Apparatus and method for automated natural language translation

• US 5,528,491 A
• Filed: 08/31/1992
• Issued: 06/18/1996
• Est. Priority Date: 08/31/1992
• Status: Expired due to Term

• Current Assignee: Cantena Service Agent Corporation
• Original Assignee: Language Engineering Corp.
• Inventors: Kuno, Susumu, Wright, Barton D.
• Primary Examiner(s): Hayes, Gail O.
• Assistant Examiner(s): Poinvil, Frantzy
• Application Number:
  US07/938,413
• Time in Patent Office: 1,387 Days
• Field of Search: 369/419.02, 369/419.03, 369/419.05, 369/419.08
• US Class Current: 704/9
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06F 40/55 Rule-based translation

Abstract

Method and apparatus for automated natural language translation. Source language textual statements are stored in computer memory, and interpretations of source language textual statements are stored in a graph. An editing interface allows for the display of altered translations based on user input, and translation operations include semantic propagation based on stored grammar rules.

Claims

1. Apparatus for translating textual information expressed in a source natural language into textual information expressed in a target natural language, comprising:

a textual input interface for receiving and storing the textual information in the source natural language;

a translation engine, comprising;

a preparser for resolving ambiguities and determining the end of a sentence, comprising;

means for dividing the input text into words, and

means for associating the words with one or more dictionary entries and a sequence of words with one or more multiword dictionary entries, the means for associating including one or more dictionaries;

a parser comprising a grammar rules data base for computing possible syntactic interpretations of a sentence by application of grammar rules to applicable sentence constituents, the syntactic possibilities being stored in a chart;

a graph making engine for expanding the chart into a graph which includes semantic interpretations of the sentence, the graph comprising;

one or more nodes, wherein one node is assigned to each different combined syntactic and semantic interpretation for a particular substring of the sentence, each node'"'"'s information being generated from a subnode(s) from which the node was formed and the grammar rule which was used to form the node; and

one or more pointers connecting a higher level node to its subnode, wherein a node without a subnode corresponds to a dictionary entry;

a storage means accessible to the graph making engine for storing each node and its semantic and syntactic information;

means for propagating the semantic information associated with each node upward through the graph;

an expert evaluator comprising a plurality of experts for evaluating the graph of stored semantic, syntactic, and other information for each node, wherein each expert assigns a numerical weight (N) to the node;

a scoring module for evaluating and combining the numerical weight (N) assigned each node by the experts, the scoring module storing the best combined numerical weight (N) scores for each node and also using the list of scores from the subnodes of the node to compute the node'"'"'s score list;

a parse extractor for creating a parse tree structure from any best numerical weight (N) represented on the list of such weights for each node;

a structural converter having conversion tables for the target language for performing a structural conversion on the parse tree structure generated by the parse extractor to obtain a translation of the source language text to the target language, wherein the structural converter further includes components which control structural conversion based on access to one or more dictionaries or other data files;

a display for displaying the translation to the user;

a user interface permitting the user to interact with the translation engine to obtain alternative translations once the apparatus has provided the user with the best translation of the source language, the user interface allowing the user to request alternative translations for a word, the entire sentence, or any sequence of words in the sentence, and allowing the user to view retranslation of the entire sentence which includes the alternatives for the portion of the sentence that the user selected; and

an editing interface module coupled to the display and responsive to the user input device and the storage means, including a graph access module responsive to input from the user input device to alter a decision by the scoring module and to display an alternate translation on the display.

2. The apparatus of claim 1 wherein said graph maker is constructed to maintain alternate meanings of words and alternate relations between words in the graph.

3. The apparatus of claim 1 wherein the editing interface further includes a constraints facility responsive to user input from the user interface to select portions of the current preferred translation of a variety of different lengths not restricted to single dictionary entries or entire sentences to be retranslated while constraining remaining portions of the current preferred translation.

4. The apparatus of claim 3 wherein the constraints facility is constructed to allow repeated differing constrained translation operations.

5. The translation engine as defined in claim 1 wherein the expert evaluator comprises:

a module for evaluating the average relative frequency of particular parts of speech of words in the textual input;

a module for evaluating the average relative frequency of particular paraphrases or meanings of words in the textual input;

a module for evaluating the average relative frequency of the use of grammar rules;

a module for evaluating the inter-relations of phrases and clauses within the sentence based on the length and grammatical role of the clauses and phrases;

a module for measuring the semantic distance between words, for use in evaluating conjunctions;

a module for evaluating words with regard to semantic relations between them, such as selectional restrictions; and

a module for using capitalization of a word to influence which meaning of the word is used.