JPH02201643A - System for discriminatingly reading isomorphic word - Google Patents

Abstract

PURPOSE:To discriminatingly read an isomorphic word without preparing a case pattern for all terminologies by obtaining the intensity of the co-occurrence of an isomorphic word and a literary word in a modification relation through the use of a co-occurrence dictionary and selecting a combination in which the coincidence is more intensive. CONSTITUTION:The above system is equipped with the co-occurrence dictionary to store co-occurrence relation between sentenses made into a clue at the time of discriminatingly read the isomorphic word (expressed in the same way but read in the different way). For example, in the execution of 'WATAKUSHIWA' (I) 'GAKKO E' (to school) 'ITTA' ('ITTA' (went) 'OKONATTA' (done)), since 'ITTA' (went) is the isomorphic word, when the co-occurrence dictionary id retrieved, the moidification relation to coincide at an index level cannot be found in the example sentence. However, the meaning code of 'school' expresses a facility, and since it coincides with 'MACHI' (town) in the hierarchy of a meaning classification one above, the intensity of the co-occurrence is made into 3 according to a prescribed rule, and as for the reading of 'OKONATTA', 'ITTA' is selected. Thus, without preparing the case pattern of all declinable words, the isomorphic word can be discriminatingly read easily.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for distinguishing isomorphic words in a text-to-speech synthesis device.

In order to distinguish between the isomorphic words of the same name, it is necessary to use other words that have some kind of relationship with the isomorphic word as clues. Therefore, in general, a method is used in which a clause that has a dependency relationship with a clause that includes a homograph is used as a clue. Regarding this, please refer to ``Automatic reading method for homographs using hierarchical word attributes''.
(Denimo Communication Society Journal 85/3 Vol. J68
-D No., 3 p, 392-p, 399). According to this, the following methods were used:

(1) If the homograph is a predicate, find the clause that is the essential path of the predicate, and select the clause whose semantic attributes of the noun in that clause match those specified by the case pattern of the predicate. .

(2) If the homographs form a parallel phrase, select a combination that is semantically close to the other noun.

(3) If homographs form an adnominal modification relationship connected by the case particle ``'no'', they can be connected before and after the word to be distinguished through the case particle ``'no''. The semantic attributes of words are given, and combinations that match them are selected.

(4) If a homograph is a case element of a predicate, select the one that matches the semantic attribute specified by the case pattern of the predicate.

(5) When a homograph has a noun clause-predicate relationship (~wa-da.'' relationship), the noun that is the predicate and the noun in the noun clause have attributes that are semantically similar. Select a combination.

(6) If the isomorphic word is modified by an adnominal, a predicate modifying the isomorphic word is created, and a noun having an attribute specified by the case pattern of the predicate is selected.

In order to realize (1), (4), and (6) of the conventional methods, case patterns must be prepared for all predicates. However, this is a very time-consuming task when the number of common words in the dictionary is large, and in order to distinguish the limited number of words called homographs, it is necessary to prepare case patterns for all the words in the dictionary. The problem is that it is not cost-effective.

Purpose The present invention was made in view of the above-mentioned circumstances.
Prepare a co-occurrence dictionary that stores typical examples of co-occurrence relations to help you distinguish between isomorphic words, and use it to determine the strength of co-occurrence between the isomorphic word and clauses in dependency relationships. By using the method of determining the combination with the strongest co-occurrence, it is possible to distinguish between isomorphic words without preparing case patterns for all terms, and when investigating the strength of co-occurrence, , by determining that a co-occurrence relationship holds to some extent, not only when the words match the co-occurrence dictionary at the heading level, but also when the words have a similar semantic classification to the words registered in the co-occurrence dictionary. The purpose of this study is to provide a method for identifying isomorphic words that is structured so that it is possible to infer the co-occurrence relationships of various expressions using only one typical example.

Furthermore, in order to achieve the two-dimensional purpose, the present invention includes a morphological analysis unit that morphologically analyzes a Japanese sentence containing kanji and kana, a dependency analysis unit that examines dependency relationships between clauses, and a dependency analysis unit that examines dependency relationships between clauses. a prosodic symbol generation unit that generates a prosodic symbol string using the results of morphological analysis and dependency analysis;

In a text-to-speech synthesizer equipped with a speech synthesis unit that converts prosodic symbol strings into synthesized speech, the co-occurrence relationship between clauses is used as a clue to distinguish between isomorphic words (words with the same spelling but different pronunciations). If there is a memorized co-occurrence dictionary, if there is a isomorphic word, the strength of co-occurrence between the clause containing the isomorphic word and the clause in a dependency relationship is searched using the co-occurrence dictionary, and the result is a co-occurrence dictionary. Selecting words that are more closely related, and
When determining the strength of co-occurrence, we consider not only the cases where the headings match the co-occurrence dictionary, but also the cases of QL words that are semantically similar to words registered in the co-occurrence dictionary. It is characterized by determining that there is. Hereinafter, the present invention will be explained based on examples.

