JPH0561905A - Sentence analyzing device - Google Patents

Abstract

PURPOSE:To analyze the sentences in real time and in sequence with no return analysis by changing the single sentence unit to the single word unit for transfer of data carried out among those devices forming a sentence analyzing device and furthermore using the voice information on the intonation, the accent and the pause, etc. CONSTITUTION:A morpheme analyzer 121 converts the character strings received sequentially from an input device 11 into the morophemes before the end of a sentence. Then the analyzer 121 performs the semantic analyses of the morphemes received sequentially before the end of the sentence to produce the semantic expression part terms with use of a semantic dictionary 15 where a semantic analyzer 13 defines the word meanings as the partial relation between the situations set before and after the occurrence of speech. Then a semantic interpreter 14 receives the semantic expression part trrms in sequence and interpretes the word meanings based on the present syntax and the knowledge except the languages.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sentence analysis device for sequentially performing sentence analysis of a natural language sentence input in real time.

[0002]

2. Description of the Related Art Conventionally, when a sentence analysis device analyzes a natural language sentence, a morpheme analysis device, a syntactic analysis device, and a semantic analysis device determine the end of a sentence with a special symbol such as a reading point, and perform morpheme analysis on a sentence-by-sentence basis. And parsing,
The parse tree that is the result of the syntactic analysis is passed to the semantic analysis device, semantic analysis is performed, a semantic expression is obtained, and the semantic expression is interpreted.

[0003]

However, in the conventional method, the minimum unit of processing of the morpheme analysis device, the syntactic analysis device, and the semantic analysis device is one sentence or a semantic expression corresponding to one sentence. Termination of sentence such as reworking in the middle, syntactically incomplete fragmentary sentence such as word stop without predicate, or voice input that does not originally include special symbol such as punctuation mark indicating the end of one sentence There is a problem that it is difficult to perform a sentence analysis process on a natural language sentence that is difficult to discriminate.

The present invention solves the above-mentioned problems of the prior art. It makes it possible to arbitrarily subdivide a sentence processing unit and configures a sentence analysis device so as not to depend on the judgment of the end of one sentence. By doing so, the object is to analyze a sentence that is difficult for the above-mentioned conventional sentence analysis device to handle.

[0005]

In order to achieve this object, the present invention firstly provides at least an input means for inputting a character expressing a natural language sentence and an output of the input means without terminating the sentence. Morphological analysis means for sequentially converting into morphemes, a semantic dictionary describing the meaning of words as a partial relationship between the situations before and after utterance, and a semantic analysis of the output of the morphological analysis means based on the semantic dictionary. The structure includes a semantic analysis means for sequentially generating meaning expression sub-terms corresponding to the meaning of a morpheme, and a meaning interpretation means for interpreting the meaning expression sub-term based on context and knowledge outside the language. There is.

Secondly, input means for inputting characters expressing a natural language sentence and voice information as auxiliary input, morphological analysis means for sequentially changing the natural language sentence into morphemes based on the voice information, and based on the voice information. The structure includes a semantic analysis unit that determines the end of one sentence and sequentially converts the output of the morphological analysis unit into a semantic expression.

[0007]

In the sentence analysis device according to the first means of the present invention, the input means inputs a character expressing a natural language sentence in real time, and the morpheme analysis means transmits a character string sequentially passed from the input means. Using a semantic dictionary, which is characterized by sequentially converting into morphemes without straddling the end of a sentence, and by means of the semantic analysis means that the meaning of a word is a partial relationship between the situations before and after the utterance. The semantic analysis of the morphemes sequentially output by the morpheme analysis means is performed without waiting for the end of the sentence to sequentially generate the semantic expression subterms, and the semantic interpretation means sequentially receives the semantic expression subterms and the current context and Sentences entered by making data transfer between input means, morpheme analysis means, semantic analysis means, and semantic interpretation means morphemes and meaning expression sub-terms by performing interpretation in light of knowledge outside the language Sentence of parsing It is possible to perform sequentially without waiting for the end.

Further, in the sentence analysis device according to the second means, the morphological analysis means performs morphological analysis using speech information such as accents and poses with the speech information as an auxiliary input, and the semantic analysis means converts into speech information such as intonation. Based on 1
By judging the end of a sentence and converting it to a semantic expression sequentially, it is possible to determine the end of a sentence, determine the type of sentence such as a question sentence or an affirmative sentence, and remove the ambiguity that occurs in morphological analysis or semantic analysis operation. It becomes possible to do it.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A first embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a conceptual diagram of a sentence analysis device according to an embodiment of the present invention. In FIG. 1, 11 is an input device, 12 is a morphological analysis device, 13 is a semantic analysis device,
Reference numeral 14 is a meaning interpreting device, 15 is a semantic dictionary, 16 is context information, 17 is a stored situation, 181 is a sentence to be input, 182 is voice information accompanying 181 and 19 is a semantic expression.

