US 6,980,949 B2Grant
Natural language processor
Issue Date:2005-12-27
•32 Claims
•17 Drawing Sheets
Abstract
A computer program product for controlling the computer's processor to perform responsive actions a natural language input has: (1) vocabulary, phrase and concept databases of words, phrase and concepts, respectively, that can be recognized in the inputted communication, wherein each of these database elements is representable by a designated semantic symbol, (2) means for searching the inputted communication to identify the words in the communication that are contained within the vocabulary database, (3) means for expressing the communication in terms of the word semantic symbols that correspond to each of the words identified in the inputted communication, (4) means for searching the communication when expressed in terms of its corresponding word semantic symbols so as to identify the phrases in the communication that are contained within the phrase database, (5) means for expressing the communication in terms of the phrase semantic symbols that correspond to each of the phrases identified in the communication, (6) means for searching the communication when expressed in terms of its corresponding phrase semantic symbols so as to identify the concepts in the communication that are contained within the concept database, and (7) means for expressing the communication in terms of the concept semantic symbols that correspond to each of the concepts identified in the inputted communication, wherein these concept semantic symbols are recognizable by the processor and can cause the processor to take action responsive to the inputted communication.
Metadata
Assignee
- Sonum Technologies, Inc.
Inventor
- W. Randolph Ford
Application Information
Application Number:US 10/390,270
Filing Date:2003-03-14
Priority Date:2003-03-14
Art Unit:2654
Patent Drawings (17 sheets)
Description
Cross-Reference to Related Application
[0001] This application is related to U.S. patent application Ser. No. 09/996,887, filed Nov. 30, 2001 and entitled “Multi-Stage Reduction System & Method For Processing Natural Language.” The teachings of this application are incorporated herein by reference to the extent that they do not conflict with the teaching herein.
Background of the Invention
[0002] 1. Field of the Invention
[0003] The present invention relates to data processing and speech signal processing. More particularly, this invention relates to methods, systems and algorithms that provide a computer interface for converting users' natural language messages, that are entered into the interface via speech and handwriting recognition software or by the use of a keyboard, into a query or command that can be processed by a computer.
[0004] 2. Description of the Related Art
[0005] The need for and advantages of an improved interface with a computer have long been recognized. Thus, many complex methods and algorithms for processing natural language have been developed.
[0006] There are two primary approaches being used today in natural language processing. The approach with the longest history is “parsing techniques.” In this approach, a user's speech input is processed in a manner very similar to the method of making sentence diagrams that are often learned in grammar school. That is, the input is scanned for a possible subject, verb, direct object, etc. in an attempt to fit the input against a recognized structure in a rule-based table.
[0007] Natural language processors using this approach are characteristically very large and require a great deal of processing power to operate. Their strength is that they allow for the identification of unknown objects in an input. For example, in the user input “Get me the phone number for Bob.” A parsing system can very efficiently identify the action (“get”), the directionality (implied by “me”), the object (“the phone number”), and the preposition (“for”). This leaves the noun “Bob” which a parsing system can easily use as an argument or parameter in searching a database without ever having seen this noun within the parsing tables. This is possible because parsing systems have the ability to eliminate all things understood through their extensive processing.
[0008] Unfortunately, the greatest weakness of parsing systems is their inefficient accommodation of users' variability of expression. There are literally thousands of ways that a user might ask for a phone number, and parsing systems are not very efficient, or even capable of accepting different users unique expressions of a concept. As such systems accommodate more and more various ways that a concept might be expressed, they increase markedly in size and in processing power required for their operation. That any such system could accommodate the user input “Would you go ahead and get me the phone number for Bob from my contact list?” is highly unlikely. Moreover, because parsing systems are very much rule-based, it is difficult for such systems to process ungrammatical or incomplete messages.
[0009] Today, many companies offering speech recognition technology attempt to circumvent the “variability of expression” problem by employing subsets of language. In one approach, spoken input is analyzed for keywords. Keywords are then matched with application functionality. In another approach, application users are expected to learn a set of specific commands that are matched with application functionality.
[0010] Systems that analyze keywords typically operate with a much higher degree of understanding than parsing systems. For example, in this keyword approach, assume the user inputs “Would you go ahead and get me the phone number for Bob from my contact list?” the system ignores all words except “get”, “phone number”, “Bob”, and “contact list”.
[0011] This approach provides a very efficient system for skirting the “variability of expression” problem, but it does have serious limitations, one academic and one pragmatic. On the academic side, more complex inputs will baffle a keyword system. For example, if the user were to express him/herself as: “Rather than get me the phone number for Bob from the contact list, for now, I only need his address,” the keyword system probably would be unable to process this request.
[0012] On the pragmatic side, a possibly greater problem in this approach is that the proper noun “Bob” must be in one's keyword tables in order to be recognized. This implies two critical issues: first, that such keyword systems are only good for information retrieval, and second, at some point, it may not be possible to accommodate all such possible nouns in one's keyword tables.
[0013] In light of these short-comings on natural language processing, a third approach has also come into use. The “educating the user” approach often takes the form of a series of restrictions on the type of verbal inputs allowed; with the result being restrictions that require one to speak his/her way through computer menus. These techniques have been available in commercial products for years, but have not achieved widespread acceptance. Many potential users continue to refuse to accept them because they apparently view such restriction methods as being inefficient, overly time consuming and not sufficiently user friendly.
[0014] In general, it can be said that when we use language with a computer, we expect the same particular levels of understanding from the computer as when we speak to another human. Correct understanding of natural language input information, which is an easy task for a human being, is clearly not being easily achieved by using any of the current conventional approaches.
[0015] Thus, despite the above noted prior art, a need continues to exist for improved methods, systems and algorithms which provide for an interface with a computer. Specifically, an efficient method is needed that can accommodate all of the various ways that a user might express a concept in trying to communicate with a computer.
Objects and Advantages
[0016] As a means to better understand the present invention, it is instructive to also consider the objects and advantages of the present invention.
[0017] It is an object of the present invention to provide an improved natural language interface that enables a computer to understand communications that are inputted into it.
[0018] It is an object of the present invention to provide an improved natural language, voice interface with a computer.
[0019] It is another object of the present invention to provide a natural language interface with a computer that possesses the ability to come close to human understanding of communications.
[0020] It is yet another object of the present invention to provide a natural language interface with a computer that can accommodate ungrammatical or incomplete sentences as input information.
[0021] It is an object of the present invention to provide a natural language interface with a computer that can resolve ambiguous information, such as which form of a word is meant when it can be used as either a noun or verb.
