JPH0827636B2 - Word spelling-phonetic symbol converter - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Outline] In a word spelling-phonetic symbol conversion device, each time an affix is separated from the spelling of a word or a stem according to a predetermined affix separation rule, the stem part from which the affix is separated is exceptional. Search for an exceptional word that sounds like. This allows
It is possible to speed up the word spelling-phonetic symbol conversion process, prevent extra affix separation, and perform word spelling-phonetic symbol conversion without error.

[Industrial applications]

The present invention relates to a word spelling-phonetic symbol conversion device that performs a process of converting word spelling into phonetic symbols at high speed and without error in a sentence-speech conversion process of converting an arbitrary sentence into speech.

When converting an arbitrary sentence into speech, first, a process of converting each word in the sentence into a symbol representing its pronunciation, that is, a phonetic symbol is performed.

However, many words have exceptional pronunciations that do not follow regular conversion rules. For example, the spelling "OVE" is usually pronounced [ouv] or [Λv], but exceptionally pronounced [u: v]. Examples of the latter include move [mu: v] and prove [pru: v].

If all dictionaries of such exceptionally pronounced words are included in the dictionary, a dictionary with a huge storage capacity is required, which is not practical. Also, if a new word is created, it cannot be accommodated. Therefore, the dictionary for exception words has only the basic form of the word (stem), and by separating the word spelling into suffixes such as suffixes and prefixes and the stem, and then searching the exception word dictionary, The storage capacity of the exception word dictionary is reduced and various exception words can be searched.

In addition, the pronunciation of the same word may change depending on the position of the accent. For example, the English spelling "x" is pronounced as [gz] when the following vowel has an accent,
In other cases, it is pronounced as [ks]. Therefore, when converting a word spelling into a phonetic symbol, it is necessary to determine the accent position of the word spelling and convert the word spelling into a phonetic symbol according to a predetermined pronunciation rule.

Therefore, when the word spelling is converted into phonetic symbols, the word spelling-phonetic symbol conversion process is efficiently performed by combining the conversion process according to the regular conversion rule and the exception word process.

[Conventional technology]

FIG. 4 is a block diagram showing the configuration of a conventional word spelling-phonetic symbol conversion system.

In FIG. 4, reference numeral 30 denotes an exceptional word processing means, which processes a word having an exceptional pronunciation. In this exceptional word processing means 30, 31 is an exceptional word dictionary, which spells words that make an exceptional pronunciation and their phonetic symbols, or spellings of words to which an exceptional accent is added and the spelling with an accent added to the spelling. And are stored as a pair.

Reference numeral 32 denotes a suffix separation rule dictionary, which stores suffix separation rules for separating suffixes from the stems of words.

Reference numeral 33 is a prefix separation rule dictionary, which stores prefix separation rules for separating the prefix from the word stem.

Reference numeral 34 is a first exception word dictionary search unit, which performs processing for searching whether the input word spelling exists in the exception word dictionary 31.

Reference numeral 35 denotes a suffix separation processing unit, which refers to the suffix separation rule dictionary 32 and applies a suffix separation rule to separate each suffix added to the word stem from the stem.

Reference numeral 36 denotes a second exception word dictionary search unit, which performs a process of searching for a word stem with a suffix separated in the exception word dictionary 31.

A prefix separation processing unit 37 refers to the prefix separation rule dictionary 33, applies a separation rule for each prefix, and separates each prefix added to the word stem from the stem.

Reference numeral 38 denotes a third exception word dictionary search unit, which performs a process of searching for a word stem with a separated prefix existing in the exception word dictionary 31.

Reference numeral 41 is an accent position estimation rule dictionary, which stores rules for adding accents to word spellings.

An accent position estimation unit 42 refers to the accent position estimation rule dictionary 41 and performs a process of estimating an accent position for an input word spelling or a spelling in which an unconverted affix spelling and phonetic symbols are combined.

Reference numeral 43 is a pronunciation rule dictionary, which stores rules for converting accented spellings into phonetic symbols.

