JPS63219067A - Dictionary retrieval device - Google Patents

Abstract

PURPOSE:To obtain an efficient dictionary retrieval device by fetching from a storage means a retrieval order and a retrieval range which a retrieval control means has extracted from a topic extraction means and which corresponds to the change of a topic and using them for the retrieval of dictionaries. CONSTITUTION:If an inputted sentence is analyzed in a natural language analysis device 1, the device 1 transmits information 100 to the topic extraction device 2 for extracting the topic. The device 2 extracts a topic 101 (is set to a subject C) in accordance with the function of a system based on the information. If the topic C which the dictionary retrieval device uses for deciding the retrieval order and the retrieval range differs with the subject 101, the topic 101 is transmitted from the device 2 to the retrieval control device 4. Said device 4 sets the topic 101, namely, the topic C as the new subject, and simultaneously fetches the retrieval order and the retrieval range with respect to the topic 101, namely the order of dictionaries D2 and D4 from a storage device 3. The retrieval order and the retrieval range is used hereafter in the retrieval of the dictionaries.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to an apparatus for searching a dictionary used in an apparatus for analyzing natural language documents or dialogues in which topics change dynamically.

BACKGROUND OF THE INVENTION Natural language documents and dialogues are composed of many topics, and the topics dynamically change according to the flow of the document and dialogue. Natural language documents and conversations use many words,
Natural language analysis devices extract information for those words by searching dictionaries, but in conventional natural language analysis devices, the dictionary search process was independent of topic trends, so The change did not change the search order or search range of the dictionary search.

One way to change the order of dictionary searches as a sentence is analyzed is to give top priority to words with the most recent ratios, as is currently widely used in word processor dictionaries. However, with this method, the search order is changed only for that word, and the effect of improving search efficiency is limited to only frequently occurring words.

The problem to be solved by the invention The dictionary contains all or almost all the words that appear in the document to be analyzed. The frequency of use of many words varies depending on the topic, and some words are rarely used depending on the topic. However, in conventional dictionary searches, dictionary searches are performed independently of the topic of the target document or conversation and changes in the topic, and therefore cannot respond to changes in the frequency of word usage due to such topics.

Furthermore, a dictionary may have a plurality of different pieces of information as options for one heading. For example, this is the case when one word has multiple meanings. When selecting one of these multiple options using natural language analysis, it is often possible to narrow down the options by topic, but with traditional dictionary searches for independent topics, it is not possible to narrow down the options at the time of search. There is a high possibility that the wrong information will be selected.

The present invention aims to solve the above-mentioned problems and realize an efficient dictionary search.

Means for Solving the Problems The dictionary search device of the present invention includes topic extraction means for extracting the topics of documents or dialogues, and a memory that stores headwords for each topic and the search order and search range of dictionary information for the headwords. and a search management means for managing the search order, search range, and dictionary used for the search, and at the same time as the topic extraction means extracts a change in topic, the search management means responds to the topic. The search order and search range obtained are retrieved from the storage means, and the search order and search range are used for subsequent dictionary searches.

Effect: The dictionary search order and dictionary search range stored in the storage means correspond to the frequency of appearance of individual words in each topic, and the dictionary search order and dictionary search range stored in the storage means correspond to the frequency of appearance of individual words in each topic. It is set to preferentially search for the most appropriate dictionary information for the topic, and not to search for information that is unlikely to be used in the topic.

Further, the preamble topic extraction means extracts the topic of the target document or dialogue, and when the topic changes, the search management means searches the storage means for a dictionary search order and dictionary search range corresponding to the new topic, A dictionary search is performed using the search order and search range.

Example First, the present invention will be explained in detail using FIG.

In the embodiment, a topic extraction device 2 receives information from a natural language analysis device 1 and extracts a topic of a sentence to be analyzed from that information, and a topic extraction device 2 extracts a topic of a sentence to be analyzed from the information, and a dictionary search order and search range for each topic is divided into topics and search order and search range. A storage device 3 that stores data in the form of a correspondence table, and a dictionary search order and dictionary corresponding to the topic of the sentence being analyzed.
A dictionary search device consisting of a search management device 4 that retrieves a dictionary search range and performs an actual search based on the search order and search range, a natural language analysis device 1 connected to the dictionary search device, and a dictionary D1 used for analysis. It consists of ~5.

In this embodiment, the dictionaries D1 to D5 are divided from the beginning, and the storage device 3 stores the dictionaries used for each topic and their search priorities. In the example, for topic A, Dl, D2, D
3. Search the dictionary in the order of D4, and for topic B, search for D2, D4,
The dictionary is searched in the order of Dl, D3, etc.

Words are sorted into different dictionaries according to their frequency of use in each topic, and searches can be performed efficiently by switching the search order and search range depending on the topic.

Next, an example of actual operation will be explained with reference to FIG.

