JP3870722B2 - Translation device, recording medium - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a speech translation apparatus and a recording medium, and more particularly to an utterance guidance type speech translation apparatus suitable for a portable translation machine for overseas travel and a recording medium for realizing the same.
[0002]
[Prior art]
Conventional translation devices are roughly classified into two types. The first type of translation device is a device that translates an arbitrary input sentence, and the second type of translation device is a device that searches and translates bilingual example sentences prepared in advance according to scenes and keywords.
[0003]
As a translation device for translating the first arbitrary input sentence, many machine translation systems have already been put into practical use for those intended for text. As a translation method, rule-based translation that performs translation based on a dictionary and grammatical rules is widely adopted. However, since it is difficult to describe rules when new sentence expressions (linguistic phenomena) that cannot be handled well by grammatical rules appear, research on example-based translation using examples is also being conducted.
[0004]
On the other hand, in the case of a translation device for speech, there is a problem that a correct translation result cannot be obtained unless speech recognition is first performed correctly. In order to solve this problem, for example, in Japanese Patent Laid-Open No. 08-328585, a correct answer is selected from a plurality of speech recognition result candidates by calculating the linguistic similarity between the input speech and all the examples in the example database. How to choose is shown.
[0005]
As an example of a translation device for a method for searching for a second prepared bilingual example sentence, Japanese Patent Laid-Open No. 56-88564 displays an example sentence when a category is designated, and when a sentence is selected from the example sentence, the selected example sentence is translated. A portable translation device is described in which sentences are output as speech.
[0006]
[Problems to be solved by the invention]
While traveling abroad, there are many occasions where someone exchanges short conversations with relatively fixed contents such as requests and questions. However, because it is an extraordinary scene of travel, there are many cases where the content, expressions, or words that are specifically spoken are not immediately conceivable in response to the speaker's intention to speak. For this reason, in the translation apparatus that translates the first arbitrary input sentence, the input sentence cannot be immediately conceived, and thus the purpose of supporting foreign language conversation during travel may not be sufficiently achieved. For example, when asking where to get off the plane and picking up checked baggage, or when presenting a baggage receipt to the attendant, "Where is the baggage claim for American Airlines 121?""This is a baggage claim card." The Japanese sentence may not come to mind immediately in its entirety. When talking directly with someone who speaks Japanese, your native language, while looking at the reaction of the other party, “Luggage you checked when you got on… Well, where… Flight number? American flight 121.” "I don't get my baggage ... but a baggage exchange card? Oh, is it a refrain when I get on it? Well, this is." If you don't enter a complete sentence such as "Where is the baggage claim for American Airlines 121?" Or "My baggage does not come out. This is my baggage voucher." Also, unlike speaking in your native language, you can not understand even if you ask the other person, such as "What is your flight number?" I need to tell.
[0007]
On the other hand, the translation device that searches for the second pre-prepared bilingual example sentence can avoid such problems of inputting arbitrary sentences. Moreover, since the processing is simple, it is easy to realize as a portable translator. However, in the conventional method of searching for example sentences classified by scene by selecting a menu, if the number of example sentences increases, the selection button is pressed repeatedly to repeatedly narrow down the scenes or scroll through many example sentences for one scene. There is a problem that it takes time to find an example sentence to be used because it needs to be displayed. Further, the conventional method of searching for word replacement candidates for one example sentence by means such as scrolling has a problem that the operability is not good when the number of word replacement candidates is increased. For example, in the example of asking the location of the baggage claim mentioned above, after finding the example sentence “Where is the baggage claim for Japan Airlines 753?”, Scroll and press the button to change the flight name to “American Airlines” Replacing it with "121 flight" is troublesome.
[0008]
An object of the present invention is to display an example sentence template suitable for a user's utterance intention and induce the user's utterance, thereby translating the advantage of the example sentence search type translation apparatus and the advantage of the arbitrary sentence input type translation apparatus. To implement the device.