Reference numeral 44 is a spelling-phonetic symbol conversion unit that refers to the pronunciation rule dictionary 43 and performs processing for converting an input accented spelling or a character string accented with a combination of the spelling and the phonetic symbol into a phonetic symbol.

Next, the word-phonetic symbol processing operation of FIG. 4 will be described.

When the word spelling data read from the sentence is input to the exceptional word processing means 30, the first exceptional word dictionary search unit 34
The exception word dictionary 31 is searched for spelling of the input word. If it exists in the exception word dictionary 31 and is paired with the phonetic symbol, the phonetic symbol is output and the exception word processing is terminated. When the spelling with accent is paired, the spelling with accent is input to the spelling-phonetic symbol conversion unit 44.

If it does not exist in the exception word dictionary 31, that is, if the search fails, the word spelling is input to the suffix separation processing unit 35.

The suffix separation processing unit 35 refers to the suffix separation rule dictionary 32 and applies the suffix separation rule to separate each suffix added to the word stem from the word stem.
When a plurality of suffixes are added to the stem, all the suffixes that have been added are separated by sequentially applying the separation rules for each suffix.

The second exception word dictionary search unit 36 performs a process of searching for a word stem whose suffix is separated in the exception word dictionary 31. If it exists in the exception word dictionary 31, the phonetic symbol of the word stem is read out and output as a character string combined with the separated suffix. This character string output is composed of phonetic symbols (stem part) and unconverted spelling (suffix part), and the accent position is undecided. Therefore, the character string output is sent to the accent position estimation unit 42 to determine the accent position.

If it does not exist in the exception word dictionary 31, that is, if the search fails, the character string in which the stem and the suffix are combined is input to the prefix separation processing unit 37.

The prefix separation processing unit 37 refers to the prefix separation rule dictionary 33 and applies a separation rule for each prefix to separate each prefix added to the word stem from the stem.
When a plurality of prefixes are added to the stem, all the added prefixes are separated by sequentially applying the separation rule for each prefix.

The third exception word dictionary search unit 38 performs a process of searching for a word stem with a separated prefix in the exception word dictionary 31. If it exists in the exception word dictionary 31, the phonetic symbol of the stem is read out and output as a character string combined with the separated prefix and suffix. This character string output consists of phonetic symbols (stem parts) and unconverted spellings (prefix and suffix parts), and the accent position is undecided. Therefore, the accent position estimation unit 42 determines the accent position. Sent.

When it does not exist in the exception word dictionary 31, that is, when the search fails, the input word spelling is a word spelling converted into a phonetic symbol according to a regular conversion rule, which is not an exceptional word. Also in this case, it is input to the accent position estimation unit 42 to determine the accent position of the unit spelling.

The accent position estimation unit 42 spells the affix spelling input from the second and third exception word dictionary search units 36 and 38 and the combined character string of phonetic symbols and the word spelling input from the third exception word dictionary search unit 245. On the other hand, referring to the accent position estimation rule dictionary 41, the process of estimating the accent position is performed, and the result is sent to the spelling-phonetic symbol conversion unit 44.

The spelling-phonetic symbol conversion unit 44 refers to the pronunciation rule dictionary 43, and performs a process of converting the input accented spelling or the spelling and the phonetic symbol to be input and a character string to which accent information is added to the phonetic symbol.

As described above, the word spelling can be converted into the phonetic symbol including the case where the input word spelling is an exceptional word and the case where the input word is converted into the phonetic symbol according to the regular conversion rule.

[Problems to be solved by the invention]

The number of affixes that connect a word is generally not high. For example, in the case of English words, the prefix is 0-3,
The suffix is 0 to 4. Therefore, among many affix separation rules, at most about seven rules can be applied to the word.

However, I don't know which rule can be applied, so
It is necessary to investigate the applicability of all rules in each suffix and prefix separation rule dictionary. For example, even if only the separation rule at the head of the dictionary is applicable and the other rules are not applicable, the applicability of all rules in the dictionary is checked.