[0022] It is another object of the present invention to provide a natural language interface with a computer that can understand an input within a time period that is comparable to that which occurs in human-to-human verbal communications.
[0023] It is yet another object of the present invention to provide a natural language interface with a computer that can reduce a user's input down to a concept that is understandable by the computer's processor.
[0024] It is also an object of the present invention to provide a natural language interface with a computer that can operate efficiently within the memory and processing restrictions imposed by current state of the art personal computers.
[0025] These and other objects and advantages of the present invention will become readily apparent, as the invention is better understood by reference to the accompanying drawings and the detailed description that follows.
Summary of the Invention
[0026] The present invention is generally directed towards mitigating the limitations and problems identified with prior methods used to provide an interface for a computer. In accordance with the present invention, a preferred embodiment of this invention takes the form of a computer program for controlling a processor to generate actions in the processor which are responsive to a natural language communication input.
[0027] This computer program includes: (1) vocabulary, phrase and concept databases of words, phrase and concepts, respectively, that can be recognized in the inputted communication, wherein each of these database elements is representable by a designated semantic symbol, (2) means for searching the inputted communication to identify the words in the communication that are contained within the vocabulary database, (3) means for expressing the communication in terms of the word semantic symbols that correspond to each of the words identified in the inputted communication, (4) means for searching the communication when expressed in terms of its corresponding word semantic symbols so as to identify the phrases in the communication that are contained within the phrase database, (5) means for expressing the communication in terms of the phrase semantic symbols that correspond to each of the phrases identified in the communication, (6) means for searching the communication when expressed in terms of its corresponding phrase semantic symbols so as to identify the concepts in the communication that are contained within the concept database, and (7) means for expressing the communication in terms of the concept semantic symbols that correspond to each of the concepts identified in the inputted communication, wherein these concept semantic symbols are recognizable by the processor and are used by the processor so as to cause the processor to take action responsive to the inputted communication.
[0028] In other preferred embodiments of the present invention, the present invention can take the form of a method for generating data in a processor usable form as a result of having received a natural language communication input, or a computer user interface that generates data in a processor usable form from a natural language input.
[0029] Thus, there has been summarized above, rather broadly, the present invention in order that the detailed description that follows may be better understood and appreciated. There are, of course, additional features of the invention that will be described hereinafter and which will form the subject matter of any eventual claims to this invention.
Brief Description of the Drawings
[0030] FIG. 1 illustrates a preferred embodiment of the present invention and the input and output devices with which it communicates.
[0031] FIGS. 2( a)–(c) illustrate portions of the words in the vocabulary database or vocabulary lookup table of the present invention and the semantic symbols that are used to represent such words.
[0032] FIGS. 3( a)–(b) illustrate portions of the phrases in the phrase database or phrase lookup table of the present invention and the semantic symbols that are used to represent such phrases.
[0033] FIG. 4 identifies the significance of the alpha numeric characters used to introduce the semantic symbols for the phrases within the phrase database of the present invention.
[0034] FIGS. 5( a)–(b) illustrate a portion of the concepts in the concept database or concept lookup table of the present invention and the semantic symbols that are used in the present invention to represent such concepts.
[0035] FIG. 6 illustrates how the present invention converts or reduces a communication input to its representation as a vector comprised of the semantic symbols that correspond to the words of the inputted communication.
[0036] FIG. 7 illustrates the order of searching an inputted vector for phrases by a preferred embodiment of the algorithm of the present invention.
[0037] FIG. 8 illustrates computer code which utilizes the present invention's identification of the concepts in an inputted communication to direct the responsive action of a computer.
[0038] FIG. 9 illustrates the printout from a display screen of the present invention which shows the intermediate steps used by the software disclosed herein for entering into a computer the complex communication of “I would like to enter a credit card transaction and I want to create a new savings account with a balance of $860.20.”
[0039] FIGS. 10( a)–(c) illustrate an example of the data space reduction that can be achieved by using a preferred form of the algorithm of the present invention.
Description of the Preferred Embodiment
[0040] Before explaining at least one embodiment of the present invention in detail, it is to be understood that the invention is not limited in its application to the arrangements of the component parts or process steps set forth in the following description or illustrated in the drawings. The invention is capable of other embodiments and of being practiced and carried out in various ways. Also, it should be understood that the phraseology and terminology employed herein are for the purpose of description and should not be regarded as limiting.
[0041] The biggest problem in creating a natural language processor is not having a processor understand the meaning of a finite set of words, or even being able to reduce a message down to the concept it conveys. It is, rather, dealing with all of the various ways that words are combined to convey a concept. In short, one must find a way to accommodate humans' variability of expression.
[0042] Working in the favor of one attempting to create a natural language processor, however, are a few facts. It has been said that the average person, while knowing tens of thousands of words and their meanings, uses fewer than two thousand unique words in the course of a day. Also, from studies of voice communication it is known that the average verbal message is between seven or eight words.
[0043] These facts can give us some idea of the size of the challenge that one faces in creating a natural language processor that can understand a person having a typical vocabulary and speaking in an average manner. Even assuming that we know which two thousand words an individual is going to use in a given day, such an individual could conceivably utter approximately 4.1×1020 distinct, nine word communications, assuming no repetition of words in these communications, (i.e., 2000×1999× . . . ×1992 communications). Thus, the task of recognizing all the possible concepts contained within such an exceedingly large number of distinct communications seems initially to be quite unmanageable.
[0044] Given the great number of ways in which humans can combine words into messages, how could one possibly create a computer program to understand natural language? I have found that the answer lies in how information is organized in the human brain.
[0045] For decades, psychologists were under the impression that many chess masters, in addition to being very skilled at the game, must have photographic memories. This belief evolved because most chess masters exhibit the ability to precisely recall many of their past games—move for move. Additionally, chess masters themselves have often reported having such photographic memory capabilities.
[0046] However, after studying the chess masters' memory abilities more closely, it was discovered that such abilities have important limitations. For example, it was discovered that this chess board memorization skill was greatly diminished and eventually lost when the chess pieces were randomly placed on a chess board.
[0047] Thus, what appeared to be a feat of limitless memory was eventually seen to be a finite encoding of memory related to stages of cognition based on the rules by which chess pieces may be moved. For example, many knowledgeable chess players can glance at a chess board in the first three moves of a game and recognize it as such. Through study, the gifted chess master can apparently remember and recognize the pieces on a chess board as being those associated with one of the many possible combinations achievable after a much higher number of moves.