[0009]
[Means for Solving the Problems]
In order to solve the above-described problem, in the translation apparatus, an example sentence file that stores a plurality of example sentence templates, a voice recognition unit that recognizes input voice, and a phrase in the example sentence template similar to the voice based on the voice It is assumed that it has a means for replacing and translating, a voice output unit for outputting the translated sentence as a voice, or a display for displaying the translated sentence. A step of recognizing a first voice by voice recognition means, searching for an example sentence template associated with the first voice and displaying it on a display; Recognizing the speech by the speech recognizing means, replacing the phrase in the searched example sentence template based on the second speech and translating, and displaying the translation result on the nest play, or It is assumed that a step of outputting sound is recorded.
[0010]
For example, the translation device is a translation device that searches for example sentences prepared in advance, and the example sentences stored in the example sentence file are described as an example sentence template including a portion (hereinafter referred to as a slot) that can be replaced with a phrase. In the example sentence template, as the information regarding the slot, a classification code (for example, information such as place and date) of a word that can appear in the slot and a representative example are described. Each example sentence template is assigned a keyword (for example, information on reservations, airplanes, tickets, etc.). It is assumed that the keyword can describe not only specific words but also classification codes.
[0011]
In the word dictionary of the translation apparatus of the present invention, pronunciation information and classification codes of words are described as information for searching for example sentence templates using voice input keywords. In addition, as information for performing template translation, headline grammar information, translated words, and translated word grammatical information are described.
Means for searching for example sentence templates include word recognition means for recognizing a keyword inputted by voice, means for searching for an example sentence template in which the recognized word and its classification code are described in the keyword information, Means for displaying an example sentence template on the screen.
[0012]
As a means for performing template translation, a sentence speech recognition means for selecting an example sentence template having a high similarity from a plurality of example sentence templates and a phonological information in a word dictionary are referred to. Word speech recognition means for recognizing words inserted into slots, means for determining translations and word forms of words inserted into slots with reference to word dictionary and translation rules, translation information of example sentence templates and insertion into slots Means for generating a translation by combining the translated phrases.
[0013]
When the user inputs a word by voice, the input word and classification code are recognized by referring to the phoneme information in the word dictionary. These are searched for example sentence templates described in the keyword information and displayed on the screen. In the example sentence template, representative examples of words that can appear in the slot are described as information about the slots included in the example sentence. Therefore, when the example sentence template is displayed on the screen, it is displayed in a state where the representative example is inserted in the slot.
[0014]
In a state where one or more example sentence templates are displayed on the screen, a sentence in which a user's desired word / phrase is inserted is input to the slot of the example sentence template. Then, the translation of the voice input sentence is executed based on the example sentence template having a high similarity among the example sentence templates displayed on the screen, and the translated sentence is displayed. For example, when the user inputs the word “phone” by voice, an example sentence template in which the word “phone” is described in the keyword information is searched. Example screen templates such as “Where is [phone]”, “Tell me how to place [local phone]” and “I want to make a [collect call] to [Japan]” are displayed on the screen. The underlined portion represents a slot. The user scrolls the screen and searches for an example sentence that can be used. When you find an example sentence that you can use, for example, "I want to make a [collect call] to [Japan]", I will speak the sentence "I want to make an international call to Canada". Then, the translator recognizes that it used the example sentence template “I would like to call [Collect Call] on [Japan]” and that “Japan” and “Collect Call” were replaced with “Canada” and “International Call”. Recognize and perform template translation.
[0015]
DETAILED DESCRIPTION OF THE INVENTION
An embodiment of the present invention will be described with reference to FIGS. This embodiment is a portable speech translation apparatus for a user whose native language is Japanese to travel in an English-speaking country. This speech translation apparatus searches and displays example sentence templates when a user inputs a keyword by speech. When the user speaks by replacing a replaceable word in the example sentence template with another word, the user recognizes which example sentence template was used to replace what word, and displays / speaks a translation for the input sentence.