Therefore, the conventional word spelling-phonetic symbol conversion method has a problem in that the number of useless searches is increased and the exceptional word processing is not efficiently performed.

Furthermore, if there is a word that gives an exceptional pronunciation only when a certain affix is connected, if you search the exception word dictionary for that root after affix separation processing, the root with the affix separated will not be an exception word. There was a problem that the dictionary search failed.

The present invention speeds up exception word processing by searching for the existence of the separated stem of the connection in the exception word dictionary each time a suffix is separated from the spelling by the affix separation rule. SUMMARY OF THE INVENTION It is an object of the present invention to provide an improved word spelling-phonetic symbol conversion device that can prevent word spelling-phonetic symbol conversion without error by preventing the failure of the exception word dictionary search by extraly separating affixes.

[Means for solving problems]

The means taken by the present invention will be described with reference to FIG.
FIG. 1 is a block diagram showing the basic configuration of the present invention.

In FIG. 1, reference numeral 10 is an exceptional word processing means for processing a word that produces an exceptional sound, and includes an affix separation processing section 11 and an exceptional word processing section 12 inside.

The affix separation processing unit 11 separates affixes from word or stem spelling according to a predetermined affix separation rule, and outputs the spelling of the word or stem when there is no affix to separate.

The exception word processing unit 12 searches whether or not the root part separated from the affix by the affix separation processing unit 11 corresponds to an exception word, and outputs the root part if it does, and affix separation processing if not. Processing for outputting to the unit 11 is performed.

[Action]

When the word spelling is input to the exception word processing means 10, the affix separation processing unit 11 separates one affix from the word spelling according to a predetermined affix separation rule, and if there is no affix to separate, spells the word or stem. Is output.

The exception word processing unit 12 searches the word spelling part in which one affix is separated for whether it corresponds to an exception word. When the search result corresponding to the exception word is obtained, the corresponding exception word is output and the exception word processing is ended. If the search fails, the affix separation processing unit 11 is notified and the next affix separation processing is performed.

Similarly, the exception word processing unit 12 is the affix separation processing unit.
Every time when one affix is separated from the stem by 11, processing is performed to search whether the separated stem part of the affix corresponds to an exceptional word, and immediately when it is found that the stem part corresponds to an exceptional word. The exception word is ended, and the affix separation processing unit 11 is caused to perform the next affix separation processing only when the exception word is not satisfied.

By doing the above, the waste of checking the applicability of all affix separation rules as in the conventional method is omitted, and the exception word processing can be sped up. In addition, each time one affix is separated, a process for searching for an exception word is performed, so it is possible to prevent an exception word from being separated by an extra affix, and it is possible to prevent a word spelling-phonetic symbol conversion process. Can be done without error.

〔Example〕

An embodiment of the present invention will be described with reference to FIG. 2 and FIG. FIG. 2 is a block diagram for explaining the configuration of an embodiment of the present invention, and FIG. 3 is a diagram for explaining a concrete example of the word spelling-phonetic symbol conversion processing according to the embodiment.

(A) Configuration of Embodiment In FIG. 2, exception word processing means 10 and affix separation processing section
11. The exception word processing unit 12 is as described in FIG.

In the affix separation processing unit 11, 111 is an affix separation rule dictionary, which stores affix separation rules for separating an affix from a word or a stem.

112 is an affix separation rule reading unit, which is an affix separation rule dictionary 1
Starting from 11, the affix separation rules are individually read in a predetermined order. Since the suffixes and prefixes of each type are arranged in a fixed order, in the case of suffixes, the suffix is read from the suffix separation rule. In the case of a prefix, it is read from the separation of the prefix at the beginning.

A rule end determination unit 113 determines whether or not all affix separation rules in the affix separation rule dictionary 111 have been read.

A rule application determination unit 114 determines whether the read affix separation rule can be applied to an input word or a stem.

An affix separation processing unit 115 applies the affix separation rule to an input word or a stem when the affix separation rule is applicable, and performs affix separation processing.