The operation of the sentence analyzing apparatus configured as described above will be described with reference to FIG. 1 as well as the data flow chart between the apparatus of the sentence analyzing apparatus shown in FIG. Figure 2
2 is a schematic diagram showing data processed by each device shown in FIG. 1 in order of time. In FIG. 2, 20 is a time axis and t
0 to t6 represent time stamps, 21 represents characters input to the input device in chronological order, 22 represents morphemes generated by the morphological analyzer, and 23 represents semantic analyzer. Semantic expression sub-items to be generated are described in chronological order, and 24 is a semantic expression generated by the meaning interpretation device in chronological order. The downward black arrows in FIG. 2 represent the transfer of data between the devices. Each arrow indicates that at the time the arrow is written, the data at the starting point of the arrow is passed from the module at the starting point of the arrow to the module at the ending point of the arrow.

FIG. 3 shows an example of means for eliminating ambiguity that occurs during morphological analysis. First, when the character string “Tanaka Sanwa Shitai” is input to the input device 11 in the time sequence as shown by 21, the character is transferred to the morphological analysis device 121 every time one character is input. The morphological analysis apparatus 121 divides the input character string into vocabulary while referring to the vocabulary data included in the vocabulary dictionary 122. Here, the word "Tanaka" is obtained from the character string "Tanaka" and passed to the semantic analysis device. Then, words such as "san" and "wa" are successively obtained and passed to the semantic analysis device.

The ambiguity regarding word segmentation that occurs during the morphological analysis process is resolved by two methods. One is a means using voice information, and the other is a semantic analysis device 13.
Alternatively, it is a method of returning from the meaning interpreting device 14.

When the morphological analysis of the character string "Mr. Tanaka" is performed only with character information, a plurality of analysis results are often obtained as shown in (1) and (2) of FIG. In the case of (2) in FIG. 3, the vocabulary matching the character string “nha” does not exist in the vocabulary dictionary 122 and is rejected, and the ambiguity is resolved.

If the ambiguity is not resolved inside the morphological analysis device 121, the morphological analysis device 121 performs a non-deterministic operation. That is, the morphological analyzer 121
Holds all the analysis results and sends all the analysis results to the semantic analysis device 13. The semantic analysis device 13 rejects a morphological analysis result that is not semantically consistent,
One candidate is selected from the plurality of word candidates generated by the morphological analysis, and the selection result is notified to the morphological analysis device 121. As a result of the notification, the morpheme analysis device 121 deletes the unnecessary portion of the word candidate held and stops the morpheme analysis based on the deleted word.

However, when voice information exists as an auxiliary input as shown in FIG. 3C, the ambiguity in morphological analysis may be resolved only inside the morphological analysis device 121. This utilizes the fact that an input character string can be decomposed into phrases or vocabulary by using accents, pauses, or intonations in voice information. In the case of (3) in FIG. 3, the start position and the end position of the phrase are determined by the accent information indicated by "*" and the pause information indicated by "p", and the character string "naka" between the start and end positions is determined. It turns out that "is a phrase" without using the vocabulary dictionary 122. Therefore, FIG.
The interpretation such as (2) in is rejected because it does not make a phrase.

This use of voice information can also be used to determine the end of a sentence. In FIG. 2, it is determined that the input character string “Tanaka Sawasui” is finished with the character “ru” because there is a time when nothing is input, that is, a long pause between the time stamps t5 and t6. be able to. Therefore, the morphological analysis apparatus 121 can pass the sentence end information to the semantic analysis apparatus 13 at the time of the time stamp t6.

If the intonation of the ending of the phrase "Iru" uttered last before the time stamp t6 when the sentence end is determined is in a rising tone, it can be determined that the sentence is a question sentence. The morphological analysis apparatus 121 also passes the information that the sentence is a question sentence to the semantic analysis apparatus 13 as auxiliary information.