[0016]
Next, the appearance of the translator of this embodiment will be described. FIG. 1 is a diagram showing the appearance and initial screen of a speech translation apparatus according to an embodiment of the present invention. The speech translation apparatus 101 includes a display screen 102, a power switch 103, a reset button 104, a speaker 105 for outputting a translated sentence as a voice, a microphone 106 for inputting a keyword for example sentence search, a sentence example search / translation / translation utterance, and the like. An execution button 107 for instructing, a “return” button 108 for returning to the previous screen, and scroll buttons 109 to 112 for moving back and forth and right and left on the screen are provided.
[0017]
Next, the hardware configuration of the translation of the present embodiment will be described. FIG. 2 shows a hardware configuration of the speech translation apparatus. The speech translation apparatus includes a CPU 201 for performing control of the apparatus, speech recognition, translation, and speech synthesis processing, a speech input device 202, a speech output device 203, a word dictionary 204, an example sentence file 205, and a memory 206. The memory 206 stores a control program 261, a word speech recognition program 262, a sentence speech recognition program 263, a template translation program 264, a speech output program 265, and a screen display program 266.
[0018]
Next, the contents of the example sentence file will be described.
[0019]
FIG. 6 shows the contents of the example sentence file. The example sentence file includes, for each example sentence template, example sentence number 601, basic example sentence number 602, detailed example sentence number 603, keyword 604, original sentence template 605, pronunciation information 606 describing pronunciation of the original sentence template, translated sentence template 607, and pronunciation of the translated sentence template. Translated pronunciation information 608 and slot information 609 are described. In the slot information 609, the number of slots 6091 is described. For each slot, a representative example 6092 of the words that enter the slot, a translation 6093 of the representative example, pronunciation information 6094 of the translation of the representative example, and what word is included in the slot And grammatical information 6096 describing the grammatical features (articles, singles, etc.) of the slot portion in the translation.
[0020]
There are basic example sentences such as “I want to make a reservation” and detailed example sentences such as “Find the flight that I can get the fastest”. For the basic example sentence, the detailed example sentence number corresponding to the basic example sentence is described in the detailed example sentence number 603. For the detailed example sentence, the example sentence number of the corresponding basic example sentence is described in the basic example sentence number 602.
[0021]
When you search for an example sentence by keyword, the basic example sentence is displayed at the top. When the user selects “I want to reserve” and instructs display of a detailed example sentence, a detailed example sentence such as “Find the flight that can be boarded first” is displayed. In the original sentence template 605 and the translation sentence template 607, the replaceable words / phrases are slots.
[0022]
When displayed on the screen, the word of the representative example 6092 is embedded in the slot of the original text template. When the user instructs the display or utterance of the translation without replacing the word, the translation 6093 of the representative example is embedded in the translation and displayed, and the pronunciation information 6094 of the translation of the representative example is embedded in the translation pronunciation information 608. To do.
[0023]
In the description of the slot semantic information 6095, words may be listed as “round trip ticket; one way ticket; ticket”, or by using a classification code such as “&# 60 place &# 62”. Also good. If any word can be replaced, no condition is described.
[0024]
An example sentence in which a classification code such as “&# 60 place &# 62” is described in the keyword 604 is searched from all words having the classification code. For example, if a user enters a word with a classification code of “&# 60 location &# 62”, such as “Post Office”, “Bus Stop”, or “Bank” as a keyword, Example sentence 1356 “Closest &# 60S1 &#"Where is 62?" Is searched and displayed.
[0025]
Next, the contents of the word dictionary will be described. FIG. 7 shows the contents of the word dictionary. In the word dictionary, for each word item, a word number 701, a heading 702, a reading 703, pronunciation information 704, grammar information 705, a classification code 706, a translation 707, a translation pronunciation 708, a translation grammar information 709, and an example sentence number 710 are stored. is described. An example sentence number 710 of a record having a classification code such as “&# 60 location &# 62” as a headline is an example sentence number of an example sentence having the classification code as a keyword.