In the exceptional word processing unit 12, 121 is an exceptional word dictionary, which is a spelling of a word that makes an exceptional pronunciation and its phonetic symbol or a spelling of a word to which an exceptional accent is added and a spelling with an accent added to the spelling. And are stored as a pair.

A second exception word dictionary search unit 122 performs a process for searching whether the input word or spelling of the word stem exists in the exception word dictionary 121.

Reference numeral 13 is a first exception word dictionary search unit, which performs a process in which the spelling of the input word exists in the exception word dictionary 121.

21 is an accent position estimation rule dictionary, which stores rules for adding accents to word spellings.

An accent position estimation unit 22 refers to the accent position estimation rule dictionary 21 and performs a process of estimating an accent for an input word spelling or a spelling in which an unconverted affix spelling and phonetic symbols are combined.

Reference numeral 23 is a pronunciation rule dictionary, which stores rules for converting spellings added to accents into phonetic symbols.

Reference numeral 24 is a spelling-phonetic symbol conversion unit which refers to the pronunciation rule dictionary 23 and converts the input spelling with accent information into phonetic symbols.

(B) Operation of the embodiment The word spelling read from the sentence is the exceptional word processing means 10
The first exception word dictionary search unit 13 searches the exception word dictionary 121 to see if the entire input word is spelled or accented. The dictionary search is performed using, for example, a hash function.

When the word spelling exists in the exception word dictionary 121, the character string of the pair and the input word spelling are replaced and the exception word processing ends. In this case, the output character string is a phonetic symbol (shown below) or a spelling with accent information (shown below). In the case of phonetic symbols, it is output as it is, and in the case of spelling with accent information, it is input to the spelling-phonetic symbol conversion unit 24.

If the word spelling does not exist in the exception word dictionary 121, the search fails and the affix separation rule reading unit 112 is notified of that.

The affix separation rule reading unit 112 is the first exception word dictionary searching unit 1
Upon receiving notification of search failure from 3, affix separation rule dictionary 1
The first affix separation rule is read according to a predetermined order from 11 and sent to the rule end determination unit 113.

The rule end determination unit 113 determines whether all the affix separation rules in the affix separation rule dictionary 111 have been read from the input affix separation rules. When all the affix separation rules have been read, it is because the input word is not an exception word, as will be apparent from the following description. Therefore, the exception word processing is terminated and the input word spelling (shown below) is accented. It is sent to the position estimation member unit 22.

If all affix separation rules have not been loaded, the loaded affix separation rules will be
Sent to 114 to determine if it is applicable to the spelling of the input word. If it is not applicable to spelling input words,
This is notified to the affix separation rule reading unit 112.

Upon receiving this notification, the connection separation rule reading unit 112 reads the next affix separation rule from the affix separation rule dictionary 111, and performs the above-described processes together with the rule end determination unit 113 and the rule application determination unit 114.

When the rule application determination unit 114 determines that the affix separation rule is applicable to the input word spelling, it sends the affix separation rule and the input word spelling to the affix separation processing unit 115 to perform the affix separation process.

The affix separation processing unit 115 performs affix separation processing on word spelling according to the input affix separation rule, and sends it to the second exception word dictionary search unit 122.

The second exception word dictionary search unit 122 searches the exception word dictionary 121 for the generated word stem portion. If the search fails, that is, if the generated stemming is not present in the exception word dictionary 121, the affix separation rule reading unit 112 is notified.

Upon receiving this notification, the affix separation rule reading unit 112 reads the next affix separation rule from the affix separation rule dictionary 111, and repeats the above-described processes.

If the search is successful, that is, if the generated spelling of the stem exists in the exceptional word dictionary 121, spelling of the character string and the stem portion of the pair is replaced, and the exceptional word processing ends. In this case, the output character string contains unconverted affix spellings (hereinafter referred to as spelling + phonetic symbols) in the phonetic symbols of the stem part, and it is necessary to determine the accent positions of these. It is sent to the accent position estimation unit 22.

In the following, referring to FIG. 3, the word spelling “improvments”
Will be explained by taking as an example the case in which is converted into its phonetic symbol [impru: vments].

