JPH07105213A - Linguistic processor - Google Patents

Abstract

PURPOSE:To take a linguistically semantic analysis of an input sentence of a general user which includes various expressions and to efficiently verify the meaning of the sentence. CONSTITUTION:A morpheme analyzing means 2 takes a morpheme analysis of the input sentence from an input means by referring to a vocabulary knowledge storage means 3 and outputs candidates for words and phrases that possibly appear to a meaning extracting means 9, which extracts plural concept elements appearing in the candidates by referring to a concept element knowledge storage means 6 and outputs a combination of concept elements constituting the meaning of the sentence to a sentence meaning establishment decision means 7; and the sentence meaning establishment decision means 7 interprets the meaning by deciding whether or not the combination of concept elements inputted from the concept element extracting means 4 is proper to express the meaning of the whole sentence.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a language processing apparatus for analyzing the meaning content of an input sentence by performing morphological analysis and semantic analysis on a sentence input in natural language.

[0002]

2. Description of the Related Art With the aim of providing an interface that is natural and easy for general users to use, it is highly desired to realize a dialogue system for operating a computer using natural language such as voice. Conventionally, morphological analysis, syntactic analysis, and semantic analysis have been used as typical language processing techniques used to interpret the meaning of an input sentence on a computer. Among them, especially in a system in which a user and a computer interactively work, a semantic analysis technique for the user to grasp what the computer wants to do plays a central role.

Regarding the semantic analysis technology, the Institute of Electronics, Information and Communication Engineers, "Japanese Information Processing" 3rd Edition (May 30, 1987) p.141-16
Among those described in 6, the semantic analysis using the case frame format that expresses the co-occurrence relation between verbs, noun phrases, prepositional phrases, etc. is generally widely used.

FIG. 7 is a block diagram showing a typical process of semantic analysis using this case frame. The input unit 1 converts the information that the user wants to input into the computer into a sentence expressed in characters, and outputs the result to the morphological analysis unit 2. The morphological analysis means 2 uses word information such as headwords and parts of speech of words stored in the word knowledge storage means 3 and grammatical rules that define units to be extracted such as words and phrases. Based on the morphological analysis of the input sentence,
The sequence of morphemes such as the extracted words and clauses is output to the meaning extraction means 9. The meaning extraction unit 9 refers to the semantic feature knowledge storage unit 5 that stores the correspondence between the morpheme and the semantic concept of the morpheme called the semantic feature, and refers the semantic feature to the morpheme input by the morpheme analysis unit 2. Correspondingly, using the case frame which is the semantic expression of the entire sentence stored in the case frame knowledge storage means 10, the semantic expression of the sentence is extracted from the morpheme with the semantic feature.

Next, the basic operation of the meaning extraction means, which is the main processing of the semantic analysis, will be described. In this method, the case frame is used as a pattern expressing the meaning of the whole sentence. Multiple case frames are stored for each verb usage,
It is composed of multiple slots that represent case elements. A slot is a part in which a noun phrase or the like that constitutes a sentence is embedded, and is prepared for each case element of a verb such as a nominative case, a subject case, and a place case. A noun phrase embedded in a slot is called a filler. As the filler, a short unit such as a word or a phrase is usually used, and a semantic dictionary storing a correspondence relationship between a word and a semantic feature is used to give each filler a semantic feature. Further, in each slot, a semantic concept allowed in the filler embedded therein, a preposition, and the like are stored, and the semantic features of the filler and the semantic compatibility with the preposition are verified.

Based on the above-mentioned semantic expression, morphemes (fillers) such as words and clauses extracted from the input sentence are
While being verified by the semantic constraints of the slots in the case frame, they are embedded in the slots that are allowed semantically. In this way, all the fillers that can form a sentence have a case frame that is semantically matched, and the meaning of the entire sentence is extracted.

[0007]

However, the above-mentioned semantic analysis method has the following problems. (1) Sentences input by general users include various phrases, and it is considered that there are a huge number of types of sentence patterns. Therefore, even if a large number of case frames expressing the meaning of the entire sentence are prepared for each sentence pattern, it may not be possible to sufficiently cover a wide variety of sentence patterns and analysis may not be possible. In other words, it was difficult to obtain verbal robustness.