[0026]
Next, the operation of the speech translation apparatus according to this embodiment will be described. FIG. 3 is a flowchart showing an outline of the operation of the speech translation apparatus. When the power button 103 is pressed (301), the program in the memory 206 is activated and an initial screen is displayed on the display screen 101 (302). When voice is input through the microphone 106 while the initial screen is displayed, keyword voice recognition is performed (303).
[0027]
Next, keyword speech recognition will be described with reference to FIG. First, when a word is input by speech (402), the word dictionary 204 is referred to, word speech recognition is performed by a speech recognition algorithm such as a hidden Markov model (HMM) (403), and word candidates are displayed on the screen ( 404).
[0028]
An example of the word candidate display screen will be described with reference to FIG. On the word candidate display screen, kana notation of the first candidate is displayed on the first line, and word candidates from the first candidate to the tenth candidate are displayed on the second and subsequent lines. Initially, the first candidate word is highlighted. By using the front / rear / left / right scroll keys, the highlighted word moves forward / backward / left / right. Thereby, the user can select the intended word. The fourth and lower candidates that are not displayed on the screen in FIG. 8 can be viewed by scrolling the screen with the down scroll key. When there is a homonym, the word candidate displays the homonym on one line, and when there is no homonym, it displays a parallel translation. “*” At the beginning of a word candidate indicates a keyword for example sentence search.
When the user selects an intended word and presses the execute button (405), the word is determined as a keyword for example sentence search (408), and the keyword recognition process is terminated (409). If the word intended by the user is not found in the word candidate, the input word is corrected at the kana level by selecting the first line of the kana notation and pressing the execution button (406), and the word dictionary search can be performed again (step 406). 407). In addition, by pressing the “return” button, it is possible to return to the keyword voice input acceptance state.
[0029]
The word candidate display screen is used by the user to confirm the keyword speech recognition result. However, since the translated word is displayed together with the word candidate, it can also be used as a bilingual dictionary. It is also possible to utter a pronunciation by selecting a translated word and pressing the execute button. In the case of homonyms, the translation is not displayed on the word candidate display screen, but if you select a word from the homonyms and press the execute button, the translation and example sentences may be displayed on the next screen. it can.
[0030]
Next, example sentence search will be described. When the input word is determined, the example sentence with the number described in the example sentence number 710 of the input word is extracted from the example sentence file 205 in the word dictionary 204 (304) and displayed on the screen (305). For words in which the classification code 706 is described in the word dictionary, an example sentence using the classification code as a keyword is also searched. For example, for the input word “bank”, in addition to an example sentence using “bank” as a keyword, an example sentence using a classification code of &# 60 location &# 62 &# 60 organization &# 62 as a keyword is also searched. The number of the example sentence having the classification code &# 60 place &# 62 as a keyword is described in the example sentence number 710 of the word dictionary record heading “&# 60 place &# 62”.
[0031]
FIG. 9 shows an example of an example sentence display screen when an example sentence search is performed using “reservation” as a keyword. In FIG. 9, reference numeral 901 denotes a searched example sentence. On the example sentence display screen, the dictionary headline and translation of the word are displayed on the first line, and the first three sentences of the searched example sentence are displayed on the second and subsequent lines. The example sentences after the fourth line can be viewed by scrolling the screen with the scroll keys. Example sentences with ★ at the beginning of the sentence indicate basic example sentences. In the example sentence display screen, the original example template 607 embedded in the slot of the original template is displayed to indicate that the phrase can be replaced with an underline. In FIG. 10, “Dallas”, “Miami”, and “14th” are replaceable phrases.
[0032]
Next, example sentence narrowing will be described. Parts of the search example that do not fit in the screen can be viewed using the up / down / left / right scroll keys, but if there are many example sentences, keywords can be added to narrow down the example sentences. This makes it easier to search for example sentences. When the user wants to narrow down example sentences, the user selects the keyword displayed on the first line of the screen with the scroll button (highlights it) and presses the execution button. Next, when an additional keyword is inputted by voice, the above-described keyword voice recognition is executed (307), the example sentences are narrowed down (308), and the narrowed down example sentences are displayed (309). FIG. 9 shows an example sentence search result using “reservation” as a keyword, and FIG. 10 shows a result of narrowing down example sentences by “flight”. When the user sees the narrowing down result and presses the “return” button, the example sentence before narrowing down is displayed (311).