In the word spelling "improvments", "im" is a prefix, "ment" and the ending "s" are suffixes,
“Prove” is the original word stem. "Prove" is an exceptional word in which the pronunciation of "o" is [u:].
Only the exception word dictionary 121 is stored.

When the word spelling "improvments" read from the sentence is input to the exception word processing means 10, the first exception word dictionary search unit 13 determines whether the entire input word spelling "improvments" is an exceptional pronunciation or accent. Exception word dictionary 121
Find out by searching. Since “improvments” does not exist in the exception word dictionary 121, the search fails and the affix separation rule reading unit 112 is notified of that.

In the input word spelling "improvments", the affix separation rules for the suffixes "s", "ment" and the prefix "im" can be applied.

First, the suffix separation rule for the suffix “s” at the end, for example, “φ * / C (CC) _S *” shown in FIG. 3 (b).
Is applied. In each affix separation rule in FIG. 3, V
Indicates a vowel letter, C indicates a consonant letter, and * at both ends indicates the ending or beginning of a word. The affix separation rule shown in FIG. 3 (b) is "s" if 1 to 3 consonant characters are followed by "s".
The rule is to insert the affix separator * before the.

By applying this suffix separation rule, the suffix "s" at the end is separated, and as shown in FIG. 3 (c), "* im
provision * .s * ”is generated.

The stem part "improv" enclosed by * in the generated character string
ments ”does not exist in the exception word dictionary 121, the second exception word dictionary search unit 122 fails in the search, and notifies the affix separation rule reading unit 112 to read the next affix separation rule.

The next applicable rule is an affix separation rule for separating the suffix "ment", and is represented by "φe *. / VC (CC) _ment *" as shown in FIG. 3 (d), for example. This affix separation rule is "vowel + 1 to 2 consonants + men
If t ", the rule is to insert e and *. before" ment ".

By applying this suffix separation rule, the suffix "ment" is separated, and as shown in Fig. 3 (e), "* impr"
ove * .ment * .s * ”is generated.

Since the word stem portion “improve” in the generated character string does not exist in the exception word dictionary 121, the second exception word dictionary search unit 122
Fails the search and notifies the affix separation rule reading unit 112 to read the next affix separation rule.

The next applicable rule is the affix separation rule that separates the prefix "im". For example, as shown in FIG.
． * / * Im_C (CC) V (V) C (CC) V ". This affix separation rule is" * im + 1 to 3 consonant + 1 to 2 vowel + 1 to 3 The rule is to insert the affix separator. * After "im" if "consonant + vowel character", that is, if "* im" precedes a character string of two or more syllables.

By applying this affix separation rule, the prefix "im" is separated, and as shown in Fig. 3 (g), "* im. * P"
rove * .ments *. * "Is generated. The above-mentioned affix separation rules are, for example, Japanese Patent Application Nos. 60-271261 and 61-61.
-12 2009 etc. are disclosed in detail.

The stem part “prov” enclosed by * in the generated character string
“E” exists in the exceptional word dictionary 121 because the vowel “o” makes an exceptional pronunciation as [u:].

When the second exception word dictionary search unit 122 succeeds in the search of [prove], the pronunciation part [pru: v] obtained by reading the stem part “prove” from the exception word dictionary 121 as shown in FIG. Replace.

"* Im. * [Pru: v] *. Ment generated in this way
* .S * ”is a combination of unconverted spellings and phonetic symbols, and the accent is undecided.
It is sent to the accent position estimation unit 22 as a phonetic symbol.

The accent position estimation unit 22 spells the word input from the rule end determination unit 113 or the second exception dictionary search unit 122.
The spelling + phonetic symbol input from the spelling-phonetic symbol conversion unit 24 is processed by adding accent information according to the rule read from the accent position estimation rule dictionary 21.

The spelling-phonetic symbol conversion unit 24 refers to the pronunciation rule dictionary 23 to convert the accented spelling input from the accent position estimation unit 22 or the first exception word dictionary search unit 13 into phonetic symbols.