When the semantic analysis device receives a word, the semantic dictionary 1
5 is used to convert a character string into a semantic representation subterm. The semantic dictionary 15 holds the meanings of words and phrases as a relationship between situations before and after the words and phrases are input. In the semantic analysis device of FIG. 2, the semantic expression subsections are expressed in abbreviated form. In this example, when the semantic analysis is performed on the character string “Tanaka” sent from the morphological analysis apparatus 121, the semantic analysis apparatus 13
Retrieves the meaning corresponding to "Tanaka" by searching the meaning dictionary 15 using the character string "Tanaka" as a heading. Semantic dictionary 15
In the above, "Tanaka" means that a humanoid individual whose name is Tanaka is introduced into the discourse. As a result, the semantic analysis device expresses the humanoid individual T as a semantic expression. Partial term and meaning expression meaning "name is Tanaka" Partial term << name, obj: T, name: "Tanaka">>
You can get two of them. Similarly, by referring to the semantic dictionary 15 below, it is possible to obtain the semantic expression subterms for the vocabulary passed by the morphological analysis apparatus 121.

When the meaning expression sub-items are obtained, the semantic analysis device 13 records the meaning expression sub-items in the meaning analysis device and at the same time sequentially transfers them to the meaning interpretation device 14. Further, the sentence end information passed from the morphological analyzer 121 and the auxiliary information that the sentence is a question sentence are also passed to the semantic interpreter 14 at the time of the time stamp t6.

Then, the meaning interpretation device 14 can receive the meaning expression sub-items from the meaning analysis device 13 and obtain the meaning expression 24 in consideration of the context information 16 and the situation 17.
If the final particle is present in the input sentence, the attitude of the speaker of the input sentence is interpreted based on the meaning expression sub-terms obtained from the input. Further, the ambiguity of the meaning that appears in the semantic analysis is resolved by referring to the context information 16 and the situation 17.

When voice information is given as an auxiliary input, information indicating the end of a sentence and the type of sentence can be obtained, and the meaning can be interpreted accordingly.

As described above, according to this embodiment, the input sentence can be sequentially analyzed and understood from the input portion without waiting for the end of the sentence. Further, when voice information is used as an auxiliary input, it is possible to reduce ambiguity in morphological analysis and semantic analysis and also to determine the end of a sentence.

[0024]

As described above, according to the present invention, the data transfer between the input means, the morpheme analysis means, the semantic analysis means, and the meaning interpretation means that constitute the sentence analysis device is made to be a morpheme or a semantic expression subterm. Thus, it is possible to sequentially analyze the input sentence without waiting for the end of the sentence. Furthermore, by accepting voice input such as intonation, pause, and accent as auxiliary input, it is possible to determine the end of a sentence, determine the type of a sentence such as a question sentence or an affirmative sentence, and remove the ambiguity that occurs in morphological analysis or semantic analysis operation. It becomes possible to do it.

[Brief description of drawings]

FIG. 1 is a conceptual diagram showing a configuration and a basic data flow of a sentence analysis device according to an embodiment of the present invention.

FIG. 2 is a conceptual diagram showing a data flow showing a sentence analysis operation in one embodiment of the present invention.

FIG. 3 is a diagram showing an example of disambiguation by using voice information according to an embodiment of the present invention.

[Explanation of symbols]

11 Input Device 121 Morphological Analysis Device 122 Vocabulary Dictionary 13 Semantic Analysis Device 14 Semantic Interpretation Device 15 Semantic Dictionary 16 Contextual Information 17 Situation 181 Input Sentence 182 Speech Information 19 Semantic Representation 20 Time Axis and Timestamp 21 Characters Input to Input Device 22 arranged along the time axis 22 words arranged by the morphological analyzer arranged along the time axis 23 meaning expression parts generated by the semantic analyzer arranged along the time axis 24 meaning Semantic expressions generated by the interpreter arranged along the time axis

Claims (2)

[Claims] 1. At least an input unit for inputting a character expressing a natural language sentence, a morphological analysis unit for sequentially converting an output of the input unit into a morpheme, and a situation between before and after uttering the meaning of a word. A semantic dictionary described as a partial relationship, and a semantic analysis unit that sequentially performs semantic analysis on the meanings of the morphemes output by the morphological analysis unit based on the semantic dictionary to sequentially generate semantic expression subterms corresponding to the meanings of the morphemes. And a meaning interpreting means for interpreting the meaning expression subterms based on context and knowledge outside the language. 2. The input means inputs characters expressing a natural language sentence and voice information as auxiliary inputs, the morphological analysis means changes the natural language sentence into morphemes based on the voice information, and the semantic analysis means based on the voice information. The sentence analysis apparatus according to claim 1, wherein the end of one sentence is determined and the output of the morphological analysis unit is sequentially converted into a semantic expression.