[0033]
When selecting an example sentence having no replaceable phrase, the example sentence may be selected by moving the cursor with a scroll key (312 to 314). When an example sentence is selected and the execution key is pressed, translation is executed (322).
[0034]
Next, detailed example sentence display will be described. When selecting an example sentence with the scroll key, selecting a star attached to the head of the basic example sentence instead of the example sentence itself and pressing the execution key displays a related detailed example sentence (314). For example, in the example sentence display screen of FIG. 9, when selecting “I want to reserve” on the third line and pressing the execution key, the translation result is displayed. Select “★” on the left side of “I want to reserve” and press the execution key. Then, a detailed example sentence “I want to reserve” as shown in FIG. 12 is displayed.
[0035]
Next, sentence speech recognition will be described. When the user utters a sentence while watching the example sentence display screen (315), sentence speech recognition is executed (316). Sentence speech recognition is performed, for example, by an automaton-controlled continuous word speech recognition algorithm using an HMM described in “Speech recognition by a probability model” (IEICE, 1988) pp46-50.
[0036]
FIG. 5 shows a flow of sentence speech recognition processing. First, a finite state automaton as shown in FIG. 14 is created for the example sentence template displayed on the screen. FIG. 14 is a finite state automaton created for the example sentence template “where is the nearest &# 60s1 &# 62”. In FIG. 14, reference numerals 1401 and 1402 denote HMMs (Hidden Markov Models) created from pronunciation information “Ichibanchikai” and “Wadkodeska”. A model of the original text template is created based on the pronunciation information 606 described in the example sentence file (502).
[0037]
The model of the slot portion is created based on pronunciation information of words that can enter the slot. For this purpose, first, by referring to the semantic information 6095 of the slot information 609 of the example sentence file (504), it is recognized which word of the word dictionary registered words can enter the slot. The words that can enter the slot are the listed word (505) when the word is listed in the slot semantic information 6095, and the word (506) having the specified classification code when the classification code is specified. If nothing is specified, the word dictionary registered word (507) is assumed. The model of each word is created based on the pronunciation information 704 of the word described in the word dictionary (508).
[0038]
Once a finite state automaton is created for the example sentence template displayed on the screen (509), speech recognition of the input sentence is performed by the automaton-controlled continuous word speech recognition algorithm by HMM, and the top 10 words are stored as slot candidate words for each slot. (510).
[0039]
When speaking while looking at example sentence templates, there may be pauses before and after the slot or immediately after the phrase containing the slot, so for each example sentence template, not only the finite state automaton for the whole sentence but also the pause. Creating a finite state automaton of phrases separated by places where they are likely to enter can provide higher recognition accuracy. For example, in the above example, in addition to “Ichibanchikaiba Stay Wadokodeska” shown in FIG. 14 and the like, a finite state automaton such as “Ichibantikai”, “Bastei”, “Wadokodeska”, “Bastiwade”, “Dokodeska” is also created, It associates with an example sentence template including a phrase. In addition, when the pose is actively incorporated and a pose is entered with `` Ichibanchikai '', it displays that it matches the `` closest '' part of `` Where is the nearest &# 60s1 &# 62 '' It is also possible to guide the user to say continuation after confirmation.
[0040]
In this embodiment, slot insertion words are recognized within the range of word dictionary registered words, but not all words can be registered in the dictionary in advance. In particular, nouns such as proper nouns and local specialties are important in travel conversation, but cannot be covered by a word dictionary. However, when the user wants to use such a word, there are many cases where the word is known in the original language, and if only the word is used, the pronunciation can be somehow approximated. Therefore, when a word of a slot portion is inputted by voice while pressing a specific button, it is conceivable that the sound inputted by the user is embedded in the translated sentence and outputted for the slot portion. As described above, if you use a method that creates a finite state automaton that corresponds to a phrase and associates it with an example sentence template that includes that phrase, you can specify the example sentence template from the phrases before and after the part that you spoke by pressing the specific button. be able to. As a result, for example, it is possible to use the example sentence template “How many minutes is waiting for this attraction” to say “How many minutes will the Back to the future wait?”. Also, for a specific slot such as &# 60s1 &# 62 in "My name is &# 60s1 &#62", it is possible to embed the sound as it is input by the user in the slot portion of the translation.
