JPH0283594A - Morpheme composition type english word dictionary constituting system - Google Patents

Abstract

PURPOSE:To reduce the size of a data base by using a morpheme deformation and composition rule base means. CONSTITUTION:Morphemes of necessary kinds of English words are registered in an English word morpheme data base 1 according to a task of a device and the rule of spelling variation, the rule of coupling order among morphemes based upon the environment of a connection between two successive morphemes, etc., are registered previously in the morpheme deformation and composition rule base 2. Then an English word composition part 3 composes an English word by deforming the morphemes phonologically or coupling them as they are and outputs the word to the English word recognizing part 4 of the device as a matching template for English word recognition. Consequently, the number of data to be registered is decreased, so the size of the data base is reducible.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a large word zero consecutive speech recognition device for unspecified speakers, and more particularly to a configuration method of a word dictionary used in the English word recognition unit.

(Prior Art) Conventionally, an English word dictionary used in this type of speech recognition device has a configuration as shown in FIG.

In FIG. 2, reference numeral 9 indicates an English word database for describing the number of English words depending on the task of the recognition device, and 10
11 is an inflection rule base for describing inflection rules for parts of speech that take regular inflections such as regular inflection verbs and nouns, and 11 is an English word matching template for word recognition while checking against rule base 10. English word derivation part for changing the ending of the basic form, 14 is English word derivation part 1
This is an English word recognition unit for outputting the results obtained in step 1.

The word registration type English word dictionary 12 and the English word database 9,
This system includes an inflection rule base 10 and an English word derivation unit 11.

(Problem to be Solved by the Invention) In the conventional English word dictionary for English word recognition described above, depending on the task of the recognition device, the second dictionary is
It is necessary to register word data in the English word database shown in the figure. Usually, word data is expressed as a combination of an alphabetic sequence or a phoneme symbol sequence, depending on the recognition method of the device. In either case, as the number of word clouds to be recognized increases, the storage capacity of the database also increases, so if the task is automatic translation between English and Japanese, the size of the database will be The downside is that it can be quite expensive.

The number of commonly used words in English is around nine or a few thousand, but in tasks such as business English, the number of English words to be registered increases even more, and it is necessary to also register deformations of irregular verbs. . For this reason, the scale of the English word database will inevitably become larger. As described above, in the conventional English word dictionary construction method of the conventional continuous speech recognition device, the size of the English word database increases linearly as the number of words becomes larger.

There are disadvantages such as a large memory capacity and a delay in recognition response time due to an increase in dictionary search time.

The purpose of the present invention is to register morphemes that are elements when composing them with English words, and to register morpheme synthesis rules that combine morphemes as they are or with phonological transformation. Based on the morphological synthesis rules, the above drawbacks are removed and the size of the database is reduced by combining the morphological texts directly or by adding phonological transformation, synthesizing English words, and outputting the results to the word recognition unit. It is an object of the present invention to provide a morpheme composition English word dictionary construction system that can be configured to scale up.

(Means for Solving the Problems) The morpheme composition form English word dictionary construction method according to the present invention includes an English word morpheme database means, a morpheme transformation synthesis rule base means, and an English word composition means. This is what I did.

The English word morpheme database means is for registering morphemes that are the core of English words.

The morpheme transformation synthesis rule base means is for registering synthesis rules for combining morphemes.

The English word synthesis means is of a type that combines and outputs morphemes directly or through transformation based on synthesis rules.

(Example) Next, the invention number will be explained with reference to the drawings.

FIG. 1 is a block diagram illustrating an embodiment of an English word dictionary construction method according to the present invention.

In FIG. 1, 1 is an English word morpheme database, 2 is a morpheme transformation synthesis rule base, 3 is an English word synthesis unit, 4 is for English word recognition, and 5 is a morpheme synthesis form English word dictionary.

Further, 6 is a terminal for connecting a host computer (not shown), F is a terminal for connecting a lower level recognition unit (not shown), and 8 is a terminal for connecting a higher level recognition unit (not shown). This is the terminal to connect.

The English word morpheme database 1 is a database for registering core morphemes that form necessary types of English words according to the tasks of the device.

The morphological transformation/synthesis rule base 2 is a rule base for pre-registering rules for changes in spelling, rules for the order of combinations of morphemes, etc. based on the connection environment before and after morphemes. The English word synthesis unit 3 transforms morphemes phonologically based on the rules described in the morpheme transformation synthesis rule base 2.

Alternatively, they can be combined as they are to synthesize English words and output to the English word identification section 4 of the device as a matching template for English word recognition.

The operation of this embodiment will be briefly explained below.

First, let's consider the problem of automatically translating an English text consisting of a continuous loud voice spoken by an unspecified speaker into a Japanese text. In the recognition of continuously uttered sentences with large words and poor sentences, the speech signal is usually recognized in units of phonemes or syllables, and then higher-level recognition of words, phrases, sentences, etc. is performed. Here, continuous English speech is recognized in units of saliva phonemes, segmentation is performed on the obtained phoneme sequences using pause information between words, and the phoneme sequences divided into words are used as word candidates for alphabetic description using phonological rules. Convert to

When the morphological composition English word dictionary 5 of the present invention is used as an English word dictionary to be referred to as a matching template for word candidates in alphabetical descriptions, the operation of English word composition is as follows.