[0048] This chess example is also appropriate because it is represents an area in which great displays of human intelligence have been apparently simulated by a computer. However, this apparent simulation proves to really be more akin to mimicking of human intelligence through the use of pattern reduction techniques. By utilizing snap shots of recognized chess patterns generated by humans, one is thereby able to give a computer an appearance of intelligence, an artificial intelligence, in its attempts at solving chess problems. I say appearance because there is in fact nothing intelligent about what the computer is doing in such chess matches (i.e., the computer chooses among the patterns in its database to best respond to any placement of the chess pieces on a chess board). The computer's behavior fails even the simplest of definitions of intelligence (e.g., Piaget: “Intelligence is that thing we use when we don't know what to do”).
[0049] The nature of intelligence, however, is not the focus of the present invention. Rather, the present invention addresses the problem of how to take all of the possible random generations of language and identify which of these are actually meaningful. Through the use of pattern recognition, intertwined with sorting and repeated simplification and classification of communications, the present invention seeks to find a meaningful and functional process by which one can reduce human communication down to its base meaning, i.e., the concepts being communicated.
[0050] Natural language processing, the subject of the present invention, has been an area of interest for the inventor for some time. His early inventive efforts in this field and an early overview of this field are given in “Natural-Language Processing By Computer: A New Approach,” the inventor's Ph.D. dissertation submitted to The Johns Hopkins University, Baltimore, Md. in 1981. The teachings and disclosure of this work are hereby incorporated by reference.
[0051] In a preferred embodiment, the present invention takes the form of a piece of computer software that is used behind a standard piece of voice or handwriting recognition software. Such recognition software is assumed to convert the words sensed by the computer's input device and to store them in a memory device in the form of the properly spelled words.
[0052] Alternatively, the language processor software of the present invention can analyze and respond to the input that is entered into a computer by one's use of a keyboard or other data entry device that results in properly spelled sentences, questions, etc. being entered into a computer.
[0053] The language processor software version of the present invention consists of two different parts. One is a knowledge base, which is made up of three look-up tables (i.e., word, phrase and concept) or databases. The other part is the actual algorithm which is used for navigating this knowledge base to reduce a user's input down to an understandable concept. See FIG. 1 .
[0054] Understanding of such concepts allows one to use them to direct a computer's actions. For example, in response to an understood question, a computer could find the answer to an understood question in a information database to which the computer has access.
[0055] Of the present invention's knowledge base, we first consider its “word” look-up table or its vocabulary database. To each word in this database we assign a semantic symbol. See FIGS. 2( a)–(c). In examining FIG. 2 , it can be noted that more than one of the words in our base vocabulary can share the same semantic symbol.
[0056] The logic behind the semantic symbols that are assigned to the various words is explained as follows:
[0057] (a) If two words can be exchanged in a message without significantly changing the concept of the message, then those two words are deemed to be semantically equivalent. For example, one can observe that “want” and “need” can be exchanged in most uses without changing the concept communicated in a message: “I want to see my balance” and “I need to see my balance” communicate similar concepts. In fact, in most cases where “want” or “need” follow the pronoun “I” the meaning of these two words is equivalent.
[0058] (b) There is no significance to the code or number used in the semantic symbol representations of the words with the present invention's vocabulary database. The only requirement for such codes or numbers is that they be unique from the other numbers or codes that are used for other “nonequivalent” words in the vocabulary database.
[0059] For the purpose of explaining the logic of the present invention, we refer to communications such as “I want to see my balance” and “I need to see my balance” as “surface structures” that can be reduced down to a single concept or “deep structure” which is usually directed at a parameter or work object. For example, in the above communications the parameter or work object is “balance.”
[0060] The words, phrases and concepts to be included in the databases of the various application-directed embodiments of the present invention are preferably identified as a result of frequency analyzing the words, phrases and concepts used in dyadic communications of the sort that occur during the goal-directed behavior (e.g., two communicators performing a task or jointly trying to solve a problem) applicable to the application or task for which a specific embodiment is sought. In some cases, the communications to be studied will be from situations where both communicators are human; in other cases, one communicator will be human and the other will be a computer. The communication to be analyzed in detail will be only that of the human.
[0061] Sources for these communications could be the technical literature that documents scientific studies of applicable communications, and voice interface simulations of human to computer interaction. Other sources could be applicable transcripts found online.
[0062] In the analyses, such communications are preferably processed through collocational algorithms to determine a number of characteristics or properties with respect to the frequency of the words, grouping of words, etc. used in these communications. For example, analysis of these communications will usually identify which words should be included in one's “word” database.
[0063] Additionally, it can be determined which words in these communications frequently appeared next to each other in chains of two words groupings, three words groupings and greater length groupings. These are identified as multi-word phrases (i.e., frequently occurring groupings of two or more words) for inclusion in our “phrase” database.
[0064] Also, if one assumes that a communication is comprised only of phrases and disregards trying to identified the various parts of speech (i.e., subjects, verbs, pronouns, etc.) of such communications, one is led to identify what I refer to as “single-word phases” (i.e., frequently verbs; words that appear in the “word” database and also in the “phrase” database) and to group adjectives with the words they modify for phrase identification purposes .
[0065] As an example, consider the following different “surface structures,” which all share the same “deep structure,” from an analysis of various recorded communications of individuals communicating with their checkbook balancing programs:
-
- [0066] “I would like to see my checkbook balance,”
- [0067] “I need to see my checkbook balance,”
- [0068] “I will need to see my checkbook balance?”
- [0069] “I am going to need to see my checkbook balance.”
[0070] From my analysis of these communications, I associate the pronoun “I” with the words that followed it. For example:
-
- [0071] “I would like to,”
- [0072] “I need to,”
- [0073] “I will need to,” and
- [0074] “I am going to need to”
[0075] Following this same sort of logic in grouping adjectives with the words they modify for phrase identification purposes, one sees that “my checkbook balance” can also be identified as a “phrase,” and thus added to our phrase database.
[0076] For these communication examples, it then follows that the verb “see” will be classified as a single-word phrase and thus will appear in both our word and phrase databases.
[0077] FIGS. 3( a)–(b) show a portion of the phase database for one preferred embodiment of the present invention. As in the word database, to each phrase in this database there is assigned a phrase semantic symbol.
[0078] The logic behind the semantic symbols identified in this phrase database is explained as follows:
[0079] (a) If two phrases can be exchanged in a verbal communication without significantly changing its concept or deep structure, then those two phrases are deemed to be semantically equivalent and given the same semantic symbol. For example, semantic representations for “I want to” and “I would like to” can be exchanged in most uses without changing a message's deep structure.
