CA2507260A1 - Device and method of predicting symbols in correlated symbol sequences - Google Patents
Device and method of predicting symbols in correlated symbol sequences Download PDFInfo
- Publication number
- CA2507260A1 CA2507260A1 CA002507260A CA2507260A CA2507260A1 CA 2507260 A1 CA2507260 A1 CA 2507260A1 CA 002507260 A CA002507260 A CA 002507260A CA 2507260 A CA2507260 A CA 2507260A CA 2507260 A1 CA2507260 A1 CA 2507260A1
- Authority
- CA
- Canada
- Prior art keywords
- memory
- agent
- symbol
- signal
- list
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 24
- 230000002596 correlated effect Effects 0.000 title claims abstract description 5
- 230000015654 memory Effects 0.000 claims abstract description 97
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000010586 diagram Methods 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000012109 statistical procedure Methods 0.000 abstract 1
- 239000003795 chemical substances by application Substances 0.000 description 125
- 239000013543 active substance Substances 0.000 description 13
- 230000008569 process Effects 0.000 description 8
- 238000003058 natural language processing Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013501 data transformation Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000002996 emotional effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/18—Speech classification or search using natural language modelling
- G10L15/183—Speech classification or search using natural language modelling using context dependencies, e.g. language models
Landscapes
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The method and the according device is intended to predict symbols in a symb ol sequence based on parts of previous symbol sequence and the device experienc e gained by the device during processing of all previous symbol sequences. The device uses heuristic statistical procedures to predict symbols in correlated symbo l sequences through automatic eliciting of dependences in such sequences. The main device components are specialized logical blocks, information agents and memories. The device can be used to improve the quality of speech recognition as well as to generate symbol sequences of human languages on the common knowledge base of a concrete language. In particular, answers to human language requests can b e generated.
Description
DEVICE AND METHOD OF PREDICTING SYMBOLS IN CORRELATED SYMBOL
SEQUENCES
BACKGROUND OF THE INVENTION
There is a problem of symbol prediction during human language speech recognition or, more generally, during recognition and processing of human language requests in any form that assumes vagueness in results. Modern speech recognition systems provide speech-to-text transformation with a great uncertainty that should be decreased by posterior processing. Mainly, speech recognition systems rely on grammar analysis to recognize words and commands that should be done after the speech recognition is complete. For example, Speech Engine 5.1 provides an application with a set of speech recognition results. Usually this set is a set of English sentences that can con-espond to input speech. A right result should be selected from the set at the application level.
A well-predicted next symbol greatly improves the quality of speech-to-text transformation in real-time because of the involvement of common language context in the recognition process. So a correct symbol can be selected in real-time from the set of probable symbols created by speech recognition system whenever a new phoneme arrives. The advantage of the actual method and the according device is its self adjustment and self adaptation to any human language. However, in spite of the declared field of application, the actual invention can be used in any process that requires prediction or generation of a language symbol sequence; for instance, in the artificial intelligence domain.
DESCRIPTION OF PRIOR ART
U.S. Patent Number 5,550,540, which issued to Furlan et al, dated August 27, 1996, for example, discloses the system for distributed coding and prediction by use of contexts. This distributed data processing system includes a plurality of data processing elements for expeditiously performing an encoding or prediction function
SEQUENCES
BACKGROUND OF THE INVENTION
There is a problem of symbol prediction during human language speech recognition or, more generally, during recognition and processing of human language requests in any form that assumes vagueness in results. Modern speech recognition systems provide speech-to-text transformation with a great uncertainty that should be decreased by posterior processing. Mainly, speech recognition systems rely on grammar analysis to recognize words and commands that should be done after the speech recognition is complete. For example, Speech Engine 5.1 provides an application with a set of speech recognition results. Usually this set is a set of English sentences that can con-espond to input speech. A right result should be selected from the set at the application level.
A well-predicted next symbol greatly improves the quality of speech-to-text transformation in real-time because of the involvement of common language context in the recognition process. So a correct symbol can be selected in real-time from the set of probable symbols created by speech recognition system whenever a new phoneme arrives. The advantage of the actual method and the according device is its self adjustment and self adaptation to any human language. However, in spite of the declared field of application, the actual invention can be used in any process that requires prediction or generation of a language symbol sequence; for instance, in the artificial intelligence domain.
DESCRIPTION OF PRIOR ART
U.S. Patent Number 5,550,540, which issued to Furlan et al, dated August 27, 1996, for example, discloses the system for distributed coding and prediction by use of contexts. This distributed data processing system includes a plurality of data processing elements for expeditiously performing an encoding or prediction function
-2-pursuant to a context-based model. This system uses some kind of decomposition and symbol segmentation to build a tree and systolic arrays for data presentation and prediction.
U.S. Patent Number 6,185,259, which issued to Dent, dated February 6, 2001, discloses the Transmitter/receiver for GMSK and offset-QAM. A transmitter encodes a number 2N data bits by using N data bits to select one of 2N levels of a cosine wave and the other N data bits to select one of 2N levels of a sine wave.
This system uses the measurement of disagreements between string symbols in left and right parts to determine a set of coefficients describing the dependence between parts.
