JPH0682376B2 - Emotion information extraction device - Google Patents

Description

TECHNICAL FIELD The present invention relates to an apparatus and system for analyzing a Japanese sentence and extracting emotional information included in the Japanese sentence.

(Prior Art) Regarding the technology for analyzing Japanese sentences, morphological analysis means,
A technique of generating a semantic structure of a sentence by a syntactic analysis unit and a semantic analysis unit is known.

For example, in the sentence "Hana ga Sata.", By morphological analysis, "Hana" is a noun, "ga" is a case particle, "Sakui" is a continuous verb of "Saku", and "ta" is a perfect auxiliary verb. It is divided into word strings that have part-of-speech information, such as the end form, and by syntactic analysis, "hana" is the noun phrase with the nominative phrase, "hanaga" is the main part,
Parsing is performed such that "Sakuta" is a sentence in the predicate verb phrase, "Hana ga Saki". Then, the semantic analysis is performed such that the phenomenon of "blooming" has occurred.

(Problems to be solved by the invention) In the dialog processing between the computer and the user, in addition to the processing by the logical information that has been conventionally performed, the user's state such as ability and psychological state of the user The processing adapted to is required. However, there is no known technique for extracting the emotional information of the user even after the analysis up to the semantic analysis.

For example, the expression "meaning that a flower has bloomed" means "the flower has bloomed.", "The flower has bloomed.", "The flower has bloomed." There are various expressions that reflect the situation of the speaker, the gender of the speaker, the relationship with the other party, and the attitude.

In conventional parsing and semantic analysis, polite words and final particles such as "masu, yo, wayo" in the above example have been ignored if they are auxiliary words that have nothing to do with the meaning of the sentence. Even when using the personal pronoun "I, me, you, you", etc., it is only possible to obtain information such as whether it is the first person or the second person from the meaning analysis, and it is not possible to handle why it was used as such. It was

However, if it is possible to obtain the information about the kindness and attitude contained in the Japanese sentence, it is possible to improve the quality of the translated sentence in the machine translation and the creation of the response sentence in the dialogue processing.

An object of the present invention is to provide a method and apparatus for analyzing a Japanese sentence and extracting emotional information included in the Japanese sentence.

(Means for Solving Problems) The emotion information extraction device of the present invention quantifies and registers emotion information in word units, and morphological analysis means for dividing an input Japanese sentence into word strings having part-of-speech information. It is characterized by comprising a word dictionary and means for extracting emotion information contained in the word string by referring to the word dictionary.

(Operation) By dividing the input Japanese sentence into word strings by the morpheme analysis means, it becomes possible to refer to the dictionary information in word units. Quantified emotion information is registered in words in the word dictionary. For each word of the word string obtained by the morphological analysis, by referring to the word dictionary,
The digitized emotion information registered in the word is taken out. The emotion information of the entire input sentence can be extracted as the emotion information of the input Japanese sentence by combining the emotion information extracted from each word.

(Example) Hereinafter, the detail of this invention is demonstrated according to an Example.

FIG. 1 is a diagram showing a configuration example of the present invention. In the figure, 1 is a Japanese sentence input device, 2 is a morphological analysis unit, 3 is a word dictionary in which emotion information is registered, 4 is an emotion information extraction unit, and 5 is an output display unit. The morphological analysis unit 2 divides the Japanese sentence input from the Japanese sentence input device 1 into word strings by referring to the word dictionary 3. The emotion information extraction unit 4 refers to the word dictionary 3 for each word of the word string obtained from the morpheme analysis unit 2 and extracts the emotion information for each word to obtain the emotion information of each word forming the entire sentence. The result obtained is output.

FIG. 2 is a diagram showing an embodiment of a word dictionary in the case of treating three kinds of emotion information, such as politeness, familiarity, and sex. In the areas of politeness, familiarity, and gender,
Numerical values indicating the degree of kindness, familiarity, and sex of each word are stored. However, the emotion information is not registered in all words, but in words corresponding to the kind of emotion such as politeness. As a value representing the degree of kindness, for example, the kindness is divided into five stages, and the kindness (+2), the kindness (+1),
Normal (0), Zozai (-1), and Rough (-2).
Similarly, the degree of familiarity is a numerical value indicating the degree of familiarity. The gender is also a numerical value indicating the degree of gender, such as whether it is a male-only word, a slightly masculine word, or a female-only word.

Next, an example of emotion information extraction will be described. For example, if the sentence "Hana has bloomed." Is input, it is divided into word units such as "Hana, G. Blossom, Masashi, Ta" by morphological analysis. The emotion information extraction unit extracts the value registered in each word for each of politeness, familiarity, and sex, and takes the sum. Here, the value is registered only for the auxiliary verb "masshi"("masunoun"), and nothing is registered for other words, so from this sentence, the politeness is +1 (somewhat good), the parent The result is 0 (normal) and unknown gender.
For the sentence "Flowers have bloomed.", You can get the results of politeness 0, friendship +2 (friendly), and gender -2 (female).