[0080] (b) The semantic symbols that are associated with these phrases are constructed such that the first digit or alphanumeric character in these symbols has a special significance and identifies sub-groupings having a common feature. The balance of these numbers or codes for these symbols only needs to be unique from the other numbers or codes for the other phrases that are not semantically equivalent. The significance of this first digit lies in the manner in which it is used for grouping various types of phrases into classes for easy identification.
[0081] During my frequency studies of communications, I observed that the phrases that preceded and followed a verb phrase, which I denote as “wrapper’ phrases, functioned to determine the “directionality” of the deep structure's verb. For example, I distinguish the following surface structures as having opposite directionalities:
-
- [0082] “I must get my balance,”
- [0083] “You need to get my balance for me.”
[0084] In the first communication, the leading phrase identifies the one to whom the information about the balance must go. In the second communication, the leading phrase identifies the one who must do the work (i.e., “you,” which is the computer being served by the language processor of the present invention) to get the information about the balance.
[0085] To distinguish these situations, I use a leading “a” to distinguish those leading wrapper phrases which identify the recipient of the computer's output, and a “b” to distinguish those leading wrapper phrases which identify the one to do the work. See FIGS. 3( a)–(b).
[0086] Similarly, I use a leading “c” to distinguish those trailing wrapper phrases which identify the recipient of the computer's output, e.g., “for me” is given the semantic representation of“c004.” A “d” is used to distinguish those trailing wrapper phrases which identify the one to do the work, e.g., “from you” is given the semantic representation “d003”.
[0087] In addition to leading wrapper and trailing wrapper phrases, my studies of communication have revealed that it proves useful to also identify pre-leading wrapper phrases that occasional occur with leading wrapper phrases. Such phrases are coded with the letter “e” such as e003 for “go ahead and”.
[0088] Continuing with the identification of such sub-grouping of phrases, I have chosen to identify verbs by the letter “f,” i.e., “get” is given the semantic representation “f110”.
[0089] As I continued to utilize such sub-grouping, it became apparent to me that these leading alphanumeric characters could be used to direct how the algorithm of the present invention processes a particular phrase in a communication. For example, leading alphanumeric characters in the range of “a” through “l” are used in the present invention to help identify how such phrases communicate a concept.
[0090] Meanwhile, phrases with a leading alphanumeric that follows later in the alphabet than “l” are to be initially ignored by the algorithm as it seeks to identify the concept contained within an imputed communication. For example, phrases identified with a leading character of“m” are set aside as parameters that identify an object to which a phrase or concept is directed (e.g., “email message”—m003). Phrases identified with a leading character of“n” are phrases that identify a subcategory of information associated with an object that has been categorized as an “m” phrase (e.g., “the subject field of an email message”—n038).
[0091] Phrases identified with a leading character of “o” are set aside as parameters for identifying a specific object, such as “the last” or “the second”. Phrases identified with a leading character of “u” are set aside as supplementary phrases, i.e., indicators of politeness (e.g., “please”—u004, “if you will”—u005) and other phrases peripheral to the initial identification of the concept of the inputted communication. See FIG. 4 for an abbreviated key that identifies the significance of the various alphanumeric characters used in the present invention.
[0092] As this sub-grouping of phrases moved forward, I discovered that I was able to easily identify and account for more and more of the variability in one's communications. Each time I identified a phrase and included it in my phrase database, I was effectively accounting for more and more of the variability of human expression. Taken to its extreme when all possible phrases are identified and placed in a phrase database, one is left with trying to analyze a communication in which the only unidentified pieces are the nouns or objects of the communication, e.g., when one is asking a computer to find certain information in its database, these nouns identify the information which is desired from the database.
[0093] Once one has a way to try to account for much of the variability of typical goal-directed communications, one is better prepared to try to identify the concepts or “deep structures’ within such communications. Upon undertaking frequency analysis of phrase semantic symbols for many communications, I again discovered significant repetition in the format of these phrase semantic symbols. This led me to create a concept database in which I try to categorize the various types of concepts being communicated in typical goal-directed communications.
[0094] FIGS. 5( a)–(b) show a portion of a preferred embodiment for a concept database or a look-up table for use with the present invention. As in the prior databases, to each concept in this database there is assigned a semantic symbol. This proves to be especially useful when we are trying to understand communications having more than one concept. Also, if it happens that two concepts can be exchanged in a verbal communication without significantly changing its meaning, then those two concepts are deemed to be semantically equivalent and given the same concept semantic symbol.
[0095] With the above descriptions of the word, phrase and concept databases, it proves useful to consider further the algorithm used to analyze a communication directed at the natural language processor of the present invention. In order for the present invention to understand the concepts within a communication, semantic symbols from the software's vocabulary look-up table are used to replace the words comprising the inputted communication.
[0096] Care is given to arrange the symbols in the same order as their corresponding words which appeared in the inputted communication. See FIG. 6 for the semantic symbol representation of the input “I would like to get my 401k account balance.”
[0097] For the purpose of this discussion, this symbolic representation of the inputted communication will be referred to as a word vector. This vector (i.e., 019 009 143 031 012 038 803 695 217) represents the semantic reduction of the words, or the first pattern reduction step in the algorithm of the present invention. It is important to note that this single vector can, in cases where there are multiple words corresponding to each semantic symbol, represent many different ways to express the same concept.
[0098] In the next step, the word vector (i.e., 019 009 143 031 012 038 803 695 217) is used to begin to search through a phrase database to create a phrase vector which is representative of the inputted communication, see FIGS. 3( a)–(b). The logic behind this search is explained as follows:
[0099] (a) The identification of phrases has a directionality to it—that is, one moves from left to right through the list of the semantic symbols that have been substituted for the words in the verbal communication of interest, wherein a semantic symbol appearing on the left indicates that its corresponding word was uttered before the word represented by the semantic symbol appearing to its right. One seeks to identify the largest or longest string possible of word semantic symbols from those phrases that have been compiled into the database of phrases.
[0100] (b) Having identified a first phrase at the beginning of our vector of semantic representations for words, our phrase identification search engine examines the semantic symbols that lie to the right of the identified phrase in order to identify any other recognizable phrases that appear in our phrase database—always trying to identify the longest possible strings of semantic representations.
[0101] Thus, for the word vector (e.g., 019 009 143 031 012 038 803 695 217) one initially searches for a match for the whole vector. Finding no match, one searches for a match for the whole vector minus the last, or furtherest right, semantic representation (i.e., 019 009 143 031 012 038 803 695), and, finding no match, again searches for a match for the whole vector minus the next-to-last semantic representation (i.e., 019 009 143 031 012 038 803). This process of searching the phrase table for phrases having ever shorter strings continues to be repeated until one finally finds a match (e.g., 019 009 143 031). See FIG. 7 .