FIG. 1(a) shows the initial state before starting analysis. In this state, no topic has been selected yet. At this time, in the dictionary search when analyzing the input sentence, the search order of the dictionary and
For the search range, set an appropriate initial value (for example, if the target sentence does not belong to any topic, the search order and search range may be stored in the storage device 3 and used.
If the initial topic is fixed, the search order and search range for that topic may be set in advance.

In the example shown in FIG. 1, the initial values are the search order and search range for topic A.

In FIG. 1(b), when the input sentence is analyzed by the natural language analysis device 1, the natural language analysis device 1 sends information 100 for topic extraction to the topic extraction device 2, and based on that information, the topic extraction device extracts the topic. Extract. There are various ways to extract topics depending on the system's functionality, such as simple methods such as keyword extraction, methods that use the syntactic structure of the input sentence as clues, and methods that extract topics by inference from the meaning of the input sentence. is possible. Furthermore, if the natural language analysis device 1 is part of a system-driven dialogue system in which the system actively switches topics, the topic can be obtained directly from the system. The information 100 sent to the topic extraction device 2 differs depending on each method.

If the topic 101 is extracted in FIG. 1(c) and the topic 101 is different from the topic used by the dictionary search device to determine the search order and search range, the topic 101 is retrieved from the topic extraction device 2. The information is sent to the management device 4. In this example, topic 101 is topic C.

In FIG. 1(d), the search management device 4 sets a topic 101 as a new topic and at the same time retrieves the search order and search range 102 for the topic 101 from the storage device 3.

In this example, the search order/search range is the order of dictionary D2 and dictionary D4.

In FIG. 1(e), the search management device 4 obtains a search order/search range 102 for a topic 101, and uses this search order/search range in subsequent dictionary searches.

Switching of the dictionary search order and search range based on topics is performed in the process described above, and this process is repeated when the target document or conversation topic changes.

Next, a method for specifying the search order and search range outside the sides of FIG. 1 will be described.

In the system shown in FIG. 2, dictionaries are completely separated for each topic, and each topic has its own dictionary. In this case, one word may exist in two or more dictionaries.

The operation when the topic changes is the same as the example shown in FIG.

In FIG. 3, the dictionary search order and search range are changed by changing the dictionary search method rather than dividing the dictionary. In the example shown in FIG.
The hash functions and hash tables used are changed according to changes in topics. Switching the search method and dividing the dictionary may be combined.

FIG. 4 is an example in which information about the topic is also included in the dictionary description. The case of searching for the word ``su'' using the example shown in FIG. 4 will be explained. The dictionary 5 includes information related to topics in addition to heading/dictionary information. The information contained in the storage device 3 is also the same topic information. During a dictionary search, only dictionary information whose topic information corresponds to the current topic matches the topic information is retrieved. Dictionary 5 has three headings for each word.

The dictionary search device 4 searches for a dictionary item having either 4 or 8 as the topic information for the current topic from among the three dictionary items, and selects the dictionary item having 4 as the topic information. Meaning b-3, which is dictionary information of this dictionary item, is passed to the natural language analysis device.

In both the examples shown in FIGS. 3 and 4, the operation when the topic changes is the same as in the example shown in FIG.

Effects of the Invention By using the above-mentioned device, it becomes possible to perform a dictionary search that corresponds to changes in the frequency of appearance of individual words in each topic, and as a result, the efficiency of the dictionary search is improved. In addition, when searching for a word that has multiple information options, the validity of the information to be searched is determined by matching the search range and search priority for each option in the storage means to the frequency of usage in the topic. increases.

[Brief explanation of drawings]

FIGS. 1(a) to 4(e) are explanatory diagrams of the switching operation of the dictionary search order and search range in accordance with a change in topic in a dictionary search device according to an embodiment of the present invention, and FIGS. 2 to 4 are diagrams showing the search FIG. 7 is an explanatory diagram of another embodiment of the dictionary search device in which the method of specifying the order and search range is changed. 1... Natural language analysis device, 2... Topic extraction device, 3... Storage device, 4... Search management device, 5... ·dictionary. Director of the Agency of Industrial Science and Technology Yukizo Iizuka Figure 1 (Q) Figure 1 (b) Figure 1 (C) Figure 1 +ch Figure 1 te+ DI D2 υ3 [
)4 LI5 Figure 2 DI, Dz 03
D4 05Figure 3Figure 4

Claims (1)

[Claims] Topic extraction means for extracting the topic of a document or dialogue from analysis of a natural language document or dialogue, and storing for each topic the search order and search range for a dictionary headword and a plurality of pieces of information that each headword has as options. It has a storage means, and a search management means for managing the order and range of dictionary searches, and the search management means changes the dictionary search order and range stored in the storage means as the topics extracted by the topic extraction means change.
A dictionary search device characterized in that a dictionary search range is automatically applied to a dictionary search during analysis.