However, the emotion information extraction unit includes the following exceptional processing function. The gender when the final particle "yo, ne" appears is determined by the part of speech of the immediately preceding word. For example, when a noun or adjective verb comes immediately before, such as "Genkiyo", it is a female expression, and when an adjective comes, "Iyoyo", it is a slightly masculine expression. For sentences that end with the verb imperative, the politeness is “rough”, the familiarity is “not very familiar”, and the gender is “male”.

In the present embodiment, the method of taking the sum of emotion information has been described as a specific example of the synthesis of emotion information, but as a method of other synthesis, a method of obtaining a simple average, a method of obtaining a weighted average, etc. are also applicable. .

Next, an embodiment of an apparatus that specifically realizes this method will be described with reference to the drawings. FIG. 3 is a diagram showing an apparatus according to one embodiment of the present invention. Input keyboard 31, display device 32, CP
The U33, the main storage device 34, and the auxiliary storage device 35 are connected via a bus 36. The auxiliary storage device 35 stores a word dictionary. A morphological analysis program and an emotion information extraction program are placed in the main memory 34, and a work area necessary for analyzing an input sentence and a dictionary buffer area for fetching necessary data from a word dictionary are provided. There is. The processing program on the main memory is executed by the CPU 33. The keyboard 31 is used for inputting instructions and Japanese sentences to the processing program. The display device 32 is used to display the input from the keyboard and the output of the processing result.

FIG. 4 is a flow chart showing the operation of the processing program. At input 41, one Japanese sentence is input. This is divided into word strings having part-of-speech information by morphological analysis 42. The details of the morphological analysis 42 are known and will not be described. Japanese sentences that have undergone morphological analysis are carefully extracted,
It is passed to three types of processing routines of familiarity extraction and gender extraction. The fact that it is branched into three at the point 43 in the figure means that the three types of processing are executed independently of each other in order.

The processing from 411 to 413 is the part where the politeness is extracted. First, in 411, it is checked whether the sentence end is a verb imperative form, and in that case, a value indicating "rough" is given to the output unit 413 as a politeness. Otherwise, at 412, the value for each word that has been registered in the word dictionary is fetched, and the sum of the values is obtained and passed to the output unit 413. The generosity for each word can be obtained by referring to the word dictionary shown in FIG. 2 and extracting the value stored in the generosity region. The output unit 413 outputs, to the display device, a word indicating the degree of generosity corresponding to the obtained numerical value.

The processing from 421 to 423 is a portion for extracting familiarity. Similarly to the politeness, if the end of a sentence is a verb command, "not very familiar" is output. If not, the sum of familiarity values for each word is calculated and the word corresponding to the sum value is output. .

431 to 435 are the parts where sex is extracted.
First, in 431, it is checked whether the end of the sentence is a verb imperative, and if so, “male” is output. Otherwise 43
In step 2, the gender value for each word is retrieved from the dictionary and added.
Further, in 433, it is checked whether the end of the sentence is the final particle "yo" or "ne", and if so, the value indicating the gender is determined by the part of speech of the word immediately before the final particle in 434, and this is determined. 43
Add to the result obtained in 2. Finally, at 435, the obtained result is converted into a word corresponding to the result and output.

In the description of the present embodiment, it is described that a Japanese sentence is input from the input unit keyboard. However, for example, a Japanese sentence previously input to an external storage device or another input means such as a communication line or a voice recognition device is used. Similar processing can be performed on the input Japanese sentence. Moreover, although the output is described as being displayed on the display device, the output destination can be passed to another processing unit that performs processes such as voice synthesis and machine translation instead of the display device.

(Effects of the Invention) By extracting emotion information of Japanese sentences, it is possible to generate a response adapted to the emotion of the interlocutor in dialogue processing in natural language, and to improve the quality of translation in machine translation. In a Japanese sentence utterance device such as a reader, it is possible to improve the quality of the generated voice such as how to put intonation.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of the present invention, FIG. 2 is a diagram showing an example of a word dictionary in the present invention, FIG. 3 is a diagram showing a configuration of one embodiment of the present invention, and FIG. It is a flowchart which shows a processing algorithm. In the figure, 1 ... Japanese sentence input device, 2 ... Morphological analysis unit, 3 ... Word dictionary, 4 ... Emotion information extraction unit, 5 ... Output display unit are shown.

Claims (1)

[Claims] 1. A morphological analysis unit that divides an input Japanese sentence into a word string having part-of-speech information, a word dictionary in which emotional information in word units is quantified and registered, and the word string is referred to with reference to the word dictionary. An emotion information extraction device comprising means for extracting emotion information contained in.