[0102] At this point, one takes the semantic symbol a002 and saves it in the first, or left-most, position of what is defined as a phrase vector.
[0103] The phrase searching process continues with the next set of semantic symbols to the right of the last symbol in the identified phrase (i.e., 012 038 803 695 217) being searched for a match in the phrase table. In the example at hand, one finds that only the single semantic symbol, 012 can be matched. It yields the phrase semantic symbol of f110.
[0104] This matching process continues until one finally find a match for the array of 038 803 695 217, which yields the phrase semantic symbol of m117. This combination of matched phrase semantic symbols is used to create the phrase vector: “a002 f110 m117.”
[0105] From searching for the concepts contained within such communications expressed by their phrase semantic symbol representations, I discovered many repeating patterns of phrases in such communications. For example, I often observed an “a” phrase preceding an “f” phrase followed by an “m” phrase” and a “d phrase,” as in “I need to get my balance from you,” which converted to phrase semantic symbols as “a002 f110 n801 d003.”
[0106] This analysis led me to realize that initially ignoring some phrases made it easier to discern the concept or “deep structure” contained within a communication. This led to my decision to denoted these initially to be ignored phrases by assigning them a leading alphanumeric that follows later in the alphabet than “I.”
[0107] When one gets to the point in the analysis of the inputted communication where they have clearly identified the recognizable phrases in an inputted communication (i.e., accounted for the possible variability within an inputted communication), one assumes that the remaining unidentifiable phrases are the specific identifying parameters for the objects within the communication. For example, the inputted query of “Get me the email address for Betty Mayfield” is translated by the present invention to the phrase vector: f110 c001 n832 “Betty Mayfield”.
[0108] Setting aside the object part of this concept vector, one is now left with: f110 c001 n832. The present invention's process of discerning the meaning of such an expression continues by searching a concept database, such as that shown in FIG. 5 .
[0109] The logic behind the manner in which the algorithm of the present invention searches these phrase vectors is given below:
[0110] (a) As with phrases, concept identification has a directionality to it—one searches from left to right through the list of semantic symbols for the phrases that comprise the inputted communication of interest and aims to identify the largest (or longest in terms of the number of phrase semantic symbols) concept possible. Having done this, the search engine of the present invention moves across the symbolic inputted communication to identify the next largest identifiable concept. In this manner our search engine is capable of identifying multiple concepts in a single communication.
[0111] (b) The semantic symbols that are associated with these concepts or “deep structures” consist of two, capitalized alphanumeric characters. There is no significance to these capitalized characters. The only requirement for them is that they be unique from the other characters that are used for other “nonequivalent” concepts in the concept database.
[0112] The algorithm of the present invention then uses the concepts identified within an inputted communication to direct the actions of the computer. For example, for the communication “Get me the email address for Betty Mayfield” and its representative phrase vector “f110 c001 n832 Betty Mayfield,” one finds that the phrase combination “f110 c001” is found in the concept table, and is represented by the semantic symbol “GM” which stands for the command “Get me.” Thus, the initial input can be expressed in our semantic symbols as “GM n832 Betty Mayfield.”
[0113] The algorithm's identification of “GM” can be used to cause the computer to search its database for the object appearing behind the identified concept, i.e., the phrase n832 which is the semantic representation for “the email address.” The computer responds by searching its database for “email address” and the fields by which these addresses are classified. If “Name” is found to be a field of classification for the emails, the computer will make an attempt to retrieve the email address for the record whose name matches the unidentified phrase “Betty Mayfield”.
[0114] As another example of how this step proceeds, consider the situation where a user asks a computer that utilizes the natural language processor software of the present invention and that is attached to a personal information manager application “What are my active to do list items?” The software of the present invention converts this input into the semantic symbols “W2 n906 m025” where “W2” represents the deep structure for “What are”, n906” represents the property “active”, and “m025” represents the object “to do list items”. This information can then be passed to a database in the form of a SQL query.
[0115] The computer code given in FIG. 8 is from just such a computer application. In it you can see that on the first line that the block of code following that line will be executed because the action does in fact equal “W2”. Because the descriptor A on the second line down does in fact equal “m025” or the object “to do list items”, the block of code next checks to see that descriptor B equals “n906” the property for “active. Given the satisfaction of these three conditions, the personal information manager application program communicates back the desired tasks that are still active.
[0116] As another example of how a preferred algorithm of the present invention operates, consider the action taken by an algorithm-directed computer in response to the following input: “I need to change the deposit date . . . it should be Jan. 22nd, 2003.”
[0117] The difficulty in interpreting this input centers around how to interpret the pronoun “it” in the second part of the communication (i.e., an “explicit pronoun”). However, the algorithm of the present invention can accommodate this input by maintaining the context within which the communication is occurring. The context is maintained by holding the results of the initially inputted messages in place while a further message is awaited. Given an initial input “I need to change the deposit date,” the algorithm converts this message into semantic symbols and searches for phrases and concepts. It represents the input as “AL m900 n808,” where AL is the symbol for the concept “to change something,” m900 is the symbol for the primary object, “deposit,” and n805 is the symbol for secondary object “date.” These are initially stored in a part of the computer's memory that is defined as the algorithm's “rolling contact manager.” They are stored as, respectively, the action or concept, the primary object, and the secondary object of the communication.
[0118] When the user follows with “it should be Jan. 22nd, 2003”, the natural language processor knows exactly what she or he is talking about with respect to “it”. With the previously stored symbolic inputs, the algorithm of the present invention understands that the communicator or natural language processor user is implicitly referring to the “date.”
[0119] The algorithm of the present invention is also seen to be able to handle communications that involve “implicit action” queries. For example, an inputted communication such as: “What is the average income for California? And the unemployment rate?, where in the second query both the deep structure “what is” and the object “California” are implicit.
[0120] The algorithm of the present invention accommodates such questions by again identifying the concepts (i.e., “what is,” which is denoted as W2) and objects (i.e., primary object of “California,” secondary object of “the average income”) within the first part of this communication and storing them in memory. Thus the context of the communication is set.
[0121] When the user follows this query with a second question, “And the unemployment rate?”, the phrase value for “unemployment rate” (another secondary object) replaces the previous secondary object “the average income”. When the natural language processor of the present invention attempts to process the input void of a concept or deep structure, it reverts make to the saved values in context and realizes that it has a complete query therein. When the user asked “And the average unemployment rate?”, the user was implicitly asking “What is the average income for California?” Everything one needs to determine this was saved in the “rolling context manager” of the present invention.
