Analyzing textual documents

• US 5,323,310 A
• Filed: 01/31/1992
• Issued: 06/21/1994
• Est. Priority Date: 02/14/1991
• Status: Expired due to Fees

• Current Assignee: The British and Foreign Bible Society
• Original Assignee: The British and Foreign Bible Society
• Inventors: Robinson, David W. C.
• Primary Examiner(s): Envall, Jr., Roy N.
• Assistant Examiner(s): Shingala, Gita
• Application Number:
  US07/830,027
• Time in Patent Office: 872 Days
• Field of Search: 364/419, 364/419.02, 364/419.05
• US Class Current: 704/2
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06F 40/58 Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation

Abstract

Two text documents, one of which is a (possibly erroneous) translation of the other are stored in the memory of a text processing system. The distributions and forms of words within the documents are compared and by a statistical process it is determined which words are likely to be correctly translated. A list of those words which seem to have been inconsistently translated is compiled.

Claims

1. A method of processing a first text document in a source language and a second text document in a target language using a computer, each of the documents being divided into segments and stored in memory means of the computer, the segments being further divided into words, the method comprising carrying out the steps of:

a) selecting a first word in the first text document and a second word in the second text document and determining a representation of the probability that the first and second words have substantially the same meaning by taking into account the result of a comparison of the distribution of the first and second words in the segments of the first text document and the second text document respectively; and

b) determining that the first and second words have substantially the same meaning when the representation of the probability that the first and second words have the same meaning is greater than a threshold.

2. A method as claimed in claim 1, wherein the first word in the first text document is regarded to have substantially the same meaning as the second word in the second text document if:

a) the representation of the probability of having substantially the same meaning as the first word in the first text document is greater for the second word in the second text document than for any other word in the second text document;

b) the ratio of the greatest representation of the probability of having substantially the same meaning as a word in the second text document to the next greatest such representation is greater for the first word in the first text document than for any other word in the first text document; and

c) that ratio is greater than one.

3. A method as claimed in claim 1 or claim 2, wherein each segment in the first text document corresponds to a single segment in the second text document and the representation of the probability that the first word in the first text document and the second word in the second test document have substantially the same meaning is determined by a process comprising determining the number of time that words identical to the first word are found in segments in the first text document which correspond to segments in the second text document containing words of the same form as the second word.

4. A method as claimed in claim 3, wherein a word is of the same form as the second word only when a word is identical to the second word.

5. A method as claimed in claim 3, wherein a word is of the same form as the second word when the word has a group of at least one or more consecutive letters in common with the second word, and the process also comprises determining the length of the group of letters.