(2) When a word or a phrase is directly treated as a filler, which is a unit of meaning understanding, a large number of hypotheses of meaning interpretation are generated due to the semantic ambiguity, and the meaning verification is not efficient. It was

The present invention solves the above problems, and linguistically robustly analyzes the meaning of an input sentence of a general user including various phrases and efficiently verifies the meaning of the sentence. The present invention provides a language processing device that can be performed in a specific manner.

[0010]

In order to solve the above problems, the present invention performs morphological analysis on a sentence input in natural language to extract a bunsetsu candidate sequence, and extracts the bunsetsu candidate sequence from the bunsetsu candidate sequence. Then, by applying a conceptual element frame with a basic concept (conceptual element) as an intermediate understanding unit by the conceptual element extraction means, a partial semantic group of sentences is extracted as a conceptual element, and the extracted plural elements are extracted. Whether the combination of conceptual elements of is suitable for expressing the meaning of the whole sentence is judged by the sentence meaning judgment means by using the knowledge about the target area and the context knowledge, and the semantic expression of the input sentence is obtained. , Is a two-stage meaning extraction method.

[0011]

According to the present invention, a concept element in which the meaning of the entire sentence having a wide variety of patterns input by a general user is not a combination of words or phrases but a semantic unit of a section longer than the phrase Is expressed as a combination of
It works to improve the expression efficiency of sentence patterns.

Further, as a target for directly applying top-down knowledge or context knowledge about the target area stored by the sentence intention determination knowledge storage means, the concept pattern is relatively easy to predict rather than the sentence pattern. To be
It makes it easier to design robust language processing.

Further, since the process of determining the establishment of the meaning of a sentence directly targets not the bunsetsu, but the conceptual element, an efficient process of semantic verification can be performed unlike the case where a word or a bunsetsu is directly treated as a filler. Like

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a functional block diagram showing an embodiment of the present invention. In FIG. 1, the input means 1 converts the information that the user wants to input into the computer into a sentence expressed in characters,
The result is output to the morphological analysis unit 2. When the user inputs a sentence to the computer using the keyboard, the input unit 1 is a keyboard, and when the user inputs a sentence by voice, the input unit 1 is a voice recognizer.

The morpheme analysis means 2 defines word information stored in the word knowledge storage means 3, such as headwords and part-of-speech of words, and word and word connections to define clauses. Morphological analysis is performed on the input sentence based on grammatical rules, all possible phrase clauses are extracted from the input sentence, and a set of these phrase candidates is output to the concept element extracting unit 4 in the meaning extracting unit 9. Here, what is extracted by the morpheme analysis unit 2 does not have to be a phrase, but may be a partial unit of a sentence.

The concept element extraction unit 4 first refers to the semantic feature knowledge storage unit 5 in which the correspondence relationship between the independent word and the semantic feature is stored and refers to each phrase candidate input from the morpheme analysis unit 2. On the other hand, it gives a semantic feature. Next, the concept element extraction unit 4 selects, from among the phrase candidates input from the morpheme analysis unit 2, a combination that can be connected in a temporal order as a sentence candidate, and the concept element knowledge about each selected phrase candidate. While verifying the semantic compatibility by referring to the storage unit 6, the concept unit frame is embedded in the slot. Finally, the concept element extraction unit 4 extracts all the concept element frames having the slots in which the phrase candidates are embedded, and determines the combination of the concept elements extracted from the combination of the phrase candidates forming the sentence candidates to determine the sentence meaning. Output to the means 7.

The sentence meaning determination means 7 refers to the sentence meaning determination knowledge storage means 8 and the combination of the concept elements input from the concept element extraction means 4 is suitable for expressing the meaning of the entire sentence. Determine if there is. Sentence determination knowledge storage means 8
The knowledge stored in is knowledge about the target area, contextual knowledge, and the like, and other knowledge sources that are effective in determining the meaning of the sentence are included here.

In the above-mentioned configuration, the operation will be described below by taking an input sentence "Tell me a clean pool hotel near the Shonan coast" as an example. The morphological analysis means 2 has "Shonan Kaigan", "Near", "Beautiful", "Pool", "Aru", "Hotel", and "Tell me" as bunsetsu candidates that can appear in the input sentence.
And so on. In addition to these, homophonetic meaning words and bunsetsu candidates (for example, “stairs” for “coast”) due to recognition errors in the case of voice input are also extracted. For these bunsetsu candidates, part-of-speech information such as "Shonan Kaigan (noun), no (case particle)" and information on the temporal position of the bunsetsu candidate in the input sentence are also extracted.