[0041]
In this embodiment, the target to be compared with the voice input sentence is limited to the example sentence template displayed on the screen, but the whole example sentence template searched by the keyword may be used. Also, example sentence templates selected in advance by the user or recently used example sentence templates may be stored in a “user example sentence file” and used as a comparison target.
[0042]
If you adopt a method that creates a finite state automaton that corresponds to a phrase and associates it with an example sentence template that includes that phrase, mark the range to be spoken with a wavy line when displaying the example sentence template, It may be possible to let people know that they don't have to speak the whole sentence. This method improves usability. For example, if the sentence is “Look for the earliest flight”, say “Look for the earliest flight” or “Fastest flight” and say “ Save.
[0043]
Next, sentence recognition result display / correction will be described. The sentence speech recognition result is indicated by moving the cursor to the recognized example sentence template and replacing the word in the slot with the input word from the representative example (317). For example, when the example sentence display screen of FIG. 10 is displayed and the user inputs a voice saying “I want a one-way ticket for 17 days from Chicago to Seattle” and the sentence speech is correctly recognized, the screen of FIG. Is displayed. Looking at the sentence recognition result, the user presses a “return” button to redo the sentence input (318) or correct the sentence recognition result as necessary (319). If the example sentence template is not selected correctly, another sentence is selected with the scroll key, and the word recognition in the slot portion is performed again using the example sentence template. If the word recognition in the slot part is not correct, the slot candidate word stored at the time of sentence recognition is displayed when the slot is selected with the scroll key and the execution button is pressed. If there is a correct slot candidate word, it is selected by selecting with the scroll key and pressing the execution button.
[0044]
Next, translation will be described. When a sentence matching the voice input sentence is confirmed, translation is instructed by pressing the execution button (320 to 321). It is also possible to view a detailed example sentence by selecting a basic example sentence by voice input, selecting ★ at the beginning of the sentence, and pressing the execution button (321). The template translation is performed by embedding the translation of the word / phrase in the slot portion in the translation of the example sentence template (322). In the example sentence template, grammatical information 6096 of the slot in the translated sentence is described. By referring to this and the grammatical information 709 of the translation in the word dictionary 204, the article or single or the like of the translation of the slot portion is determined. For example, if you want to embed the word `` apple apple '' in which the grammatical information `` Initial vowel '' is described in the slot designated `` indefinite article, singular '', add the indefinite article To embed the word “candy” in the designated slot, embed the plural model “dies”.
[0045]
In the translation result display step (323), the translation result display screen shown in FIG. 13 is displayed. If the user's instruction (324) is a “return” button, the example sentence displayed before execution of translation is displayed (325). If the translation is selected and the execution button is selected, the translation is uttered (326), and the original sentence is selected. If it is an execution button, the next utterance example sentence is displayed (327). The example sentence displayed as the next utterance example is a detailed example sentence when the previously translated sentence is a basic example sentence, and the example sentence displayed before the translation is executed when the previously translated sentence is a detailed example sentence. is there. This is because a set of detailed example sentences searched with the same keyword is a related example sentence, and it is considered that there is a possibility that a plurality of sentences will be used continuously from these example sentences. As an alternative, keywords related to the scene, such as “hotel”, “front”, “shopping”, etc., are assigned to each example template as “scene information” as information separate from the keyword, and only example sentences with common scene information Can be displayed as an example sentence for the next utterance. In addition, even when a new keyword search is performed, the scene information that matches the example sentence used immediately before is preferentially displayed on the upper side, or the same scene continues until it is reset, and the example sentence in the scene information It is conceivable to display with narrowing down.