[0122] As yet another example of how a preferred algorithm of the present invention operates, consider the action taken by our algorithm-directed computer in response to the following input: “Can you tell me Betty's phone number?,” where our natural language process must equate “Betty's phone number” with “the phone number of/for Betty.”
[0123] In considering how the algorithm of the present invention can accommodate this “possessive noun” challenge, it is instructive to first consider what sort of “preprocessing” of the inputted communication is conducted by the present invention. For example, when, in examining the words that are inputted, the algorithm of the present invention encounters a contraction (e.g., I'll, you've), it converts such inputs to their non-contracted form's (e.g., I will, you have). This converted message is then processed as if it had never employed a contraction.
[0124] Similarly, this preprocessing step also entails converting possessive nouns and pronouns (e.g., Betty's, its) to specifically defined equivalents (e.g., of Betty, of it) which indicate the possessive without using an “s” or a “'s ”. Such converted messages are then processed as if it had never employed a possessive using an “s” or a “s”.
[0125] Another task which the algorithm of the present invention accomplishes is some general selective editing, which can prove to be especially valuable when the communications are being verbally inputted into a computer. In such circumstances, it can often happen that the users of natural language processors will direct their verbal communications to others in the vicinity besides just the computer's input microphone (e.g., workers collaborating on a project could be discussing not only the work at hand, but also could pepper their conversation with comments about Sunday's football game, remarks directed at mocking their boss, and complaints about spouses).
[0126] The computer's audio input is then challenged to understand verbal communications that may have apparently disjointed thoughts and concepts. To resolve this situation, a preferred embodiment for the algorithm of the present invention will ignore that part of the verbal input which is outside of the context of the concept database that will have been created for a specific applicational use of the present invention. When the input is specific to the designed applicational use of the present invention, this natural language processor will search its “deep structures” or concept database and their associated objects to identify the proper action to be taken by the computer in response to the relevant or non-ignored input.
[0127] As previously mentioned, a preferred embodiment of the algorithm of the present invention can accommodate communications having multiple concepts. The means by which this is accomplished is illustrated in FIG. 9 . This printout from a display screen of the present invention shows the intermediate steps used by the software disclosed herein for entering into a computer the complex communication: “I would like to enter a credit card transaction and I want to create a new savings account with a balance of $860.20.”
[0128] The display screen is seen to have lines that show the semantic symbols generated at each stage of the algorithm's action on this input. On the line labeled “Final” one can see that the language processor of the present invention has correctly parsed this input as having two key concepts (i.e., enter and create), two primary objects (i.e., “credit card transaction” and account) and a secondary object (i.e., balance). The name of the “account” is the unrecognized word “savings,” and the amount of the balance is the unrecognized “$680.20.”
[0129] To better appreciate the interpretative power of the present invention, consider the power of the search conducted with just the phrases “a002 f110.” When reviewing the information shown in the phrase database, one discovers that the phrases “a002 f110” represents over three hundred difference ways that a user might ask for his/her 401k account balance. Multiply this times the six ways the primary object “401k balance” can be represented and we have accommodated 1,800 different ways a user might ask for a specific piece of information with a single table entry and a complementary object. See FIGS. 10( a)–(c) for this illustrative example.
[0130] In an early demo application of the present invention, a specific application had an object vocabulary that consisted of 46 unique entries. For this application, we find that there 82,800 different ways of asking for these 46 unique pieces of information from the generalized concept symbolized by AA. Assuming that there are in fact 43 different table entries in the concept database for AA, the retrieval operation of the present invention can understand 3,560,400 different ways to ask for these 46 unique objects, all from a database with fewer than 1,500 symbolic entries across words, phrases, and concepts.
[0131] With thirty-one entries of different semantic symbols for various concepts, one finds that a table with 1,500 entries can represent over 100 million user inputs, in 1.2 megabyte of space. The present invention is thus seen to efficiently accommodate a wide range of variability in how a user might attempt to communicate a particular thought or concept.
[0132] Although the foregoing disclosure relates to preferred embodiments of the invention, it is understood that these details have been given for the purposes of clarification only. Various changes and modifications of the invention will be apparent, to one having ordinary skill in the art, without departing from the spirit and scope of the invention as hereinafter set forth in the claims.
Claims
1. A method for converting a natural language communication inputted into a processor into data that can be processed by said processor, said method comprising the steps of:
establishing a vocabulary database of words, wherein each of said database words is representable by a designated word semantic symbol,
establishing a phrase database of phrases, wherein each of said database phrases being representable by a designated phrase semantic symbol,
establishing a concept database of concepts, wherein each of said database concepts being representable by a designated concept semantic symbol,
searching said inputted communication to identify the words in said communication that are contained within said vocabulary database,
expressing said inputted communication in terms of the word semantic symbols that correspond to each of said words identified in said inputted communication,
searching said inputted communication when expressed in terms of word semantic symbols so as to identify the phrases in said communication that are contained within said phrase database,
expressing said inputted communication in terms of the phrase semantic symbols that correspond to each of said phrases identified in said inputted communication,
searching said inputted communication when expressed in terms of phrase semantic symbols so as to identify the concepts in said communication that are contained within said concept database,
expressing said inputted communication in terms of the concept semantic symbols that correspond to each of said concepts identified in said inputted communication, and upon identifying said concepts in said inputted communication, causing said processor to take action responsive to said identified concepts.
2. A method as recited in claim 1, wherein: words not identifiable in said vocabulary database are used as the objects of said inputted communication.
3. A method as recited in claim 1, wherein: words in said vocabulary database that are exchangeable in an inputted communication without changing the concept of said communication share the same word semantic symbol, phrases in said phrase database that are exchangeable in an inputted communication without changing the concept of said communication share the same phrase semantic symbol, and concepts in said concept database that are exchangeable in an inputted communication without changing the meaning of said communication share the same concept semantic symbol.
4. A method as recited in claim 3, wherein: said phrase semantic symbols include a digit that is used to identify sub-groupings of said phrases that share a common feature which is used to aid in said search for a concept within said inputted communication.
5. A method as recited in claim 3, wherein: said search for phrases of said inputted communication when expressed in terms of word semantic symbols is conducted by examining initially the longest string of words that lie to the right of an identified phrase to determine whether this grouping of words comprises a phrase that is within said phrase database; if not, the next shorter string that is formed by dropping the last word in the original string is examined to see whether said next shorter string is in said phrase database, with this process of discarding the last word and examining the remaining string of words being continued until all phrases have been found.
6. A method as recited in claim 4, wherein: said search for phrases of said inputted communication when expressed in terms of word semantic symbols is conducted by examining initially the longest string of words that lie to the right of an identified phrase to determine whether this grouping of words comprises a phrase that is within said phrase database; if not, the next shorter string that is formed by dropping the last word in the original string is examined to see whether said next shorter string is in said phrase database, with this process of discarding the last word and examining the remaining string of words being continued until all phrases have been found.
7. A method as recited in claim 4, wherein: words, phrases and concepts in said respective word, phrase and concept databases are selected as a result of frequency analysis of communications.
8. A method as recited in claim 7, further comprising the step of: preprocessing said inputted communication to convert contractions appearing therein to their non-contracted form and to convert possessive nouns and pronouns appearing therein to equivalent forms which utilize the preposition “of”.
9. A user interface for a computer that generates data in a processor usable form from a natural language communication inputted into said processor, said interface comprising:
a vocabulary database of words that could be recognized in said inputted communication, wherein each of said database words is representable by a designated word semantic symbol,
a phrase database of phrases that could be recognized in said inputted communication, wherein each of said database phrases is composed of words from said word database and can be represented as the collection of said semantic symbols corresponding to the words contained within said phrase, wherein each of said database phrases being representable by a designated phrase semantic symbol,
a concept database of concepts that could be recognized in said inputted communication, wherein each of said database concepts is composed of phrases from said phrase database and can be represented as the collection of said semantic symbols corresponding to the phrases contained within said concept, wherein each of said database concepts being representable by a designated concept semantic symbol which is recognizable by said processor and can be used by said processor so as to cause said processor to take action responsive to said inputted communication,
a means for searching said inputted communication to identify the words in said communication that are contained within said vocabulary database,
a means for expressing said inputted communication in terms of the word semantic symbols that correspond to each of said words identified in said inputted communication,
a means for searching said inputted communication when expressed in terms of word semantic symbols so as to identify the phrases in said communication that are contained within said phrase database,
a means for expressing said inputted communication in terms of the phrase semantic symbols that correspond to each of said phrases identified in said inputted communication,
a means for searching said inputted communication when expressed in terms of phrase semantic symbols so as to identify the concepts in said communication that are contained within said concept database,
a means for expressing said inputted communication in terms of the concept semantic symbols that correspond to each of said concepts identified in said inputted communication, and
a means for causing said processor to take action responsive to the identification of said concepts.
10. A user interface for a computer as recited in claim 9, wherein: words not identifiable in said vocabulary database are used as the objects of said inputted communication.
11. A user interface for a computer as recited in claim 9, wherein: words in said vocabulary database that are exchangeable in an inputted communication without changing the concept of said communication share the same word semantic symbol, phrases in said phrase database that are exchangeable in an inputted communication without changing the concept of said communication share the same phrase semantic symbol, and concepts in said concept database that are exchangeable in an inputted communication without changing the meaning of said communication share the same concept semantic symbol.
12. A user interface for a computer as recited in claim 11, wherein: said phrase semantic symbols include a digit that is used to identify sub-groupings of said phrases that share a common feature which is used to aid in said search for a concept within said inputted communication.
13. A user interface for a computer as recited in claim 11, wherein: said search for phrases of said inputted communication when expressed in terms of word semantic symbols is conducted by examining initially the longest string of words that lie to the right of an identified phrase to determine whether this grouping of words comprises a phrase that is within said phrase database; if not, the next shorter string that is formed by dropping the last word in the original string is examined to see whether this string which is shorter by one word is in the phrase database, with this process of discarding the last word and examining the remaining string of words being continued until all phrases have been found.
14. A user interface for a computer as recited in claim 12, wherein: said search for phrases of said inputted communication when expressed in terms of word semantic symbols is conducted by examining initially the longest string of words that lie to the right of an identified phrase to determine whether this grouping of words comprises a phrase that is within said phrase database; if not, the next shorter string that is formed by dropping the last word in the original string is examined to see whether this string which is shorter by one word is in the phrase database, with this process of discarding the last word and examining the remaining string of words being continued until all phrases have been found.
15. A user interface for a computer as recited in claim 12, wherein: words, phrases and concepts in said respective word, phrase and concept databases are selected as a result of frequency analysis of communications.
16. A user interface for a computer as recited in claim 15, further comprising: a means for preprocessing said inputted communication to convert contractions appearing therein to their non-contracted form and to convert possessive nouns and pronouns appearing therein to equivalent forms which utilize the preposition “of”.
17. A computer program product in a computer readable memory for generating data in a processor usable form from a natural language communication inputted into said processor, said product comprising:
a vocabulary database of words that can be recognized in said inputted communication, wherein each of said database words is representable by a designated word semantic symbol,
a phrase database of phrases that can be recognized in said inputted communication, wherein each of said database phrases is composed of words from said word database and can be represented as the collection of said semantic symbols corresponding to the words contained within said phrase, wherein each of said database phrases being representable by a designated phrase semantic symbol,
a concept database of concepts that can be recognized in said inputted communication, wherein each of said database concepts is composed of phrases from said phrase database and can be represented as the collection of said semantic symbols corresponding to the phrases contained within said concept, wherein each of said database concepts being representable by a designated concept semantic symbol which is recognizable by said processor and can be used by said processor so as to cause said processor to take action responsive to said inputted communication,
means for searching said inputted communication to identify the words in said communication that are contained within said vocabulary database,
means for expressing said inputted communication in terms of the word semantic symbols that correspond to each of said words identified in said inputted communication,
means for searching said inputted communication when expressed in terms of word semantic symbols so as to identify the phrases in said communication that are contained within said phrase database,
means for expressing said inputted communication in terms of the phrase semantic symbols that correspond to each of said phrases identified in said inputted communication,
means for searching said inputted communication when expressed in terms of phrase semantic symbols so as to identify the concepts in said communication that are contained within said concept database, and
means for expressing said inputted communication in terms of the concept semantic symbols that correspond to each of said concepts identified in said inputted communication.
18. A computer program product as recited in claim 17, wherein: words not identifiable in said vocabulary database are used as the objects of said inputted communication.
19. A computer program product as recited in claim 17, wherein: words in said vocabulary database that are exchangeable in an inputted communication without changing the concept of said communication share the same word semantic symbol, phrases in said phrase database that are exchangeable in an inputted communication without changing the concept of said communication share the same phrase semantic symbol, and concepts in said concept database that are exchangeable in an inputted communication without changing the meaning of said communication share the same concept semantic symbol.
20. A computer program product as recited in claim 19, wherein: said phrase semantic symbols include a digit that is used to identify sub-groupings of said phrases that share a common feature which is used to aid in said search for a concept within said inputted communication.
21. A computer program product as recited in claim 19, wherein: said search for phrases of said inputted communication when expressed in terms of word semantic symbols is conducted by examining initially the longest string of words that lie to the right of an identified phrase to determine whether this grouping of words comprises a phrase that is within said phrase database; if not, the next shorter string that is formed by dropping the last word in the original string is examined to see whether this string which is shorter by one word is in the phrase database, with this process of discarding the last word and examining the remaining string of words being continued until all phrases have been found.
22. A computer program product as recited in claim 20, wherein: said search for phrases of said inputted communication when expressed in terms of word semantic symbols is conducted by examining initially the longest string of words that lie to the right of an identified phrase to determine whether this grouping of words comprises a phrase that is within said phrase database; if not, the next shorter string that is formed by dropping the last word in the original string is examined to see whether this string which is shorter by one word is in the phrase database, with this process of discarding the last word and examining the remaining string of words being continued until all phrases have been found.
23. A computer program product as recited in claim 20, wherein: words, phrases and concepts in said respective word, phrase and concept databases are selected as a result of frequency analysis of communications.
24. A computer program product as recited in claim 23, further comprising: a means for preprocessing said inputted communication to convert contractions appearing therein to their non-contracted form and to convert possessive nouns and pronouns appearing therein to equivalent forms which utilize the preposition “of”.
25. A computer program product in a computer readable memory for generating data in a processor usable form from a natural language communication inputted into said processor, said product comprising:
a vocabulary database of words that could be recognized in said inputted communication, wherein each of said database words is representable by a designated word semantic symbol,
a phrase database of phrases that could be recognized in said inputted communication, wherein each of said database phrases is composed of words from said word database and can be represented as the collection of said semantic symbols corresponding to the words contained within said phrase, wherein each of said database phrases being representable by a designated phrase semantic symbol which is recognizable by said processor and upon said recognition causes said processor to take action responsive to said inputted communication,
means for searching said inputted communication to identify the words in said communication that are contained within said vocabulary database,
means for expressing said inputted communication in terms of the word semantic symbols that correspond to each of said words identified in said inputted communication,
means for searching said inputted communication when expressed in terms of word semantic symbols so as to identify the phrases in said communication that are contained within said phrase database,
means for expressing said inputted communication in terms of the phrase semantic symbols that correspond to each of said phrases identified in said inputted communication, and
means for, upon identifying said phrases, causing said processor to take action responsive to said expressed phrase semantic symbols.
26. A computer program product as recited in claim 25, wherein: words not identifiable in said vocabulary database are used as the objects of said inputted communication.
27. A computer program product as recited in claim 25, wherein: words in said vocabulary database that are exchangeable in an inputted communication without changing the concept of said communication share the same word semantic symbol, and phrases in said phrase database that are exchangeable in an inputted communication without changing the concept of said communication share the same phrase semantic symbol.
28. A computer program product as recited in claim 27, wherein: said phrase semantic symbols include a digit that is used to identify sub-groupings of said phrases that share a common feature which is used to aid in converting said inputted communication into processable information.
29. A computer program product as recited in claim 27, wherein: said search for phrases of said inputted communication when expressed in terms of word semantic symbols is conducted by examining initially the longest string of words that lie to the right of an identified phrase to determine whether this grouping of words comprises a phrase that is within said phrase database; if not, the next shorter string that is formed by dropping the last word in the original string is examined to see whether said next shorter string is in the phrase database, with this process of discarding the last word and examining the remaining string of words being continued until all phrases have been found.
30. A computer program product as recited in claim 28, wherein: said search for phrases of said inputted communication when expressed in terms of word semantic symbols is conducted by examining initially the longest string of words that lie to the right of an identified phrase to determine whether this grouping of words comprises a phrase that is within said phrase database; if not, the next shorter string that is formed by dropping the last word in the original string is examined to see whether said next shorter string is in the phrase database, with this process of discarding the last word and examining the remaining string of words being continued until all phrases have been found.
31. A computer program product as recited in claim 28, wherein: words and phrases in said respective word and phrase databases are selected as a result of frequency analysis of communications.
32. A computer program product as recited in claim 31, further comprising of: means for preprocessing said inputted communication to convert contractions appearing therein to their non-contracted form and to convert possessive nouns and pronouns appearing therein to equivalent forms which utilize the preposition “of”.
Patent Citations (27)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US4742481(A) | 1988-05-01 | Yoshimura | Examiner |
| US4887212(A) | 1989-12-01 | Zamora et al. | Applicant |
| US5083268(A) | 1992-01-01 | Hemphill et al. | Applicant |
| US5608624(A) | 1997-03-01 | Luciw | Applicant |
| US5644774(A) | 1997-07-01 | Fukumochi et al. | Examiner |
| US5694523(A) | 1997-12-01 | Wical | Applicant |
| US5774845(A) | 1998-06-01 | Ando et al. | Applicant |
| US5794050(A) | 1998-08-01 | Dahlgren et al. | Applicant |
| US5878386(A) | 1999-03-01 | Coughlin | Applicant |
| US5963940(A) | 1999-10-01 | Liddy et al. | Applicant |
| US6026388(A) | 2000-02-01 | Liddy et al. | Examiner |
| US6081774(A) | 2000-06-01 | De Hita et al. | Applicant |
| US6108620(A) | 2000-08-01 | Richardson et al. | Applicant |
| US6112168(A) | 2000-08-01 | Corston et al. | Applicant |
| US6188977(B1) | 2001-02-01 | Hirota | Applicant |
| US6219643(B1) | 2001-04-01 | Cohen et al. | Applicant |
| US6236959(B1) | 2001-05-01 | Weise | Applicant |
| US6275791(B1) | 2001-08-01 | Weise | Applicant |
| US6292767(B1) | 2001-09-01 | Jackson et al. | Applicant |
| US6314411(B1) | 2001-11-01 | Armstrong | Applicant |
| US6321190(B1) | 2001-11-01 | Bernardes et al. | Applicant |
| US6393428(B1) | 2002-05-01 | Miller | Applicant |
| US6434524(B1) | 2002-08-01 | Weber | Applicant |
| US6442522(B1) | 2002-08-01 | Carberry | Applicant |
| US6466899(B1) | 2002-10-01 | Yano et al. | Applicant |
| US6499013(B1) | 2002-12-01 | Weber | Applicant |
| US6505157(B1) | 2003-01-01 | Elworthy | Applicant |
Non-Patent Literature (2)
- Inter. Search Report, PCT/US04/07056.Applicant
- W. R. Ford, “Natural-Language Processing By Computer: A New Approach,” Ph. D. dissertation submitted to The Johns Hopkins University, 1981.Applicant