The set of these bunsetsu candidates is input to the concept element extraction means 4, and the correspondence information between the independent word information in each bunsetsu candidate and the independent words and semantic features stored in the semantic feature knowledge storage means 5. The information gives semantic features as shown in FIG. This semantic feature is for verifying the semantic constraints stored in each slot of the conceptual element frame, and whether a certain clause candidate can semantically satisfy a certain slot of the conceptual element frame. Is used to determine

Next, as shown in FIG. 2, the conceptual element frame is applied to the combination of the phrase candidates that can be connected in the temporal order as the sentence candidates, and the phrase candidates are grouped into conceptual element units. A concrete definition of a conceptual element is stored as a conceptual element frame as shown in FIG. The example of FIG. 3 (1) is a conceptual element frame that stores a concept expressing the perspective of a place such as “close to a station” or “away from a downtown area” as a basic concept that can appear in a sentence. In addition, the example of FIG. 3 (2) shows a conceptual element frame expressing equipment such as "there is a pool" and "there is a parking lot". In each slot, a semantic feature and a particle are stored as constraints that the filler must satisfy. Each concept element stores a slot that can sufficiently cover a phrase for expressing only that concept element.

Since the unit of understanding the meaning of a sentence is not a case frame expressing the whole sentence but a partial conceptual element of the sentence, it is not necessary to memorize all the expressions of the sentence,
The wording need only be stored in units of each conceptual element. Therefore, it is clear that the expression efficiency of wording is higher than that of the case frame in which the entire sentence is processed. The description of these conceptual element frames is stored in the conceptual element knowledge storage unit 6.

FIG. 4 shows an example in which the phrase candidates shown in FIG. 2 are collected in units of conceptual elements in this way. In this example, the concept element
"Shonan Kaigan" and "near" are summarized as expressions that represent (the perspective of a place), and similarly, "clean", "pool", and "is" as the expressions that represent conceptual elements (equipment). , "Hotel" as an expression that represents a conceptual element (accommodation), and "Tell me" as an expression that represents a conceptual element (intention)
But each is summarized. The meaning of the whole sentence is expressed by the combination of these conceptual elements.

Therefore, when considering the semantic pattern of a sentence, the temporal order of appearance of conceptual elements is arbitrary, and addition of another conceptual element is also optional. Needless to say, it is possible to robustly analyze the meaning of input sentences of general users including.

Next, with respect to the combination of the concept elements input from the concept element extracting means 4, the sentence meaning determination means 7 determines whether or not the combination of the concept elements is appropriate to establish the meaning of the whole sentence. Is determined by referring to the sentence meaning determination knowledge storage means 8. Hereinafter, three types of methods for determining the meaning of a sentence performed in this embodiment will be described.

The first sentence intention determination method is a method of utilizing the user's intention included in the input sentence as top-down knowledge about the concept element extracted from the concept element extraction means 4. The intention expresses what the user wants the system to do in the question answering system, and specifically refers to the type in which the question sentence is classified. Taking the example of booking a hotel, the question text is
Can be categorized as cancellation, WH question, YN question, search, etc. The type of these question sentences can be determined by the information of the phrase candidates in the concept element frame extracted from the concept element extraction unit 4, and the existence condition of the concept element in each intention can be set.

FIG. 5 shows a semantic structure when reserving a hotel. "Hotel reservation" in the figure indicates the intention,
"Hotel name", "date and time", "room", "number of people", etc.
This section shows the concept element as essential information for hotel reservation. Other conceptual elements indicate the attributes of the hotel. By referring to such a hierarchical semantic structure, the intention is inferred from the information of the clause candidates in the concept element.

Specifically, as shown in FIG. 6, predicates (reserved, changed, stopped, searched, etc.) that are deeply related to each intention are used as keywords for utterance intention extraction, and the predicates are omitted. In this case, (1) Reservation essential information for hotel reservation (hotel name, date and time, room type,
The intention is inferred based on the presence / absence of the number and the number of persons, and (2) the presence / absence of attribute information of the hotel hierarchically expressed by the concept element.