[0046]
In this embodiment, the template translation is performed using the bilingual example sentence template, but the true purpose of the example sentence template is to induce the user's utterance. Therefore, the translation process may be performed by machine translation based on a dictionary and grammatical rules. In that case, the example sentence template does not necessarily need to be translated, and only the example sentence template of the input language is sufficient.
[0047]
As mentioned above, although the embodiment as a portable speech translation apparatus has been described, the present invention records the flows described in FIGS. 3, 4, and 5 for realizing the translation apparatus as described above. It may be a recording medium. As such a recording medium, a step of recognizing a speech input phrase by a speech recognition means, a step of searching an example sentence template associated with the phrase in the example sentence file, and displaying the searched example sentence file on a display A step of recognizing a sentence spoken by the user based on the displayed example sentence template, a step of comparing the recognized sentence with the searched example sentence template, and recognizing the most similar one A computer-readable recording of the steps to replace and translate replaceable words / phrases in the example sentence template based on the sentence spoken by the user, and to display the translation result on the display or to output the voice Recording media can be considered.
[0048]
Another type of translation apparatus may have a different user interface. For example, in a speech translation apparatus provided at a window of a hospital, a government office, a hotel, etc., it is conceivable that a screen is used as a touch panel and only a replacement word is uttered by touching a slot with a finger. In this embodiment, considering the small display screen and use with one hand, the example sentence template search is performed by keyword voice input. However, the scene and utterance intention (question, request, explanation, etc.) It is also possible to classify example sentence templates with, and narrow down by menu display / selection.
[0049]
In the present embodiment, the speech translation apparatus that translates from Japanese to English has been described, but the present invention can also be applied to other language pairs. In this embodiment, in the bilingual example sentence template, information for recognition is described for the original sentence template and information for generation for the translated sentence template is described. It is also possible to execute translation in both directions by describing the information and the information for generation. When bi-directional translation is performed, it is effective to associate the example sentence template of the answer with the example sentence template of the question sentence in the same manner as the basic example sentence and the detailed example sentence are associated with each other in this embodiment.
[0050]
【The invention's effect】
ADVANTAGE OF THE INVENTION According to this invention, the example sentence template suitable for a user's utterance intention can be searched quickly from the keyword input by speech, and a user's utterance can be induced | guided | derived. Users can talk while looking at example sentence templates, so they can easily say what they want to say in the form of a complete sentence, and they can speak any word or phrase as expected in the slot of the example sentence template. There is no troublesome replacement. Accordingly, it is possible to realize a translation apparatus that has the advantages of the example sentence search type translation apparatus and the advantage of the arbitrary sentence input type translation apparatus.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating an appearance and an initial screen of an embodiment of a speech translation apparatus according to the present invention.
FIG. 2 is a block diagram showing a configuration of an embodiment of a speech translation apparatus according to the present invention.
FIG. 3 is a flowchart showing the operation of the embodiment of the speech translation apparatus of the present invention.
FIG. 4 is a flowchart showing a procedure of word speech recognition.
FIG. 5 is a flowchart showing a procedure of input sentence speech recognition.
FIG. 6 is a diagram showing the contents of an example sentence file.
FIG. 7 is a diagram showing the contents of a word dictionary.
FIG. 8 is a diagram showing a screen display example in a word candidate display step.
FIG. 9 is a diagram showing a screen display example in an example sentence display step.
FIG. 10 is a diagram showing a screen display example in a narrowed example sentence display step.
FIG. 11 is a diagram showing a screen display example in a sentence recognition result display step.
FIG. 12 is a diagram showing a screen display example in a detailed example sentence display step.
FIG. 13 is a diagram showing a screen display example in a translation result display step.
FIG. 14 is a diagram illustrating an example of a finite state automaton for an example sentence template.
[Explanation of symbols]
201: CPU
202: Voice input device
203: Audio output device
204: Word dictionary
205: Example sentence file
206: Memory
261: Control program
262: Word speech recognition program
263: Sentence speech recognition program
H.264: Template translation program
265: Audio output program
266: Screen display program.

