = 1. , FIG. 9 is a diagram showing a configuration example of the word length calculation circuit 8. In the figure, 1 ... Word dictionary memory, 2 ... Shift register (i-th candidate shift register), 3 ... Comparison circuit (j
(Character comparison circuit), 4 ... Address counter, 5 ... Input device, 6 ... Judgment circuit, 7 ... Controller, 8 ...
Word length detection circuit, 30 ... Match signal, 31 ... Residual detection signal, 50 ... Input clock, 70 ... Shift clock, 71 ...
… Counter clock, 72 …… Judgment clock.

Claims (1)

(57) [Claims] 1. First to mth candidates (m is m) for each character
An input device for a character string having m types of candidates up to ≧ 2) and one word notation at each address having a data width of n characters (n is an integer where n ≧ 1) are stored. N
A word dictionary memory (the number of registered words ≧ 2) in which predetermined residual symbols are filled in a portion less than the number of characters, and one shift clock each time m types of candidates for one character are input by the input device. A controller that generates a determination clock and a counter clock the number of times corresponding to the total number of words in the word dictionary memory, and an address of the word dictionary memory that performs reset in synchronization with the shift clock and count-up in synchronization with the counter clock. A counter and a first character string input by the input device
Second, ... ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, The mth candidate shift register for n characters each of which is sequentially forwarded one character at a time corresponding to the mth candidate and synchronized with the shift clock. 1 of data for n characters read from the word dictionary memory
The character, the second character, ..., The character at the corresponding position corresponding to the nth character is the character at the same position in any of the first, second, ... When the first character, the second character, ..., And the nth character comparing circuit that outputs a coincidence signal and a residual detection signal, and the first character, the second character, ... In synchronization with the determination clock, ...... ・
a determination circuit for determining that a word existing in the word dictionary memory appears in the character string input by the input device when a match signal is detected from all the n-th character comparison circuits; and the residual detection signal. And a word length calculating circuit for calculating a word length based on the above.