The English word synthesis section 3 first searches the English word morpheme database 1 for morphemes that can be matched with the alphabetical description word candidates input to the English word identification section 4. For morphemes with increased matching, the English word synthesis unit 3 refers to the rules written in the morpheme transformation synthesis rule base 21C, and selects morphemes that can be connected before and after the morpheme from the English word morpheme database 1.

The selected tiger morphemes are connected before and after the well-matched morphemes in the English word synthesis section 3, with the spelling changed if necessary based on the spelling change rules described in the morpheme transformation and synthesis rule base 2. The connected morpheme sequence is output to the English word recognition unit 4 as an English word matching template. In this unit, a matching test is performed with all alphabetical description word candidates.

As a result of the matching test, if the word candidate cannot be completely matched with the word candidate, the word candidate is considered to have been correctly accepted, and the morpheme number sequence of the matching template is passed from the English word synthesis unit 3 to the English word recognition unit 4. The result of the English word recognition is sent to the higher level recognition unit 8.

As a result of the test, if a complete match cannot be made with the word candidate, a U ) lie signal is sent to the English word synthesis unit 3,
Connectable morphemes are selected again, different English words are combined using matching templates, and the result is output to the English word recognition unit 4.

Even if the above IJ) lie operation is repeated, perfect alignment cannot be achieved.%
``!9 If there are no more connectable morpheme combinations,
The word recognition section 4 reports this to the host computer 6 as a recognition error in the lower level recognition unit 7, and stops the recognition operation.

FIG. 8 shows a specific example of combining morphemes to synthesize an English word. First, FIG. 8(a) shows how derivative English words are synthesized around the core morpheme %nati01l. That is, the morpheme %a which is an adjectivalization suffix to the morpheme %nation'
When llf is connected, 1n&-tonal' is synthesized, and %national' is prefixed with the morpheme %
If you connect 1nter', %1nternati
onal' is synthesized and %1nter-natio
When the morpheme 'ize', which is a verbization suffix, is connected to nal', %1internationalize' is synthesized and %'quantity internationalfze'of'
If you transform lze' into %iz l and connect the morpheme %at-ion', which is each recommended suffix, you get % 1nterna.
tiona11zati-on' is synthesized.

FIG. 3 shows the basic feature of the present invention, in which there is no need to register individual words in this morphological composition English word dictionary.

In Fig. 8(b'), unlike the example in Fig. 3(a), morphemes are not continuously connected to one morpheme, but morphemes are connected in parallel with one morpheme as the core. It shows how different English words can be connected and synthesized. That is, %5urv around the morpheme 'viv'
lve “revive” 5urvival
This indicates that English words such as ``revival'' and t vivld'' are combined.

FIG. 8(c) K:b shows an example in which the spelling changes depending on the phonological environment of the morphemes connected before and after the morphemes. Here, an example will be given in which the prefix morpheme %ln l causes a change in spelling depending on the spelling of the morpheme to be connected afterwards.

In other words, if K"% 1 # comes after, it changes to 1111, if Vc1bl 1pl comes after, it changes to %1m', if %rl comes after, it changes to %1rI, and if it is any other spelling, it changes to %lnl. It represents a change to.

If such regular spelling changes are described on a rule basis, the number of morphemes that need to be registered in the database will be reduced, and the size of the English word dictionary will be smaller.

(Effect of the invention) As explained above, the present invention registers morphemes instead of English words themselves, and automatically synthesizes English words from the morphemes by referring to the spelling change rules of the morphemes and the connection rules of the morphemes. By using it as a matching template for English word recognition, the amount of data to be registered is significantly reduced compared to when registering the English words themselves, so the size of the database can be reduced. effective.

Furthermore, as the size of the database becomes smaller, the time required for data retrieval is reduced, which has the effect of shortening the time required for recognition and speeding up the response time.

Furthermore, since English words usually consist of one or more morphemes, adding a recognized word type requires fewer morphemes than when registering the English word itself. There is. This effect is particularly remarkable for compound words such as new words.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the morphological composition form English word dictionary q7I construction method according to the present invention. FIGS. 2(a) to 2(c) are explanatory diagrams each showing an example of connection of morphemes according to the present invention. FIG. 8 is a block diagram showing an example of a configuration method of a formal Iwakura molded English word dictionary according to the prior art. 1・-Φ English word morpheme database 2 Lun・kai morpheme transformation synthesis rule pace 3...-English word synthesis section 4 fist-・English single @ recognition section 6-8 10 ・ 11 ・ 12 # Morphology old molded English word dictionary・Terminal English word database One inflection rule base, English word derivation part Sand word registration form English word dictionary

Claims (1)

[Claims] English word morpheme database means for registering morphemes that are the core of English words; morpheme transformation synthesis rule base means for registering composition rules for combining the morphemes; A morpheme composition English word dictionary composition system characterized by comprising an English word composition means for combining and outputting morpheme composition forms directly or by transforming them.