As described above, the input word spelling can be converted into the corresponding phonetic symbol.

Next, the exceptional word processing efficiency between the word spelling-phonetic symbol conversion method of the present invention and the conventional method shown in FIG. 4 will be described.

The types of input words are (a) exception words that the affix does not connect to,
(B) Exception words connected by affixes, (c) regular words not connected by affixes, (d) regular words connected by affixes
Types can be considered. For each word, the time required for exceptional word processing is Ta1, Tb1, Tc1, Td for the conventional method.
1. In the case of the method of the present invention, Ta2, Tb2, Tc2, Td2.

(I) When the affix is an exception word that is not connected, both expressions have an exception word processing unit first, so the processing time is equal to the search time of the exception word dictionary, so long as the same dictionary and the same search method are used. , Ta1 = Ta2.

(Ii) In the case of a regular word with no affix connection, spelling matching will be performed with all affix separation rules, and as long as the same affix separation rule dictionary and the same matching algorithm are used, the processing time will be The method is also the same, and Tc1 = Tc2.

(Iii) When the affixes are connected regular words, Td1 = time required for spelling matching with all rules + exception word dictionary search time Td2 = time required for spelling matching with all rules + exception word dictionary Search time x number of affixes. However, if a high-speed algorithm is used for the exceptional word dictionary search, as described above, the number of affixes connected to one word is about 7 at most, and the number of exceptional word dictionary searches is about 7 times. Therefore, Td1 and Td2 are considered to be almost equal.

(Iv) When the affix is an connected exception word, Tb1 = time required for matching with all rules and spelling + exception word dictionary search time Tb2 = time required for matching with rules up to applicable rules and spelling + Exception word dictionary search time. If the rule applicable to spelling is at the end of the rule dictionary, Tb1 = Tb2, but otherwise Tb1> Tb2.

Therefore, Ta1 + Tb1 + Tc1 + Td1 ≧ Ta2 + Tb2 + Tc2 + Td2, and the method of the present invention can perform exceptional word processing faster than the conventional method.

Although the embodiment of converting English word spelling into phonetic symbols has been described above, the present invention is also applicable to a word spelling-phonetic symbol conversion system for languages other than English.

〔The invention's effect〕

As described above, according to the present invention, the following effects can be obtained.

(B) Each time the affix is separated from the spelling of the word or the stem, it is searched whether the stem part becomes an exceptional pronunciation or accent, and if the search is successful, the example external processing ends immediately. Can also speed up exception word processing.

(B) Since processing to search for an exception word is performed each time an affix is separated, it is possible to prevent an affix from being separated excessively and fail to search for an exception word. Word spelling-phonetic symbol conversion Processing can be performed without error.

[Brief description of drawings]

FIG. 1 is an explanatory diagram of the basic configuration of the present invention, FIG. 2 is an explanatory diagram of the configuration of an embodiment of the present invention, and FIG. 3 is an explanatory diagram of a specific example of word spelling-phonetic symbol conversion processing according to the same embodiment. FIG. 4 is an explanatory diagram of a conventional word spelling-phonetic symbol conversion method. In FIGS. 1 and 2, 10 ... Exception word processing means, 11 ... Affix separation processing section, 12 ... Exception word processing section, 13 ... First exception word dictionary searching section, 21 ... Accent position estimation rule dictionary, 22 ... Accent position estimation unit, 23 ... Pronunciation rule dictionary, 25 ... Spelling-phonetic symbol conversion unit.

Claims (1)

[Claims] 1. A word spelling-phonetic symbol converter for converting an arbitrary sentence into speech, comprising an exceptional word processing means for processing spelling of words of an exceptional word that makes an exceptional pronunciation. (A) an affix separation processing unit that separates the affix from the spelling of the word or the stem according to a predetermined affix separation rule, and outputs the spelling of the word or the stem if there is no separating affix; The affix separation processing unit searches for a root part whose affix is separated corresponds to an exception word, and outputs a root part if it does, and outputs an exception word processing unit to the affix separation processing unit if not. , A word spelling-phonetic symbol conversion device.