Figure 1 is for explaining the isomorphic word classification method according to the present invention, and shows the structure of a co-occurrence dictionary, and Figure 2 shows the meaning classification of each word, which is defined hierarchically. .

In this embodiment, the homograph reading process is performed immediately after the dependency analysis in a series of text-to-speech synthesis processes. Next, how the isomorphic word recognition process is performed will be specifically explained. The rules for determining the strength of co-occurrence are defined as shown in Table 1 below.

Example 1 “°°“°゛゛°°゛°°゛°°“′°゛°°゛″””
'1r"""""""""'"′] 謹"""'1 He conducted (did/conducted) an experiment today.

In this example, “gone” is the isomorphic word.

There are three phrases related to ``did'': ``he did n'', ``today'', ``experiment''. When you search for ``did'' in the co-occurrence dictionary, you get
The following two co-occurrence relationships are obtained.

``I went to the town.'' ``I conducted an experiment.'' In the example 10 sentence, there is a dependency relationship such as ``I conducted an experiment,'' which is exactly the same as the entry in the co-occurrence dictionary. The strength of the occurrence is 5, and in this case, ``Ogadatta'' is selected as the reading of ``Ogata''.

Example 2: “I went to school” in the example “゛pa°゛°°“”””’1 謹”””’1 I went to school (I went/did), because “I went” is a homograph. , = 7 If you search the co-occurrence dictionary, you will get the same search results as Example 1. This time, there is no dependency relationship that matches at the heading level in the example sentence. However, the meaning code for "school" is 2 in Figure 2. , 1.1 represents the facility, and “town” and 1 of the semantic classification
Since they match in the upper hierarchy, the strength of co-occurrence is 3, and the reading of ``Ichita'' is selected as ``Ichita''.

Example 3 r”””””””””””””°゛°°°°゛°“”
``1 I left Hachidai/Yatsushiro.

In this example, since "Yatsushiro" is a homograph, searching the co-occurrence dictionary yields "Arrived at Yatsushiro". Since "departed" has the same semantic classification as "arrived", the strength of co-occurrence is 2, and "Yatsushiro" is selected as the reading of "Yashiro".

As is clear from the above explanation, according to the present invention, by using a co-occurrence dictionary, isomorphic words can be easily distinguished without having to prepare case patterns for all words. In addition, by considering the agreement of the semantic classification level when determining the strength of co-occurrence, even if only typical examples are registered in the co-occurrence dictionary, co-occurrence relationships of quite diverse expressions can be realized. can also be handled.

[Brief explanation of the drawing]

FIG. 1 is a diagram for explaining the isomorphic word classification method according to the present invention, and is a diagram showing the structure of a co-occurrence dictionary, and FIG. 2 is a diagram showing hierarchically defined meaning classifications. Patent applicant Ricoh Co., Ltd.

Claims (1)

[Scope of Claims] 1. A morphological analysis unit that morphologically analyzes a Japanese sentence containing kanji and kana, a dependency analysis unit that examines dependency relationships between clauses, and a system that uses the results of the morphological analysis and dependency analysis. In a text-to-speech synthesizer equipped with a prosodic symbol generator that generates a prosodic symbol string and a speech synthesizer that converts the prosodic symbol string into synthesized speech, the co-occurrence relationship between clauses is used as a clue when distinguishing isomorphic words. If there is a isomorphic word, the co-occurrence dictionary is used to search for the strength of co-occurrence between a clause containing the isomorphic word and a clause in a dependency relationship, and the result is , is an isomorphic reading method characterized by selecting words with a stronger co-occurrence relationship. 2. When determining the strength of co-occurrence, we consider not only the cases where the heading matches the co-occurrence dictionary, but also the cases of words that are semantically similar to words registered in the co-occurrence dictionary. 2. The isomorphism reading method according to claim 1, wherein it is determined that there is a relationship.