Claims (9)

In a translation apparatus having a CPU, a voice input unit, a display, an example sentence file for storing an example sentence template in association with its pronunciation information, and a storage means for storing a word dictionary for storing words in association with their pronunciation information A translation method,
A first step of displaying an example sentence template stored in the example sentence file on the display, the example sentence template including a replaceable part and having pronunciation information stored in advance in a memory;
A second step of receiving voice input from the voice input unit;
In the CPU,
A speech model generated using pronunciation information of the displayed example sentence template stored in the storage means and pronunciation information of a word that can enter a replaceable part of the example sentence template is compared with the voice input A third step in which an example sentence template including a speech input portion corresponding to the replaceable portion is a recognition result of the speech input;
A fourth step of outputting a recognition result of the voice input portion;
Translation method characterized by having a fifth step of outputting the translated the recognition result. The translation method according to claim 1,
In the first step, a plurality of the example sentence templates are displayed,
Wherein in a third step, as compared with phonetic information of a plurality of example sentences templates the speech input is the display, the translation method characterized by identifying a sentence template corresponding to speech input. The translation method according to claim 1,
In the first step, displaying information indicating the characteristics of the words to be inserted into the replaceable part of the example sentence template,
Wherein in a third step, the recognition of the voice input portion, the translation method and performing with phonetic information word corresponding to characteristics of the displayed words of the word dictionary. A translation device,
An example sentence template including a replaceable part, an example sentence file including pronunciation information thereof, and a memory storing a word dictionary storing a plurality of words and voice information thereof;
A display that displays information about the example sentence template and the replaceable portion stored in the memory;
An input unit for receiving voice input;
CPU and
The CPU stores the speech model generated using the pronunciation information of the displayed example sentence template stored in the memory and the pronunciation information of words that can enter a replaceable portion of the example sentence template, and the voice input. A translation apparatus characterized in that an example sentence template including a speech input portion corresponding to the replaceable portion is compared as a recognition result of the speech input , and the recognition result is translated and output. The translation device according to claim 4,
The display displays a plurality of the example sentence templates,
Wherein the CPU, the audio input compared with phonetic information of said displayed plurality of example sentences templates, translation apparatus and specifying a sentence template corresponding to speech input. The translation device according to claim 4,
The display displays information indicating the characteristics of words inserted in the replaceable part of the example sentence template,
The translation apparatus according to claim 1, wherein the CPU recognizes the voice input portion by using pronunciation information of a word corresponding to a feature of the displayed phrase in the word dictionary. A program storage medium that stores a processing unit, a voice input unit, a display, an example sentence file that stores an example sentence template in association with the pronunciation information, and a word dictionary that stores words in association with the pronunciation information A translation device having storage means;
A first step of displaying an example sentence template stored in the example sentence file on the display, the example sentence template including a replaceable part and having pronunciation information stored in advance in a memory;
A second step of receiving voice input from the voice input unit;
In the processing unit, the speech model generated using the pronunciation information of the displayed example sentence template stored in the storage unit and the pronunciation information of a word that can enter a replaceable part of the example sentence template, and the voice input And a third step in which an example sentence template including a speech input portion corresponding to the replaceable portion is used as a recognition result of the speech input, and a fourth step of translating the recognition result and outputting the translation result And the steps
A program storage medium for storing a program characterized in that the program is executed. 8. The program storage medium according to claim 7, wherein the program is
In the first step, a plurality of example sentence templates are displayed,
The program storage medium characterized in that, in the third step, the voice input is compared with pronunciation information of the plurality of displayed example sentence templates to identify an example sentence template corresponding to the voice input. 8. The program storage medium according to claim 7, wherein the program is
In the first step, information indicating the characteristics of the words to be inserted into the replaceable part of the example sentence template is displayed;
The program storage medium characterized in that in the third step, the voice input portion is recognized using pronunciation information of a word corresponding to a feature of the displayed phrase in the word dictionary.

Images