Next, in the second method of determining the meaning of a sentence, with respect to the concept elements extracted from the concept element extracting means 4, a slot having a central semantic role of each concept element is defined as an essential slot, This is a method of preventing a conceptual element frame consisting of only other slots from being established independently. This method eliminates semantic elements that are semantically wrong when they are established by a single slot.

The third method of determining the meaning of a sentence is a method of assuming that all phrase candidates are temporally connected in each conceptual element. This is equivalent to extending each conceptual element temporally while adapting semantically to the input sentence,
A combination of conceptual elements in which the same conceptual element appears in different positions in the input sentence is excluded.

In the sentence meaning determination means 7 and the sentence determination knowledge storage means 8, not only the three types of sentence determination methods described above but also other knowledge such as context knowledge and common sense are widely applied. You can

[0031]

As described above, according to the present invention, as compared with the conventional semantic analysis method using the case frame based on the usage of verbs, the unit of meaning understanding of a sentence is Instead, by using a partial conceptual element of the sentence,
It is not necessary to store all phrases as sentences, and phrases may be stored only in units of each conceptual element. Therefore,
This has the effect that the expression efficiency of wording is higher than in case frames in which the entire sentence is processed. This effect becomes more remarkable as the number of wording types of the target sentence increases.

Further, with the effect that the expression efficiency of the wording is high, the process of determining the establishment of the meaning of the sentence directly targets not the bunsetsu, but the conceptual element, so that a word or a bunsetsu is directly treated as a filler. Unlike the above, there is also an effect that efficient semantic verification processing can be performed.

On the other hand, when considering the semantic pattern of a sentence, the temporal order of appearance of conceptual elements is arbitrary, and it is also arbitrary that another conceptual element is added. It has the effect of semantically analyzing the input sentence of a general user including verbally.

Further, the definition of the conceptual element is more general than the definition of the sentence pattern with respect to the change of the target area, and when changing the target area of the processing, it is easy to reconstruct the basic concept element. Has the effect of expressing a new target area.

[Brief description of drawings]

FIG. 1 is a functional block configuration diagram of a language processing device showing an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an example of adding semantic feature information according to an embodiment of the present invention.

FIG. 3 is an explanatory diagram showing an example of a conceptual element frame according to an embodiment of the present invention.

FIG. 4 is an explanatory diagram showing an example of conceptual element extraction by a conceptual element frame according to an embodiment of the present invention.

FIG. 5 is an explanatory diagram showing a semantic structure of a target area in one embodiment of the present invention.

FIG. 6 is an explanatory diagram showing an example of intention extraction knowledge as sentence determination knowledge in one embodiment of the present invention.

FIG. 7 is a functional block configuration diagram of a semantic analysis method using a conventional case frame.

[Explanation of symbols]

 1 Input Means 2 Morphological Analysis Means 3 Words Knowledge Storage Means 4 Conceptual Membrane Extraction Means 5 Semantic Feature Knowledge Storage Means 6 Conceptual Element Knowledge Storage Means 7 Sentence Determining Judgment Means 8 Sentence Determining Judgment Knowledge Storage Means 9 Meaning Extraction Means 10 Case frame knowledge storage means

Claims (1)

[Claims] 1. Input means for inputting a sentence expressing a natural language, morphological analysis means for performing morphological analysis on a sentence input by the input means, and outputting candidates such as words and phrases, and words and phrases. Semantic feature knowledge storage means for storing semantic feature information for, and semantic relations between words and clauses are stored, expressed as a frame structure as a partial semantic unit of a sentence, and stored as a concept element. By referring to the concept element knowledge storage means and the semantic feature knowledge storage means, a semantic feature is given to each candidate output by the morpheme analysis means, and the concept element knowledge storage is performed for a candidate group forming a sentence part. A concept element extraction means for extracting a concept element by referring to a means, and a sentence intention that stores a condition for a combination of concept elements for recognizing a semantic expression of the whole sentence as a concept expression combination With respect to the combination of the vertical determination knowledge storage means and the conceptual element extracted by the conceptual element extraction means, a combination of conceptual elements that meets the condition for the combination of conceptual elements stored in the sentence determination success knowledge storage means is output. A language processing device, comprising: