US20070094225A1 - Method and system for using natural language input to provide customer support - Google Patents

Method and system for using natural language input to provide customer support Download PDF

Info

Publication number
US20070094225A1
US20070094225A1 US11/503,938 US50393806A US2007094225A1 US 20070094225 A1 US20070094225 A1 US 20070094225A1 US 50393806 A US50393806 A US 50393806A US 2007094225 A1 US2007094225 A1 US 2007094225A1
Authority
US
United States
Prior art keywords
nodes
natural language
semantic
context
contextual
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Abandoned
Application number
US11/503,938
Inventor
Lawrence Au
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Datacloud Technologies LLC
Chartoleaux KG LLC
Original Assignee
Individual
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from US09/085,830 external-priority patent/US6778970B2/en
Application filed by Individual filed Critical Individual
Priority to US11/503,938 priority Critical patent/US20070094225A1/en
Assigned to Q-PHRASE LLC reassignment Q-PHRASE LLC ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: AU, LAWRENCE
Publication of US20070094225A1 publication Critical patent/US20070094225A1/en
Assigned to QPS TECH. LIMITED LIABILITY COMPANY reassignment QPS TECH. LIMITED LIABILITY COMPANY ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: Q-PHRASE, LLC
Assigned to INTELLECTUAL VENTURES ASSETS 151 LLC reassignment INTELLECTUAL VENTURES ASSETS 151 LLC ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: CHARTOLEAUX KG LIMITED LIABILITY COMPANY
Assigned to DATACLOUD TECHNOLOGIES, LLC reassignment DATACLOUD TECHNOLOGIES, LLC ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: INTELLECTUAL VENTURES ASSETS 151 LLC
Abandoned legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • G06F16/3331Query processing
    • G06F16/334Query execution
    • G06F16/3344Query execution using natural language analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/237Lexical tools
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/30Semantic analysis
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00Speech recognition
    • G10L15/08Speech classification or search
    • G10L15/18Speech classification or search using natural language modelling
    • G10L15/1815Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning

Definitions

  • Examples of natural language forms include written and spoken words, gestures and face expressions which clarify the meaning of words, intonation patterns indicating whether a sentence is a question, and other auditory and visual contextual aids for conveying meaning.
  • Each of these forms may convey multiple meanings which may appear ambiguous when forms occur out of context. For instance, there are many dictionary meanings of the word ‘dash’, such as a race or a small amount of seasoning. Since multiple meanings for words are so common, lexicographers refer to a single word's diversity of meanings as polysemy. To disambiguate polysemy, all the of the above natural language forms must be correlated in order to infer the context which identified specific meaning from a list of polysemous potential meanings.
  • the prior art detects such inputs on a piecemeal basis, but to provide a comprehensive user interface for computer systems, all of the above inputs must be correlated, as efficiently as a person can correlate them.
  • the prior art has failed to correlate even a fraction of such inputs as efficiently as people do.
  • designers of natural language processing systems have greatly restricted the range of conversation recognized by natural language processing, to increase accuracy at the expense of flexibility.
  • the word ‘dash’ can mean at least four different nouns and three different verbs: dash as in a small quantity mixed into something (a dash of salt), dash as in a race(a fifty yard dash), dash as in car dashboard, dash as in a Morse Code (a signal longer that a dot), dash as in to mix together, dash as in to ruin something (dash hopes), dash as in to move quickly.
  • dash may be needed and added later to a natural language processing system. For instance, dash as in to splatter, or dash as in a short sudden movement.
  • Prior art uses grammar or statistics to disambiguate polysemy.
  • Prior art using a grammar analysis to choose the correct meaning of ‘dash’ based in grammar can only determine whether a noun or verb meaning is best. Such a system would not disambiguate whether dash means a Morse code or a small quantity of something.
  • Prior art using statistics of usage can only choose a meaning which was chosen most of the time in a context. Thus statistical methods disable any other meanings and prevent the acquisition of new meanings for a symbol within any context.
  • Human linguistic abilities are clearly less limited, in addition to being more accurate than prior art. Besides acquiring new meanings for words within a conversational context, humans also create new meaningful contexts through self-organizing linguistic abilities. For instance, as humans converse about subjects, subjects acquire new semantic meanings which evolve into new conversational contexts. In contrast, prior art computer systems only acquire new semantic contexts through laborious programming and data-entry efforts. The encumbrance of such programming and data-entry prevents prior art computer systems acquiring semantic knowledge on a real-time basis during natural language conversations with humans. Although a large number of programming languages have been created for inputting semantic knowledge, none of which have the flexibility and general utility of a natural language such as English. Languages such as Prolog, SQL or Lisp cannot match the convenience of conversing in plain English or other natural languages.
  • Context is impractical to define logically because logic requires enumeration of logical inputs and outputs.
  • the number of contexts which can be defined by a natural language input is limited only by every possible shade of meaning of every symbol which occurs in every possible natural language sentence. Since such a large number of combinations cannot be enumerated, logical natural language processors store a subset to the full set of possible contexts as a logical approximation: Each natural language symbol is stored with its own set of logical data, and with rules for combining its logical data with other symbols.
  • N For a semantic dictionary of N symbols, in a language where the maximum number of symbols strung together is M, the number of logical combinations is N to the M power. For a vocabulary of 1,000 words combined in short four word sentences, the number of logical combinations is 1,000 raised to the fourth power, which equals 1,000,000,000,000. If a team lexicographers attempts to define and test a semantic dictionary of this small size there would be 100,000,000,000 testing hours required if each test takes 1/10 th of an hour. If 500 testers each work 2000 hours a year, the team can work 1,000,000 hours per year, and the testing will be complete in 100 years. By that time, the dictionary will surely be obsolete and require re-testing. For longer sentences and larger dictionaries, this drawback quickly grows exponentially worse.
  • a natural language processing system must track shifts in overall meaning of phrases to remain accurate. For instance, the meaning of a celebrity's name shifts as that name is used in major news reports, particularly if their fame is new. To logically represent such shifts in meaning, the rules describing how to combine a celebrity's name in various contexts must be extended to handle each new conversational use of the celebrity's name explicitly. Using logical methods, all possible combinations of phrases and contexts must be defined and tested.
  • the present invention is a computer implemented data processor system to automatically disambiguate a contextual meaning of natural language input streams of symbols, stored as new input stream nodes within a semantic network, whose various candidate meanings are stored as nodes in the semantic network, and whose natural language context is stored as a set of context nodes in the semantic network.
  • the present invention can automatically choose candidate nodes closest to the context node set to represent a contextual meaning of the input stream.
  • the present invention can create new semantic inheritance links between the input stream nodes and the candidate nodes closest to the context nodes. These new links create precise records of contextual meaning to be stored for later retrieval by subsequent natural language queries.
  • the present invention thus enables the semantic network to function as a natural language database by storing data as semantic network nodes, with all semantic categories needed for accurate natural language database design automatically created and linked entirely from natural language input streams.
  • the present invention thus automatically provides precise storage and retrieval of detailed, intersecting levels of information, comparable to relational database storage and retrieval capability, without the labor cost and inconvenience of designing a database, to enable flexible and efficient natural language interfaces to computers, household appliances and hand-held devices. Additional advantages and novel features of the invention will be set forth in part in the description that follows, and in part will become more apparent to those skilled in the art upon examination of the following or upon learning by practice of the invention.
  • FIG. 1 shows a natural language processing system deployed on a central computer shared by a variety of customer support staff and remote customers connected by Internet connections, in accordance with an embodiment of the present invention.
  • FIG. 2 shows a natural language processing system deployed on computers connected by peer-to-peer Internet connections, to provide technical support to customers, in accordance with an embodiment of the present invention.
  • FIG. 3 shows a natural language processing system deployed on a cellphone connected to the Internet, to provide a natural language user interface for dialing, storing and retrieving names, phone numbers from personal address book information, in accordance with an embodiment of the present invention.
  • FIG. 4 shows a natural language processing system deployed on a microwave oven home appliance, to provide a natural language user interface for operating oven controls, in accordance with an embodiment of the present invention.
  • FIG. 5 shows a natural language processing system having inputs, storage, and outputs in the form of images, sounds and texts, in accordance with an embodiment of the present invention.
  • FIG. 6 shows a database system having inputs of SQL commands, a relational database processor, and output in the form of images, sounds and texts, in accordance with an embodiment of the present invention.
  • FIG. 7 shows the database system of FIG. 2 . with a natural language processor added for converting natural language text input to SQL commands, in accordance with an embodiment of the present invention.
  • FIG. 8 shows an example of an SQL database processing a query corresponding to the natural language request “fascinating art not ugly,” in accordance with an embodiment of the present invention.
  • FIG. 9 shows a simple image processing method for identifying emotional content of face expressions by analyzing the contours of the mouth, in accordance with an embodiment of the present invention.
  • FIG. 10 show an overview of sound processing methods to identify words from sounds by analyzing phonemes comprised of transitions from one frequency bands to another, in accordance with an embodiment of the present invention.
  • FIG. 11 shows a semantic network storing the logical relations of FIG. 8 as semantic nodes and links, in accordance with an embodiment of the present invention.
  • FIG. 12 shows a method for combining topological categorization with topological contextual filtering, in accordance with an embodiment of the present invention.
  • FIG. 13 shows a table mapping from filter types of FIG. 12 to filter actions which screen out input nodes of FIG. 12 , in accordance with an embodiment of the present invention.
  • FIG. 14 shows a table mapping from topological categorization input category types to their corresponding semantic network traversal functions, in accordance with an embodiment of the present invention.
  • FIG. 15 shows a table mapping from topological contextual filtering input context types to their corresponding best contextual distance functions, in accordance with an embodiment of the present invention.
  • FIG. 16 shows an example of parsing the natural language text “fascinating art” using the method of FIG. 12 , in accordance with an embodiment of the present invention.
  • FIG. 17 shows an example of parsing “fascinating art not ugly” using the method of FIG. 12 to filter the output of FIG. 16 , in accordance with an embodiment of the present invention.
  • FIG. 18 shows an example of a method applicable to step 7 of FIG. 16 to assess a visual response, in accordance with an embodiment of the present invention.
  • FIG. 19 shows an example of a method applicable to step 7 of FIG. 17 to assess a visual response, in accordance with an embodiment of the present invention.
  • FIG. 20 shows an example of a method for disambiguating the meaning of “ugly” as defined in the semantic network of FIG. 11 from a combination of text and image inputs, in accordance with an embodiment of the present invention.
  • FIG. 20 shows an the example of a method in FIG. 20 for disambiguating the meaning of “ugly” as defined in the semantic network of FIG. 11 , from a different combination of text and image inputs, in accordance with an embodiment of the present invention.
  • FIG. 22 shows a method for combining topological categorization with topological contextual filtering for connecting phrases to complementary phrases within a parse tree, in accordance with an embodiment of the present invention.
  • FIG. 23 shows a small semantic network dictionary for storing semantics related to conversations about ordering food in restaurants, in accordance with an embodiment of the present invention.
  • FIG. 24 shows a parse tree created from a sentence “She would like salad and not steak and potatoes, in accordance with an embodiment of the present invention.
  • FIG. 25 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 23 to process the parse tree of FIG. 24 , in accordance with an embodiment of the present invention.
  • FIG. 26 shows the parse tree of FIG. 24 after processing by the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 23 , in accordance with an embodiment of the present invention.
  • FIG. 27 shows a parse tree created for the sentence “She would like salad not steak and blue cheese dressing,” in accordance with an embodiment of the present invention.
  • FIG. 28 shows the parse tree of FIG. 27 after processing by the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 23 , in accordance with an embodiment of the present invention.
  • FIG. 29 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 23 to process the parse tree of FIG. 27 , in accordance with an embodiment of the present invention.
  • FIG. 30 shows the semantic network dictionary of FIG. 23 after augmentation with nodes from the parse tree of FIG. 28 , in accordance with an embodiment of the present invention.
  • FIG. 31 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 30 to parse “She would like a lot of cheese,” in accordance with an embodiment of the present invention.
  • FIG. 32 shows the semantic network dictionary of FIG. 23 after augmentation with nodes from a parse tree created by parsing “She would like a lot of,” in accordance with an embodiment of the present invention.
  • FIG. 33 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 32 to disambiguate the meaning of ‘cheese,’ in accordance with an embodiment of the present invention.
  • FIG. 34 shows a small semantic network dictionary for storing semantics related to pronouns, in accordance with an embodiment of the present invention.
  • FIG. 35 shows a parse tree created for the sentences “Amy should be here. Harold said her food was ready,” in accordance with an embodiment of the present invention.
  • FIG. 36 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 34 to disambiguate the meaning of ‘her’ in the parse tree of FIG. 35 , in accordance with an embodiment of the present invention.
  • FIG. 37 shows a parse tree created for the sentences “Amy should be here. Harold said his food was ready,” in accordance with an embodiment of the present invention.
  • FIG. 38 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 34 to disambiguate the meaning of ‘his’ in the parse tree of FIG. 37 , in accordance with an embodiment of the present invention.
  • FIG. 39 shows a small semantic network dictionary for storing semantics related to the transitive relation of ‘bigger than,’ in accordance with an embodiment of the present invention.
  • FIG. 40 shows a variation of the method of FIG. 12 , amended to categorize nodes by spelling and by a complex traversal function, to disambiguate the meaning of “Chicago bigger than a bread box?”, in accordance with an embodiment of the present invention.
  • FIG. 41 shows the method of FIG. 40 used in conjunction with the semantic network dictionary of FIG. 39 to disambiguate the meaning of ‘Chicago’ in “Chicago bigger than a breadbox?”, in accordance with an embodiment of the present invention.
  • FIG. 42 shows an example of a dialog between a person and natural language processor where the natural language processor remembers and retrieves dialog semantic information, in accordance with an embodiment of the present invention.
  • FIG. 43 shows an example of a method to create semantic is-a type dictionary links from parse phrase trees containing ‘is’ and ‘a’ phrasing, in accordance with an embodiment of the present invention.
  • FIG. 44 shows a small semantic network for storing semantics related to personal address book information, in accordance with an embodiment of the present invention.
  • FIG. 45 shows the semantic network of FIG. 44 augmented by the first sentence of the dialog in FIG. 42 , after applying step 1 of FIG. 43 , in accordance with an embodiment of the present invention.
  • FIG. 46 shows the semantic network of FIG. 45 after applying step 2 of FIG. 43 to create an is-a link between programmer: 2 and Hal: 2 , in accordance with an embodiment of the present invention.
  • FIG. 47 shows the semantic network of FIG. 46 after applying step 2 a of FIG. 43 to remove redundant semantic information, in accordance with an embodiment of the present invention.
  • FIG. 48 shows the semantic network of FIG. 47 after applying step 4 after step 3 of FIG. 43 to add semantic information from the second and third sentences of FIG. 42 , in accordance with an embodiment of the present invention.
  • FIG. 49 shows the semantic network of FIG. 48 after applying step 2 of FIG. 43 to copy the new phrase tree's “good trumpet player” portion to be inherited by an is-a link by Hal: 2 , in accordance with an embodiment of the present invention.
  • FIG. 50 shows the semantic network of FIG. 49 after applying step 2 a of FIG. 43 to remove redundant semantic information, in accordance with an embodiment of the present invention.
  • FIG. 51 shows an example of a method to detect and answer a question encountered in a dialog input phrase tree, in accordance with an embodiment of the present invention.
  • FIG. 52 shows an example of a table mapping from an input phrase tree's input node types to allowable links and traversal starting points of step 2 in FIG. 51 , in accordance with an embodiment of the present invention.
  • FIG. 53 shows the semantic network of FIG. 50 after adding the input phrase tree from the question “Who is a good musician friend,” in accordance with an embodiment of the present invention.
  • FIG. 54 shows the method of FIG. 50 applied to answer the input phrase tree question in FIG. 53 , in accordance with an embodiment of the present invention.
  • FIG. 55 shows an example of a dialog in which an answer to question has been answered affirmatively, indicating that the answer would be useful to store in a semantic network, in accordance with an embodiment of the present invention.
  • FIG. 56 shows an example of a method to stored the confirmed answer to a question, using the inversion method of FIG. 43 , in accordance with an embodiment of the present invention.
  • FIG. 57 shows an example of a table of input phrase tree's input node types and corresponding connecting links used to connect interrogative nodes in step 2 of FIG. 56 , in accordance with an embodiment of the present invention.
  • FIG. 58 shows the method of FIG. 56 applied to the semantic network of FIG. 53 to store the confirmed answer to the question “Who is a good musician friend,” in accordance with an embodiment of the present invention.
  • FIG. 59 shows the semantic network of FIG. 53 after step 2 of FIG. 56 has added a new is-a link between Who: 2 and Hal: 2 , in accordance with an embodiment of the present invention.
  • FIG. 60 shows the semantic network of FIG. 59 after step 3 of FIG. 56 has applied the inversion method of FIG. 43 to invert the input phrase tree for “good musician friend,” in accordance with an embodiment of the present invention.
  • FIG. 61 shows a method for identifying nodes in a candidate node set having a minimum average distance between a candidate node and a context node set within a semantic network, in accordance with an embodiment of the present invention.
  • FIG. 62 shows a method for identifying nodes in a candidate node set having a maximum average distance between a candidate node and a context node set within a semantic network, in accordance with an embodiment of the present invention.
  • FIG. 63 shows a method for identifying nodes in a candidate node set having a minimum distance between a candidate node and a context node set within a semantic network, in accordance with an embodiment of the present invention.
  • FIG. 64 shows a method for identifying nodes in a candidate node set having an infinite minimum distance between a candidate node and a context node set within a semantic network, in accordance with an embodiment of the present invention.
  • topological methods disambiguate polysemy by measuring the relative distance between symbols in a semantic network.
  • the distance between symbols is proportional to the closeness in meaning between symbols. Consequently the contextual meaning of a symbol can be found by locating the closest symbol to a set of symbols defining the context.
  • Both new symbols and new meanings can be recorded simply by adding new symbol nodes to existing topologies. Shifts in meaning can be recorded simply by shifts in topologies.
  • the first problem is that the topological distance between a pair of symbol nodes in a semantic network may not reflect closeness in semantic meaning between the pair of symbol nodes. This problem can be avoided by adjusting automatically adjusting distance between each pair of nodes to accurately reflect semantic meaning by regrouping, adding, and removing nodes. Relevant methods for automatically regrouping, adding and removing nodes are described in patent application 09/085,830 “topological methods to organize semantic network data flows for conversation applications.”
  • the second problem with topological methods occurs when representing and disambiguating the meaning of negation.
  • some semantic path will generally connect the ‘A’ and candidate symbols.
  • closeness of the ‘A’ to candidate symbols cannot represent closeness to desired meaning. Rather, it represents closeness to opposite meaning. Therefore, when seeking a candidate symbol which is NOT ‘A’, some other metric besides closeness must be applied to find appropriate candidate symbols.
  • a person might request ‘fascinating art’ from an information retrieval system.
  • topological methods such as the method of ‘most specific common abstraction’ (Kolodner, pages 346-347)
  • a retrieval system could find nodes closest to ‘fascinating art’ by measuring semantic path distances from every candidate node to both ‘fascinating’ and ‘art’.
  • the retrieval system would return candidate nodes having minimal distance to both ‘fascinating’ and ‘art’.
  • logical methods employ sets of rules to identify sets of candidate nodes preferred for a given logical context, and other sets of rules to rank the set of candidate nodes for the given logical context.
  • Prior art logical methods have been preferred for their processing speed and economy with storage space. However such advantages have lost relevance as computer processors has become faster and computer memory becomes cheaper.
  • logic have cannot handle the combinatorial complexity of natural language when there are more than a thousand symbol nodes. In addition, logic cannot resolve contradictions in meaning which arise naturally in general-purpose natural language systems.
  • a person might request ‘fascinating art that is not ugly’ from an information retrieval system.
  • the system might find, categorized under art, a set of candidate nodes. For instance, there might be two such nodes.
  • the first semantic node ‘stained glass statues of people’ might inherit from ‘art’, ‘beautiful’ and ‘stained glass’.
  • the second semantic node ‘dancers made of children's blocks’ might inherited from ‘art’ and ‘strange’.
  • a first such rule might be that inheritors of ‘beautiful’ are not ugly.
  • a second such rule might be that toxic substances are ugly. Since ‘stained glass’ might inherit from ‘lead’ which in turn might inherit from ‘toxic substance’, the first rule might conflict with the second rule. To resolve such conflicts, semantic rules must be ranked to determine which rule about ‘ugly’ is more significant.
  • the relative distances between nodes are used in place of logic to resolve logical contradictions.
  • contexts can be defined with various degrees of focus.
  • General contexts can be defined by a few high-level nodes, and specific, narrowly focused contexts can be defined by a large number of low level nodes.
  • logical contradictions can be resolved by topology recorded in the local neighborhood of that narrow context.
  • fuzzy logic algorithms were developed to locally resolve logical contradictions by using statistical functions. Since some things can be both beautiful and ugly, there is a probability that beautiful implies not ugly, and this is compared to the probability of the toxic quality of lead implies ugly.
  • statistical functions can resolve logical contradictions between frequently competing candidate nodes, they cannot resolve logical contractions for new descriptions or old descriptions relative to new semantic nodes, because, as mentioned above, all statistics require a prior set of trial events upon to base a statistical distribution. If ‘fascinating art that is not ugly’ was a new request, there would be no events upon which to base a statistical distribution and statistical methods would be of no use.
  • relational databases enable large amounts of data to be searched quickly and efficiently.
  • translating requests into the form of logical operators is laborious, and translating data into the form of logically correct statements is even more laborious, if an application contains a limited variety of data, the translation can be completed within a reasonable budget.
  • databases grow to become data warehouses they require considerable manual intervention to maintain their logical consistency as new data is added. The budget for such maintenance can often become unaffordable. For large databases such as Internet web search engine portals, such maintenance costs run into millions of dollars.
  • the adjective ‘attractive’ is commonly attributed to the noun phrase ‘movie star’.
  • some movie stars base their appeal upon a kind of curiosity with unattractiveness, by playing villainous or anti-heroic roles.
  • the meaning of attractive might have to be confined to a star's “ability to attract box-office sales”.
  • a natural language processor must efficiently store criteria for differentiating between the shades of meaning which might be attributed to the word ‘attractive’, even if these shades of meaning cannot be immediately mapped to a logical context.
  • Natural language processing systems must also disambiguate the meaning of requests containing logical prepositions. In such cases, logic must be combined with contextual semantic meanings to determine which noun phrase to link to a prepositional phrase. Thus the disambiguation of meaning of natural language logical prepositions requires a precise method for combining contextual shades of meaning which themselves may not be logically defined.
  • it is especially important to accurately respond to the meaning of such prepositions. For instance, ‘and’ and ‘or’ as well as ‘from’, ‘to’, and ‘after’ are important prepositions used to zero in on specific information.
  • the meaning of prepositions vary in subtle ways within natural language discourse.
  • SQL Structured Query Language
  • Methods for improved detection of contextual and prepositional meaning in natural language processing systems would enable a broad range of applications where a user interface on a direct human conversation level would be preferred to a menu-based or graphical user interface.
  • help desk services handle a large volume of customer requests. Such requests must often by handled by a human dispatcher for a significant cost on labor, or an automated menu-driven voice-response system which is inexpensive but difficult for customers to interact with a user interface on a direct human conversation level would be preferred to voice-response systems which force customers to navigate a menu tree.
  • Simple appliances such as washing machines and microwave ovens could also benefit from imbedded natural language processors to accept direct conversational commands.
  • More complex appliances such as cellphones could have imbedded natural language systems to complement or even supplant menu-driven screen displays.
  • Topological methods for semantic networks measure distances between nodes by counting the number of links along paths between nodes. In order for these distances to be meaningful, the number of links along paths between nodes must correspond to actual closeness of semantic meanings.
  • U.S. patent application Ser. No. 09/085830 by Au describes methods for enforcing such correspondence. Enforcing such correspondence ensures that topological distances are a reliable indicator of closeness of meaning between any two semantic network nodes.
  • a semantic network representing closeness in meaning as distances records the closeness in meaning for any two nodes in the semantic network.
  • each path through the semantic network records a picture of why two nodes are a particular distance. For instance, in a small hierarchy of nodes in which retriever, beagle and puppy inherit from dog, the path from retriever to retriever to beagle has to two links. If dog in turn inherits directly from mammal, then retriever, beagle and puppy are each two links away from mammal. The inheritance links attached to dog are thus in turn inherited by retriever, beagle and puppy, thus automatically recording any closeness in meaning added to dog on behalf of retriever, beagle and puppy.
  • the word retriever could be linked to inherit from another meaning for retriever which has a more abstract meaning, that of something which fetches an object.
  • the shorter distance between meanings can be used to disambiguate meaning. For instance, for the second sentence, the distance between puppy and the concrete meaning of retriever is only two links, whereas the abstract meaning of retriever is three links away. For the second sentence, the more concrete meaning of retriever would thus be chosen.
  • each word in a sentence relates to a set of paths through a semantic network.
  • the multiple words in a sentence can be used as a redundant pointers to the most appropriate meaning of each word in the sentence.
  • This semantic redundancy models the human ability to ignore noise and grammatical errors in natural language discourse.
  • the present invention stores and retrieves the meaning of negation as the relative absence of a close connection.
  • the absence of a close connection is computed relative to a contextual set of nodes.
  • traversal functions for traversing between nodes can be used to constrain search to a requested category.
  • Such topological methods can identify a node closest to a particular context, even if that context is a combination of many individual nodes.
  • the average or minimum distance to the set of nodes can be used to measure closeness.
  • the present invention combines topological methods for categorization with topological methods for contextual filtering, to disambiguate the meaning of negated symbols, as well as symbols relative to other prepositions such as ‘and’ and ‘or’.
  • topological methods for categorization to disambiguate the meaning of negated symbols, as well as symbols relative to other prepositions such as ‘and’ and ‘or’.
  • By constraining contextual searches to specific categories precise logical relationships implied by semantic networks can be accurately detected.
  • precise logical functionality of database languages such as SQL can be provided by the present invention while eliminating the cost of maintaining data-relationships and queries by hand.
  • semantic networks can be used in place of relational databases, to bypass the tedious and error-prone process of translating natural language data into data relations.
  • fine shades of contextual meaning can be automatically stored for each phrase-context pair without explicitly defining a relational join for each phrase-context pair.
  • Another advantage of using a natively natural language database comes from the elegance with which new meanings can be stored directly as parse trees of conversational input. Even while parsing a conversational flow of natural language, new shades of semantic meaning can be stored. Phrases encountered in the beginning of a paragraph can be stored as new shades of meaning which help disambiguate the meaning of phrases encountered later in a conversation.
  • the present invention also contains methods to transform meanings encoded as parse trees into standard semantic inheritance trees, thus automatically updating shades of meaning during conversational input with a user.
  • FIG. 1 shows a fully networked support system for customer support, in which various supporting activities are coordinated by a natural language processor in a shared central computer, in accordance with one embodiment of the present invention.
  • FIG. 1 At the top of FIG. 1 is a customer, shown on the left, interfacing through a camcorder (or web-cam) to a networked personal computer. In this manner, a customer's image and verbal requests can be relayed across the Internet to a private network, shown at the middle of FIG. 1 . Other customers may input verbal request over devices such as the cellphone and telephone shown in FIG. 1 .
  • requests When requests reach the private network, they could be monitored and handled by support staff, also equipped with camcorder (or web-cam) to assist customers over a kind of a video-phone connection. Alternatively, requests might be dispatched to the appropriate support staff by a natural language processor, as shown in FIG. 1 .
  • Supporting functions which may be orchestrated by a natural language processor as shown in FIG. 1 include scanning of archives, stored in micro film and requesting attention from operators of archives (all shown in the lower left of FIG. 1 ). Archives may include email messages which can be scanned in terms of their natural language meaning, to correlate previous current requests with previous requests and responses. For instance, requests for explanation of bills may be correlated to billing records and emails described returned merchandise.
  • requests for news clippings could be correlated with previous requests for specific kinds of news of interest and requests to ignore other kinds of news.
  • Other applications include matching of open job descriptions to resumes and matching of email messages to topics requested by email list subscribers, and indexing of email archives by topic.
  • a natural language processor to orchestrate ail of these functions could greatly increase the efficiency of customer service for the above businesses.
  • FIG. 2 shows a natural language processing system deployed on computers connected by peer-to-peer Internet connections, to provide technical support to customers, in accordance with one embodiment of the present invention.
  • a conversational natural language interface can be supported by a natural language processor in each computer, to exchange semantic information across the Internet as shown.
  • a camcorder or web-cam
  • a camcorder or web-cam
  • a camcorder is used to store pre-recorded images and sounds for playback as orchestrated by the natural language processor in the lower computer.
  • the images and sounds sent back to the user can be synthesized by either the lower or upper computer as needed.
  • Peer-to-peer natural language computer systems with these configurations could be useful ways to interactively share and disseminate information that otherwise would have to be interactively disseminated in more labor-intensive lectures, workshops and video-conference sessions.
  • FIG. 3 shows a natural language processing system deployed on a cellphone connected to the Internet, to provide a natural language user interface for dialing, storing and retrieving names, phone numbers from personal address book information, in accordance with one embodiment of the present invention.
  • a natural language processing system deployed on a cellphone connected to the Internet, to provide a natural language user interface for dialing, storing and retrieving names, phone numbers from personal address book information, in accordance with one embodiment of the present invention.
  • Such a systems would be capable of searching, downloading and uploading information on the Internet, as well as monitoring the progress of requests to buy, sell, or inquire about merchandise or services on the Internet.
  • Using a natural language interface bypasses the difficulty in learning menu-driven and graphical user interfaces for such activities.
  • FIG. 4 shows a natural language processing system deployed on a microwave oven home appliance, to provide a natural language user interface for operating oven controls.
  • a natural language interface bypasses the difficulty in learning menu-driven and graphical user interfaces for operating such appliances, as well as provides a convenient way to report status of operations back to the user.
  • an appliance with such a natural language interface can adapt to a style of conversation convenient for individual users.
  • the present invention uses natural language reasoning, parsing and pattern matching to process natural language in a natural language format.
  • Components of natural language such as images, sound and text are inputs which can be correlated to semantic meanings by the present invention.
  • FIG. 4 shows these inputs in the upper part of the figure, in accordance with one embodiment of the present invention.
  • the present invention can then retrieve the stored versions of the inputs using uses natural language reasoning, parsing and pattern matching.
  • FIG. 4 shows these storage and retrieval operations as the pair of arrows in the middle of the figure. The confirmation and results of these operations are produced as output such as images, sound and text.
  • FIG. 4 shows these inputs in the lower part of the figure.
  • FIG. 5 shows these structures as inputs in the upper part of the figure, in accordance with one embodiment of the present invention.
  • FIG. 5 shows data relationships as curved lines labeled as ‘relationship-a’, relationship-b’, relationship-c’.
  • the present invention bypasses the need for logical consistency by representing data in a semantic network where meaning is recorded redundantly by multiple paths over semantic links connecting semantic network nodes.
  • FIG. 5 shows the database system of FIG. 5 , but showing in the upper left hand corner of FIG. 6 a natural language processor similar to FIG. 4 .
  • This architecture allows the general public to access data, but not to maintain the database. When tables or relationships must be added to the database, such modifications must be done by people skilled in formal logic and programming languages.
  • FIG. 7 An example of the complex and esoteric structure required by the prior art to process simple natural language requests is shown in FIG. 7 .
  • a prior art system as shown in FIG. 7 might get a request for “fascinating art not ugly”.
  • this simple phrase would translate into the SQL query shown in the upper part of FIG. 8 .
  • the query comprises three clauses: a ‘select’ clause, a ‘from’ clause, and a ‘where’ clause.
  • the beginning of each clause is shown in bold.
  • the processing order of SQL clauses is ‘from’ clause, then ‘where’ clause, then ‘select’ clause.
  • the ‘from’ clause requests data from two tables, art and artAdjective.
  • the ‘select’ clause selects values from two columns from the art table, ‘num’ and ‘artDescription’. These are displayed for the joined row returned by the query, shown in the bottom of FIG. 8 .
  • FIG. 11 shows a small semantic network dictionary, in accordance with one embodiment of the present invention, for representing the nouns and adjectives of FIG. 8 .
  • the semantic network of FIG. 11 succinctly maps adjectives like ‘fascinating’ and ‘ugly’ to noun phrases like “stained glass statues of people.” Since semantic links can be used to store various relationships such as ‘is-a’, ‘part-of’ and ‘because-of’, most semantic networks label the link lines with similar labels.
  • FIG. 11 shows a small semantic network dictionary, in accordance with one embodiment of the present invention, for representing the nouns and adjectives of FIG. 8 .
  • FIG. 11 shows each ‘is-a’ links with a thin dark line, and each link of any other type with a fat gray line.
  • ‘is-a’ links are strict categorization links. For instance, ‘art’ is a ‘noun’, and ‘desirable’ is an adjective.
  • the gray links are less strict in meaning.
  • lead is linked to stained glass. Strictly speaking, lead ‘is-a-component-of’ stained glass, along with tint and glass.
  • FIG. 11 does not label each direct link above stained glass with ‘is-a-component-of’ because of the clutter that would cause. Instead, simple gray lines are used to show that some aspect of lead, tint, and glass informs the nature of stained glass.
  • semantic network can be implemented with just two link types, one for storing ‘is-a’ links, and the other for storing ‘informs’ links.
  • FIG. 9 shows a high level overview of a prior art image processor for mapping face expressions to semantic nodes signifying a person's emotions, in accordance with one embodiment of the present invention.
  • the mapping takes place in four steps.
  • the first step abstracts information from the pixels of an image by tracing simple contours around the mouth.
  • the second step abstracts information from contour shapes by measuring the direction and magnitude of their curvature.
  • the last step abstracts information from contour directions and magnitudes by mapping from classes of contour directions and magnitudes to specific semantic nodes of “happy face”, “sad face”, “quizzical face” and “angry face”.
  • FIG. 10 shows an example of a the prior art sound processor design for mapping sounds to words, in accordance with one embodiment of the present invention.
  • sounds are input as air pressure amplitude fluctuations over time.
  • bands of vibration frequencies are identified from fluctuations in amplitude by a fast Fourier transformation. Each band is identified as a range of frequencies defined in cycles per second also known as Hertz (HZ). The persistence of each detected band of frequencies is mapped over time by the fast Fourier transformation.
  • HZ Hertz
  • the identification of phonemes is abstracted from specific combinations of frequency bands. For instance, ‘sh’ is a broad band of frequencies, whereas the consonants ‘o’ and ‘a’ are narrower bands.
  • the identification of word nodes is abstracted from sequences of phonemes.
  • the semantic network of FIG. 11 stores both the semantic image nodes from FIG. 9 and semantic word nodes FIG. 10 .
  • the semantic network of FIG. 11 also stores implied logical relationships. These relationships are implied by the semantic links. For instance, ‘fascinating’ is linked via an inheritor link to “stained glass statues of people”. This stores the same meaning as the row of artAdjective table “30—fascinating” joined by relation-b to the row of art table “3-30—stained glass statues of people” in FIG. 8 .
  • the method of FIG. 12 combines topological categorization with topological contextual filtering.
  • the method of FIG. 12 starts by automatically mapping inputs to specific semantic nodes in a semantic network, in accordance with one embodiment of the present invention.
  • FIG. 12 step 1 automatically identifies any existing semantic nodes which can be which can be mapped to input data, as well as creating new semantic nodes for input which cannot be mapped to existing semantic nodes.
  • the semantic nodes identified by step 1 are presented to filters which determine what role each identified node can play in steps 2 and 3 of the method.’ Filters which supply nodes for these roles are shown as ‘traversal type filter’, ‘category node filter’, ‘context type filter’, and ‘context node filter’.
  • FIG. 13 is a table showing acceptable nodes for each filter type, at various stages of a natural language conversation, in accordance with one embodiment of the present invention.
  • FIG. 13 For the start of a conversation, FIG. 13 has rows where stage is marked ‘start’.
  • stage For mid-conversation input, when dialog has already led to some agreement, FIG. 13 has rows where stage is marked ‘middle’. This allows filtering to shift to take advantage of information which emerges about a conversation, as a conversation develops.
  • the transition from start to middle of the conversation can be automatically detected when the context node set grows to an optimal size, typically about seven nodes.
  • the context node set can be diminished or entirely flushed, to transit the conversational state from middle back to start.
  • FIG. 13 defines the filters themselves in terms of the semantic network of FIG. 11 .
  • ‘traversal type filter’ accepts only nodes which are traversal types.
  • the semantic network of FIG. 11 shows three nodes which inherit from ‘traversal type’: ‘category hierarchy’, ‘general hierarchy’ and ‘list of nodes’. These three nodes would thus be the only nodes acceptable by the traversal type filter.
  • the category set filter accepts only nodes which inherit from ‘adjective’ or ‘adverb’ in FIG. 11 .
  • Default values are indicated for ‘input traversal type’ and ‘input context type”. In case no nodes pass through the filters to perform these roles, default nodes are substituted. For instance, if no input traversal type nodes are found in the input, the default node ‘general hierarchy’ substitutes for the input traversal type node. Similarly, if no input context type nodes are found in the input, the default node ‘and’ substitutes for the input context type.
  • a priority-based dispatching mechanism could be used as an alternative to filtering.
  • a dispatching mechanism would be more complex to maintain but could be more accurate. For instance, if the symbols ‘fast run’ were input, fast could either be an adverb (fast: 1 ) for the input category set or a verb (fast: 2 ) for the context node set. On the other hand, run could either be an noun (run: 2 meaning a small brook or stream) or a verb (run: 1 ) both for the context node set. To identify both an input category set and a context node set, fast could be identified with fast: 1 and run identified with run: 1 or run: 2 . The dispatcher could reject the alternative of identifying fast with fast: 2 because this would prevent identification of an input category set.
  • both an input traversal type node and an input category node set must be elicited from step 1 .
  • a default traversal type may be substituted for step 2 instead.
  • FIG. 12 shows a default traversal type of ‘general hierarchy’.
  • other default traversal types may be chosen instead, such as ‘category hierarchy’ when semantics are more formal.
  • Step 2 of FIG. 12 automatically collects a set of candidate nodes, traversing from an input category node set according to traversal functions described in FIG. 14 , in accordance with one embodiment of the present invention.
  • FIG. 14 shows a corresponding traversal function.
  • the default traversal type ‘general hierarchy’ corresponds to the traversal function ‘any inherited links’.
  • the contextual meaning chosen by the overall of FIG. 12 . is limited to the set of candidate nodes returned by the traversal function.
  • Step 3 of FIG. 12 automatically computes a best contextual distance for each candidate node, given a context type passed from the context type filter.
  • the default context type is ‘and’ which in FIG. 15 . corresponds to a best contextual candidate function of ‘minimum average distance.’
  • Step 3 uses this best contextual candidate function to choose a subset from the set of candidate nodes returned by step 2 .
  • FIG. 15 is an example of a table of other context types and their corresponding best contextual distance functions, in accordance with one embodiment of the present invention.
  • the input context type ‘and’ corresponds to a best contextual distance of ‘minimum average distance’.
  • a minimum average distance can be calculated from the topology of a semantic network, by determining the shortest path, in semantic links, between each candidate node and each context node, then averaging those shortest paths and choosing a subset of nodes having minimum average shortest path length to context.
  • FIG. 61 is a flowchart showing how minimum average distance can be calculated, in accordance with one embodiment of the present invention. Step 3 a of FIG. 61 is a loop which iterates through each node of the candidate node set.
  • step 3 a For each pass through that loop, the shortest path from a single candidate node to each node of the context node set is identified.
  • step 3 a calculates the average length of those shortest paths and stores that average number for the candidate node.
  • Step 3 a then returns to the top of the loop to process the next candidate node until all candidate nodes have been processed.
  • Step 3 b determines the minimum average shortest path length recorded by Step 3 a over all candidate nodes, and chooses the subset of candidate nodes having that minimum average shortest path length to context. By choosing the minimum average, the method of FIG. 62 finds the closest nodes to context, in accordance with one embodiment of the present invention.
  • FIG. 62 is a flowchart showing how maximum average distance can be calculated. This calculation is identical to minimum average distance except for the last step.
  • step 3 b of FIG. 62 instead of determining the minimum average shortest path length recorded by FIG. 61 step 3 b over all candidate nodes, FIG. 62 step 3 b determines the maximum average shortest path distance recorded over all candidate nodes. By choosing the maximum rather than minimum average, the method of FIG. 62 finds the farthest nodes from context, to satisfy requests for symbols which are NOT like the context.
  • FIG. 63 is a flowchart showing how minimum distance can be calculated, in accordance with one embodiment of the present invention.
  • Step 3 a of FIG. 63 records only the shortest path length between a candidate node and of any node of the context node set, rather than the average shortest path length between a candidate node and all nodes of the context node set as is done in step 3 a of FIG. 61 and FIG. 62 .
  • the method of FIG. 63 emphasizes candidate nodes which are close to some context nodes even if they are very far from other context node.
  • step 4 a identifies the candidate nodes having the shortest path length to any context node.
  • FIG. 64 is a flowchart showing how infinite minimum distance can be calculated, in accordance with one embodiment of the present invention.
  • Step 3 a of FIG. 64 determines only if there is any path from a candidate node to any context node, unlike step 3 a of FIG. 62 which determines the shortest path from a candidate node to each context node.
  • Step 4 b of FIG. 64 chooses only candidate nodes which have no path to any context node, unlike step 4 b of FIG. 62 which chooses the candidate nodes having maximum average path to context nodes.
  • FIG. 64 step 3 thus chooses a subset of the nodes chosen by FIG. 62 .
  • Best contextual distance functions can also be made more stringent or more lenient by different methods of tallying distances along paths between candidate nodes and context nodes. Different types of inheritance links can be given different weightings when counting the links in a path to tally the distance along the path. For instance, in FIG. 15 , the input context type ‘and equivalent’ has the best contextual distance function ‘minimum average distance, not counting is-a links.’ This best contextual distance function would be identical to that outline by FIG. 61 , except that when computing the candidate node's average shortest path length, is-a type inheritance links would not be counted, so that they would have no effect on the path length. By not counting is-a links, all nodes in a is-a inheritance hierarchy are considered equivalent to each other in meaning, for the context type ‘and equivalent’. Such a distance function would thus measure contextual equivalency of meaning of candidate nodes.
  • Step 3 and 4 of FIG. 12 can be automatically performed by any of the contextual distance functions suggested above.
  • step 5 of FIG. 12 can generate a summary of that subset for presentation to a user of the computer system.
  • the user's response to that presentation can be acquired, to be assessed in step 7 of FIG. 12 .
  • the assessment can be made using a contextual distance function as outlined above by FIG. 61, 62 , 63 or 64 .
  • the assessment step categorizes which of the four possible categories of response the user's response best matches: deny, ask, confirm, or complain.
  • FIG. 12 shows branches leading from step 7 for each of these categories of response.
  • the response can be processed by passing along with the additional input received in step 6 along with the current set of candidate nodes to the beginning of another pass through the method of FIG. 12 .
  • This branch flows into the line along the left edge of FIG. 12 .
  • the response can be processed by passing the additional input received but without the current set of candidate nodes. Dropping the current set of candidate nodes, allows the user to start from scratch without having to deal with parsing problems of the previous input. This branch also flows into the line along the left edge of FIG. 12 .
  • the subset of candidate nodes returned by step 4 can be stored in a semantic network along with the processed input.
  • new semantic meanings can be later retrieved by other input fed to step 1 of FIG. 12 .
  • the method of FIG. 12 enables new semantic meanings to be stored and later retrieved by other natural language input.
  • the method of FIG. 12 can supercede relational databases and other databases by enabling greater ease of use and better tools for handling of diverse kinds of data.
  • categories of response may be needed to handle user responses. For instance, rather than responding directly, the user may respond with a further question. To handle such responses, an additional category response node ‘digress’ can be added to the ‘response type’ hierarchy of FIG. 11 , and a corresponding branch of action leading from step 7 in FIG. 12 .
  • FIGS. 16 and 17 show the method of FIG. 12 handling specific input data similar to the SQL Query shown in FIG. 8 .
  • the eight line SQL query at the top of FIG. 8 selects information from two joined tables, seeking ‘fascinating’ from one table and avoiding ‘ugly’ in the other table.
  • FIGS. 16 and 17 show how the method of FIG. 12 can more elegantly retrieve similar information from a semantic network corresponding to ‘fascinating art not ugly’.
  • FIG. 16 shows the retrieval of nodes corresponding to ‘fascinating art’ and FIG. 17 shows now these nodes can be winnowed down to correspond to ‘not ugly’.
  • FIGS. 16 and 17 also handle visual inputs of a user's face, to show how image processing as shown in FIG. 9 can be useful when disambiguating the meaning of user input.
  • the user input text ‘fascinating art’ and a visual input node of ‘quizzical face’ are received.
  • these inputs are passed to four filters: a traversal type filter, a category filter, a context type filter, and a context node filter, whose filtering actions are directed by the table in FIG. 13 for the start of a conversation.
  • the traversal type filter does find any node inputs in FIG.
  • the category set filter finds the node ‘fascinating’ in FIG. 11 inheriting from ‘adjective’, allowing ‘fascinating’ to pass on as the input category set.
  • the other input nodes ‘art’ and ‘quizzical face’ do not inherit from ‘adjective’ or ‘adverb’ in FIG. 11 , so they are not passed on as nodes of the input category set.
  • the context type filter does not find any input nodes inheriting from ‘and’, ‘or’, or ‘not’, so it passes the default context type of ‘and’.
  • the context node filter finds two input nodes ‘art’ and ‘quizzical face’ which inherit from ‘noun’ in FIG. 11 , so it passes them on as an input context node set.
  • step 2 receives the input category set ‘fascinating’ and the input category set ‘general hierarchy’.
  • Step 2 maps ‘general hierarchy’ via FIG. 14 to a corresponding traversal function ‘any inherited links’.
  • Step 2 uses this traversal function to traverse from the node ‘fascinating’ in FIG. 11 downward to each inheritor of ‘fascinating’: ‘charming’, ‘delicate figurines of pasta’ and ‘stained glass statues of people.’
  • These three nodes are then collected into the set of candidate nodes passed to step 3 .
  • FIG. 16 step 3 loops through the set of three candidate nodes, computing for each candidate node a best contextual distance to the input context node set. Step 3 uses the table of FIG. 15 and the context type passed into step 3 of ‘and’ to determine that the best contextual distance function is ‘minimum average distance’.
  • FIG. 61 shows the steps in applying the minimum average distance function.
  • FIG. 16 step 3 shows results of applying FIG. 61 step 3 a to the semantic network of FIG. 11 . From the candidate node ‘charming’, the shortest path to ‘art’ has 2 links: from ‘art’ to ‘delicate figurines of pasta’ to ‘charming’. From the candidate node ‘charming’ to the other context node ‘quizzical face’ the shortest path has 6 links.
  • two other candidate nodes have shorter average shortest paths.
  • the shortest path from ‘delicate figurines of pasta’ is just one link long to ‘art’.
  • the average shortest path from ‘stained glass statues of people’ to the two context nodes ‘art’ and ‘quizzical face’ is also 3.5 links.
  • a subset of the candidate nodes are chosen having best contextual distance. Since candidate nodes ‘delicate figurines of pasta’ and ‘stained glass statues of people’ have the minimal average distance, as described by FIG. 15 for the input context type ‘and’, they are chosen, eliminating the candidate node ‘charming’. Elimination of the candidate node ‘charming’ results from the relatively distant position of ‘charming’ from ‘art’ in the semantic network of FIG. 11 . From a linguistic point of view, this corresponds to the notion that art is less directly connected to the adjective captivating than it is to specific instances of art. In larger semantic networks, structured according to methods outline by was described in patent application Ser. No. 09/085830 by Au, ‘art’ may head a larger subtree, increasing the distance from ‘art’ to various instances of art, but the distance between ‘art’ and the adjective ‘charming’ would also increase, remaining greater.
  • step 5 of FIG. 16 the subset of candidate nodes chosen in step 4 are described as two nodes texts connected by ‘or’ for presentation to a user.
  • Other presentations to the user may also be useful, such as a simple scrolling list or a list integrated into a graphical user interface relating parsed user input items to list elements.
  • Step 6 of both FIG. 12 and FIG. 16 gathers from the user a response to the presentation of step 4 .
  • the purpose of gathering a response is to validate that the candidate nodes displayed by step 5 are desired.
  • Step 6 of FIG. 16 shows an example of a response, which includes the text ‘not ugly’ and the visual input node of ‘unhappy face’. This response is assessed in terms of its input nodes in step 7 by a method outlined in FIG. 18 .
  • the method in FIG. 18 is a variation on the topological classification of an input response shown in FIG. 12 .
  • the method of FIG. 18 begins with of a set of response nodes in step 7 a, and, by determining which response type node most closely corresponds to the set of input nodes.
  • the method of FIG. 18 begins by checking for polysemy in the input text. Polysemy can be detected as an input word or input phrase which can be associated with two or more nodes of the same spelling in a semantic network. For instance, the word ugly in the input text of step 6 FIG. 16 can be associated with two nodes both spelled ‘ugly’ in FIG. 11 : ugly: 1 and ugly: 2 .
  • Step 7 c of FIG. 18 disambiguates which of these two nodes is most meaningful in the context of the input.
  • Step 7 d returns this set of most meaningful nodes.
  • Step 7 e returns to FIG. 12 to perform actions such as storing summaries of candidate nodes of FIG. 12 or dropping the current set of candidate nodes.
  • FIG. 19 in accordance with one embodiment of the present invention, shows an the method of FIG. 18 with a different set of inputs having no text, only an image of a happy face. Since there is no polysemy detected, semantic distances are only compared between candidate nodes and the context node “happy face”.
  • FIG. 20 shows details of step 7 a of FIGS. 18 and 19 .
  • pre-processors map inputs to nodes and pass these nodes to category set, context type, and context node filters. After filtering, candidate nodes are chosen by which node has the best contextual distance, to disambiguate polysemy by reducing the set of candidate nodes to the subset having best contextual distance.
  • FIG. 20 shows how an input of a happy face maps to a best contextual distance for the candidate node ‘ugly: 2 ’ which is ugly in the sense of desirable, scary and entertaining. In contrast, FIG.
  • FIGS. 19 and 20 show how subtle shades of meaning recorded in the topology of a semantic network guide disambiguation of polysemy, by applying best contextual distance functions.
  • FIGS. 9 through 20 show how the present invention can be used for natural language processing systems engaged in some form of dialog with a user.
  • the conversational context of a dialog can be detected and confirmed.
  • other contextually driven meanings can be determined from the stream of a natural language input.
  • the use of a word to refer to a preceding phrase is highly dependent upon context.
  • the grammatical term for this is anaphora.
  • Common examples of anaphora involve the words it and do. For instance in the sentence “The time has come, I know it” the it refers to the phrase “The time has come.” In natural language processing anaphora are difficult to map without using contextual clues.
  • FIG. 22 shows a method for automatically disambiguating anaphora by connecting phrases to the branch of a phrase tree having the best match in term of a contextual distance function, in accordance with one embodiment of the present invention. Unlike the method of FIG. 12 , the method of FIG. 22 does not involve user feedback, since it relies upon inputs and context information already established. Thus the method of FIG. 22 can be used for text-stream parsing and automatic summarization and translation of natural language texts.
  • the method of FIG. 22 can also be incorporated into a methods such as FIG. 12 to support a robust parsing of input text.
  • FIG. 22 begins with inputs of a parsed phrase tree and an anaphora which is a subset of that phrase tree.
  • FIG. 23 shows a small semantic network from which best contextual distances can be measured for the method of FIG. 22 .
  • FIG. 23 contains a dictionary of vocabulary related to ordering food in a restaurant.
  • step 1 a of FIG. 22 as after step 1 of FIG. 12 , context type and corresponding best contextual distance functions are identified using filtering or dispatching methods as discussed earlier for FIG. 12 .
  • step 1 b of FIG. 22 context node set is identified also using filtering or dispatching methods as discussed in FIG. 12 , but in FIG. 22 filters or dispatches from the input phrase tree of nodes rather than an input set of nodes as in FIG. 12 .
  • the method of FIG. 22 resolves anaphora by looking for a best possible branch of the parsed phrase tree to connect to the anaphora subset of the phrase tree. Upon resolution, the anaphora subset will be relocated to link directly to the best possible branch of the phrase tree.
  • the candidate node set in FIG. 22 comes from the set of nodes in the input phrase tree to which the anaphora subset can be connected; the candidate notes are NOT in the anaphora subset. Step 1 c identifies these candidate nodes.
  • Step 2 of FIG. 22 identifies the best candidate branch node of the phrase tree according to the optimal contextual distance function for the context type.
  • the best contextual distance function is chosen according to the table in FIG. 15 .
  • the best contextual distance function is ‘minimum average distance’.
  • the prepositional keywords ‘not’ has the best contextual distance function ‘maximum average distance.’
  • FIG. 22 Step 3 reconfigures the phrase tree to connect the anaphora subset to the best candidate node, thus recording the meaning of the anaphora.
  • FIG. 24 shows a simple phrase tree in which all the nodes are in a single inheritance list, in accordance with one embodiment of the present invention.
  • the structure of the input is often just a list which is then processed into a more detailed tree structure reflecting grammatical and syntactic meanings.
  • FIG. 24 shows an example of the beginning of such a processing operation. Scanning for a keyword from the end of the list, the prepositional keyword ‘and’ is found. Prepositional keywords such as ‘and’, ‘or’ and ‘not’ are significant because each prepositional phrase must be properly connected to prepositional subject. Other anaphora keywords such as ‘his’, ‘her’, ‘its’ and ‘this’ or ‘that’ may also be sought when scanning for keywords.
  • the method of FIG. 22 is used to find the best branch node of the phrase tree for that connection.
  • the anaphoric subset of FIG. 24 thus becomes the phase ‘and potatoes’ for which the method of FIG. 22 will find a best prepositional subject in terms of a best candidate node.
  • FIG. 25 shows the method of FIG. 22 applied to the parse tree of FIG. 24 .
  • step 1 a of FIG. 25 filtering the input phrase tree subset ‘and pototoes’ produces the prepositional keyword and context type ‘and’ and the corresponding best contextual distance function of ‘minimum average distance’ from the table on FIG. 13 .
  • step 1 b filtering produces the context node set ‘potatotes’
  • step 1 c the candidate node set is produced from the nodes of the phrase tree which are NOT in the anaphora subset of ‘and potatoes’.
  • This candidate nodes of ‘She’, ‘would like’, ‘salad’, ‘not’ and ‘steak’ are then measured in step 2 to find the candidate with best average contextual distance to each node in the context node set.
  • FIG. 25 there is one node in the context set, ‘potatoes’.
  • Other drawings will show how more nodes in the context nodes set affects the average best contextual distances measured.
  • step 3 since according to the semantic network of FIG. 23 the closest contextual distance from potatoes to steak, the anaphora subset of ‘and potatoes’ is connected to ‘steak’ in the parse tree.
  • FIG. 26 shows the parse tree of FIG. 24 after reconnecting the anaphora subset of ‘and potatoes’ to steak. Since the preposition ‘and’ refers to both steak and potatoes, the keyword ‘and’ node heads the subtree to which steak and potatoes are connected.
  • FIG. 27 shows a contrasting input phrase tree in which ‘and blue cheese dressing’ is in place of ‘and potatoes’.
  • FIG. 28 in accordance with one embodiment of the present invention, shows the method of FIG. 22 applied to the parse tree of FIG. 27 using the same dictionary of FIG. 23 .
  • FIG. 29 shows, in accordance with one embodiment of the present invention, in step 1 b the context node set produced consists of two nodes: “blue cheese” and “dressing”. In step 3 of FIG. 29 , distances to this two node context node set is are evaluated for each of the candidate node. Since the dictionary of FIG.
  • step 3 connects the anaphora subset phrase “and blue cheese dressing” to “salad”.
  • FIG. 28 shows the parse tree of FIG. 27 after step 3 of FIG. 29 has reconnected “salad” and “blue cheese dressing” under the keyword “and”.
  • FIG. 23 shows a semantic network dictionary isolated from the parse trees of FIGS. 24, 26 , 27 and 28 . Since parse trees contain valuable semantic information, they can be stored directly in a semantic network together with dictionary symbol nodes. By recording how dictionary symbol nodes are actually used in parsed trees, the meaning of dictionary symbol nodes is more complete and up-to-date. At the same time, the meaning of parsed tree nodes is more accurately tied to the dictionary.
  • FIG. 30 shows the semantic network dictionary of FIG. 23 augmented by the parse trees of FIGS. 26 and 28 . Since the parse trees of FIGS. 26 and 28 both begin with “She would like”, they have been automatically combined under the header ‘She would like’ to compact their storage. The meaning of the parsed node dressing: 2 inherited directly from the dictionary node ‘dressing’ to record that the parser has successfully disambiguated the meaning of dressing.
  • the method of FIG. 22 can adjust automatically to new meanings associated with parsed natural language input.
  • the meaning of subsequent natural language input can be detected using meanings automatically recorded in this way.
  • FIG. 31 shows how meanings automatically recorded in FIG. 30 are useful when processing subsequent natural language text.
  • the natural language input text of “She would like a lot of cheese” correctly maps the meaning of cheese to cheese: 2 in the dictionary of FIG. 30 .
  • Cheese: 2 is the cheese which more closely associated with salad dressing, as opposed to cheese: 1 which is associated with pizza.
  • FIG. 33 in accordance with one embodiment of the present invention, shows the same natural language input text of “She would like a lot of cheese” and same method steps outlined by FIG. 22 but using the semantic network of FIG. 23 .
  • the meaning of cheese is correctly mapped to cheese: 1 in FIG. 33 because cheese: 1 of a pizza is more closely associated with ‘a lot of’ in the dictionaries of FIGS. 23 and 30 .
  • FIG. 32 shows the semantic network dictionary of FIG. 23 after adding parsed nodes from the input phrase “She would like a lot of cheese.”
  • FIG. 32 shows how differently the meaning of the parsed nodes She: 2 and ‘would like: 2 ’ are recorded when the input “She would like a lot of cheese” precedes “She would like salad not steak and potatoes.”
  • the present invention uses them to more accurately attribute meanings to subsequent natural language input.
  • General methods for compacting them are outlined in patent application Ser. No. 09/085,830. Specific methods for compacting parsed nodes are outlined in drawing FIGS. 43 through 60 in the present invention.
  • FIG. 34 through 38 Examples using the method of FIG. 22 to disambiguate the meaning of pronoun anaphora are shown in FIG. 34 through 38 , in accordance with one embodiment of the present invention.
  • FIG. 34 a small semantic network dictionary is shown.
  • FIG. 35 shows an input parse tree in the form of a list, in which the anaphora pronoun keyword ‘her’ has been marked to delimit the start of an anaphora subset phrase of ‘her food’.
  • the other delimiter is the verb ‘was’ which marks the end of the anaphora subset phrase.
  • FIG. 36 shows how the steps of FIG. 22 can be used with the dictionary of FIG. 34 to map the anaphora subset of ‘her food’ to best branch node of FIG. 35 .
  • step 3 reconnects ‘her’ and ‘food’ to ‘Amy’.
  • FIG. 37 shows an input parse tree in the form of a list, in which the anaphora pronoun keyword ‘his’ has been marked to delimit the start of an anaphora subset of ‘his food’.
  • the other delimiter is the verb ‘was’ which marks the end of the anaphora subset phrase.
  • FIG. 38 shows the steps of the method in FIG. 22 used with the dictionary of FIG. 34 and the input phrase tree of FIG. 37 . Since the name ‘Harold’ is closer to ‘his’ in the dictionary than the other candidates of ‘Amy’, ‘should be’, ‘her’, ‘.’, ‘Harold’, ‘said’, ‘was’ and ‘ready’, FIG. 38 step 3 reconnects ‘his’ and ‘food’ to ‘Harold’.
  • FIG. 39 in accordance with one embodiment of the present invention, shows a small semantic network dictionary emphasizing how long chains of categorization hierarchy links can classify various meanings for the symbol ‘Chicago’.
  • FIG. 40 in accordance with one embodiment of the present invention shows a variation of the method of FIG. 12 , amended to categorize nodes by spelling and by a complex traversal function, to disambiguate the meaning of the phrase “Chicago bigger than a bread box?”
  • Step 2 of FIG. 40 describes the complex traversal function, which travels long chains of categorization hierarchy links to retrieve a set of candidate nodes. These candidate nodes are compared using a best contextual distance function as in FIG. 12 .
  • FIG. 41 in accordance with one embodiment of the present invention is an example of how the method of FIG. 40 can be used to map candidates of Chicago: 1 , Chicago: 2 and Chicago: 3 to a context node set of “bread box”, correctly identifying Chicago: 2 as the best meaning of Chicago in “Chicago bigger than bread box.”
  • FIG. 43 shows a method to convert parsed node subtrees to corresponding is-a link dictionary nodes.
  • Is-a links are standard classification hierarchy links, shown as narrow black lines in FIG. 44 , as opposed to the more vague ‘informs’ links shown as wide gray lines in FIG. 44 .
  • step 1 of FIG. 43 parse trees marked for compaction are detected by their inheritance link to the key node ‘inversion’ (shown in bold).
  • step 2 of FIG. 43 the subjects of these parse tree are marked, and semantic links to the subjects are reconfigured to inherit from the predicate via direct is-a link. Since the parse tree has now been recorded as a standard dictionary node, step 3 of FIG. 43 deletes the redundant nodes from the parse tree to compact the semantic network. Step 3 then returns to steps 1 as a loop, to automatically compact other parse trees.
  • FIG. 42 in accordance with one embodiment of the present invention shows an example of a series of natural language inputs to be recorded by a semantic network by methods of FIG. 22 and FIG. 43 .
  • FIG. 44 in accordance with one embodiment of the present invention, shows a small semantic network dictionary for use by methods of FIGS. 22 and 43 to record the inputs shown in FIG. 42 .
  • FIG. 45 shows the input sentence “My friend Hal was a programmer” recorded into the semantic network of FIG. 44 by method of FIG. 22 .
  • the node was: 3 is shown in bold to indicate that it inherits from ‘inversion’ and therefore will be detected by the method of FIG. 43 for inversion to dictionary format.
  • FIG. 46 shows how the subtree of was: 3 is inverted to dictionary format node programmer: 2 inheriting from was: 1 and programmer: 1 . Since step 1 also links Hal: 2 to inherit from programmer: 2 , the subtree of was: 3 has now been fully inverted to dictionary format. Hal: 2 now inherits the meaning of programmer: 2 directly from an is-a link, exactly as a standard dictionary node instead of a parse tree node.
  • the redundant subtree nodes was: 3 , a: 4 and programmer: 3 nodes are deleted, to compact the semantic network.
  • the next two lines of input from FIG. 42 are recorded by the method of FIG. 22 to augment the semantic network of FIG. 47 .
  • These inputs are “His number is (402) 287-0101. His address is 404 Fesser Street in Detroit.”
  • This augmented semantic network is shown in FIG. 48 , in accordance with one embodiment of the present invention. Since number and address as predicate concepts are not intrinsic to their subjects, the is: 1 node associated with number: 1 and address: 1 are do not inherit from ‘inversion’. This prevents the method of FIG. 43 from improperly inverting parse node subtrees which inherit from is: 1 . Since is: 1 is closer in the semantic network to number: 1 and address: 1 , the method of FIG. 22 connects the parse node is: 4 to is: 1 rather than is: 2 .
  • FIG. 49 shows the semantic network of FIG. 48 after applying the method of FIG. 22 on the fourth input line of FIG. 42 : “He is a good trumpet player.” Since is: 2 is closer to player: 1 than is: 1 in FIG. 48 , the method of FIG. 22 connects the parse node is: 5 to is: 2 rather than is: 1 . This permits the method of FIG. 43 to correctly invert the parse node subtree is: 5 to standard dictionary node is-a links, shown in FIG. 50 .
  • FIG. 43 through 50 in accordance with one embodiment of the present invention thus show how semantic information can be automatically parsed into standard semantic network dictionary format, to automatically create large semantic network dictionaries without manual intervention by lexicographers and linguists. These labor-saving methods can be extended to also automatically record semantic information contained questions posed to a natural language processor.
  • FIG. 51 through 60 in accordance with one embodiment of the present invention, show how parsed questions can be stored in a semantic network dictionary just as parsed statements can be stored. The parsed questions can be marked for later linking to their answers, as shown in FIG. 59 , and entire question-answer parsed subtrees can be inverted to standard dictionary is-a link nodes as shown in FIG. 60 . The method shown in FIG.
  • FIG. 58 shows how to invert questions into is-a links.
  • FIG. 51 shows how to generate an answer to a question, by describing nodes which best in the sense of a best contextual distance to the question nodes.
  • FIG. 52 shows how to map question node types to traversal functions to fetch candidate nodes for answering questions. A set of candidate nodes having best contextual distance to the question nodes can then be chosen as an answer to questions.
  • FIG. 53 shows how the method of FIG. 51 answers the question from FIG. 42 by creating nodes in the semantic network of FIG. 50 . Since the present invention automatically acquires semantic network meanings from natural language statements, questions and answers to questions, the present invention can automatically acquire all type of natural language meanings from natural language dialog. This automatic language acquisition functionality, together with the accuracy of the retrieval methods of the present invention makes the comparatively laborious data base maintenance work of relational database obsolete.

Abstract

A computer implemented data processor system automatically disambiguates a contextual meaning of natural language symbols to enable precise meanings to be stored for later retrieval from a natural language database, so that natural language database design is automatic, to enable flexible and efficient natural language interfaces to computers, household appliances and hand-held devices.

Description

  • This application is a continuation-in-part of U. S. patent application Ser. No. 09/085,830 filed May 28, 1998, and claims priority to U.S. Provisional Patent 60/342,360 filed Dec. 27, 2001, each of which are hereby incorporated by reference. Topological methods described in Ser. No. 09/085,830 traverse semantic networks in ways useful for computer-implemented natural language processing, including methods for constraining search to intersections of semantic categories.
  • BACKGROUND OF THE INVENTION
  • Examples of natural language forms include written and spoken words, gestures and face expressions which clarify the meaning of words, intonation patterns indicating whether a sentence is a question, and other auditory and visual contextual aids for conveying meaning. Each of these forms may convey multiple meanings which may appear ambiguous when forms occur out of context. For instance, there are many dictionary meanings of the word ‘dash’, such as a race or a small amount of seasoning. Since multiple meanings for words are so common, lexicographers refer to a single word's diversity of meanings as polysemy. To disambiguate polysemy, all the of the above natural language forms must be correlated in order to infer the context which identified specific meaning from a list of polysemous potential meanings.
  • The prior art detects such inputs on a piecemeal basis, but to provide a comprehensive user interface for computer systems, all of the above inputs must be correlated, as efficiently as a person can correlate them. However, the prior art has failed to correlate even a fraction of such inputs as efficiently as people do. As a result, designers of natural language processing systems have greatly restricted the range of conversation recognized by natural language processing, to increase accuracy at the expense of flexibility.
  • Thus prior art computer systems correlate small ranges of natural language inputs, but the vast majority of inputs that a human could easily correlate remain useless to such systems. For instance, a raised eyebrow can indicate a question and a smile may indicate agreement. and a wave of a hand might indicate dismissal or indicate of a change in subject. All of these gestures are useful and meaningful for people, but the prior art cannot reliably correlate simple gestures with other inputs such as spoken or written words.
  • The inability of computer systems to correlate a wide variety of inputs has hampered the ability of computers to participate in conversations with people. In particular, this inability to correlate prevents computers from recognizing important contextual shades of meaning which are needed for parsing natural language. In human conversation, contexts shift fluidly and dramatically to constrain the meaning of gestures and other symbols which can have varying meanings depending on precise context. For instance, the word ‘dash’ can mean at least four different nouns and three different verbs: dash as in a small quantity mixed into something (a dash of salt), dash as in a race(a fifty yard dash), dash as in car dashboard, dash as in a Morse Code (a signal longer that a dot), dash as in to mix together, dash as in to ruin something (dash hopes), dash as in to move quickly. For clarity, more meanings for dash may be needed and added later to a natural language processing system. For instance, dash as in to splatter, or dash as in a short sudden movement.
  • Prior art uses grammar or statistics to disambiguate polysemy. Prior art using a grammar analysis to choose the correct meaning of ‘dash’ based in grammar can only determine whether a noun or verb meaning is best. Such a system would not disambiguate whether dash means a Morse code or a small quantity of something. Prior art using statistics of usage can only choose a meaning which was chosen most of the time in a context. Thus statistical methods disable any other meanings and prevent the acquisition of new meanings for a symbol within any context.
  • Human linguistic abilities are clearly less limited, in addition to being more accurate than prior art. Besides acquiring new meanings for words within a conversational context, humans also create new meaningful contexts through self-organizing linguistic abilities. For instance, as humans converse about subjects, subjects acquire new semantic meanings which evolve into new conversational contexts. In contrast, prior art computer systems only acquire new semantic contexts through laborious programming and data-entry efforts. The encumbrance of such programming and data-entry prevents prior art computer systems acquiring semantic knowledge on a real-time basis during natural language conversations with humans. Although a large number of programming languages have been created for inputting semantic knowledge, none of which have the flexibility and general utility of a natural language such as English. Languages such as Prolog, SQL or Lisp cannot match the convenience of conversing in plain English or other natural languages.
  • General computer-implemented methods to process natural language have been based in either logic, statistics, or topology. Logic has been the dominant method in the prior art. However, logical ambiguities inherent in natural language have foiled the prior art attempts which rely upon logic as a basis for processing natural language. For instance, one of the most important aspects of human conversation is called polysemy: the re-use of identical symbols to mean different things, depending upon context. For instance, the word run could mean a verb meaning to step quickly, but it could also mean the verb to campaign for office or to mean the noun for a small brook. In any specific conversation, run would signify just one of these meanings, unless a pun was intended, in which case run might signify two such meanings. By sensing natural language contexts to determine which polysemous meanings are most within context, humans recognize which meanings are signified.
  • Context is impractical to define logically because logic requires enumeration of logical inputs and outputs. The number of contexts which can be defined by a natural language input is limited only by every possible shade of meaning of every symbol which occurs in every possible natural language sentence. Since such a large number of combinations cannot be enumerated, logical natural language processors store a subset to the full set of possible contexts as a logical approximation: Each natural language symbol is stored with its own set of logical data, and with rules for combining its logical data with other symbols.
  • For instance, when the context is politics, to run would mean to campaign for office. However, many of these rules will break when a combination of contexts is pertinent. For instance, if the context is a political appointee who runs for elected office and also runs a government agency, the meaning of run remains logically ambiguous.
  • The larger the semantic system, the more frequently contextually defined semantic rules are broken. For vocabularies larger than ten thousand words the frequency of flaws from broken rules easily overwhelms the accuracy of a natural language processing system, as demonstrated by problems in the CYC project. Even when attempting to define a small static semantic dictionary, logical contradictions emerge during testing which cannot be resolved without creating a new logical category for each possible combination of symbols. The combinatorial complexity of language makes testing these categories generally impractical.
  • For a semantic dictionary of N symbols, in a language where the maximum number of symbols strung together is M, the number of logical combinations is N to the M power. For a vocabulary of 1,000 words combined in short four word sentences, the number of logical combinations is 1,000 raised to the fourth power, which equals 1,000,000,000,000. If a team lexicographers attempts to define and test a semantic dictionary of this small size there would be 100,000,000,000 testing hours required if each test takes 1/10th of an hour. If 500 testers each work 2000 hours a year, the team can work 1,000,000 hours per year, and the testing will be complete in 100 years. By that time, the dictionary will surely be obsolete and require re-testing. For longer sentences and larger dictionaries, this drawback quickly grows exponentially worse.
  • Even worse, as phrases are used within new conversations, they immediately acquire new shades of meaning from these new conversations. A natural language processing system must track shifts in overall meaning of phrases to remain accurate. For instance, the meaning of a celebrity's name shifts as that name is used in major news reports, particularly if their fame is new. To logically represent such shifts in meaning, the rules describing how to combine a celebrity's name in various contexts must be extended to handle each new conversational use of the celebrity's name explicitly. Using logical methods, all possible combinations of phrases and contexts must be defined and tested.
  • Because the testing of logical methods is so impractical for large vocabularies, statistical methods have instead been dominant in natural language processing systems, particularly in speech recognition. In the prior art, statistical probability functions have been used to map from inputs to most likely contexts. Statistics, however, only apply to sets of previously occurring events.
  • All statistics require the collection of a set of prior events from which to calculate a statistical aggregate. For new events no such set exists and no statistical aggregate can be calculated. Unfortunately, natural language is full of new events, such as newly concatenated phrases, each having a unique contextual shade of meaning. For instance, a person might request “fascinating art—not ugly.” A person would have no trouble combining the definitions of fascinating, art, not and ugly to make some sense of such a request, even if that person had never before heard the phrase ‘fascinating art—not ugly.’ A statistical natural language processing system, on the other hand, would have no statistical event set from which to disambiguate the meaning of a new combination of words such as ‘fascinating art—not ugly.’
  • Another problem with statistics is that once an event set has been collected to describe the meanings for a symbol, statistical functions prefer frequently chosen meanings over rarely chosen meanings, rendering the system insensitive to new meanings conveyed by new events. Thus as statistical natural language systems acquire semantic knowledge, their ability to distinguish new information diminishes.
  • The above drawbacks can be avoided by topological methods for processing natural language.
  • SUMMARY OF THE INVENTION
  • The present invention is a computer implemented data processor system to automatically disambiguate a contextual meaning of natural language input streams of symbols, stored as new input stream nodes within a semantic network, whose various candidate meanings are stored as nodes in the semantic network, and whose natural language context is stored as a set of context nodes in the semantic network. By precisely measuring semantic distances in terms of semantic links when traversing between various candidate meanings and the set of context nodes, the present invention can automatically choose candidate nodes closest to the context node set to represent a contextual meaning of the input stream. To store this contextual meaning, the present invention can create new semantic inheritance links between the input stream nodes and the candidate nodes closest to the context nodes. These new links create precise records of contextual meaning to be stored for later retrieval by subsequent natural language queries. The present invention thus enables the semantic network to function as a natural language database by storing data as semantic network nodes, with all semantic categories needed for accurate natural language database design automatically created and linked entirely from natural language input streams. The present invention thus automatically provides precise storage and retrieval of detailed, intersecting levels of information, comparable to relational database storage and retrieval capability, without the labor cost and inconvenience of designing a database, to enable flexible and efficient natural language interfaces to computers, household appliances and hand-held devices. Additional advantages and novel features of the invention will be set forth in part in the description that follows, and in part will become more apparent to those skilled in the art upon examination of the following or upon learning by practice of the invention.
  • BRIEF DESCRIPTION OF THE DRAWINGS
  • FIG. 1 shows a natural language processing system deployed on a central computer shared by a variety of customer support staff and remote customers connected by Internet connections, in accordance with an embodiment of the present invention.
  • FIG. 2 shows a natural language processing system deployed on computers connected by peer-to-peer Internet connections, to provide technical support to customers, in accordance with an embodiment of the present invention.
  • FIG. 3 shows a natural language processing system deployed on a cellphone connected to the Internet, to provide a natural language user interface for dialing, storing and retrieving names, phone numbers from personal address book information, in accordance with an embodiment of the present invention.
  • FIG. 4 shows a natural language processing system deployed on a microwave oven home appliance, to provide a natural language user interface for operating oven controls, in accordance with an embodiment of the present invention.
  • FIG. 5 shows a natural language processing system having inputs, storage, and outputs in the form of images, sounds and texts, in accordance with an embodiment of the present invention.
  • FIG. 6 shows a database system having inputs of SQL commands, a relational database processor, and output in the form of images, sounds and texts, in accordance with an embodiment of the present invention.
  • FIG. 7 shows the database system of FIG. 2. with a natural language processor added for converting natural language text input to SQL commands, in accordance with an embodiment of the present invention.
  • FIG. 8 shows an example of an SQL database processing a query corresponding to the natural language request “fascinating art not ugly,” in accordance with an embodiment of the present invention.
  • FIG. 9 shows a simple image processing method for identifying emotional content of face expressions by analyzing the contours of the mouth, in accordance with an embodiment of the present invention.
  • FIG. 10 show an overview of sound processing methods to identify words from sounds by analyzing phonemes comprised of transitions from one frequency bands to another, in accordance with an embodiment of the present invention.
  • FIG. 11 shows a semantic network storing the logical relations of FIG. 8 as semantic nodes and links, in accordance with an embodiment of the present invention.
  • FIG. 12 shows a method for combining topological categorization with topological contextual filtering, in accordance with an embodiment of the present invention.
  • FIG. 13 shows a table mapping from filter types of FIG. 12 to filter actions which screen out input nodes of FIG. 12, in accordance with an embodiment of the present invention.
  • FIG. 14 shows a table mapping from topological categorization input category types to their corresponding semantic network traversal functions, in accordance with an embodiment of the present invention.
  • FIG. 15 shows a table mapping from topological contextual filtering input context types to their corresponding best contextual distance functions, in accordance with an embodiment of the present invention.
  • FIG. 16 shows an example of parsing the natural language text “fascinating art” using the method of FIG. 12, in accordance with an embodiment of the present invention.
  • FIG. 17 shows an example of parsing “fascinating art not ugly” using the method of FIG. 12 to filter the output of FIG. 16, in accordance with an embodiment of the present invention.
  • FIG. 18 shows an example of a method applicable to step 7 of FIG. 16 to assess a visual response, in accordance with an embodiment of the present invention.
  • FIG. 19 shows an example of a method applicable to step 7 of FIG. 17 to assess a visual response, in accordance with an embodiment of the present invention.
  • FIG. 20 shows an example of a method for disambiguating the meaning of “ugly” as defined in the semantic network of FIG. 11 from a combination of text and image inputs, in accordance with an embodiment of the present invention.
  • FIG. 20 shows an the example of a method in FIG. 20 for disambiguating the meaning of “ugly” as defined in the semantic network of FIG. 11, from a different combination of text and image inputs, in accordance with an embodiment of the present invention.
  • FIG. 22 shows a method for combining topological categorization with topological contextual filtering for connecting phrases to complementary phrases within a parse tree, in accordance with an embodiment of the present invention.
  • FIG. 23 shows a small semantic network dictionary for storing semantics related to conversations about ordering food in restaurants, in accordance with an embodiment of the present invention.
  • FIG. 24 shows a parse tree created from a sentence “She would like salad and not steak and potatoes, in accordance with an embodiment of the present invention.
  • FIG. 25 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 23 to process the parse tree of FIG. 24, in accordance with an embodiment of the present invention.
  • FIG. 26 shows the parse tree of FIG. 24 after processing by the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 23, in accordance with an embodiment of the present invention.
  • FIG. 27 shows a parse tree created for the sentence “She would like salad not steak and blue cheese dressing,” in accordance with an embodiment of the present invention.
  • FIG. 28 shows the parse tree of FIG. 27 after processing by the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 23, in accordance with an embodiment of the present invention.
  • FIG. 29 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 23 to process the parse tree of FIG. 27, in accordance with an embodiment of the present invention.
  • FIG. 30 shows the semantic network dictionary of FIG. 23 after augmentation with nodes from the parse tree of FIG. 28, in accordance with an embodiment of the present invention.
  • FIG. 31 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 30 to parse “She would like a lot of cheese,” in accordance with an embodiment of the present invention.
  • FIG. 32 shows the semantic network dictionary of FIG. 23 after augmentation with nodes from a parse tree created by parsing “She would like a lot of,” in accordance with an embodiment of the present invention.
  • FIG. 33 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 32 to disambiguate the meaning of ‘cheese,’ in accordance with an embodiment of the present invention.
  • FIG. 34 shows a small semantic network dictionary for storing semantics related to pronouns, in accordance with an embodiment of the present invention.
  • FIG. 35 shows a parse tree created for the sentences “Amy should be here. Harold said her food was ready,” in accordance with an embodiment of the present invention.
  • FIG. 36 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 34 to disambiguate the meaning of ‘her’ in the parse tree of FIG. 35, in accordance with an embodiment of the present invention.
  • FIG. 37 shows a parse tree created for the sentences “Amy should be here. Harold said his food was ready,” in accordance with an embodiment of the present invention.
  • FIG. 38 shows the method of FIG. 22 used in conjunction with the semantic network dictionary of FIG. 34 to disambiguate the meaning of ‘his’ in the parse tree of FIG. 37, in accordance with an embodiment of the present invention.
  • FIG. 39 shows a small semantic network dictionary for storing semantics related to the transitive relation of ‘bigger than,’ in accordance with an embodiment of the present invention.
  • FIG. 40 shows a variation of the method of FIG. 12, amended to categorize nodes by spelling and by a complex traversal function, to disambiguate the meaning of “Chicago bigger than a bread box?”, in accordance with an embodiment of the present invention.
  • FIG. 41 shows the method of FIG. 40 used in conjunction with the semantic network dictionary of FIG. 39 to disambiguate the meaning of ‘Chicago’ in “Chicago bigger than a breadbox?”, in accordance with an embodiment of the present invention.
  • FIG. 42 shows an example of a dialog between a person and natural language processor where the natural language processor remembers and retrieves dialog semantic information, in accordance with an embodiment of the present invention.
  • FIG. 43 shows an example of a method to create semantic is-a type dictionary links from parse phrase trees containing ‘is’ and ‘a’ phrasing, in accordance with an embodiment of the present invention.
  • FIG. 44 shows a small semantic network for storing semantics related to personal address book information, in accordance with an embodiment of the present invention.
  • FIG. 45 shows the semantic network of FIG. 44 augmented by the first sentence of the dialog in FIG. 42, after applying step 1 of FIG. 43, in accordance with an embodiment of the present invention.
  • FIG. 46 shows the semantic network of FIG. 45 after applying step 2 of FIG. 43 to create an is-a link between programmer:2 and Hal:2, in accordance with an embodiment of the present invention.
  • FIG. 47 shows the semantic network of FIG. 46 after applying step 2 a of FIG. 43 to remove redundant semantic information, in accordance with an embodiment of the present invention.
  • FIG. 48 shows the semantic network of FIG. 47 after applying step 4 after step 3 of FIG. 43 to add semantic information from the second and third sentences of FIG. 42, in accordance with an embodiment of the present invention.
  • FIG. 49 shows the semantic network of FIG. 48 after applying step 2 of FIG. 43 to copy the new phrase tree's “good trumpet player” portion to be inherited by an is-a link by Hal:2, in accordance with an embodiment of the present invention.
  • FIG. 50 shows the semantic network of FIG. 49 after applying step 2 a of FIG. 43 to remove redundant semantic information, in accordance with an embodiment of the present invention.
  • FIG. 51 shows an example of a method to detect and answer a question encountered in a dialog input phrase tree, in accordance with an embodiment of the present invention.
  • FIG. 52 shows an example of a table mapping from an input phrase tree's input node types to allowable links and traversal starting points of step 2 in FIG. 51, in accordance with an embodiment of the present invention.
  • FIG. 53 shows the semantic network of FIG. 50 after adding the input phrase tree from the question “Who is a good musician friend,” in accordance with an embodiment of the present invention.
  • FIG. 54 shows the method of FIG. 50 applied to answer the input phrase tree question in FIG. 53, in accordance with an embodiment of the present invention.
  • FIG. 55 shows an example of a dialog in which an answer to question has been answered affirmatively, indicating that the answer would be useful to store in a semantic network, in accordance with an embodiment of the present invention.
  • FIG. 56 shows an example of a method to stored the confirmed answer to a question, using the inversion method of FIG. 43, in accordance with an embodiment of the present invention.
  • FIG. 57 shows an example of a table of input phrase tree's input node types and corresponding connecting links used to connect interrogative nodes in step 2 of FIG. 56, in accordance with an embodiment of the present invention.
  • FIG. 58 shows the method of FIG. 56 applied to the semantic network of FIG. 53 to store the confirmed answer to the question “Who is a good musician friend,” in accordance with an embodiment of the present invention.
  • FIG. 59 shows the semantic network of FIG. 53 after step 2 of FIG. 56 has added a new is-a link between Who:2 and Hal:2, in accordance with an embodiment of the present invention.
  • FIG. 60 shows the semantic network of FIG. 59 after step 3 of FIG. 56 has applied the inversion method of FIG. 43 to invert the input phrase tree for “good musician friend,” in accordance with an embodiment of the present invention.
  • FIG. 61 shows a method for identifying nodes in a candidate node set having a minimum average distance between a candidate node and a context node set within a semantic network, in accordance with an embodiment of the present invention.
  • FIG. 62 shows a method for identifying nodes in a candidate node set having a maximum average distance between a candidate node and a context node set within a semantic network, in accordance with an embodiment of the present invention.
  • FIG. 63 shows a method for identifying nodes in a candidate node set having a minimum distance between a candidate node and a context node set within a semantic network, in accordance with an embodiment of the present invention.
  • FIG. 64 shows a method for identifying nodes in a candidate node set having an infinite minimum distance between a candidate node and a context node set within a semantic network, in accordance with an embodiment of the present invention.
  • DETAILED DESCRIPTION OF THE INVENTION
  • In accordance with the present invention, topological methods disambiguate polysemy by measuring the relative distance between symbols in a semantic network. In a well constructed semantic network, the distance between symbols is proportional to the closeness in meaning between symbols. Consequently the contextual meaning of a symbol can be found by locating the closest symbol to a set of symbols defining the context. Both new symbols and new meanings can be recorded simply by adding new symbol nodes to existing topologies. Shifts in meaning can be recorded simply by shifts in topologies.
  • However, among others, two problems with topological methods, which have prevented them from succeeding in prior art, are overcome with the present invention. The first problem is that the topological distance between a pair of symbol nodes in a semantic network may not reflect closeness in semantic meaning between the pair of symbol nodes. This problem can be avoided by adjusting automatically adjusting distance between each pair of nodes to accurately reflect semantic meaning by regrouping, adding, and removing nodes. Relevant methods for automatically regrouping, adding and removing nodes are described in patent application 09/085,830 “topological methods to organize semantic network data flows for conversation applications.”
  • The second problem with topological methods occurs when representing and disambiguating the meaning of negation. When seeking a candidate symbol in terms of being NOT some symbol ‘A’, some semantic path will generally connect the ‘A’ and candidate symbols. However, closeness of the ‘A’ to candidate symbols cannot represent closeness to desired meaning. Rather, it represents closeness to opposite meaning. Therefore, when seeking a candidate symbol which is NOT ‘A’, some other metric besides closeness must be applied to find appropriate candidate symbols.
  • For instance, a person might request ‘fascinating art’ from an information retrieval system. Using topological methods such as the method of ‘most specific common abstraction’ (Kolodner, pages 346-347), a retrieval system could find nodes closest to ‘fascinating art’ by measuring semantic path distances from every candidate node to both ‘fascinating’ and ‘art’. The retrieval system would return candidate nodes having minimal distance to both ‘fascinating’ and ‘art’.
  • For requests containing negation, such a minimal distance function would not suffice. Similarly, requests containing other logical operators like ‘or’ as well as prepositions like ‘from’, ‘to’, and ‘after’ demand other logical-syntactic meanings beyond the capability of a simple minimal distance function to model.
  • For instance, simple distance functions have difficulty interpreting a request such as ‘fascinating art that is not ugly’ because of the complexity of possible negative connections between ‘art’ and ‘ugly’: malicious art, art that is harsh, art that is crude, art made from toxic substances. Any of these negative concepts connecting ‘art’ to ‘ugly’ might be considered when retrieving meaning corresponding to ‘fascinating art that is not ugly’. Yet depending upon context, some of these negative concepts might be important and some might be irrelevant.
  • For instance, if the overall context of a request concerned environmental pollution, then art made from toxic substances would be a more important meaning of ‘ugly’ to avoid and others meanings might not matter. Because of the logical meaning imbedded in the word ‘not’, prior art logical methods have typically defined negative concepts for a particular context. For instance, prior art has typically contrived rules such that when the overall context of a request includes ‘pollution’ then ‘ugly’ signifies ‘toxic substances’.
  • In prior art, logical methods employ sets of rules to identify sets of candidate nodes preferred for a given logical context, and other sets of rules to rank the set of candidate nodes for the given logical context. Prior art logical methods have been preferred for their processing speed and economy with storage space. However such advantages have lost relevance as computer processors has become faster and computer memory becomes cheaper. As outlined above, logic have cannot handle the combinatorial complexity of natural language when there are more than a thousand symbol nodes. In addition, logic cannot resolve contradictions in meaning which arise naturally in general-purpose natural language systems.
  • For example, a person might request ‘fascinating art that is not ugly’ from an information retrieval system. The system might find, categorized under art, a set of candidate nodes. For instance, there might be two such nodes. The first semantic node ‘stained glass statues of people’, might inherit from ‘art’, ‘beautiful’ and ‘stained glass’. The second semantic node ‘dancers made of children's blocks’ might inherited from ‘art’ and ‘strange’. To determine which of the two nodes is most fascinating and least ugly, a logical semantic processor would apply rules. A first such rule might be that inheritors of ‘beautiful’ are not ugly. A second such rule might be that toxic substances are ugly. Since ‘stained glass’ might inherit from ‘lead’ which in turn might inherit from ‘toxic substance’, the first rule might conflict with the second rule. To resolve such conflicts, semantic rules must be ranked to determine which rule about ‘ugly’ is more significant.
  • In the present invention, the relative distances between nodes are used in place of logic to resolve logical contradictions. By using sets nodes to define context, contexts can be defined with various degrees of focus. General contexts can be defined by a few high-level nodes, and specific, narrowly focused contexts can be defined by a large number of low level nodes. By minimizing distances to a large number of low level nodes in a narrowly focused context, logical contradictions can be resolved by topology recorded in the local neighborhood of that narrow context.
  • By contrast, in prior art, fuzzy logic algorithms were developed to locally resolve logical contradictions by using statistical functions. Since some things can be both beautiful and ugly, there is a probability that beautiful implies not ugly, and this is compared to the probability of the toxic quality of lead implies ugly. Although statistical functions can resolve logical contradictions between frequently competing candidate nodes, they cannot resolve logical contractions for new descriptions or old descriptions relative to new semantic nodes, because, as mentioned above, all statistics require a prior set of trial events upon to base a statistical distribution. If ‘fascinating art that is not ugly’ was a new request, there would be no events upon which to base a statistical distribution and statistical methods would be of no use. Even after recording an event set showing a preference for one instance of art over another, there would be no statistical method to relate that event set to a new node was added to the semantic net under the category of art, for instance a node representing ‘charming figurines of edible pasta’.
  • In large-scale database applications, pure logic is still used to retrieve semantic information, even though logic cannot resolve the inconsistencies which naturally occur in large semantic systems. The simplicity of implementing logical operators such as AND, OR and NOT has supported their popularity when disambiguating simple data, particularly when such logical operations are nested in parenthesis to build a tree-structure of logical relationships. For instance, a natural language request ‘fascinating art’ can be recast as a logical request ‘art AND fascinating’. This request might return candidate rows of data which could then be manually winnowed down by new request ‘art AND fascinating AND (NOT ugly)’ which would return a more focused set of candidate rows.
  • By compounding logical restrictions using relational table joining operators, relational databases enable large amounts of data to be searched quickly and efficiently. Although translating requests into the form of logical operators is laborious, and translating data into the form of logically correct statements is even more laborious, if an application contains a limited variety of data, the translation can be completed within a reasonable budget. However, as databases grow to become data warehouses, they require considerable manual intervention to maintain their logical consistency as new data is added. The budget for such maintenance can often become unaffordable. For large databases such as Internet web search engine portals, such maintenance costs run into millions of dollars.
  • For convenience and efficiency, it would be better to automatically store and retrieve information using natural language, bypassing the labor needed to translate requests into logical operators, bypassing the labor needed to translate data into logically correct statements and bypassing the cost of manually maintaining logical consistency. At the same time, because large natural languages contain paradoxical logical contradictions which must be resolved by choosing specific contextual shades of meaning, people need a system for storing criteria by which contextual shades of meaning can be disambiguated.
  • For instance, the adjective ‘attractive’ is commonly attributed to the noun phrase ‘movie star’. However, some movie stars base their appeal upon a kind of fascination with unattractiveness, by playing villainous or anti-heroic roles. In such contexts, the meaning of attractive might have to be confined to a star's “ability to attract box-office sales”. To remain accurate in such contexts, a natural language processor must efficiently store criteria for differentiating between the shades of meaning which might be attributed to the word ‘attractive’, even if these shades of meaning cannot be immediately mapped to a logical context.
  • Natural language processing systems must also disambiguate the meaning of requests containing logical prepositions. In such cases, logic must be combined with contextual semantic meanings to determine which noun phrase to link to a prepositional phrase. Thus the disambiguation of meaning of natural language logical prepositions requires a precise method for combining contextual shades of meaning which themselves may not be logically defined. When servicing natural language requests for specific information from large databases, it is especially important to accurately respond to the meaning of such prepositions. For instance, ‘and’ and ‘or’ as well as ‘from’, ‘to’, and ‘after’ are important prepositions used to zero in on specific information. However, the meaning of prepositions vary in subtle ways within natural language discourse.
  • The prior art uses rules and statistics can such prepositions, but only after drastically limiting the scope of input. To handle the full complete meaning of natural language, computer-implemented natural language processors require better methods to disambiguate prepositional contextual shades of meaning.
  • Because of the above difficulties in dealing with prepositional contextual shades of meaning, there are a number of commercial uses for such improved methods of detecting prepositional contextual shades of meaning. For example, database languages which store and retrieve information could be bypassed in favor of direct natural language user interfaces for storing and retrieving information.
  • For instance, Structured Query Language (SQL) was developed to manipulate sets of data using compounded formal logic. The success of SQL attests to the ability of logic to retrieve massive data sets. However, SQL has drawbacks when handling complex data relations. Since SQL databases require that each data relation be explicitly queried by a logical expression selecting logically validated data, both the queries and the validated data require considerable expertise to create. Typically, a data base specification and a process for acquiring and validating data must be created before actual use of a database. Because of the semantic limitations of SQL, both specification and data acquisition are typically specified by a natural language, such as English. In addition, during the maintenance phase of a database, bug fixes and corrections for bad data are also typically specified in a natural language.
  • Since natural language typically specifies database implementation efforts throughout the life of a database, database implementation would be cheaper, faster and more accurate if databases used natural language constructs to store and retrieve information directly, rather than translating natural language specifications into logical data relations. To store and retrieve information natively in English via natural language processing methods would eliminate expenses of translating English to a database language such as SQL. However, the prior natural language processing methods have lacked the precision of SQL when retrieving requests containing prepositional meaning. SQL can easily handle logical preposition requests by selecting rows from specific tables using constraints on specific values of columns of those rows. The prior natural language processing methods do not have a similarly succinct method for responding to the precise prepositional meaning of a request.
  • Although natural languages such as English are perfectly capable of conveying prepositional meaning to the average person, the prior natural language processing methods have not been able to accurately process requests containing logical prepositions.
  • Methods for improved detection of contextual and prepositional meaning in natural language processing systems would enable a broad range of applications where a user interface on a direct human conversation level would be preferred to a menu-based or graphical user interface.
  • For instance, help desk services handle a large volume of customer requests. Such requests must often by handled by a human dispatcher for a significant cost on labor, or an automated menu-driven voice-response system which is inexpensive but difficult for customers to interact with a user interface on a direct human conversation level would be preferred to voice-response systems which force customers to navigate a menu tree.
  • Simple appliances such as washing machines and microwave ovens could also benefit from imbedded natural language processors to accept direct conversational commands. More complex appliances such as cellphones could have imbedded natural language systems to complement or even supplant menu-driven screen displays.
  • However, none of these desirable natural language processing applications can succeed without flexible and precise methods for detecting contextual variations in prepositional meaning.
  • Topological methods for semantic networks measure distances between nodes by counting the number of links along paths between nodes. In order for these distances to be meaningful, the number of links along paths between nodes must correspond to actual closeness of semantic meanings. U.S. patent application Ser. No. 09/085830 by Au describes methods for enforcing such correspondence. Enforcing such correspondence ensures that topological distances are a reliable indicator of closeness of meaning between any two semantic network nodes.
  • A semantic network representing closeness in meaning as distances records the closeness in meaning for any two nodes in the semantic network. At the same time, each path through the semantic network records a picture of why two nodes are a particular distance. For instance, in a small hierarchy of nodes in which retriever, beagle and puppy inherit from dog, the path from retriever to retriever to beagle has to two links. If dog in turn inherits directly from mammal, then retriever, beagle and puppy are each two links away from mammal. The inheritance links attached to dog are thus in turn inherited by retriever, beagle and puppy, thus automatically recording any closeness in meaning added to dog on behalf of retriever, beagle and puppy. This transitivity of closeness in distance models the human ability to quickly associate new information with existing symbolic relationships. For instance, if a new link were to make dog inherit directly from pet, then retriever, beagle and puppy would also inherit from pet, each at a distance of only two links. This redundancy of links in a semantic network provides accuracy when disambiguating meaning.
  • For instance, the word retriever could be linked to inherit from another meaning for retriever which has a more abstract meaning, that of something which fetches an object. A sentence of “A database as a better retriever of information than a card catalog,” refers to the more abstract meaning of retriever. On the other hand, the sentence “My golden retriever is a puppy,” would refer to the more concrete meaning of retriever which inherits from dog. By using a semantic network to measure the average distance to other words in a sentence, the shorter distance between meanings can be used to disambiguate meaning. For instance, for the second sentence, the distance between puppy and the concrete meaning of retriever is only two links, whereas the abstract meaning of retriever is three links away. For the second sentence, the more concrete meaning of retriever would thus be chosen. Since a sentence is made of multiple words, each word in a sentence relates to a set of paths through a semantic network. By choosing the shortest paths from that set, the multiple words in a sentence can be used as a redundant pointers to the most appropriate meaning of each word in the sentence. This semantic redundancy models the human ability to ignore noise and grammatical errors in natural language discourse.
  • Since the number of links along paths between nodes is a positive number, distances between nodes are also positive numbers; there are no negative distances. Some prior art semantic networks contain negation links to represent negative connections between ideas. By combining negation links with non-negation links, some prior art logical rules estimate the degree to which nodes are negations of each other, but such rules are not reliable for general-purpose semantic networks where degrees of negation varies by context. There are additional problems with prior art: many paths will contain two or more negation links. The meaning of paths containing two or more negation links varies from context to context, and as noted above, natural language contains too many contexts to model logically.
  • To accurately store and retrieve negation concepts, the present invention stores and retrieves the meaning of negation as the relative absence of a close connection. The absence of a close connection is computed relative to a contextual set of nodes. Thus the problem of how to consistently represent negation is solved by measuring relative distances within a context. As described by U.S. patent application Ser. No. 09/085830 by Au, traversal functions for traversing between nodes can be used to constrain search to a requested category. Such topological methods can identify a node closest to a particular context, even if that context is a combination of many individual nodes. For contexts composed of multiple nodes, the average or minimum distance to the set of nodes can be used to measure closeness.
  • The present invention combines topological methods for categorization with topological methods for contextual filtering, to disambiguate the meaning of negated symbols, as well as symbols relative to other prepositions such as ‘and’ and ‘or’. By constraining contextual searches to specific categories, precise logical relationships implied by semantic networks can be accurately detected. Thus the precise logical functionality of database languages such as SQL can be provided by the present invention while eliminating the cost of maintaining data-relationships and queries by hand.
  • With an accurate basis for storing and retrieving semantic network information according to logical prepositions, semantic networks can be used in place of relational databases, to bypass the tedious and error-prone process of translating natural language data into data relations. In addition, fine shades of contextual meaning can be automatically stored for each phrase-context pair without explicitly defining a relational join for each phrase-context pair.
  • Another advantage of using a natively natural language database comes from the elegance with which new meanings can be stored directly as parse trees of conversational input. Even while parsing a conversational flow of natural language, new shades of semantic meaning can be stored. Phrases encountered in the beginning of a paragraph can be stored as new shades of meaning which help disambiguate the meaning of phrases encountered later in a conversation. The present invention also contains methods to transform meanings encoded as parse trees into standard semantic inheritance trees, thus automatically updating shades of meaning during conversational input with a user.
  • This ability to update shades of meaning helps greatly when processing explanatory dialog, defined as texts in which terms are introduced early in the dialog and later used as building blocks of meaning later in the dialog.
  • The present invention improves computer-implemented natural languages processors, so that they can be used as user interfaces to a variety of commercial applications. FIG. 1 shows a fully networked support system for customer support, in which various supporting activities are coordinated by a natural language processor in a shared central computer, in accordance with one embodiment of the present invention.
  • At the top of FIG. 1 is a customer, shown on the left, interfacing through a camcorder (or web-cam) to a networked personal computer. In this manner, a customer's image and verbal requests can be relayed across the Internet to a private network, shown at the middle of FIG. 1. Other customers may input verbal request over devices such as the cellphone and telephone shown in FIG. 1.
  • When requests reach the private network, they could be monitored and handled by support staff, also equipped with camcorder (or web-cam) to assist customers over a kind of a video-phone connection. Alternatively, requests might be dispatched to the appropriate support staff by a natural language processor, as shown in FIG. 1. Supporting functions which may be orchestrated by a natural language processor as shown in FIG. 1 include scanning of archives, stored in micro film and requesting attention from operators of archives (all shown in the lower left of FIG. 1). Archives may include email messages which can be scanned in terms of their natural language meaning, to correlate previous current requests with previous requests and responses. For instance, requests for explanation of bills may be correlated to billing records and emails described returned merchandise. Alternatively, requests for news clippings could be correlated with previous requests for specific kinds of news of interest and requests to ignore other kinds of news. Other applications include matching of open job descriptions to resumes and matching of email messages to topics requested by email list subscribers, and indexing of email archives by topic. A natural language processor to orchestrate ail of these functions could greatly increase the efficiency of customer service for the above businesses.
  • FIG. 2 shows a natural language processing system deployed on computers connected by peer-to-peer Internet connections, to provide technical support to customers, in accordance with one embodiment of the present invention. To bypass automated-support user interfaces such as graphical user interfaces and menu-driven interfaces, a conversational natural language interface can be supported by a natural language processor in each computer, to exchange semantic information across the Internet as shown. To provide audio and visual linguistic input to the natural language processor in the upper computer, a camcorder (or web-cam) is shown. To provide similar audio and linguistic input back to the user shown in the upper half of FIG. 2, a camcorder (or web-cam) is used to store pre-recorded images and sounds for playback as orchestrated by the natural language processor in the lower computer. Alternatively, the images and sounds sent back to the user can be synthesized by either the lower or upper computer as needed. Peer-to-peer natural language computer systems with these configurations could be useful ways to interactively share and disseminate information that otherwise would have to be interactively disseminated in more labor-intensive lectures, workshops and video-conference sessions.
  • FIG. 3 shows a natural language processing system deployed on a cellphone connected to the Internet, to provide a natural language user interface for dialing, storing and retrieving names, phone numbers from personal address book information, in accordance with one embodiment of the present invention. Such a systems would be capable of searching, downloading and uploading information on the Internet, as well as monitoring the progress of requests to buy, sell, or inquire about merchandise or services on the Internet. Using a natural language interface bypasses the difficulty in learning menu-driven and graphical user interfaces for such activities.
  • FIG. 4 shows a natural language processing system deployed on a microwave oven home appliance, to provide a natural language user interface for operating oven controls. Using a natural language interface bypasses the difficulty in learning menu-driven and graphical user interfaces for operating such appliances, as well as provides a convenient way to report status of operations back to the user. By learning new vocabulary from the user, through conversation, an appliance with such a natural language interface can adapt to a style of conversation convenient for individual users.
  • The present invention uses natural language reasoning, parsing and pattern matching to process natural language in a natural language format. Components of natural language such as images, sound and text are inputs which can be correlated to semantic meanings by the present invention. FIG. 4 shows these inputs in the upper part of the figure, in accordance with one embodiment of the present invention. Using a semantic network to store parsed versions of the inputs, the present invention can then retrieve the stored versions of the inputs using uses natural language reasoning, parsing and pattern matching. FIG. 4 shows these storage and retrieval operations as the pair of arrows in the middle of the figure. The confirmation and results of these operations are produced as output such as images, sound and text. FIG. 4 shows these inputs in the lower part of the figure.
  • In contrast, the prior art database systems rely upon logic, query parsing and relational indexing to process inputs and produce outputs. When operating on data from a broad range of contexts, logic must be supported by numerous structures. These structures include structured queries, typically using Structured Query Language (SQL) and database maintenance commands, such as ‘create table’, ‘update table’ and ‘create index’. FIG. 5 shows these structures as inputs in the upper part of the figure, in accordance with one embodiment of the present invention. Once these structured inputs have been received by a data base processor, they are parsed into a query plan structure used to operate upon table structures categorizing data into rows and columns of symbols. In order for structured queries to succeed, the symbols in these rows and columns must be logically ordered. The standard for this logical ordering is commonly called database normalization, and it requires that relationships between data be logically consistent. FIG. 5 shows data relationships as curved lines labeled as ‘relationship-a’, relationship-b’, relationship-c’. Unfortunately, logical inconsistencies are so common that most database systems sacrifice the quality of logical ordering to achieve a reduction in database cost. The magnitude of this sacrifice is generally larger for larger databases, because the relationships between rows of data create a combinatorial explosion of logical inconsistencies between data rows.
  • The present invention bypasses the need for logical consistency by representing data in a semantic network where meaning is recorded redundantly by multiple paths over semantic links connecting semantic network nodes.
  • The input structures shown in the upper portion of FIG. 5 are also difficult to use. They require users to know formal logic and programming languages. To make the data in database more accessible to the general public, natural language processors have been added as pre-processors to database inputs. FIG. 6 shows the database system of FIG. 5, but showing in the upper left hand corner of FIG. 6 a natural language processor similar to FIG. 4. This architecture allows the general public to access data, but not to maintain the database. When tables or relationships must be added to the database, such modifications must be done by people skilled in formal logic and programming languages.
  • An example of the complex and esoteric structure required by the prior art to process simple natural language requests is shown in FIG. 7. A prior art system as shown in FIG. 7 might get a request for “fascinating art not ugly”. Using prior art natural language processing techniques, this simple phrase would translate into the SQL query shown in the upper part of FIG. 8.
  • The query comprises three clauses: a ‘select’ clause, a ‘from’ clause, and a ‘where’ clause. The beginning of each clause is shown in bold. The processing order of SQL clauses is ‘from’ clause, then ‘where’ clause, then ‘select’ clause. The ‘from’ clause requests data from two tables, art and artAdjective.
  • The first line of the ‘where’ clause joins rows from the art table to rows in the artAdjective table, finding matching pairs of rows where the artAdjectiveNum column matches the num column. For instance, the first two rows of artAdjective have a num of 10 which matches the first row of art which also an artAdjectiveNum of 10. Similarly, the third through sixth rows of artAdjective match the second row of art, and the seventh through eleventh rows of artAdjective match the third row of art. Each of these joined rows is filtered by the second logical requirement in the ‘where’ clause, specifying that adjective=fascinating. This filters all but the two of the joined rows, shown in bold in FIG. 8: “20—fascinating (joining) 2-20—delicately charming figures of pasta” and “30—fascinating (joining) 3-30—stained glass statues of people”. The third line of the ‘where clause’ filters out rows where artAdjectiveNum is among a proscribed the set of artAdjectiveNum values related to the adjective ‘ugly’. Since the artAdjectiveNum of 30 shown in italics in FIG. 8 is therefore proscribed, that artAdjectiveNum's corresponding joined row “30—fascinating (joining) 3-30—stained glass statues of people” is filtered out, leaving only the joined row: “20—fascinating (joining) 2-20—delicately charming figures of pasta”.
  • The ‘select’ clause selects values from two columns from the art table, ‘num’ and ‘artDescription’. These are displayed for the joined row returned by the query, shown in the bottom of FIG. 8.
  • Many problems arise from logical representations of semantic relationships as shown in the prior art of FIG. 8. Since the strength of a relation between an adjective and a noun varies by context, a separate table is required for each context. For instance, the meaning of ‘ugly’ might be less significant if the context were industrial art instead of art. Since there are as many different contexts as there are combinations of nouns, adjectives and verbs, there is no practical way to represent all the different contexts as logical table structures. Other problems with logical representation arise from subjectivity. One person's opinion relating ‘ugly’ to a noun will vary from another person's opinion. Logic does not easily model differences in subjective opinion, because the subjectivity is based mostly on implied relationships for which no explicit logical relationship has been defined.
  • These problems in logical representations can be bypassed by representing semantic relationships directly in a semantic network dictionary. FIG. 11 shows a small semantic network dictionary, in accordance with one embodiment of the present invention, for representing the nouns and adjectives of FIG. 8. By replacing table relations with direct links between nodes, the semantic network of FIG. 11 succinctly maps adjectives like ‘fascinating’ and ‘ugly’ to noun phrases like “stained glass statues of people.” Since semantic links can be used to store various relationships such as ‘is-a’, ‘part-of’ and ‘because-of’, most semantic networks label the link lines with similar labels. However, for simplicity, FIG. 11 shows each ‘is-a’ links with a thin dark line, and each link of any other type with a fat gray line. Generally ‘is-a’ links are strict categorization links. For instance, ‘art’ is a ‘noun’, and ‘desirable’ is an adjective. The gray links are less strict in meaning. For instance, lead is linked to stained glass. Strictly speaking, lead ‘is-a-component-of’ stained glass, along with tint and glass. FIG. 11 does not label each direct link above stained glass with ‘is-a-component-of’ because of the clutter that would cause. Instead, simple gray lines are used to show that some aspect of lead, tint, and glass informs the nature of stained glass. Similarly, as semantic network can be implemented with just two link types, one for storing ‘is-a’ links, and the other for storing ‘informs’ links.
  • To correlate visual and audio input with written input, natural language processor systems often rely upon input processors to map raw input data to specific semantic nodes. FIG. 9 shows a high level overview of a prior art image processor for mapping face expressions to semantic nodes signifying a person's emotions, in accordance with one embodiment of the present invention. The mapping takes place in four steps. The first step abstracts information from the pixels of an image by tracing simple contours around the mouth. The second step abstracts information from contour shapes by measuring the direction and magnitude of their curvature. The last step abstracts information from contour directions and magnitudes by mapping from classes of contour directions and magnitudes to specific semantic nodes of “happy face”, “sad face”, “quizzical face” and “angry face”.
  • FIG. 10 shows an example of a the prior art sound processor design for mapping sounds to words, in accordance with one embodiment of the present invention. In step 1, sounds are input as air pressure amplitude fluctuations over time. In step 2, bands of vibration frequencies are identified from fluctuations in amplitude by a fast Fourier transformation. Each band is identified as a range of frequencies defined in cycles per second also known as Hertz (HZ). The persistence of each detected band of frequencies is mapped over time by the fast Fourier transformation. In step 3, the identification of phonemes is abstracted from specific combinations of frequency bands. For instance, ‘sh’ is a broad band of frequencies, whereas the consonants ‘o’ and ‘a’ are narrower bands. In step 4, the identification of word nodes is abstracted from sequences of phonemes.
  • The semantic network of FIG. 11 stores both the semantic image nodes from FIG. 9 and semantic word nodes FIG. 10. The semantic network of FIG. 11 also stores implied logical relationships. These relationships are implied by the semantic links. For instance, ‘fascinating’ is linked via an inheritor link to “stained glass statues of people”. This stores the same meaning as the row of artAdjective table “30—fascinating” joined by relation-b to the row of art table “3-30—stained glass statues of people” in FIG. 8. To retrieve nodes corresponding to logical relationships from the semantic network of FIG. 11, the method of FIG. 12 combines topological categorization with topological contextual filtering.
  • The method of FIG. 12 starts by automatically mapping inputs to specific semantic nodes in a semantic network, in accordance with one embodiment of the present invention. FIG. 12 step 1 automatically identifies any existing semantic nodes which can be which can be mapped to input data, as well as creating new semantic nodes for input which cannot be mapped to existing semantic nodes. After step 1, the semantic nodes identified by step 1 are presented to filters which determine what role each identified node can play in steps 2 and 3 of the method.’ Filters which supply nodes for these roles are shown as ‘traversal type filter’, ‘category node filter’, ‘context type filter’, and ‘context node filter’. FIG. 13 is a table showing acceptable nodes for each filter type, at various stages of a natural language conversation, in accordance with one embodiment of the present invention. For the start of a conversation, FIG. 13 has rows where stage is marked ‘start’. For mid-conversation input, when dialog has already led to some agreement, FIG. 13 has rows where stage is marked ‘middle’. This allows filtering to shift to take advantage of information which emerges about a conversation, as a conversation develops. Alternatively, the transition from start to middle of the conversation can be automatically detected when the context node set grows to an optimal size, typically about seven nodes. In other implementations of the present invention, at user request, the context node set can be diminished or entirely flushed, to transit the conversational state from middle back to start.
  • FIG. 13 defines the filters themselves in terms of the semantic network of FIG. 11. For instance, ‘traversal type filter’ accepts only nodes which are traversal types. The semantic network of FIG. 11 shows three nodes which inherit from ‘traversal type’: ‘category hierarchy’, ‘general hierarchy’ and ‘list of nodes’. These three nodes would thus be the only nodes acceptable by the traversal type filter. Similarly, the category set filter accepts only nodes which inherit from ‘adjective’ or ‘adverb’ in FIG. 11.
  • Default values are indicated for ‘input traversal type’ and ‘input context type”. In case no nodes pass through the filters to perform these roles, default nodes are substituted. For instance, if no input traversal type nodes are found in the input, the default node ‘general hierarchy’ substitutes for the input traversal type node. Similarly, if no input context type nodes are found in the input, the default node ‘and’ substitutes for the input context type.
  • So that priority could be given toward filling each specific role, a priority-based dispatching mechanism could be used as an alternative to filtering. A dispatching mechanism would be more complex to maintain but could be more accurate. For instance, if the symbols ‘fast run’ were input, fast could either be an adverb (fast:1) for the input category set or a verb (fast:2) for the context node set. On the other hand, run could either be an noun (run:2 meaning a small brook or stream) or a verb (run:1) both for the context node set. To identify both an input category set and a context node set, fast could be identified with fast:1 and run identified with run:1 or run:2. The dispatcher could reject the alternative of identifying fast with fast:2 because this would prevent identification of an input category set.
  • To perform step 2 of FIG. 12, both an input traversal type node and an input category node set must be elicited from step 1. In cases where no input traversal type node passes the traversal type filter, a default traversal type may be substituted for step 2 instead. For instance, FIG. 12 shows a default traversal type of ‘general hierarchy’. In cases when special semantics are involved, other default traversal types may be chosen instead, such as ‘category hierarchy’ when semantics are more formal.
  • Step 2 of FIG. 12 automatically collects a set of candidate nodes, traversing from an input category node set according to traversal functions described in FIG. 14, in accordance with one embodiment of the present invention. For each possible input traversal type, FIG. 14 shows a corresponding traversal function. For instance, the default traversal type ‘general hierarchy’ corresponds to the traversal function ‘any inherited links’. The contextual meaning chosen by the overall of FIG. 12. is limited to the set of candidate nodes returned by the traversal function.
  • Step 3 of FIG. 12 automatically computes a best contextual distance for each candidate node, given a context type passed from the context type filter. The default context type is ‘and’ which in FIG. 15. corresponds to a best contextual candidate function of ‘minimum average distance.’ Step 3 uses this best contextual candidate function to choose a subset from the set of candidate nodes returned by step 2.
  • FIG. 15 is an example of a table of other context types and their corresponding best contextual distance functions, in accordance with one embodiment of the present invention. For instance, the input context type ‘and’ corresponds to a best contextual distance of ‘minimum average distance’. A minimum average distance can be calculated from the topology of a semantic network, by determining the shortest path, in semantic links, between each candidate node and each context node, then averaging those shortest paths and choosing a subset of nodes having minimum average shortest path length to context. FIG. 61 is a flowchart showing how minimum average distance can be calculated, in accordance with one embodiment of the present invention. Step 3 a of FIG. 61 is a loop which iterates through each node of the candidate node set. For each pass through that loop, the shortest path from a single candidate node to each node of the context node set is identified. When the shortest paths to each context node have been identified for the candidate node, step 3 a calculates the average length of those shortest paths and stores that average number for the candidate node. Step 3 a then returns to the top of the loop to process the next candidate node until all candidate nodes have been processed. Step 3 b then determines the minimum average shortest path length recorded by Step 3 a over all candidate nodes, and chooses the subset of candidate nodes having that minimum average shortest path length to context. By choosing the minimum average, the method of FIG. 62 finds the closest nodes to context, in accordance with one embodiment of the present invention.
  • Other input context types correspond to other best contextual distance functions, according to the meaning of that context type. For instance, the input context type ‘not’ corresponds to a best contextual distance of ‘maximum average distance’. FIG. 62 is a flowchart showing how maximum average distance can be calculated. This calculation is identical to minimum average distance except for the last step. In step 3 b of FIG. 62, instead of determining the minimum average shortest path length recorded by FIG. 61 step 3 b over all candidate nodes, FIG. 62 step 3 b determines the maximum average shortest path distance recorded over all candidate nodes. By choosing the maximum rather than minimum average, the method of FIG. 62 finds the farthest nodes from context, to satisfy requests for symbols which are NOT like the context.
  • There are other context types which correspond to different best contextual distance functions. The context type ‘or’ corresponds to a best contextual distance function of ‘minimum distance’. FIG. 63 is a flowchart showing how minimum distance can be calculated, in accordance with one embodiment of the present invention. Step 3 a of FIG. 63, records only the shortest path length between a candidate node and of any node of the context node set, rather than the average shortest path length between a candidate node and all nodes of the context node set as is done in step 3 a of FIG. 61 and FIG. 62. By recording the shortest rather than average path length, the method of FIG. 63 emphasizes candidate nodes which are close to some context nodes even if they are very far from other context node. The result of this emphasis allows divergent contextual semantic meanings to be identified. The methods of FIGS. 61 and 62, in contrast, emphasize convergent semantic meanings, by penalizing candidate nodes which are far from some context nodes even if they are near to other context nodes. In FIG. 63, step 4 a identifies the candidate nodes having the shortest path length to any context node.
  • Other context types require highly stringent best contextual distance functions. The context type ‘absolutely not’ is a more stringent then the context type ‘not’. Its corresponding best contextual distance function ‘infinite minimum distance’ is also more focused. FIG. 64 is a flowchart showing how infinite minimum distance can be calculated, in accordance with one embodiment of the present invention. Step 3 a of FIG. 64 determines only if there is any path from a candidate node to any context node, unlike step 3 a of FIG. 62 which determines the shortest path from a candidate node to each context node. Step 4 b of FIG. 64 chooses only candidate nodes which have no path to any context node, unlike step 4 b of FIG. 62 which chooses the candidate nodes having maximum average path to context nodes. By choosing only nodes which have no path to context, FIG. 64 step 3 thus chooses a subset of the nodes chosen by FIG. 62.
  • Best contextual distance functions can also be made more stringent or more lenient by different methods of tallying distances along paths between candidate nodes and context nodes. Different types of inheritance links can be given different weightings when counting the links in a path to tally the distance along the path. For instance, in FIG. 15, the input context type ‘and equivalent’ has the best contextual distance function ‘minimum average distance, not counting is-a links.’ This best contextual distance function would be identical to that outline by FIG. 61, except that when computing the candidate node's average shortest path length, is-a type inheritance links would not be counted, so that they would have no effect on the path length. By not counting is-a links, all nodes in a is-a inheritance hierarchy are considered equivalent to each other in meaning, for the context type ‘and equivalent’. Such a distance function would thus measure contextual equivalency of meaning of candidate nodes.
  • Step 3 and 4 of FIG. 12 can be automatically performed by any of the contextual distance functions suggested above. After step 4 has returned a subset of candidate nodes tied for best contextual distance, step 5 of FIG. 12 can generate a summary of that subset for presentation to a user of the computer system. Then in step 6, the user's response to that presentation can be acquired, to be assessed in step 7 of FIG. 12. The assessment can be made using a contextual distance function as outlined above by FIG. 61, 62, 63 or 64. The assessment step categorizes which of the four possible categories of response the user's response best matches: deny, ask, confirm, or complain. FIG. 12 shows branches leading from step 7 for each of these categories of response.
  • For user responses categorized as ‘deny’ or ‘ask’, the response can be processed by passing along with the additional input received in step 6 along with the current set of candidate nodes to the beginning of another pass through the method of FIG. 12. This branch flows into the line along the left edge of FIG. 12.
  • For responses categorized as ‘complain’, the response can be processed by passing the additional input received but without the current set of candidate nodes. Dropping the current set of candidate nodes, allows the user to start from scratch without having to deal with parsing problems of the previous input. This branch also flows into the line along the left edge of FIG. 12.
  • For responses categorized as ‘confirm’, the subset of candidate nodes returned by step 4 can be stored in a semantic network along with the processed input. By generating new semantic links to store all of these nodes in the semantic network, new semantic meanings can be later retrieved by other input fed to step 1 of FIG. 12. Thus the method of FIG. 12 enables new semantic meanings to be stored and later retrieved by other natural language input. By storing and retrieving natural language information directly as semantic meanings, the method of FIG. 12 can supercede relational databases and other databases by enabling greater ease of use and better tools for handling of diverse kinds of data.
  • Other categories of response may be needed to handle user responses. For instance, rather than responding directly, the user may respond with a further question. To handle such responses, an additional category response node ‘digress’ can be added to the ‘response type’ hierarchy of FIG. 11, and a corresponding branch of action leading from step 7 in FIG. 12.
  • FIGS. 16 and 17, in accordance with one embodiment of the present invention, show the method of FIG. 12 handling specific input data similar to the SQL Query shown in FIG. 8. To retrieve records from a relational database corresponding to ‘fascinating art not ugly’, the eight line SQL query at the top of FIG. 8 selects information from two joined tables, seeking ‘fascinating’ from one table and avoiding ‘ugly’ in the other table. FIGS. 16 and 17 show how the method of FIG. 12 can more elegantly retrieve similar information from a semantic network corresponding to ‘fascinating art not ugly’. FIG. 16 shows the retrieval of nodes corresponding to ‘fascinating art’ and FIG. 17 shows now these nodes can be winnowed down to correspond to ‘not ugly’.
  • FIGS. 16 and 17 also handle visual inputs of a user's face, to show how image processing as shown in FIG. 9 can be useful when disambiguating the meaning of user input. At the top of FIG. 16, the user input text ‘fascinating art’ and a visual input node of ‘quizzical face’ are received. In step 1, similarly to FIG. 12, these inputs are passed to four filters: a traversal type filter, a category filter, a context type filter, and a context node filter, whose filtering actions are directed by the table in FIG. 13 for the start of a conversation. The traversal type filter does find any node inputs in FIG. 11 inheriting from the node ‘traversal type’, so the traversal filter passes on its default traversal node of ‘general hierarchy.’ The category set filter finds the node ‘fascinating’ in FIG. 11 inheriting from ‘adjective’, allowing ‘fascinating’ to pass on as the input category set. The other input nodes ‘art’ and ‘quizzical face’ do not inherit from ‘adjective’ or ‘adverb’ in FIG. 11, so they are not passed on as nodes of the input category set. The context type filter does not find any input nodes inheriting from ‘and’, ‘or’, or ‘not’, so it passes the default context type of ‘and’. The context node filter finds two input nodes ‘art’ and ‘quizzical face’ which inherit from ‘noun’ in FIG. 11, so it passes them on as an input context node set.
  • FIG. 16 step 2 receives the input category set ‘fascinating’ and the input category set ‘general hierarchy’. Step 2 maps ‘general hierarchy’ via FIG. 14 to a corresponding traversal function ‘any inherited links’. Step 2 uses this traversal function to traverse from the node ‘fascinating’ in FIG. 11 downward to each inheritor of ‘fascinating’: ‘charming’, ‘delicate figurines of pasta’ and ‘stained glass statues of people.’ These three nodes are then collected into the set of candidate nodes passed to step 3.
  • FIG. 16 step 3 loops through the set of three candidate nodes, computing for each candidate node a best contextual distance to the input context node set. Step 3 uses the table of FIG. 15 and the context type passed into step 3 of ‘and’ to determine that the best contextual distance function is ‘minimum average distance’. FIG. 61 shows the steps in applying the minimum average distance function. FIG. 16 step 3 shows results of applying FIG. 61 step 3 a to the semantic network of FIG. 11. From the candidate node ‘charming’, the shortest path to ‘art’ has 2 links: from ‘art’ to ‘delicate figurines of pasta’ to ‘charming’. From the candidate node ‘charming’ to the other context node ‘quizzical face’ the shortest path has 6 links. Thus the average shortest path from ‘charming’ to the two context nodes is (2+6)/2=8/2=4 links. In contrast, two other candidate nodes have shorter average shortest paths. The shortest path from ‘delicate figurines of pasta’ is just one link long to ‘art’. The shortest path from ‘delicate figurines of pasta’ to ‘quizzical face’ is six links long. So the average shortest path from ‘delicate figurines of pasta’ to the two context nodes ‘art’ and ‘quizzical face’ is (1+6)/2=7/2=3.5 links. Similarly, the average shortest path from ‘stained glass statues of people’ to the two context nodes ‘art’ and ‘quizzical face’ is also 3.5 links. These shortest paths are outline in step 3 of FIG. 16.
  • In step 4 of FIG. 16, a subset of the candidate nodes are chosen having best contextual distance. Since candidate nodes ‘delicate figurines of pasta’ and ‘stained glass statues of people’ have the minimal average distance, as described by FIG. 15 for the input context type ‘and’, they are chosen, eliminating the candidate node ‘charming’. Elimination of the candidate node ‘charming’ results from the relatively distant position of ‘charming’ from ‘art’ in the semantic network of FIG. 11. From a linguistic point of view, this corresponds to the notion that art is less directly connected to the adjective charming than it is to specific instances of art. In larger semantic networks, structured according to methods outline by was described in patent application Ser. No. 09/085830 by Au, ‘art’ may head a larger subtree, increasing the distance from ‘art’ to various instances of art, but the distance between ‘art’ and the adjective ‘charming’ would also increase, remaining greater.
  • In step 5 of FIG. 16, the subset of candidate nodes chosen in step 4 are described as two nodes texts connected by ‘or’ for presentation to a user. Other presentations to the user may also be useful, such as a simple scrolling list or a list integrated into a graphical user interface relating parsed user input items to list elements.
  • Step 6 of both FIG. 12 and FIG. 16 gathers from the user a response to the presentation of step 4. The purpose of gathering a response is to validate that the candidate nodes displayed by step 5 are desired. Step 6 of FIG. 16 shows an example of a response, which includes the text ‘not ugly’ and the visual input node of ‘unhappy face’. This response is assessed in terms of its input nodes in step 7 by a method outlined in FIG. 18.
  • The method in FIG. 18, in accordance with one embodiment of the present invention, is a variation on the topological classification of an input response shown in FIG. 12. The method of FIG. 18 begins with of a set of response nodes in step 7 a, and, by determining which response type node most closely corresponds to the set of input nodes. The method of FIG. 18 begins by checking for polysemy in the input text. Polysemy can be detected as an input word or input phrase which can be associated with two or more nodes of the same spelling in a semantic network. For instance, the word ugly in the input text of step 6 FIG. 16 can be associated with two nodes both spelled ‘ugly’ in FIG. 11: ugly:1 and ugly:2. By measuring distances in the semantic network to find the set of nodes with best contextual distance, Step 7 c of FIG. 18 disambiguates which of these two nodes is most meaningful in the context of the input. Step 7 d returns this set of most meaningful nodes. Step 7 e returns to FIG. 12 to perform actions such as storing summaries of candidate nodes of FIG. 12 or dropping the current set of candidate nodes. FIG. 19, in accordance with one embodiment of the present invention, shows an the method of FIG. 18 with a different set of inputs having no text, only an image of a happy face. Since there is no polysemy detected, semantic distances are only compared between candidate nodes and the context node “happy face”.
  • FIG. 20, in accordance with one embodiment of the present invention, shows details of step 7 a of FIGS. 18 and 19. Similarly to the method of FIG. 12, pre-processors map inputs to nodes and pass these nodes to category set, context type, and context node filters. After filtering, candidate nodes are chosen by which node has the best contextual distance, to disambiguate polysemy by reducing the set of candidate nodes to the subset having best contextual distance. FIG. 20 shows how an input of a happy face maps to a best contextual distance for the candidate node ‘ugly:2’ which is ugly in the sense of desirable, scary and entertaining. In contrast, FIG. 21, in accordance with one embodiment of the present invention, shows how an input of a sad face maps to a best contextual distance for the candidate node ‘ugly:1’ which is ugly in the sense of an undesirable toxic substance. FIGS. 19 and 20 show how subtle shades of meaning recorded in the topology of a semantic network guide disambiguation of polysemy, by applying best contextual distance functions.
  • FIGS. 9 through 20 show how the present invention can be used for natural language processing systems engaged in some form of dialog with a user. By repeatedly responding to a user's input while refining a context node set, the conversational context of a dialog can be detected and confirmed. Once a conversational context has been established, other contextually driven meanings can be determined from the stream of a natural language input. The use of a word to refer to a preceding phrase is highly dependent upon context. The grammatical term for this is anaphora. Common examples of anaphora involve the words it and do. For instance in the sentence “The time has come, I know it” the it refers to the phrase “The time has come.” In natural language processing anaphora are difficult to map without using contextual clues. Not only key words like he, she, it, do demand mapping, but also his, her, its, was and even ellipsis, where the actual anaphora's key word has been omitted. By leaving out the ‘it’ the previous example becomes an example of ellipsis combined with anaphora: “The time has come, I know.”
  • FIG. 22 shows a method for automatically disambiguating anaphora by connecting phrases to the branch of a phrase tree having the best match in term of a contextual distance function, in accordance with one embodiment of the present invention. Unlike the method of FIG. 12, the method of FIG. 22 does not involve user feedback, since it relies upon inputs and context information already established. Thus the method of FIG. 22 can be used for text-stream parsing and automatic summarization and translation of natural language texts.
  • The method of FIG. 22 can also be incorporated into a methods such as FIG. 12 to support a robust parsing of input text.
  • The method of FIG. 22 begins with inputs of a parsed phrase tree and an anaphora which is a subset of that phrase tree. Similarly to FIG. 11, FIG. 23, in accordance with one embodiment of the present invention, shows a small semantic network from which best contextual distances can be measured for the method of FIG. 22. FIG. 23 contains a dictionary of vocabulary related to ordering food in a restaurant. In step 1 a of FIG. 22, as after step 1 of FIG. 12, context type and corresponding best contextual distance functions are identified using filtering or dispatching methods as discussed earlier for FIG. 12. In step 1 b of FIG. 22, context node set is identified also using filtering or dispatching methods as discussed in FIG. 12, but in FIG. 22 filters or dispatches from the input phrase tree of nodes rather than an input set of nodes as in FIG. 12.
  • The method of FIG. 22 resolves anaphora by looking for a best possible branch of the parsed phrase tree to connect to the anaphora subset of the phrase tree. Upon resolution, the anaphora subset will be relocated to link directly to the best possible branch of the phrase tree. Thus the candidate node set in FIG. 22 comes from the set of nodes in the input phrase tree to which the anaphora subset can be connected; the candidate notes are NOT in the anaphora subset. Step 1 c identifies these candidate nodes.
  • Step 2 of FIG. 22 identifies the best candidate branch node of the phrase tree according to the optimal contextual distance function for the context type. As in FIG. 12, the best contextual distance function is chosen according to the table in FIG. 15. For the pronoun keywords ‘his’, ‘her’ and the preposition keyword ‘and’ the best contextual distance function is ‘minimum average distance’. In contrast, the prepositional keywords ‘not’ has the best contextual distance function ‘maximum average distance.’ FIG. 22 Step 3 reconfigures the phrase tree to connect the anaphora subset to the best candidate node, thus recording the meaning of the anaphora.
  • FIG. 24 shows a simple phrase tree in which all the nodes are in a single inheritance list, in accordance with one embodiment of the present invention. As an input stream of natural language texts is processed, the structure of the input is often just a list which is then processed into a more detailed tree structure reflecting grammatical and syntactic meanings. FIG. 24 shows an example of the beginning of such a processing operation. Scanning for a keyword from the end of the list, the prepositional keyword ‘and’ is found. Prepositional keywords such as ‘and’, ‘or’ and ‘not’ are significant because each prepositional phrase must be properly connected to prepositional subject. Other anaphora keywords such as ‘his’, ‘her’, ‘its’ and ‘this’ or ‘that’ may also be sought when scanning for keywords. Since these connections are influenced by natural language context, the method of FIG. 22 is used to find the best branch node of the phrase tree for that connection. The anaphoric subset of FIG. 24 thus becomes the phase ‘and potatoes’ for which the method of FIG. 22 will find a best prepositional subject in terms of a best candidate node.
  • FIG. 25, in accordance with one embodiment of the present invention, shows the method of FIG. 22 applied to the parse tree of FIG. 24. In step 1 a of FIG. 25, filtering the input phrase tree subset ‘and pototoes’ produces the prepositional keyword and context type ‘and’ and the corresponding best contextual distance function of ‘minimum average distance’ from the table on FIG. 13. In FIG. 25 step 1 b, filtering produces the context node set ‘potatotes’ In step 1 c, the candidate node set is produced from the nodes of the phrase tree which are NOT in the anaphora subset of ‘and potatoes’. This candidate nodes of ‘She’, ‘would like’, ‘salad’, ‘not’ and ‘steak’ are then measured in step 2 to find the candidate with best average contextual distance to each node in the context node set. In FIG. 25 there is one node in the context set, ‘potatoes’. Other drawings will show how more nodes in the context nodes set affects the average best contextual distances measured. In step 3, since according to the semantic network of FIG. 23 the closest contextual distance from potatoes to steak, the anaphora subset of ‘and potatoes’ is connected to ‘steak’ in the parse tree.
  • FIG. 26, in accordance with one embodiment of the present invention, shows the parse tree of FIG. 24 after reconnecting the anaphora subset of ‘and potatoes’ to steak. Since the preposition ‘and’ refers to both steak and potatoes, the keyword ‘and’ node heads the subtree to which steak and potatoes are connected.
  • FIG. 27, in accordance with one embodiment of the present invention, shows a contrasting input phrase tree in which ‘and blue cheese dressing’ is in place of ‘and potatoes’. FIG. 28, in accordance with one embodiment of the present invention, shows the method of FIG. 22 applied to the parse tree of FIG. 27 using the same dictionary of FIG. 23. In FIG. 29 shows, in accordance with one embodiment of the present invention, in step 1 b the context node set produced consists of two nodes: “blue cheese” and “dressing”. In step 3 of FIG. 29, distances to this two node context node set is are evaluated for each of the candidate node. Since the dictionary of FIG. 23 shows that blue cheese and dressing are closer to salad than the other candidates, step 3 connects the anaphora subset phrase “and blue cheese dressing” to “salad”. FIG. 28 shows the parse tree of FIG. 27 after step 3 of FIG. 29 has reconnected “salad” and “blue cheese dressing” under the keyword “and”.
  • FIG. 23 shows a semantic network dictionary isolated from the parse trees of FIGS. 24, 26, 27 and 28. Since parse trees contain valuable semantic information, they can be stored directly in a semantic network together with dictionary symbol nodes. By recording how dictionary symbol nodes are actually used in parsed trees, the meaning of dictionary symbol nodes is more complete and up-to-date. At the same time, the meaning of parsed tree nodes is more accurately tied to the dictionary.
  • FIG. 30, in accordance with one embodiment of the present invention, shows the semantic network dictionary of FIG. 23 augmented by the parse trees of FIGS. 26 and 28. Since the parse trees of FIGS. 26 and 28 both begin with “She would like”, they have been automatically combined under the header ‘She would like’ to compact their storage. The meaning of the parsed node dressing:2 inherited directly from the dictionary node ‘dressing’ to record that the parser has successfully disambiguated the meaning of dressing.
  • By combining parse trees with a dictionary as in FIG. 30, the method of FIG. 22 can adjust automatically to new meanings associated with parsed natural language input. The meaning of subsequent natural language input can be detected using meanings automatically recorded in this way.
  • FIG. 31, in accordance with one embodiment of the present invention, shows how meanings automatically recorded in FIG. 30 are useful when processing subsequent natural language text. Using the same method steps outlined by FIG. 22 and using the semantic network of FIG. 30, the natural language input text of “She would like a lot of cheese” correctly maps the meaning of cheese to cheese:2 in the dictionary of FIG. 30. Cheese:2 is the cheese which more closely associated with salad dressing, as opposed to cheese:1 which is associated with pizza. In contrast, FIG. 33, in accordance with one embodiment of the present invention, shows the same natural language input text of “She would like a lot of cheese” and same method steps outlined by FIG. 22 but using the semantic network of FIG. 23. The meaning of cheese is correctly mapped to cheese:1 in FIG. 33 because cheese:1 of a pizza is more closely associated with ‘a lot of’ in the dictionaries of FIGS. 23 and 30.
  • As an alternative to the semantic network dictionary of FIG. 30, FIG. 32, in accordance with one embodiment of the present invention, shows the semantic network dictionary of FIG. 23 after adding parsed nodes from the input phrase “She would like a lot of cheese.” FIG. 32 shows how differently the meaning of the parsed nodes She:2 and ‘would like:2’ are recorded when the input “She would like a lot of cheese” precedes “She would like salad not steak and potatoes.” As parsed nodes accumulate in a semantic network dictionary, the present invention uses them to more accurately attribute meanings to subsequent natural language input. However, if too many parsed nodes accumulate, their topology needs to be compacted. General methods for compacting them are outlined in patent application Ser. No. 09/085,830. Specific methods for compacting parsed nodes are outlined in drawing FIGS. 43 through 60 in the present invention.
  • Examples using the method of FIG. 22 to disambiguate the meaning of pronoun anaphora are shown in FIG. 34 through 38, in accordance with one embodiment of the present invention. In FIG. 34, a small semantic network dictionary is shown. FIG. 35 shows an input parse tree in the form of a list, in which the anaphora pronoun keyword ‘her’ has been marked to delimit the start of an anaphora subset phrase of ‘her food’. The other delimiter is the verb ‘was’ which marks the end of the anaphora subset phrase. FIG. 36 shows how the steps of FIG. 22 can be used with the dictionary of FIG. 34 to map the anaphora subset of ‘her food’ to best branch node of FIG. 35. Since the name ‘Amy’ is closer to ‘her’ in the dictionary than the other candidates of ‘should be’, ‘here’, ‘.’, ‘Harold’, ‘said’, ‘was’ and ‘ready’, step 3 reconnects ‘her’ and ‘food’ to ‘Amy’.
  • In contrast, FIG. 37 shows an input parse tree in the form of a list, in which the anaphora pronoun keyword ‘his’ has been marked to delimit the start of an anaphora subset of ‘his food’. The other delimiter is the verb ‘was’ which marks the end of the anaphora subset phrase. FIG. 38 shows the steps of the method in FIG. 22 used with the dictionary of FIG. 34 and the input phrase tree of FIG. 37. Since the name ‘Harold’ is closer to ‘his’ in the dictionary than the other candidates of ‘Amy’, ‘should be’, ‘her’, ‘.’, ‘Harold’, ‘said’, ‘was’ and ‘ready’, FIG. 38 step 3 reconnects ‘his’ and ‘food’ to ‘Harold’.
  • FIG. 39, in accordance with one embodiment of the present invention, shows a small semantic network dictionary emphasizing how long chains of categorization hierarchy links can classify various meanings for the symbol ‘Chicago’. FIG. 40, in accordance with one embodiment of the present invention shows a variation of the method of FIG. 12, amended to categorize nodes by spelling and by a complex traversal function, to disambiguate the meaning of the phrase “Chicago bigger than a bread box?”
  • Step 2 of FIG. 40 describes the complex traversal function, which travels long chains of categorization hierarchy links to retrieve a set of candidate nodes. These candidate nodes are compared using a best contextual distance function as in FIG. 12. FIG. 41, in accordance with one embodiment of the present invention is an example of how the method of FIG. 40 can be used to map candidates of Chicago:1, Chicago:2 and Chicago:3 to a context node set of “bread box”, correctly identifying Chicago:2 as the best meaning of Chicago in “Chicago bigger than bread box.”
  • As shown earlier, in FIG. 30, a semantic network dictionary combined from dictionary node symbols and parsed natural language node symbols can be used to automatically add parsed semantic information to a large semantic dictionary. When a large number of parsed natural language nodes is added to a dictionary, the topology of the dictionary grows irregular and often repeats the same parse tree fragments. To compact the dictionary, FIG. 43, in accordance with one embodiment of the present invention, shows a method to convert parsed node subtrees to corresponding is-a link dictionary nodes. In accordance with one embodiment of the present invention, Is-a links are standard classification hierarchy links, shown as narrow black lines in FIG. 44, as opposed to the more vague ‘informs’ links shown as wide gray lines in FIG. 44. In step 1 of FIG. 43, parse trees marked for compaction are detected by their inheritance link to the key node ‘inversion’ (shown in bold). In step 2 of FIG. 43, the subjects of these parse tree are marked, and semantic links to the subjects are reconfigured to inherit from the predicate via direct is-a link. Since the parse tree has now been recorded as a standard dictionary node, step 3 of FIG. 43 deletes the redundant nodes from the parse tree to compact the semantic network. Step 3 then returns to steps 1 as a loop, to automatically compact other parse trees.
  • FIG. 42, in accordance with one embodiment of the present invention shows an example of a series of natural language inputs to be recorded by a semantic network by methods of FIG. 22 and FIG. 43. FIG. 44, in accordance with one embodiment of the present invention, shows a small semantic network dictionary for use by methods of FIGS. 22 and 43 to record the inputs shown in FIG. 42. FIG. 45 shows the input sentence “My friend Hal was a programmer” recorded into the semantic network of FIG. 44 by method of FIG. 22. The node was:3 is shown in bold to indicate that it inherits from ‘inversion’ and therefore will be detected by the method of FIG. 43 for inversion to dictionary format.
  • FIG. 46, in accordance with one embodiment of the present invention, shows how the subtree of was:3 is inverted to dictionary format node programmer:2 inheriting from was:1 and programmer:1. Since step 1 also links Hal:2 to inherit from programmer:2, the subtree of was:3 has now been fully inverted to dictionary format. Hal:2 now inherits the meaning of programmer:2 directly from an is-a link, exactly as a standard dictionary node instead of a parse tree node. In FIG. 47, in accordance with one embodiment of the present invention, the redundant subtree nodes was:3, a:4 and programmer:3 nodes are deleted, to compact the semantic network.
  • The next two lines of input from FIG. 42 are recorded by the method of FIG. 22 to augment the semantic network of FIG. 47. These inputs are “His number is (402) 287-0101. His address is 404 Fesser Street in Detroit.” This augmented semantic network is shown in FIG. 48, in accordance with one embodiment of the present invention. Since number and address as predicate concepts are not intrinsic to their subjects, the is:1 node associated with number:1 and address:1 are do not inherit from ‘inversion’. This prevents the method of FIG. 43 from improperly inverting parse node subtrees which inherit from is:1. Since is:1 is closer in the semantic network to number:1 and address:1, the method of FIG. 22 connects the parse node is:4 to is:1 rather than is:2.
  • FIG. 49, in accordance with one embodiment of the present invention, shows the semantic network of FIG. 48 after applying the method of FIG. 22 on the fourth input line of FIG. 42: “He is a good trumpet player.” Since is:2 is closer to player:1 than is:1 in FIG. 48, the method of FIG. 22 connects the parse node is:5 to is:2 rather than is:1. This permits the method of FIG. 43 to correctly invert the parse node subtree is:5 to standard dictionary node is-a links, shown in FIG. 50.
  • FIG. 43 through 50, in accordance with one embodiment of the present invention thus show how semantic information can be automatically parsed into standard semantic network dictionary format, to automatically create large semantic network dictionaries without manual intervention by lexicographers and linguists. These labor-saving methods can be extended to also automatically record semantic information contained questions posed to a natural language processor. FIG. 51 through 60, in accordance with one embodiment of the present invention, show how parsed questions can be stored in a semantic network dictionary just as parsed statements can be stored. The parsed questions can be marked for later linking to their answers, as shown in FIG. 59, and entire question-answer parsed subtrees can be inverted to standard dictionary is-a link nodes as shown in FIG. 60. The method shown in FIG. 58 shows how to invert questions into is-a links. FIG. 51 shows how to generate an answer to a question, by describing nodes which best in the sense of a best contextual distance to the question nodes. FIG. 52 shows how to map question node types to traversal functions to fetch candidate nodes for answering questions. A set of candidate nodes having best contextual distance to the question nodes can then be chosen as an answer to questions.
  • FIG. 53 shows how the method of FIG. 51 answers the question from FIG. 42 by creating nodes in the semantic network of FIG. 50. Since the present invention automatically acquires semantic network meanings from natural language statements, questions and answers to questions, the present invention can automatically acquire all type of natural language meanings from natural language dialog. This automatic language acquisition functionality, together with the accuracy of the retrieval methods of the present invention makes the comparatively laborious data base maintenance work of relational database obsolete.

