System and method for parsing natural language by unifying lexical features of words

• US 5,083,268 A
• Filed: 08/27/1990
• Issued: 01/21/1992
• Est. Priority Date: 10/15/1986
• Status: Expired due to Term

• Current Assignee: Texas Instruments, Inc.
• Original Assignee: Texas Instruments, Inc.
• Inventors: Hemphill, Charles T., Vlach, Frank
• Primary Examiner(s): Shaw, Gareth D.
• Assistant Examiner(s): Kriess, Kevin A.
• Application Number:
  US07/574,557
• Time in Patent Office: 512 Days
• Field of Search: 364/200 MS File, 364/900 MS File, 364/513, 364/419
• US Class Current: 706/11
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: Y10S 706/902 Application using ai with detail of the ai system

Abstract

A method for parsing for natural languages includes a grammar and a lexicon. A knowledge base may be used to define elements in the lexicon. A processor receives single words input by a user and adds them to a sentence under construction. Valid next words are predicted after each received input word. The preferred system has two major components: a parser and a predictor. The predictor accesses only the lexicon and the knowledge base, if one is used, to determine the valid next input words. The parser constructs sentences which are valid according to the grammar out of words accepted by the predictor.

Claims

1. A method for recognizing a sentence having a plurality of words, performed by a computer and comprising the steps of:

(a) accepting a predefined grammar having terminal and non-terminal elements for defining allowable sentence structues;

(b) accepting a predefined lexicon having entries for defining terminal elements of the grammar in terms of linguistic or semantic features;

(c) predicting valid next lexical entries for any current partial sentences according to the grammar;

(d) reading input words;

(e) comparing features of the input words with the features of the valid next lexical entries;

(f) rejecting input words which have features which do not match valid next lexical entries;

(g) accepting input words which have features which match a valid next lexical entry;

(h) generating partial sentences by combining accepted input words consistent with the grammar and lexicon;

(i) unifying the lexical features of the words forming the partial sentences so that the partial sentences are self-consistent with respect to such features; and

(j) repeating steps (c) through (i) until a complete sentence is formed as defined in the grammar.

2. A method for recognizing a sentence according to claim 1, further comprising the steps of:

(k) before step (c), providing a knowledge base having a plurality of words and associated lexical features; and

wherein step (e) comprises the steps of;

(l) comparing the input words to those contained in the knowledge base;

(m) defining input words which are not contained in the knowledge base as not matching a valid next lexical entry; and

(n) for words contained in the knowledge base, comparing their associated lexical features with the features of the valid next lexical entries.

3. A method for recognizing a sentence according to claim 1, wherein step (h) comprises the steps of:

(o) appending each accepted input word to the end of any existing partial sentence;

(p) examining the grammar to determine a set of all valid terminal elements which can follow a terminal element represented by the accepted input word; and

(q) proposing the terminal elements determined in step (p) as valid continuations of the partial sentence.

4. A method for recognizing a sentence according to claim 3, wherein step (c) comprises the steps of:

(r) for any terminal element proposed in step (q), determining from the lexicon the features defined therefor; and

(s) determining word and partial sentence features which are set by the current partial sentence.

5. A method for recognizing a sentence according to claim 1, wherein the number of features for the words is substantially less than the number of words which make the valid next lexical entries.

6. A method for recognizing a sentence according to claim 1, wherein the features compared in step (e) are lexical features.

7. A method for recognizing a sentence according to claim 6, wherein the lexical features are taken from the set spec, lex, stem, predicate, subject, object, number, sound, and definiteness.

8. A system, having a grammar and a lexicon, for recognizing a natural language sentence having a plurality of words, comprising:

input means for reading words;

a parser comprising;

means for retrieving rules from said grammar for combining words into partial sentences,

means for generating sets of a valid next elements for the partial sentences according to said rules, and

means for determining whether the partial sentences are complete as defined by said rules; and

a predictor coupled to said input means, comprising;

means for retrieving features from said lexicon,

means for predicting the features of said valid next elements,

means for comparing words read by said input means with words in said lexicon,

means for rejecting words having features supplied by said lexicon which are not consistent with said predicted features of a valid next element, and

means for unifying lexical features of the words forming said partial sentences so that said partial sentences are self-consistent with respect to such lexical features.

9. A system according to claim 8, further comprising a knowledge base for supplying features of words, wherein said predictor is coupled to said knowledge base and wherein at least some words read by said input means are compared with words in said knowledge base, and wherein, for those words, features which must be consistent with those of a valid next element are taken from said knowledge base.

10. A system according to claim 8, further comprising means for generating chart, wherein the chart is accessed by said parser and said predictor for storing intermediate results.

11. A system according to claim 10, wherein the chart comprises edges, vertices, and pending edges of symbols to be transformed into edges.

12. A system according to claim 8, wherein the features of words and grammatical elements comprise lexical features.

13. A system according to claim 8, wherein said input means reads words serially, and wherein said parser and said predictor are invoked between each input word read by said input means.

14. A method for recognizing a sentence having a plurality of words, performed by a computer and comprising the steps of:

(a) accepting lexical features for words;

(b) predicting a set of valid next input words according to a grammar;

(c) reading an input word;

(d) for a valid input word, appending the input word to a partial sentence to generate a new partial sentence;

(e) unifying the lexical features of the words forming the new partial sentence so that the new partial sentence is self-consistent with respect to such lexical features; and

(f) repeating steps (b) through (e) until a complete sentence is formed according to the grammar.

15. A method according to claim 14, wherein step (e) comprises the step of:

(g) for all lexical features which are yet undefined by the partial sentence, and which are defined by the valid input word, defining such features for the new partial sentence to be the same as defined for the valid input word.

16. A system, having a grammar and a lexicon, for parsing a natural language sentence having a plurality of words, comprising:

input means for reading words;

means for retrieving rules from said grammar;

means for retrieving features from said lexicon;

means coupled to said input means and said feature retrieving means for instantiating the features of words read by said input means at a sentence creation time; and

means for combining the words into sentences according to said rules retrieved from said grammar.

17. A system according to claim 16, wherein said input means reads words serially, and wherein said instantiation means is invoked between each input word read by said input means.

18. A system according to claim 16, wherein said input means supplies a partially instantiated feature-set, and wherein said instantiation means is invoked once on such partially instantiated feature-set, whereby a valid sentence according to said grammar is generated.

19. A system according to claim 16, further comprising:

a knowledge base coupled to said instantiating means for defining features of the elements of said grammar, wherein the elements are fully defined by a combination of said lexicon and said knowledge base.