Boundary extracting system from a sentence

• US 5,479,563 A
• Filed: 12/28/1993
• Issued: 12/26/1995
• Est. Priority Date: 09/07/1990
• Status: Expired due to Fees

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Yamaguchi, Yukiko
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Doerrler, Michelle
• Application Number:
  US08/174,629
• Time in Patent Office: 728 Days
• Field of Search: 395/2.41-2.62, 395/22, 381/41-45
• US Class Current: 704/232
• CPC Class Codes: G10L 15/05 Word boundary detection
• CPC Class Codes: G10L 15/16 using artificial neural networks

Abstract

The present invention extracts boundaries from a sentence with no need for linguistic knowledge or complicated grammatical rules. Upon extracting a clause/phrase boundary, words are classified according to part-of-speech numbers of words which form inputted sentence information. Then, an input pattern representing part-of-speech numbers of a target word is checked to determine whether a clause/phrase boundary exists before or after the target word; a plurality of words before and after the target words is then applied to a neural network. Among units in the output layer of the neural network, a unit having the output larger than a threshold is determined to refer to a clause/phrase boundary of the target word. Upon extracting a subject-predicate boundary, words are classified in word number, and an input pattern corresponding to a plurality of words are applied to the neural network. The neural network comprises output units for a subject and a predicate, and a boundary is extracted by an inputted pattern which changes the output of these units.

Claims

1. An intra-sentence boundary extracting system for extracting a boundary in a structure of a sentence located either before or after a target word included in words forming inputted sentence data, said system comprising:

inputted word classifying means for classifying a first number of words forming the inputted sentence data to produce an input pattern indicating a second number of classification results of the words; and

boundary position data output means for receiving the classification results from said inputted word classifying means and for outputting at least one of clause and phrase boundary position data, said boundary position data output means including a neural network formed of

an input layer having a third number of units, each unit coupled to said inputted word classifying means to receive each part of the input pattern for the first number of words including the target word, n preceding words before the target word and m succeeding words after the target word, where n and m are at least one, the third number corresponding to the first number times the second number of classification results of the words;

more than one intermediate layer coupled to said input layer; and

an output layer, coupled to at least one of said more than one intermediate layer, to output the boundary position data on the structure of the sentence, corresponding to a boundary of at least one of a clause, a noun phrase, a verb phrase, a preposition phrase and an infinitive phrase either before or after the target word.

2. An intra-sentence boundary extracting system according to claim 1, further comprising:

boundary determining means for outputting a boundary determination result for the target word based on the boundary position data outputted by said boundary position data output means indicating whether at least one of a clause and a phrase boundary exists at least one of before and after the target word.

3. An intra-sentence boundary extracting system according to claim 2, wherein said boundary determining means comprises:

a comparing unit, operatively connected to said boundary position information output means, for comparing the classification results output by said inputted word classifying means with a predetermined threshold, and then outputting a comparison result as a boundary determination result for the target word.

4. An intra-sentence boundary extracting system according to claim 3, further comprising:

a boundary sign converting unit operatively connected to receive the comparison result from said comparing unit, for outputting, using the comparison result output by said comparing unit, a boundary sign for the target word; and

an output control unit, operatively connected to said boundary position data output means to receive the target word and to said boundary sign converting unit, to add the boundary sign outputted by said boundary sign converting unit to the target word, and to output sentence data with the boundary sign.

5. An intra-sentence boundary extracting system according to claim 1, wherein said inputted word classifying means comprises:

a word dictionary for storing part-of-speech numbers representing respective parts of speech; and

a part-of-speech number selecting unit for selecting and using from said word dictionary, each part-of-speech number representing one of the words forming the inputted sentence data, and then outputting the part-of-speech number as the classification results for each of the words.

6. An intra-sentence boundary extracting system according to claim 1, wherein said boundary position data output means further comprises an input pattern generating unit, operatively connected to said neural network and said inputted word classifying means, to generate in response to operation of said neural network, the input pattern using the classification results output by said inputted word classifying means.

7. An intra-sentence boundary extracting system according to claim 6, wherein said input pattern generating unit comprises:

a word sequence generating unit, operatively connected to said inputted word classifying means, for generating a word-sequence as a sequence of part-of-speech numbers of the n preceding words, the target word and the m succeeding words; and

an input control unit, operatively connected to said word sequence generating unit and said neural network, for outputting the input pattern to said neural network using the word-sequence output by said word sequence generating unit.

8. An intra-sentence boundary extracting system for extracting a boundary either before or after a word included in a sentence data, said system comprising:

inputted word number retrieving means for receiving inputted sentence data and for retrieving a word number of a target word included in the inputted sentence data;

boundary position data output means for receiving each word number retrieved by said inputted word number retrieving means and for outputting boundary position data, said boundary position data output means including a neural network formed of

an input layer having a first number of units, each unit coupled to said inputted word number retrieving means to receive each word number in an input pattern representing a second number of words including the target word, n preceding words before the target word and m succeeding words after the target word, where n and m are at least one, the first number being equal to the second number times a third number of possible word numbers;

more than one intermediate layer coupled to said input layer; and

an output layer, coupled to at least one of said more than one intermediate layer, to output the boundary position data indicating a boundary either before or after the target word depending on whether the target word is a subject or a verb; and

boundary determining means for outputting, according to the boundary position data outputted by said boundary position data output means, a boundary determination result for the target word.

9. An intra-sentence boundary extracting system according to claim 8, wherein said inputted word number retrieving means comprises:

a word dictionary for storing word numbers corresponding one-to-one to dictionary words,

a collating unit for selecting, from said word dictionary, a corresponding word number and outputting the corresponding word number as a retrieval result for each word in the inputted sentence data.

10. An intra-sentence boundary extracting system according to claim 8, wherein said boundary position data output means comprises:

a maximum value extracting unit, operatively connected to said neural network, for extracting a maximum value of the boundary position data output by the output layer of said neural network; and

an input pattern generating unit, operatively connected to said inputted word number retrieving means and said neural network, for generating, in response to operation of said neural network, the input pattern using the word numbers output by said inputted word number retrieving means.

11. An intra-sentence boundary extracting system according to claim 10, wherein said input pattern generating unit comprises:

a word sequence generating unit, operatively connected to said inputted word classifying means, for generating as a word-sequence the n preceding words, the target word and the m succeeding words; and

an input control unit, operatively connected to said word sequence generating unit and said neural network, for outputting the input pattern to said neural network using the word-sequence output by said word sequence generating unit.

12. An intra-sentence boundary extracting system according to claim 8, wherein said boundary determining means comprises:

a comparing unit, operatively connected to said boundary position data output means, for comparing the boundary position data output by said boundary position data output means with a predetermined threshold and for outputting a comparison result as the boundary determination result for the target word.

13. An intra-sentence boundary extracting system according to claim 8, wherein words represented by the sentence data are sequentially inputted to said inputted word number retrieving means, starting with one of a first and a last word represented by the sentence data.