Claims (25)

1-19. (canceled)
20. A method of providing customer support using natural language input, the method comprising:
receiving a natural language search input, wherein the natural language input includes at least one word;
analyzing each of the at least one word of the natural language input using a semantic inheritance network; and
determining a contextual meaning for the natural language input using each of the at least one word of the natural language input analyzed using the semantic inheritance network.
21. The method of claim 20, wherein analyzing each of the at least one word of the natural language input using a semantic inheritance network includes:
analyzing each of the at least one word of the natural language input using a plurality of candidate nodes and at least one contextual distance; and
determining at least one preferred contextual distance.
22. The method of claim 21, wherein determining a contextual meaning for the natural language input using each of the at least one word of the natural language input analyzed using the semantic inheritance network includes:
applying the at least one contextual distance to the plurality of candidate nodes to produce a contextual meaning for the natural language input.
23. The method of claim 22, wherein the semantic inheritance network includes a stored natural language context including a plurality of context nodes, and wherein analyzing each of the at least one word of the natural language input using a plurality of candidate nodes and at least one contextual distance includes:
identifying a plurality of links, wherein each of the plurality of links traverses from one of the plurality of candidate nodes to one of the plurality of context nodes; and
computing each of the at least one contextual distance for each of the plurality of candidate nodes, each of the at least one contextual distance occurring between one of the plurality of candidate nodes and one of the plurality of context nodes.
24. The method of claim 23, wherein determining at least one preferred contextual distance includes:
comparing each of the at least one contextual distance to one another.
25. The method of claim 24, wherein determining at least one preferred contextual distance further includes:
selecting a subset of candidate nodes from the plurality of candidate nodes.
26. The method of claim 25, wherein determining at least one preferred contextual distance further includes:
determining a preferred contextual meaning corresponding to the natural language input;
wherein each of the subset of candidate nodes has an associated contextual distance approximating an optimal contextual distance, such that each of the input nodes has at least one candidate node selected from the subset of candidate nodes, the at least one associated candidate node being the preferred contextual meaning for the corresponding one from the set of input nodes.
27. The method of claim 26, wherein the semantic inheritance network includes a plurality of semantic nodes, and wherein each of the subset of candidate nodes inherits from a common semantic network node, the common semantic network node being selected from the plurality of semantic nodes.
28. The method of claim 26, wherein determining at least one preferred contextual distance further includes:
determining a greatest minimal distance among each of the plurality of candidate nodes and each of the plurality of context nodes.
29. The method of claim 26, wherein, if no links are identified as traversing among one of the plurality of candidate nodes and any one of the plurality of context nodes, estimating a distance between each of the plurality of candidate nodes and each of the plurality of context nodes.
30. The method of claim 26, wherein each of the plurality of links has a corresponding one of a plurality of semantic link types.
31. The method of claim 26, wherein no links are identified as traversing from one of the plurality of candidate nodes to one from the plurality of context nodes for at least one of the plurality of semantic link types.
32. The method of claim 26, wherein identifying a plurality of links includes:
excluding each of the plurality of links having the at least one of the plurality of semantic link types for which no links are identified.
33. The method of claim 20, wherein the network is the Internet.
34. The method of claim 20, wherein the method is performed by a processor.
35. The method of claim 34, wherein the processor is contained in a terminal.
36. The method of claim 35, wherein the terminal is selected from a group consisting of a personal computer, a minicomputer, a microcomputer, a main frame computer, a telephone device, and a hand-held device.
37. The method of claim 34, wherein the processor is imbedded in a device having voice synthesis capability.
38. The method of claim 22, wherein the plurality of candidate nodes have inheritance sibling nodes and links among the inheritance sibling nodes in a topology, the method further comprising:
applying an abstractness analysis to the topology for each of the candidate nodes and to at least one alternative topology for each of the candidate nodes; and
selecting at least one preferred topology based on the abstractness analysis.
39. The method of claim 38, wherein selecting at least one preferred topology based on the abstractness analysis includes:
weighting the topology and each of the at least one alternative topology.
40. The method of claim 22, further comprising:
identifying a customer support query most closely corresponding to the subset of the plurality of possible words having the contextual meaning meeting the preferred contextual meaning threshold.
41. The method of claim 40, further comprising:
providing a customer support function corresponding to the identified customer support query.
42. The method of claim 41, wherein the customer support function is a text or sound response to the customer support query.
43. The method of claim 42, wherein the sound response includes voice information.
US11/503,938 1998-05-28 2006-08-15 Method and system for using natural language input to provide customer support Abandoned US20070094225A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US11/503,938 US20070094225A1 (en) 1998-05-28 2006-08-15 Method and system for using natural language input to provide customer support

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US09/085,830 US6778970B2 (en) 1998-05-28 1998-05-28 Topological methods to organize semantic network data flows for conversational applications
US34236001P 2001-12-27 2001-12-27
US10/329,402 US7711672B2 (en) 1998-05-28 2002-12-27 Semantic network methods to disambiguate natural language meaning
US11/503,938 US20070094225A1 (en) 1998-05-28 2006-08-15 Method and system for using natural language input to provide customer support

Related Parent Applications (1)

Application Number Title Priority Date Filing Date
US10/329,402 Continuation US7711672B2 (en) 1998-05-28 2002-12-27 Semantic network methods to disambiguate natural language meaning

Publications (1)

Publication Number Publication Date
US20070094225A1 true US20070094225A1 (en) 2007-04-26

Family

ID=37986485

Family Applications (5)

Application Number Title Priority Date Filing Date
US10/329,402 Expired - Fee Related US7711672B2 (en) 1998-05-28 2002-12-27 Semantic network methods to disambiguate natural language meaning
US11/503,938 Abandoned US20070094225A1 (en) 1998-05-28 2006-08-15 Method and system for using natural language input to provide customer support
US12/577,160 Expired - Fee Related US8135660B2 (en) 1998-05-28 2009-10-10 Semantic network methods to disambiguate natural language meaning
US12/577,159 Expired - Fee Related US8204844B2 (en) 1998-05-28 2009-10-10 Systems and methods to increase efficiency in semantic networks to disambiguate natural language meaning
US12/715,968 Expired - Fee Related US8200608B2 (en) 1998-05-28 2010-03-02 Semantic network methods to disambiguate natural language meaning

Family Applications Before (1)

Application Number Title Priority Date Filing Date
US10/329,402 Expired - Fee Related US7711672B2 (en) 1998-05-28 2002-12-27 Semantic network methods to disambiguate natural language meaning

Family Applications After (3)

Application Number Title Priority Date Filing Date
US12/577,160 Expired - Fee Related US8135660B2 (en) 1998-05-28 2009-10-10 Semantic network methods to disambiguate natural language meaning
US12/577,159 Expired - Fee Related US8204844B2 (en) 1998-05-28 2009-10-10 Systems and methods to increase efficiency in semantic networks to disambiguate natural language meaning
US12/715,968 Expired - Fee Related US8200608B2 (en) 1998-05-28 2010-03-02 Semantic network methods to disambiguate natural language meaning

Country Status (1)

Country Link
US (5) US7711672B2 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050010561A1 (en) * 2003-04-28 2005-01-13 France Telecom System for generating queries
US7698266B1 (en) * 1999-11-01 2010-04-13 Google Inc. Meaning-based advertising and document relevance determination

Families Citing this family (339)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2001013255A2 (en) 1999-08-13 2001-02-22 Pixo, Inc. Displaying and traversing links in character array
US8645137B2 (en) 2000-03-16 2014-02-04 Apple Inc. Fast, language-independent method for user authentication by voice
US20020059204A1 (en) * 2000-07-28 2002-05-16 Harris Larry R. Distributed search system and method
US7013308B1 (en) 2000-11-28 2006-03-14 Semscript Ltd. Knowledge storage and retrieval system and method
JP2002288201A (en) * 2001-03-23 2002-10-04 Fujitsu Ltd Question-answer processing method, question-answer processing program, recording medium for the question- answer processing program, and question-answer processor
ITFI20010199A1 (en) 2001-10-22 2003-04-22 Riccardo Vieri SYSTEM AND METHOD TO TRANSFORM TEXTUAL COMMUNICATIONS INTO VOICE AND SEND THEM WITH AN INTERNET CONNECTION TO ANY TELEPHONE SYSTEM
US6907417B2 (en) * 2002-03-28 2005-06-14 International Business Machines Corporation System and method for converting node-and-link knowledge representations to outline format
US7873223B2 (en) * 2002-10-15 2011-01-18 Definiens Ag Cognition integrator and language
US20040167883A1 (en) * 2002-12-06 2004-08-26 Attensity Corporation Methods and systems for providing a service for producing structured data elements from free text sources
US7669134B1 (en) 2003-05-02 2010-02-23 Apple Inc. Method and apparatus for displaying information during an instant messaging session
US7958132B2 (en) * 2004-02-10 2011-06-07 Microsoft Corporation Voting based scheme for electronic document node reuse
KR100590553B1 (en) * 2004-05-21 2006-06-19 삼성전자주식회사 Method and apparatus for generating dialog prosody structure and speech synthesis method and system employing the same
US7337170B2 (en) * 2005-01-18 2008-02-26 International Business Machines Corporation System and method for planning and generating queries for multi-dimensional analysis using domain models and data federation
JP4423327B2 (en) * 2005-02-08 2010-03-03 日本電信電話株式会社 Information communication terminal, information communication system, information communication method, information communication program, and recording medium recording the same
US8849860B2 (en) 2005-03-30 2014-09-30 Primal Fusion Inc. Systems and methods for applying statistical inference techniques to knowledge representations
US9378203B2 (en) 2008-05-01 2016-06-28 Primal Fusion Inc. Methods and apparatus for providing information of interest to one or more users
US9104779B2 (en) 2005-03-30 2015-08-11 Primal Fusion Inc. Systems and methods for analyzing and synthesizing complex knowledge representations
US10002325B2 (en) 2005-03-30 2018-06-19 Primal Fusion Inc. Knowledge representation systems and methods incorporating inference rules
US7849090B2 (en) 2005-03-30 2010-12-07 Primal Fusion Inc. System, method and computer program for faceted classification synthesis
US9177248B2 (en) 2005-03-30 2015-11-03 Primal Fusion Inc. Knowledge representation systems and methods incorporating customization
US7548849B2 (en) * 2005-04-29 2009-06-16 Research In Motion Limited Method for generating text that meets specified characteristics in a handheld electronic device and a handheld electronic device incorporating the same
US8666928B2 (en) 2005-08-01 2014-03-04 Evi Technologies Limited Knowledge repository
US8677377B2 (en) 2005-09-08 2014-03-18 Apple Inc. Method and apparatus for building an intelligent automated assistant
US7633076B2 (en) 2005-09-30 2009-12-15 Apple Inc. Automated response to and sensing of user activity in portable devices
US20070143273A1 (en) * 2005-12-08 2007-06-21 Knaus William A Search engine with increased performance and specificity
US8229733B2 (en) * 2006-02-09 2012-07-24 John Harney Method and apparatus for linguistic independent parsing in a natural language systems
US9318108B2 (en) 2010-01-18 2016-04-19 Apple Inc. Intelligent automated assistant
EP2076838A4 (en) * 2006-10-05 2010-03-17 Brainwave Applic Ltd A novel database
US7555478B2 (en) * 2006-12-05 2009-06-30 Yahoo! Inc. Search results presented as visually illustrative concepts
US9754273B2 (en) * 2006-12-19 2017-09-05 Microsoft Technology Licensing, Llc Enterprise resource tracking of knowledge
US8977255B2 (en) 2007-04-03 2015-03-10 Apple Inc. Method and system for operating a multi-function portable electronic device using voice-activation
US7899666B2 (en) 2007-05-04 2011-03-01 Expert System S.P.A. Method and system for automatically extracting relations between concepts included in text
US8069188B2 (en) * 2007-05-07 2011-11-29 Applied Technical Systems, Inc. Database system storing a data structure that includes data nodes connected by context nodes and related method
ITFI20070177A1 (en) 2007-07-26 2009-01-27 Riccardo Vieri SYSTEM FOR THE CREATION AND SETTING OF AN ADVERTISING CAMPAIGN DERIVING FROM THE INSERTION OF ADVERTISING MESSAGES WITHIN AN EXCHANGE OF MESSAGES AND METHOD FOR ITS FUNCTIONING.
US8712758B2 (en) * 2007-08-31 2014-04-29 Microsoft Corporation Coreference resolution in an ambiguity-sensitive natural language processing system
US9053089B2 (en) 2007-10-02 2015-06-09 Apple Inc. Part-of-speech tagging using latent analogy
US8165886B1 (en) 2007-10-04 2012-04-24 Great Northern Research LLC Speech interface system and method for control and interaction with applications on a computing system
US8838659B2 (en) 2007-10-04 2014-09-16 Amazon Technologies, Inc. Enhanced knowledge repository
US8595642B1 (en) 2007-10-04 2013-11-26 Great Northern Research, LLC Multiple shell multi faceted graphical user interface
US8364694B2 (en) 2007-10-26 2013-01-29 Apple Inc. Search assistant for digital media assets
US7844555B2 (en) * 2007-11-13 2010-11-30 Microsoft Corporation Ranker selection for statistical natural language processing
US8620662B2 (en) 2007-11-20 2013-12-31 Apple Inc. Context-aware unit selection
US10002189B2 (en) 2007-12-20 2018-06-19 Apple Inc. Method and apparatus for searching using an active ontology
US9330720B2 (en) 2008-01-03 2016-05-03 Apple Inc. Methods and apparatus for altering audio output signals
US8327272B2 (en) 2008-01-06 2012-12-04 Apple Inc. Portable multifunction device, method, and graphical user interface for viewing and managing electronic calendars
US20090198488A1 (en) * 2008-02-05 2009-08-06 Eric Arno Vigen System and method for analyzing communications using multi-placement hierarchical structures
US8065143B2 (en) 2008-02-22 2011-11-22 Apple Inc. Providing text input using speech data and non-speech data
US8289283B2 (en) 2008-03-04 2012-10-16 Apple Inc. Language input interface on a device
US8996376B2 (en) 2008-04-05 2015-03-31 Apple Inc. Intelligent text-to-speech conversion
WO2009132442A1 (en) 2008-05-01 2009-11-05 Sweeney Peter Method, system, and computer program for user-driven dynamic generation of semantic networks and media synthesis
US9361365B2 (en) 2008-05-01 2016-06-07 Primal Fusion Inc. Methods and apparatus for searching of content using semantic synthesis
US8676732B2 (en) 2008-05-01 2014-03-18 Primal Fusion Inc. Methods and apparatus for providing information of interest to one or more users
US10496753B2 (en) 2010-01-18 2019-12-03 Apple Inc. Automatically adapting user interfaces for hands-free interaction
US8464150B2 (en) 2008-06-07 2013-06-11 Apple Inc. Automatic language identification for dynamic text processing
US20100030549A1 (en) 2008-07-31 2010-02-04 Lee Michael M Mobile device having human language translation capability with positional feedback
CA2734756C (en) 2008-08-29 2018-08-21 Primal Fusion Inc. Systems and methods for semantic concept definition and semantic concept relationship synthesis utilizing existing domain definitions
US8768702B2 (en) 2008-09-05 2014-07-01 Apple Inc. Multi-tiered voice feedback in an electronic device
US8898568B2 (en) 2008-09-09 2014-11-25 Apple Inc. Audio user interface
US8712776B2 (en) 2008-09-29 2014-04-29 Apple Inc. Systems and methods for selective text to speech synthesis
US8396714B2 (en) 2008-09-29 2013-03-12 Apple Inc. Systems and methods for concatenation of words in text to speech synthesis
US8352272B2 (en) 2008-09-29 2013-01-08 Apple Inc. Systems and methods for text to speech synthesis
US8355919B2 (en) 2008-09-29 2013-01-15 Apple Inc. Systems and methods for text normalization for text to speech synthesis
US8583418B2 (en) 2008-09-29 2013-11-12 Apple Inc. Systems and methods of detecting language and natural language strings for text to speech synthesis
US8352268B2 (en) 2008-09-29 2013-01-08 Apple Inc. Systems and methods for selective rate of speech and speech preferences for text to speech synthesis
US8676904B2 (en) 2008-10-02 2014-03-18 Apple Inc. Electronic devices with voice command and contextual data processing capabilities
US9959870B2 (en) 2008-12-11 2018-05-01 Apple Inc. Speech recognition involving a mobile device
US8862252B2 (en) 2009-01-30 2014-10-14 Apple Inc. Audio user interface for displayless electronic device
US9805089B2 (en) * 2009-02-10 2017-10-31 Amazon Technologies, Inc. Local business and product search system and method
US8380507B2 (en) 2009-03-09 2013-02-19 Apple Inc. Systems and methods for determining the language to use for speech generated by a text to speech engine
US9858925B2 (en) 2009-06-05 2018-01-02 Apple Inc. Using context information to facilitate processing of commands in a virtual assistant
US10241752B2 (en) 2011-09-30 2019-03-26 Apple Inc. Interface for a virtual digital assistant
US10540976B2 (en) 2009-06-05 2020-01-21 Apple Inc. Contextual voice commands
US10241644B2 (en) 2011-06-03 2019-03-26 Apple Inc. Actionable reminder entries
US20120311585A1 (en) 2011-06-03 2012-12-06 Apple Inc. Organizing task items that represent tasks to perform
US9431006B2 (en) 2009-07-02 2016-08-30 Apple Inc. Methods and apparatuses for automatic speech recognition
US20110060645A1 (en) * 2009-09-08 2011-03-10 Peter Sweeney Synthesizing messaging using context provided by consumers
US20110060644A1 (en) * 2009-09-08 2011-03-10 Peter Sweeney Synthesizing messaging using context provided by consumers
US9292855B2 (en) * 2009-09-08 2016-03-22 Primal Fusion Inc. Synthesizing messaging using context provided by consumers
US9262520B2 (en) 2009-11-10 2016-02-16 Primal Fusion Inc. System, method and computer program for creating and manipulating data structures using an interactive graphical interface
US8682649B2 (en) 2009-11-12 2014-03-25 Apple Inc. Sentiment prediction from textual data
US8712759B2 (en) * 2009-11-13 2014-04-29 Clausal Computing Oy Specializing disambiguation of a natural language expression
US20110119047A1 (en) * 2009-11-19 2011-05-19 Tatu Ylonen Oy Ltd Joint disambiguation of the meaning of a natural language expression
US8504355B2 (en) * 2009-11-20 2013-08-06 Clausal Computing Oy Joint disambiguation of syntactic and semantic ambiguity
US8914401B2 (en) * 2009-12-30 2014-12-16 At&T Intellectual Property I, L.P. System and method for an N-best list interface
US8600743B2 (en) 2010-01-06 2013-12-03 Apple Inc. Noise profile determination for voice-related feature
US8381107B2 (en) 2010-01-13 2013-02-19 Apple Inc. Adaptive audio feedback system and method
US8311838B2 (en) 2010-01-13 2012-11-13 Apple Inc. Devices and methods for identifying a prompt corresponding to a voice input in a sequence of prompts
US10679605B2 (en) 2010-01-18 2020-06-09 Apple Inc. Hands-free list-reading by intelligent automated assistant
US10276170B2 (en) 2010-01-18 2019-04-30 Apple Inc. Intelligent automated assistant
US10705794B2 (en) 2010-01-18 2020-07-07 Apple Inc. Automatically adapting user interfaces for hands-free interaction
US10553209B2 (en) 2010-01-18 2020-02-04 Apple Inc. Systems and methods for hands-free notification summaries
WO2011089450A2 (en) 2010-01-25 2011-07-28 Andrew Peter Nelson Jerram Apparatuses, methods and systems for a digital conversation management platform
US8682667B2 (en) 2010-02-25 2014-03-25 Apple Inc. User profiling for selecting user specific voice input processing information
US8694304B2 (en) 2010-03-26 2014-04-08 Virtuoz Sa Semantic clustering and user interfaces
US9378202B2 (en) 2010-03-26 2016-06-28 Virtuoz Sa Semantic clustering
US9299025B1 (en) * 2010-04-06 2016-03-29 The United States Of America As Represented By The Secretary Of The Navy Case-based reasoning system using case generalization method
US9141690B2 (en) 2010-05-14 2015-09-22 Salesforce.Com, Inc. Methods and systems for categorizing data in an on-demand database environment
US9110882B2 (en) 2010-05-14 2015-08-18 Amazon Technologies, Inc. Extracting structured knowledge from unstructured text
US8639516B2 (en) 2010-06-04 2014-01-28 Apple Inc. User-specific noise suppression for voice quality improvements
US9235806B2 (en) 2010-06-22 2016-01-12 Primal Fusion Inc. Methods and devices for customizing knowledge representation systems
US10474647B2 (en) 2010-06-22 2019-11-12 Primal Fusion Inc. Methods and devices for customizing knowledge representation systems
US11068657B2 (en) * 2010-06-28 2021-07-20 Skyscanner Limited Natural language question answering system and method based on deep semantics
US8713021B2 (en) 2010-07-07 2014-04-29 Apple Inc. Unsupervised document clustering using latent semantic density analysis
US20120016661A1 (en) * 2010-07-19 2012-01-19 Eyal Pinkas System, method and device for intelligent textual conversation system
US9104670B2 (en) 2010-07-21 2015-08-11 Apple Inc. Customized search or acquisition of digital media assets
US8719006B2 (en) 2010-08-27 2014-05-06 Apple Inc. Combined statistical and rule-based part-of-speech tagging for text-to-speech synthesis
US9405848B2 (en) * 2010-09-15 2016-08-02 Vcvc Iii Llc Recommending mobile device activities
US8719014B2 (en) 2010-09-27 2014-05-06 Apple Inc. Electronic device with text error correction based on voice recognition data
US9524291B2 (en) * 2010-10-06 2016-12-20 Virtuoz Sa Visual display of semantic information
WO2012061252A2 (en) 2010-11-04 2012-05-10 Dw Associates, Llc. Methods and systems for identifying, quantifying, analyzing, and optimizing the level of engagement of components within a defined ecosystem or context
US8566273B2 (en) * 2010-12-15 2013-10-22 Siemens Aktiengesellschaft Method, system, and computer program for information retrieval in semantic networks
US10515147B2 (en) 2010-12-22 2019-12-24 Apple Inc. Using statistical language models for contextual lookup
US10762293B2 (en) 2010-12-22 2020-09-01 Apple Inc. Using parts-of-speech tagging and named entity recognition for spelling correction
US11294977B2 (en) 2011-06-20 2022-04-05 Primal Fusion Inc. Techniques for presenting content to a user based on the user's preferences
US8781836B2 (en) 2011-02-22 2014-07-15 Apple Inc. Hearing assistance system for providing consistent human speech
US20120239381A1 (en) 2011-03-17 2012-09-20 Sap Ag Semantic phrase suggestion engine
US8527451B2 (en) 2011-03-17 2013-09-03 Sap Ag Business semantic network build
US9262612B2 (en) 2011-03-21 2016-02-16 Apple Inc. Device access using voice authentication
US8996359B2 (en) 2011-05-18 2015-03-31 Dw Associates, Llc Taxonomy and application of language analysis and processing
US8972240B2 (en) * 2011-05-19 2015-03-03 Microsoft Corporation User-modifiable word lattice display for editing documents and search queries
US8538742B2 (en) * 2011-05-20 2013-09-17 Google Inc. Feed translation for a social network
US8725760B2 (en) * 2011-05-31 2014-05-13 Sap Ag Semantic terminology importer
US10672399B2 (en) 2011-06-03 2020-06-02 Apple Inc. Switching between text data and audio data based on a mapping
US10057736B2 (en) 2011-06-03 2018-08-21 Apple Inc. Active transport based notifications
US9098575B2 (en) 2011-06-20 2015-08-04 Primal Fusion Inc. Preference-guided semantic processing
US8812294B2 (en) 2011-06-21 2014-08-19 Apple Inc. Translating phrases from one language into another using an order-based set of declarative rules
US8952796B1 (en) 2011-06-28 2015-02-10 Dw Associates, Llc Enactive perception device
US20130069768A1 (en) * 2011-07-20 2013-03-21 Maitreya Visweswara Madhyastha Systems, devices, methods and computer-readable storage media that facilitate control of battery-powered devices
US8838434B1 (en) * 2011-07-29 2014-09-16 Nuance Communications, Inc. Bootstrap call router to other languages using selected N-best translations
US8706472B2 (en) 2011-08-11 2014-04-22 Apple Inc. Method for disambiguating multiple readings in language conversion
US8935230B2 (en) 2011-08-25 2015-01-13 Sap Se Self-learning semantic search engine
US8994660B2 (en) 2011-08-29 2015-03-31 Apple Inc. Text correction processing
US8762156B2 (en) 2011-09-28 2014-06-24 Apple Inc. Speech recognition repair using contextual information
US9785628B2 (en) 2011-09-29 2017-10-10 Microsoft Technology Licensing, Llc System, method and computer-readable storage device for providing cloud-based shared vocabulary/typing history for efficient social communication
US9256595B2 (en) * 2011-10-28 2016-02-09 Sap Se Calculating term similarity using a meta-model semantic network
US9779135B2 (en) 2011-11-03 2017-10-03 Sap Se Semantic related objects
US9269353B1 (en) 2011-12-07 2016-02-23 Manu Rehani Methods and systems for measuring semantics in communications
JP5783024B2 (en) * 2011-12-08 2015-09-24 富士通株式会社 Information display program, information display method, and information processing apparatus
US9020807B2 (en) 2012-01-18 2015-04-28 Dw Associates, Llc Format for displaying text analytics results
US10134385B2 (en) 2012-03-02 2018-11-20 Apple Inc. Systems and methods for name pronunciation
US9483461B2 (en) 2012-03-06 2016-11-01 Apple Inc. Handling speech synthesis of content for multiple languages
US9280610B2 (en) 2012-05-14 2016-03-08 Apple Inc. Crowd sourcing information to fulfill user requests
US8775442B2 (en) 2012-05-15 2014-07-08 Apple Inc. Semantic search using a single-source semantic model
US10417037B2 (en) 2012-05-15 2019-09-17 Apple Inc. Systems and methods for integrating third party services with a digital assistant
US20130325757A1 (en) * 2012-06-05 2013-12-05 Sap Ag Cascading learning system as semantic search
US9721563B2 (en) 2012-06-08 2017-08-01 Apple Inc. Name recognition system
US10019994B2 (en) 2012-06-08 2018-07-10 Apple Inc. Systems and methods for recognizing textual identifiers within a plurality of words
US9495129B2 (en) 2012-06-29 2016-11-15 Apple Inc. Device, method, and user interface for voice-activated navigation and browsing of a document
US9424233B2 (en) 2012-07-20 2016-08-23 Veveo, Inc. Method of and system for inferring user intent in search input in a conversational interaction system
US20140032574A1 (en) * 2012-07-23 2014-01-30 Emdadur R. Khan Natural language understanding using brain-like approach: semantic engine using brain-like approach (sebla) derives semantics of words and sentences
US9465833B2 (en) 2012-07-31 2016-10-11 Veveo, Inc. Disambiguating user intent in conversational interaction system for large corpus information retrieval
US9280520B2 (en) * 2012-08-02 2016-03-08 American Express Travel Related Services Company, Inc. Systems and methods for semantic information retrieval
US9576574B2 (en) 2012-09-10 2017-02-21 Apple Inc. Context-sensitive handling of interruptions by intelligent digital assistant
US9547647B2 (en) 2012-09-19 2017-01-17 Apple Inc. Voice-based media searching
US8935167B2 (en) 2012-09-25 2015-01-13 Apple Inc. Exemplar-based latent perceptual modeling for automatic speech recognition
US9547095B2 (en) * 2012-12-19 2017-01-17 Westerngeco L.L.C. MEMS-based rotation sensor for seismic applications and sensor units having same
KR20230137475A (en) 2013-02-07 2023-10-04 애플 인크. Voice trigger for a digital assistant
US9569425B2 (en) * 2013-03-01 2017-02-14 The Software Shop, Inc. Systems and methods for improving the efficiency of syntactic and semantic analysis in automated processes for natural language understanding using traveling features
US9454957B1 (en) * 2013-03-05 2016-09-27 Amazon Technologies, Inc. Named entity resolution in spoken language processing
US9269354B2 (en) * 2013-03-11 2016-02-23 Nuance Communications, Inc. Semantic re-ranking of NLU results in conversational dialogue applications
US9171542B2 (en) * 2013-03-11 2015-10-27 Nuance Communications, Inc. Anaphora resolution using linguisitic cues, dialogue context, and general knowledge
US9361884B2 (en) * 2013-03-11 2016-06-07 Nuance Communications, Inc. Communicating context across different components of multi-modal dialog applications
US9761225B2 (en) 2013-03-11 2017-09-12 Nuance Communications, Inc. Semantic re-ranking of NLU results in conversational dialogue applications
US10652394B2 (en) 2013-03-14 2020-05-12 Apple Inc. System and method for processing voicemail
US9977779B2 (en) 2013-03-14 2018-05-22 Apple Inc. Automatic supplementation of word correction dictionaries
US10642574B2 (en) 2013-03-14 2020-05-05 Apple Inc. Device, method, and graphical user interface for outputting captions
US10572476B2 (en) 2013-03-14 2020-02-25 Apple Inc. Refining a search based on schedule items
US9733821B2 (en) 2013-03-14 2017-08-15 Apple Inc. Voice control to diagnose inadvertent activation of accessibility features
US9368114B2 (en) 2013-03-14 2016-06-14 Apple Inc. Context-sensitive handling of interruptions
US9922642B2 (en) 2013-03-15 2018-03-20 Apple Inc. Training an at least partial voice command system
US11151899B2 (en) 2013-03-15 2021-10-19 Apple Inc. User training by intelligent digital assistant
US10748529B1 (en) 2013-03-15 2020-08-18 Apple Inc. Voice activated device for use with a voice-based digital assistant
WO2014168730A2 (en) 2013-03-15 2014-10-16 Apple Inc. Context-sensitive handling of interruptions
WO2014144579A1 (en) 2013-03-15 2014-09-18 Apple Inc. System and method for updating an adaptive speech recognition model
US9031933B2 (en) * 2013-04-03 2015-05-12 International Business Machines Corporation Method and apparatus for optimizing the evaluation of semantic web queries
US10152538B2 (en) 2013-05-06 2018-12-11 Dropbox, Inc. Suggested search based on a content item
EP3640938A1 (en) 2013-05-07 2020-04-22 Veveo, Inc. Incremental speech input interface with real time feedback
USD805535S1 (en) 2013-06-04 2017-12-19 Abbyy Production Llc Display screen or portion thereof with a transitional graphical user interface
USD802609S1 (en) 2013-06-04 2017-11-14 Abbyy Production Llc Display screen with graphical user interface
WO2014197336A1 (en) 2013-06-07 2014-12-11 Apple Inc. System and method for detecting errors in interactions with a voice-based digital assistant
US9582608B2 (en) 2013-06-07 2017-02-28 Apple Inc. Unified ranking with entropy-weighted information for phrase-based semantic auto-completion
WO2014197334A2 (en) 2013-06-07 2014-12-11 Apple Inc. System and method for user-specified pronunciation of words for speech synthesis and recognition
WO2014197335A1 (en) 2013-06-08 2014-12-11 Apple Inc. Interpreting and acting upon commands that involve sharing information with remote devices
EP3937002A1 (en) 2013-06-09 2022-01-12 Apple Inc. Device, method, and graphical user interface for enabling conversation persistence across two or more instances of a digital assistant
US10176167B2 (en) 2013-06-09 2019-01-08 Apple Inc. System and method for inferring user intent from speech inputs
KR101809808B1 (en) 2013-06-13 2017-12-15 애플 인크. System and method for emergency calls initiated by voice command
US20140379753A1 (en) * 2013-06-25 2014-12-25 Hewlett-Packard Development Company, L.P. Ambiguous queries in configuration management databases
US9342608B2 (en) * 2013-08-01 2016-05-17 International Business Machines Corporation Clarification of submitted questions in a question and answer system
JP6163266B2 (en) 2013-08-06 2017-07-12 アップル インコーポレイテッド Automatic activation of smart responses based on activation from remote devices
US10102851B1 (en) * 2013-08-28 2018-10-16 Amazon Technologies, Inc. Incremental utterance processing and semantic stability determination
US10296160B2 (en) 2013-12-06 2019-05-21 Apple Inc. Method for extracting salient dialog usage from live data
RU2665239C2 (en) 2014-01-15 2018-08-28 Общество с ограниченной ответственностью "Аби Продакшн" Named entities from the text automatic extraction
US9620105B2 (en) 2014-05-15 2017-04-11 Apple Inc. Analyzing audio input for efficient speech and music recognition
US10592095B2 (en) 2014-05-23 2020-03-17 Apple Inc. Instantaneous speaking of content on touch devices
JP5737641B2 (en) * 2014-05-24 2015-06-17 洋彰 宮崎 Autonomous thinking pattern generator
US9502031B2 (en) 2014-05-27 2016-11-22 Apple Inc. Method for supporting dynamic grammars in WFST-based ASR
EP3480811A1 (en) 2014-05-30 2019-05-08 Apple Inc. Multi-command single utterance input method
US9430463B2 (en) 2014-05-30 2016-08-30 Apple Inc. Exemplar-based natural language processing
US9842101B2 (en) 2014-05-30 2017-12-12 Apple Inc. Predictive conversion of language input
US9633004B2 (en) 2014-05-30 2017-04-25 Apple Inc. Better resolution when referencing to concepts
US9734193B2 (en) 2014-05-30 2017-08-15 Apple Inc. Determining domain salience ranking from ambiguous words in natural speech
US10078631B2 (en) 2014-05-30 2018-09-18 Apple Inc. Entropy-guided text prediction using combined word and character n-gram language models
US9785630B2 (en) 2014-05-30 2017-10-10 Apple Inc. Text prediction using combined word N-gram and unigram language models
US9715875B2 (en) 2014-05-30 2017-07-25 Apple Inc. Reducing the need for manual start/end-pointing and trigger phrases
US10289433B2 (en) 2014-05-30 2019-05-14 Apple Inc. Domain specific language for encoding assistant dialog
US9760559B2 (en) 2014-05-30 2017-09-12 Apple Inc. Predictive text input
US10170123B2 (en) 2014-05-30 2019-01-01 Apple Inc. Intelligent assistant for home automation
US10659851B2 (en) 2014-06-30 2020-05-19 Apple Inc. Real-time digital assistant knowledge updates
US9338493B2 (en) 2014-06-30 2016-05-10 Apple Inc. Intelligent automated assistant for TV user interactions
US10446141B2 (en) 2014-08-28 2019-10-15 Apple Inc. Automatic speech recognition based on user feedback
US9818400B2 (en) 2014-09-11 2017-11-14 Apple Inc. Method and apparatus for discovering trending terms in speech requests
US10789041B2 (en) 2014-09-12 2020-09-29 Apple Inc. Dynamic thresholds for always listening speech trigger
JP2016061970A (en) * 2014-09-18 2016-04-25 株式会社東芝 Speech dialog device, method, and program
US10074360B2 (en) 2014-09-30 2018-09-11 Apple Inc. Providing an indication of the suitability of speech recognition
US10127911B2 (en) 2014-09-30 2018-11-13 Apple Inc. Speaker identification and unsupervised speaker adaptation techniques
US9646609B2 (en) 2014-09-30 2017-05-09 Apple Inc. Caching apparatus for serving phonetic pronunciations
US9668121B2 (en) 2014-09-30 2017-05-30 Apple Inc. Social reminders
US9886432B2 (en) 2014-09-30 2018-02-06 Apple Inc. Parsimonious handling of word inflection via categorical stem + suffix N-gram language models
US10552013B2 (en) 2014-12-02 2020-02-04 Apple Inc. Data detection
US9711141B2 (en) 2014-12-09 2017-07-18 Apple Inc. Disambiguating heteronyms in speech synthesis
US10409909B2 (en) 2014-12-12 2019-09-10 Omni Ai, Inc. Lexical analyzer for a neuro-linguistic behavior recognition system
US10409910B2 (en) 2014-12-12 2019-09-10 Omni Ai, Inc. Perceptual associative memory for a neuro-linguistic behavior recognition system
US9852136B2 (en) 2014-12-23 2017-12-26 Rovi Guides, Inc. Systems and methods for determining whether a negation statement applies to a current or past query
US9854049B2 (en) 2015-01-30 2017-12-26 Rovi Guides, Inc. Systems and methods for resolving ambiguous terms in social chatter based on a user profile
RU2584457C1 (en) * 2015-02-03 2016-05-20 Общество с ограниченной ответственностью "Аби ИнфоПоиск" System and method of creating and using user semantic dictionaries for processing user text in natural language
RU2592396C1 (en) 2015-02-03 2016-07-20 Общество с ограниченной ответственностью "Аби ИнфоПоиск" Method and system for machine extraction and interpretation of text information
RU2596599C2 (en) * 2015-02-03 2016-09-10 Общество с ограниченной ответственностью "Аби ИнфоПоиск" System and method of creating and using user ontology-based patterns for processing user text in natural language
US10152299B2 (en) 2015-03-06 2018-12-11 Apple Inc. Reducing response latency of intelligent automated assistants
US9865280B2 (en) 2015-03-06 2018-01-09 Apple Inc. Structured dictation using intelligent automated assistants
US10567477B2 (en) 2015-03-08 2020-02-18 Apple Inc. Virtual assistant continuity
US9721566B2 (en) 2015-03-08 2017-08-01 Apple Inc. Competing devices responding to voice triggers
US9886953B2 (en) 2015-03-08 2018-02-06 Apple Inc. Virtual assistant activation
US9899019B2 (en) 2015-03-18 2018-02-20 Apple Inc. Systems and methods for structured stem and suffix language models
RU2610241C2 (en) 2015-03-19 2017-02-08 Общество с ограниченной ответственностью "Аби ИнфоПоиск" Method and system for text synthesis based on information extracted as rdf-graph using templates
US9842105B2 (en) 2015-04-16 2017-12-12 Apple Inc. Parsimonious continuous-space phrase representations for natural language processing
US10460227B2 (en) 2015-05-15 2019-10-29 Apple Inc. Virtual assistant in a communication session
US10083688B2 (en) 2015-05-27 2018-09-25 Apple Inc. Device voice control for selecting a displayed affordance
US10127220B2 (en) 2015-06-04 2018-11-13 Apple Inc. Language identification from short strings
US9578173B2 (en) 2015-06-05 2017-02-21 Apple Inc. Virtual assistant aided communication with 3rd party service in a communication session
US10101822B2 (en) 2015-06-05 2018-10-16 Apple Inc. Language input correction
US10255907B2 (en) 2015-06-07 2019-04-09 Apple Inc. Automatic accent detection using acoustic models
US11025565B2 (en) 2015-06-07 2021-06-01 Apple Inc. Personalized prediction of responses for instant messaging
US10186254B2 (en) 2015-06-07 2019-01-22 Apple Inc. Context-based endpoint detection
US20160378747A1 (en) 2015-06-29 2016-12-29 Apple Inc. Virtual assistant for media playback
US10747498B2 (en) 2015-09-08 2020-08-18 Apple Inc. Zero latency digital assistant
US10671428B2 (en) 2015-09-08 2020-06-02 Apple Inc. Distributed personal assistant
US9697820B2 (en) 2015-09-24 2017-07-04 Apple Inc. Unit-selection text-to-speech synthesis using concatenation-sensitive neural networks
US10366158B2 (en) 2015-09-29 2019-07-30 Apple Inc. Efficient word encoding for recurrent neural network language models
US11010550B2 (en) 2015-09-29 2021-05-18 Apple Inc. Unified language modeling framework for word prediction, auto-completion and auto-correction
US11587559B2 (en) 2015-09-30 2023-02-21 Apple Inc. Intelligent device identification
US10691473B2 (en) 2015-11-06 2020-06-23 Apple Inc. Intelligent automated assistant in a messaging environment
US10049668B2 (en) 2015-12-02 2018-08-14 Apple Inc. Applying neural network language models to weighted finite state transducers for automatic speech recognition
US10223066B2 (en) 2015-12-23 2019-03-05 Apple Inc. Proactive assistance based on dialog communication between devices
US10963956B1 (en) * 2016-03-04 2021-03-30 United Services Automobile Association (Usaa) Semantic computer language for exchanging platform-agnostic financial information
US10417021B2 (en) * 2016-03-04 2019-09-17 Ricoh Company, Ltd. Interactive command assistant for an interactive whiteboard appliance
US10409550B2 (en) 2016-03-04 2019-09-10 Ricoh Company, Ltd. Voice control of interactive whiteboard appliances
US10446143B2 (en) 2016-03-14 2019-10-15 Apple Inc. Identification of voice inputs providing credentials
US9934775B2 (en) 2016-05-26 2018-04-03 Apple Inc. Unit-selection text-to-speech synthesis based on predicted concatenation parameters
US9972304B2 (en) 2016-06-03 2018-05-15 Apple Inc. Privacy preserving distributed evaluation framework for embedded personalized systems
US11227589B2 (en) 2016-06-06 2022-01-18 Apple Inc. Intelligent list reading
US10249300B2 (en) 2016-06-06 2019-04-02 Apple Inc. Intelligent list reading
US10049663B2 (en) 2016-06-08 2018-08-14 Apple, Inc. Intelligent automated assistant for media exploration
DK179309B1 (en) 2016-06-09 2018-04-23 Apple Inc Intelligent automated assistant in a home environment
US10067938B2 (en) 2016-06-10 2018-09-04 Apple Inc. Multilingual word prediction
US10490187B2 (en) 2016-06-10 2019-11-26 Apple Inc. Digital assistant providing automated status report
US10586535B2 (en) 2016-06-10 2020-03-10 Apple Inc. Intelligent digital assistant in a multi-tasking environment
US10192552B2 (en) 2016-06-10 2019-01-29 Apple Inc. Digital assistant providing whispered speech
US10509862B2 (en) 2016-06-10 2019-12-17 Apple Inc. Dynamic phrase expansion of language input
DK179415B1 (en) 2016-06-11 2018-06-14 Apple Inc Intelligent device arbitration and control
DK179343B1 (en) 2016-06-11 2018-05-14 Apple Inc Intelligent task discovery
DK201670540A1 (en) 2016-06-11 2018-01-08 Apple Inc Application integration with a digital assistant
DK179049B1 (en) 2016-06-11 2017-09-18 Apple Inc Data driven natural language event detection and classification
US10394950B2 (en) * 2016-08-22 2019-08-27 International Business Machines Corporation Generation of a grammatically diverse test set for deep question answering systems
US10133724B2 (en) * 2016-08-22 2018-11-20 International Business Machines Corporation Syntactic classification of natural language sentences with respect to a targeted element
US10474753B2 (en) 2016-09-07 2019-11-12 Apple Inc. Language identification using recurrent neural networks
US10043516B2 (en) 2016-09-23 2018-08-07 Apple Inc. Intelligent automated assistant
US11281993B2 (en) 2016-12-05 2022-03-22 Apple Inc. Model and ensemble compression for metric learning
US10593346B2 (en) 2016-12-22 2020-03-17 Apple Inc. Rank-reduced token representation for automatic speech recognition
CN110168544A (en) * 2016-12-27 2019-08-23 夏普株式会社 Answering device, the control method of answering device and control program
US11204787B2 (en) 2017-01-09 2021-12-21 Apple Inc. Application integration with a digital assistant
US10417266B2 (en) 2017-05-09 2019-09-17 Apple Inc. Context-aware ranking of intelligent response suggestions
DK201770383A1 (en) 2017-05-09 2018-12-14 Apple Inc. User interface for correcting recognition errors
US10395654B2 (en) 2017-05-11 2019-08-27 Apple Inc. Text normalization based on a data-driven learning network
US10726832B2 (en) 2017-05-11 2020-07-28 Apple Inc. Maintaining privacy of personal information
DK201770439A1 (en) 2017-05-11 2018-12-13 Apple Inc. Offline personal assistant
DK179496B1 (en) 2017-05-12 2019-01-15 Apple Inc. USER-SPECIFIC Acoustic Models
DK179745B1 (en) 2017-05-12 2019-05-01 Apple Inc. SYNCHRONIZATION AND TASK DELEGATION OF A DIGITAL ASSISTANT
DK201770428A1 (en) 2017-05-12 2019-02-18 Apple Inc. Low-latency intelligent automated assistant
US11301477B2 (en) 2017-05-12 2022-04-12 Apple Inc. Feedback analysis of a digital assistant
DK201770432A1 (en) 2017-05-15 2018-12-21 Apple Inc. Hierarchical belief states for digital assistants
DK201770431A1 (en) 2017-05-15 2018-12-20 Apple Inc. Optimizing dialogue policy decisions for digital assistants using implicit feedback
US20180336275A1 (en) 2017-05-16 2018-11-22 Apple Inc. Intelligent automated assistant for media exploration
US10311144B2 (en) 2017-05-16 2019-06-04 Apple Inc. Emoji word sense disambiguation
US20180336892A1 (en) 2017-05-16 2018-11-22 Apple Inc. Detecting a trigger of a digital assistant
DK179560B1 (en) 2017-05-16 2019-02-18 Apple Inc. Far-field extension for digital assistant services
US10403278B2 (en) 2017-05-16 2019-09-03 Apple Inc. Methods and systems for phonetic matching in digital assistant services
US10552426B2 (en) 2017-05-23 2020-02-04 International Business Machines Corporation Adaptive conversational disambiguation system
US10657328B2 (en) 2017-06-02 2020-05-19 Apple Inc. Multi-task recurrent neural network architecture for efficient morphology handling in neural language modeling
CN109271621B (en) * 2017-07-18 2023-04-18 腾讯科技(北京)有限公司 Semantic disambiguation processing method, device and equipment
US10445429B2 (en) 2017-09-21 2019-10-15 Apple Inc. Natural language understanding using vocabularies with compressed serialized tries
US10755051B2 (en) 2017-09-29 2020-08-25 Apple Inc. Rule-based natural language processing
WO2019100167A1 (en) * 2017-11-27 2019-05-31 Retailcommon Inc. Method and system for syntactic searching
US10636424B2 (en) 2017-11-30 2020-04-28 Apple Inc. Multi-turn canned dialog
US10733982B2 (en) 2018-01-08 2020-08-04 Apple Inc. Multi-directional dialog
US10733375B2 (en) 2018-01-31 2020-08-04 Apple Inc. Knowledge-based framework for improving natural language understanding
US10789959B2 (en) 2018-03-02 2020-09-29 Apple Inc. Training speaker recognition models for digital assistants
US10592604B2 (en) 2018-03-12 2020-03-17 Apple Inc. Inverse text normalization for automatic speech recognition
US10818288B2 (en) 2018-03-26 2020-10-27 Apple Inc. Natural assistant interaction
US10909331B2 (en) 2018-03-30 2021-02-02 Apple Inc. Implicit identification of translation payload with neural machine translation
US11145294B2 (en) 2018-05-07 2021-10-12 Apple Inc. Intelligent automated assistant for delivering content from user experiences
US10928918B2 (en) 2018-05-07 2021-02-23 Apple Inc. Raise to speak
US10984780B2 (en) 2018-05-21 2021-04-20 Apple Inc. Global semantic word embeddings using bi-directional recurrent neural networks
DK201870355A1 (en) 2018-06-01 2019-12-16 Apple Inc. Virtual assistant operation in multi-device environments
DK179822B1 (en) 2018-06-01 2019-07-12 Apple Inc. Voice interaction at a primary device to access call functionality of a companion device
US11386266B2 (en) 2018-06-01 2022-07-12 Apple Inc. Text correction
DK180639B1 (en) 2018-06-01 2021-11-04 Apple Inc DISABILITY OF ATTENTION-ATTENTIVE VIRTUAL ASSISTANT
US10892996B2 (en) 2018-06-01 2021-01-12 Apple Inc. Variable latency device coordination
US11076039B2 (en) 2018-06-03 2021-07-27 Apple Inc. Accelerated task performance
US11010561B2 (en) 2018-09-27 2021-05-18 Apple Inc. Sentiment prediction from textual data
US10839159B2 (en) 2018-09-28 2020-11-17 Apple Inc. Named entity normalization in a spoken dialog system
US11462215B2 (en) 2018-09-28 2022-10-04 Apple Inc. Multi-modal inputs for voice commands
US11170166B2 (en) 2018-09-28 2021-11-09 Apple Inc. Neural typographical error modeling via generative adversarial networks
US11475898B2 (en) 2018-10-26 2022-10-18 Apple Inc. Low-latency multi-speaker speech recognition
US11638059B2 (en) 2019-01-04 2023-04-25 Apple Inc. Content playback on multiple devices
US11348573B2 (en) 2019-03-18 2022-05-31 Apple Inc. Multimodality in digital assistant systems
US11307752B2 (en) 2019-05-06 2022-04-19 Apple Inc. User configurable task triggers
US11423908B2 (en) 2019-05-06 2022-08-23 Apple Inc. Interpreting spoken requests
US11217251B2 (en) 2019-05-06 2022-01-04 Apple Inc. Spoken notifications
US11475884B2 (en) 2019-05-06 2022-10-18 Apple Inc. Reducing digital assistant latency when a language is incorrectly determined
US10902854B1 (en) 2019-05-17 2021-01-26 Eyeballs Financial, LLC Systems and methods for generating responses to questions about user accounts
US11140099B2 (en) 2019-05-21 2021-10-05 Apple Inc. Providing message response suggestions
DK201970510A1 (en) 2019-05-31 2021-02-11 Apple Inc Voice identification in digital assistant systems
DK180129B1 (en) 2019-05-31 2020-06-02 Apple Inc. User activity shortcut suggestions
US11289073B2 (en) 2019-05-31 2022-03-29 Apple Inc. Device text to speech
US11496600B2 (en) 2019-05-31 2022-11-08 Apple Inc. Remote execution of machine-learned models
US11360641B2 (en) 2019-06-01 2022-06-14 Apple Inc. Increasing the relevance of new available information
WO2021056255A1 (en) 2019-09-25 2021-04-01 Apple Inc. Text detection using global geometry estimators
WO2022043675A2 (en) * 2020-08-24 2022-03-03 Unlikely Artificial Intelligence Limited A computer implemented method for the automated analysis or use of data
US20230394241A1 (en) * 2020-10-15 2023-12-07 The Regents Of The University Of California System and method for bridging graphics and linguistics

Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5299125A (en) * 1990-08-09 1994-03-29 Semantic Compaction Systems Natural language processing system and method for parsing a plurality of input symbol sequences into syntactically or pragmatically correct word messages
US5479563A (en) * 1990-09-07 1995-12-26 Fujitsu Limited Boundary extracting system from a sentence
US5802508A (en) * 1996-08-21 1998-09-01 International Business Machines Corporation Reasoning with rules in a multiple inheritance semantic network with exceptions
US5809269A (en) * 1992-10-06 1998-09-15 Sextant Avionique Method and device for the analysis of a message given by interaction means to a man/machine dialog system
US6006221A (en) * 1995-08-16 1999-12-21 Syracuse University Multilingual document retrieval system and method using semantic vector matching
US6269335B1 (en) * 1998-08-14 2001-07-31 International Business Machines Corporation Apparatus and methods for identifying homophones among words in a speech recognition system
US6304864B1 (en) * 1999-04-20 2001-10-16 Textwise Llc System for retrieving multimedia information from the internet using multiple evolving intelligent agents
US6430531B1 (en) * 1999-02-04 2002-08-06 Soliloquy, Inc. Bilateral speech system
US6442522B1 (en) * 1999-10-12 2002-08-27 International Business Machines Corporation Bi-directional natural language system for interfacing with multiple back-end applications
US20030028367A1 (en) * 2001-06-15 2003-02-06 Achraf Chalabi Method and system for theme-based word sense ambiguity reduction
US20030037073A1 (en) * 2001-05-08 2003-02-20 Naoyuki Tokuda New differential LSI space-based probabilistic document classifier
US6609091B1 (en) * 1994-09-30 2003-08-19 Robert L. Budzinski Memory system for storing and retrieving experience and knowledge with natural language utilizing state representation data, word sense numbers, function codes and/or directed graphs
US6684201B1 (en) * 2000-03-31 2004-01-27 Microsoft Corporation Linguistic disambiguation system and method using string-based pattern training to learn to resolve ambiguity sites
US20040117352A1 (en) * 2000-04-28 2004-06-17 Global Information Research And Technologies Llc System for answering natural language questions
US20050149510A1 (en) * 2004-01-07 2005-07-07 Uri Shafrir Concept mining and concept discovery-semantic search tool for large digital databases

Family Cites Families (81)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4468728A (en) * 1981-06-25 1984-08-28 At&T Bell Laboratories Data structure and search method for a data base management system
US4429385A (en) * 1981-12-31 1984-01-31 American Newspaper Publishers Association Method and apparatus for digital serial scanning with hierarchical and relational access
US4677550A (en) * 1983-09-30 1987-06-30 Amalgamated Software Of North America, Inc. Method of compacting and searching a data index
US4769772A (en) * 1985-02-28 1988-09-06 Honeywell Bull, Inc. Automated query optimization method using both global and parallel local optimizations for materialization access planning for distributed databases
JPS61220027A (en) * 1985-03-27 1986-09-30 Hitachi Ltd Information memory system
US4774657A (en) * 1986-06-06 1988-09-27 International Business Machines Corporation Index key range estimator
US4914569A (en) * 1987-10-30 1990-04-03 International Business Machines Corporation Method for concurrent record access, insertion, deletion and alteration using an index tree
US4914590A (en) * 1988-05-18 1990-04-03 Emhart Industries, Inc. Natural language understanding system
US5043872A (en) * 1988-07-15 1991-08-27 International Business Machines Corporation Access path optimization using degrees of clustering
US4905163A (en) * 1988-10-03 1990-02-27 Minnesota Mining & Manufacturing Company Intelligent optical navigator dynamic information presentation and navigation system
US5111398A (en) * 1988-11-21 1992-05-05 Xerox Corporation Processing natural language text using autonomous punctuational structure
JPH02159674A (en) * 1988-12-13 1990-06-19 Matsushita Electric Ind Co Ltd Method for analyzing meaning and method for analyzing syntax
US5829002A (en) * 1989-02-15 1998-10-27 Priest; W. Curtiss System for coordinating information transfer and retrieval
SE466029B (en) * 1989-03-06 1991-12-02 Ibm Svenska Ab DEVICE AND PROCEDURE FOR ANALYSIS OF NATURAL LANGUAGES IN A COMPUTER-BASED INFORMATION PROCESSING SYSTEM
US5056021A (en) * 1989-06-08 1991-10-08 Carolyn Ausborn Method and apparatus for abstracting concepts from natural language
US5123057A (en) * 1989-07-28 1992-06-16 Massachusetts Institute Of Technology Model based pattern recognition
US5202986A (en) * 1989-09-28 1993-04-13 Bull Hn Information Systems Inc. Prefix search tree partial key branching
US5155825A (en) * 1989-12-27 1992-10-13 Motorola, Inc. Page address translation cache replacement algorithm with improved testability
US5752016A (en) * 1990-02-08 1998-05-12 Hewlett-Packard Company Method and apparatus for database interrogation using a user-defined table
US5095458A (en) * 1990-04-02 1992-03-10 Advanced Micro Devices, Inc. Radix 4 carry lookahead tree and redundant cell therefor
DE69032349T2 (en) * 1990-07-31 1998-10-01 Hewlett Packard Co Object based system
JP3009215B2 (en) * 1990-11-30 2000-02-14 株式会社日立製作所 Natural language processing method and natural language processing system
US5598560A (en) * 1991-03-07 1997-01-28 Digital Equipment Corporation Tracking condition codes in translation code for different machine architectures
US5694590A (en) * 1991-09-27 1997-12-02 The Mitre Corporation Apparatus and method for the detection of security violations in multilevel secure databases
US5826256A (en) * 1991-10-22 1998-10-20 Lucent Technologies Inc. Apparatus and methods for source code discovery
US5778223A (en) * 1992-03-17 1998-07-07 International Business Machines Corporation Dictionary for encoding and retrieving hierarchical data processing information for a computer system
US5721895A (en) * 1992-03-17 1998-02-24 International Business Machines Corporation Computer program product and program storage device for a data transmission dictionary for encoding, storing, and retrieving hierarchical data processing information for a computer system
US5434777A (en) 1992-05-27 1995-07-18 Apple Computer, Inc. Method and apparatus for processing natural language
US5528491A (en) * 1992-08-31 1996-06-18 Language Engineering Corporation Apparatus and method for automated natural language translation
JPH06176081A (en) * 1992-12-02 1994-06-24 Hitachi Ltd Hierarchical structure browsing method and device
US5628011A (en) * 1993-01-04 1997-05-06 At&T Network-based intelligent information-sourcing arrangement
US5630125A (en) * 1994-05-23 1997-05-13 Zellweger; Paul Method and apparatus for information management using an open hierarchical data structure
US5794050A (en) * 1995-01-04 1998-08-11 Intelligent Text Processing, Inc. Natural language understanding system
US5625767A (en) * 1995-03-13 1997-04-29 Bartell; Brian Method and system for two-dimensional visualization of an information taxonomy and of text documents based on topical content of the documents
GB9511748D0 (en) 1995-06-09 1995-08-02 Cobalt Systems Limited Oven
GB2302420A (en) * 1995-06-19 1997-01-15 Ibm Semantic network
US5963940A (en) * 1995-08-16 1999-10-05 Syracuse University Natural language information retrieval system and method
US5894554A (en) * 1996-04-23 1999-04-13 Infospinner, Inc. System for managing dynamic web page generation requests by intercepting request at web server and routing to page server thereby releasing web server to process other requests
US6179491B1 (en) * 1997-02-05 2001-01-30 International Business Machines Corporation Method and apparatus for slicing class hierarchies
JP3159242B2 (en) 1997-03-13 2001-04-23 日本電気株式会社 Emotion generating apparatus and method
US5937400A (en) 1997-03-19 1999-08-10 Au; Lawrence Method to quantify abstraction within semantic networks
US5901100A (en) * 1997-04-01 1999-05-04 Ramtron International Corporation First-in, first-out integrated circuit memory device utilizing a dynamic random access memory array for data storage implemented in conjunction with an associated static random access memory cache
US6154213A (en) 1997-05-30 2000-11-28 Rennison; Earl F. Immersive movement-based interaction with large complex information structures
US6233575B1 (en) * 1997-06-24 2001-05-15 International Business Machines Corporation Multilevel taxonomy based on features derived from training documents classification using fisher values as discrimination values
US5974412A (en) * 1997-09-24 1999-10-26 Sapient Health Network Intelligent query system for automatically indexing information in a database and automatically categorizing users
US6263352B1 (en) * 1997-11-14 2001-07-17 Microsoft Corporation Automated web site creation using template driven generation of active server page applications
US6778970B2 (en) * 1998-05-28 2004-08-17 Lawrence Au Topological methods to organize semantic network data flows for conversational applications
EP0962873A1 (en) * 1998-06-02 1999-12-08 International Business Machines Corporation Processing of textual information and automated apprehension of information
US6256623B1 (en) * 1998-06-22 2001-07-03 Microsoft Corporation Network search access construct for accessing web-based search services
US7152031B1 (en) 2000-02-25 2006-12-19 Novell, Inc. Construction, manipulation, and comparison of a multi-dimensional semantic space
US6446061B1 (en) * 1998-07-31 2002-09-03 International Business Machines Corporation Taxonomy generation for document collections
DE19914326A1 (en) 1999-03-30 2000-10-05 Delphi 2 Creative Tech Gmbh Procedure for using fractal semantic networks for all types of databank applications to enable fuzzy classifications to be used and much more flexible query procedures to be used than conventional databank structures
CA2272739C (en) * 1999-05-25 2003-10-07 Suhayya Abu-Hakima Apparatus and method for interpreting and intelligently managing electronic messages
US6356906B1 (en) * 1999-07-26 2002-03-12 Microsoft Corporation Standard database queries within standard request-response protocols
US6405162B1 (en) * 1999-09-23 2002-06-11 Xerox Corporation Type-based selection of rules for semantically disambiguating words
US6675205B2 (en) * 1999-10-14 2004-01-06 Arcessa, Inc. Peer-to-peer automated anonymous asynchronous file sharing
US6665658B1 (en) * 2000-01-13 2003-12-16 International Business Machines Corporation System and method for automatically gathering dynamic content and resources on the world wide web by stimulating user interaction and managing session information
US6931397B1 (en) * 2000-02-11 2005-08-16 International Business Machines Corporation System and method for automatic generation of dynamic search abstracts contain metadata by crawler
US6446083B1 (en) * 2000-05-12 2002-09-03 Vastvideo, Inc. System and method for classifying media items
AU2001271891A1 (en) * 2000-07-07 2002-01-21 Criticalpoint Software Corporation Methods and system for generating and searching ontology databases
US6463430B1 (en) * 2000-07-10 2002-10-08 Mohomine, Inc. Devices and methods for generating and managing a database
US6675159B1 (en) 2000-07-27 2004-01-06 Science Applic Int Corp Concept-based search and retrieval system
US20030217052A1 (en) 2000-08-24 2003-11-20 Celebros Ltd. Search engine method and apparatus
US20020133347A1 (en) 2000-12-29 2002-09-19 Eberhard Schoneburg Method and apparatus for natural language dialog interface
US6778975B1 (en) * 2001-03-05 2004-08-17 Overture Services, Inc. Search engine for selecting targeted messages
US7426505B2 (en) * 2001-03-07 2008-09-16 International Business Machines Corporation Method for identifying word patterns in text
US20030041047A1 (en) 2001-08-09 2003-02-27 International Business Machines Corporation Concept-based system for representing and processing multimedia objects with arbitrary constraints
EP1288794A1 (en) 2001-08-29 2003-03-05 Tarchon BV Methods of ordering and of retrieving information from a corpus of documents and database system for the same
NO316480B1 (en) 2001-11-15 2004-01-26 Forinnova As Method and system for textual examination and discovery
US6826568B2 (en) * 2001-12-20 2004-11-30 Microsoft Corporation Methods and system for model matching
US20100100437A1 (en) 2002-09-24 2010-04-22 Google, Inc. Suggesting and/or providing ad serving constraint information
GB0306877D0 (en) 2003-03-25 2003-04-30 British Telecomm Information retrieval
US7107264B2 (en) 2003-04-04 2006-09-12 Yahoo, Inc. Content bridge for associating host content and guest content wherein guest content is determined by search
US7395256B2 (en) 2003-06-20 2008-07-01 Agency For Science, Technology And Research Method and platform for term extraction from large collection of documents
US8014997B2 (en) * 2003-09-20 2011-09-06 International Business Machines Corporation Method of search content enhancement
KR100650404B1 (en) 2003-11-24 2006-11-28 엔에이치엔(주) On-line Advertising System And Method
US20050210009A1 (en) 2004-03-18 2005-09-22 Bao Tran Systems and methods for intellectual property management
US7428530B2 (en) 2004-07-01 2008-09-23 Microsoft Corporation Dispersing search engine results by using page category information
US7493333B2 (en) 2004-09-03 2009-02-17 Biowisdom Limited System and method for parsing and/or exporting data from one or more multi-relational ontologies
JP4654745B2 (en) 2005-04-13 2011-03-23 富士ゼロックス株式会社 Question answering system, data retrieval method, and computer program
US7603351B2 (en) 2006-04-19 2009-10-13 Apple Inc. Semantic reconstruction

Patent Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5299125A (en) * 1990-08-09 1994-03-29 Semantic Compaction Systems Natural language processing system and method for parsing a plurality of input symbol sequences into syntactically or pragmatically correct word messages
US5479563A (en) * 1990-09-07 1995-12-26 Fujitsu Limited Boundary extracting system from a sentence
US5809269A (en) * 1992-10-06 1998-09-15 Sextant Avionique Method and device for the analysis of a message given by interaction means to a man/machine dialog system
US6609091B1 (en) * 1994-09-30 2003-08-19 Robert L. Budzinski Memory system for storing and retrieving experience and knowledge with natural language utilizing state representation data, word sense numbers, function codes and/or directed graphs
US6006221A (en) * 1995-08-16 1999-12-21 Syracuse University Multilingual document retrieval system and method using semantic vector matching
US5802508A (en) * 1996-08-21 1998-09-01 International Business Machines Corporation Reasoning with rules in a multiple inheritance semantic network with exceptions
US6269335B1 (en) * 1998-08-14 2001-07-31 International Business Machines Corporation Apparatus and methods for identifying homophones among words in a speech recognition system
US6430531B1 (en) * 1999-02-04 2002-08-06 Soliloquy, Inc. Bilateral speech system
US6304864B1 (en) * 1999-04-20 2001-10-16 Textwise Llc System for retrieving multimedia information from the internet using multiple evolving intelligent agents
US6442522B1 (en) * 1999-10-12 2002-08-27 International Business Machines Corporation Bi-directional natural language system for interfacing with multiple back-end applications
US6684201B1 (en) * 2000-03-31 2004-01-27 Microsoft Corporation Linguistic disambiguation system and method using string-based pattern training to learn to resolve ambiguity sites
US20040117352A1 (en) * 2000-04-28 2004-06-17 Global Information Research And Technologies Llc System for answering natural language questions
US20030037073A1 (en) * 2001-05-08 2003-02-20 Naoyuki Tokuda New differential LSI space-based probabilistic document classifier
US20030028367A1 (en) * 2001-06-15 2003-02-06 Achraf Chalabi Method and system for theme-based word sense ambiguity reduction
US20050149510A1 (en) * 2004-01-07 2005-07-07 Uri Shafrir Concept mining and concept discovery-semantic search tool for large digital databases

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9710825B1 (en) 1999-09-22 2017-07-18 Google Inc. Meaning-based advertising and document relevance determination
US7698266B1 (en) * 1999-11-01 2010-04-13 Google Inc. Meaning-based advertising and document relevance determination
US9135239B1 (en) 1999-11-01 2015-09-15 Google Inc. Meaning-based advertising and document relevance determination
US20050010561A1 (en) * 2003-04-28 2005-01-13 France Telecom System for generating queries

Also Published As

Publication number Publication date
US8200608B2 (en) 2012-06-12
US7711672B2 (en) 2010-05-04
US20030130976A1 (en) 2003-07-10
US8204844B2 (en) 2012-06-19
US20100161317A1 (en) 2010-06-24
US8135660B2 (en) 2012-03-13
US20100030723A1 (en) 2010-02-04
US20100030724A1 (en) 2010-02-04

Similar Documents

Publication Publication Date Title
US7711672B2 (en) Semantic network methods to disambiguate natural language meaning
US7536374B2 (en) Method and system for using voice input for performing device functions
Kumar Natural language processing
CN109684448B (en) Intelligent question and answer method
US9213687B2 (en) Compassion, variety and cohesion for methods of text analytics, writing, search, user interfaces
JP7089330B2 (en) Adaptive processing components
Oostdijk Corpus linguistics and the automatic analysis of English
US8463594B2 (en) System and method for analyzing text using emotional intelligence factors
US9152623B2 (en) Natural language processing system and method
Shamsfard et al. The state of the art in ontology learning: a framework for comparison
US7685118B2 (en) Method using ontology and user query processing to solve inventor problems and user problems
US20140280072A1 (en) Method and Apparatus for Human-Machine Interaction
US20070294229A1 (en) Chat conversation methods traversing a provisional scaffold of meanings
JP2012520528A (en) System and method for automatic semantic labeling of natural language text
KR20120001053A (en) System and method for anaylyzing document sentiment
Tucker Automatic summarising and the CLASP system
Copestake Natural language processing
Chang et al. Mining semantics for large scale integration on the web: evidences, insights, and challenges
Jwalapuram et al. Handling Multi-Sentence Queries in a Domain Independent Dialogue System
WEGDERES T SENTIMENT MINING AND ASPECT BASED SUMMARIZATION OF OPINIONATED AFAAN OROMOO NEWS TEXT
FERRERI et al. Designing and validating conversational agents for data exploration
Estigarribia Automatic detection of syntactic patterns from texts with application to Spanish clitic doubling
Reitter et al. Hybrid natural language processing in a customer-care environment
Gaudio Automatic extraction of definitions
Walker et al. 5. Theory-driven and corpus-driven computational linguistics, and the use o corpora

Legal Events

Date Code Title Description
AS Assignment

Owner name: Q-PHRASE LLC, VIRGINIA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:AU, LAWRENCE;REEL/FRAME:018511/0068

Effective date: 20061031

AS Assignment

Owner name: QPS TECH. LIMITED LIABILITY COMPANY, DELAWARE

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:Q-PHRASE, LLC;REEL/FRAME:019843/0338

Effective date: 20070906

Owner name: QPS TECH. LIMITED LIABILITY COMPANY,DELAWARE

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:Q-PHRASE, LLC;REEL/FRAME:019843/0338

Effective date: 20070906

STCB Information on status: application discontinuation

Free format text: ABANDONED -- FAILURE TO RESPOND TO AN OFFICE ACTION

AS Assignment

Owner name: INTELLECTUAL VENTURES ASSETS 151 LLC, DELAWARE

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:CHARTOLEAUX KG LIMITED LIABILITY COMPANY;REEL/FRAME:050914/0969

Effective date: 20191031

AS Assignment

Owner name: DATACLOUD TECHNOLOGIES, LLC, GEORGIA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:INTELLECTUAL VENTURES ASSETS 151 LLC;REEL/FRAME:051463/0934

Effective date: 20191115