Canadian Patent Number CA 2151370, which is issued to Strong, dated February 15, 2005, discloses speech recognition system that uses partitioning for speech recognition rules for generation of a language model and interpretation. Also a generation of a language model may be performed and used for symbol prediction.
U.S. Patent Number 6,058,365, which issued to Nagai et al, dated May 2, 2002, discloses the system for speech processing using an expanded left to right parser. This system provides speech recognition by selecting among hypotheses, consisting of candidates of symbol strings obtained by connecting phonemes corresponding to a Hidden Markov Model having the highest probability, by referring to a phoneme context dependent type from input speech. This system uses a grammar parser of LR and RL stack type, which cannot exactly describe human languages, to analyze and predict phonemes.
U.S. Patent Number 6,665,644, which issued to Kanevsky et al, dated December 16, 2003, discloses a method of Conversational data mining. This method is intended for collecting data associated with the voice of a voice system user includes conducting a plurality of conversations with a plurality of voice system users.
A speech waveform is captured and digitized, and at least one acoustic feature is extracted. The features are correlated with attributes such as gender, age, accent, native language, dialect, educational level and emotional state that are important in some applications.
U.S. Patent Number 6,185,259, which issued to Dent, dated February 6, 2001, discloses the Transmitter/receiver for GMSK and offset-QAM. A transmitter encodes a number 2N data bits by using N data bits to select one of 2N levels of a cosine wave and the other N data bits to select one of 2N levels of a sine wave.
This system uses the measurement of disagreements between string symbols in left and right parts to determine a set of coefficients describing the dependence between parts.
Canadian Patent Number CA 2151370, which is issued to Strong, dated February 15, 2005, discloses speech recognition system that uses partitioning for speech recognition rules for generation of a language model and interpretation. Also a generation of a language model may be performed and used for symbol prediction.
U.S. Patent Number 6,058,365, which issued to Nagai et al, dated May 2, 2002, discloses the system for speech processing using an expanded left to right parser. This system provides speech recognition by selecting among hypotheses, consisting of candidates of symbol strings obtained by connecting phonemes corresponding to a Hidden Markov Model having the highest probability, by referring to a phoneme context dependent type from input speech. This system uses a grammar parser of LR and RL stack type, which cannot exactly describe human languages, to analyze and predict phonemes.
U.S. Patent Number 6,665,644, which issued to Kanevsky et al, dated December 16, 2003, discloses a method of Conversational data mining. This method is intended for collecting data associated with the voice of a voice system user includes conducting a plurality of conversations with a plurality of voice system users.
A speech waveform is captured and digitized, and at least one acoustic feature is extracted. The features are correlated with attributes such as gender, age, accent, native language, dialect, educational level and emotional state that are important in some applications.
-3-U.S. Patent Number 6,161,090, which issued to Kanevsky et al, dated December 12, 2000, discloses apparatus and methods for speaker verification /
identification / classification employing non-acoustic and/or acoustic models and databases. This method is based on generating a score corresponding to the accuracy of the decoded answer and the closeness of the match between the voice sample and the model, and comparing the score to a predetermined threshold value. This invention mainly provides an apparatus for collecting data associated with the voice of a user and a real-time-modifiable voice system for interaction with a user using restricted language set.
U.S. Patent Number 5,793,933, which issued to Iwamasa et al, dated August 11, 1998, discloses the computer-implemented system and method for constructing a system. This system is intended for helping to construct a model-based diagnostic system which allows a user to customize the model-based diagnostic system according to the nature of various diagnosis targets. This invention presents a solution for concrete case of diagnostic systems.
U.S. Patent Number 6,584,464, which issued to Warthen, dated June 24, 2003, discloses the grammar template query system. Information server is the main part of this invention. The information server includes a query input processor, a question processor and an answer processor. The query input processor is used for accepting an initial user query. The question processor processes the initial user query to identify a set of possible well-formed questions selected from the question database, where a well-formed question is a question in the database that is coupled to at least one answer reference. The answer reference is typically either an answer or a pointer to a possible location of an answer. This system uses general mathematical linguistics approach that is common for Internet search systems that implies collecting huge amount of natural language samples and possible rules of language object transformation.
It is apparent from the foregoing that the prior art deals with isolated cases of general problem of effective automatic speech recognition when it is taken in a complex with a problem of computer human language understanding and generation.
identification / classification employing non-acoustic and/or acoustic models and databases. This method is based on generating a score corresponding to the accuracy of the decoded answer and the closeness of the match between the voice sample and the model, and comparing the score to a predetermined threshold value. This invention mainly provides an apparatus for collecting data associated with the voice of a user and a real-time-modifiable voice system for interaction with a user using restricted language set.
U.S. Patent Number 5,793,933, which issued to Iwamasa et al, dated August 11, 1998, discloses the computer-implemented system and method for constructing a system. This system is intended for helping to construct a model-based diagnostic system which allows a user to customize the model-based diagnostic system according to the nature of various diagnosis targets. This invention presents a solution for concrete case of diagnostic systems.
U.S. Patent Number 6,584,464, which issued to Warthen, dated June 24, 2003, discloses the grammar template query system. Information server is the main part of this invention. The information server includes a query input processor, a question processor and an answer processor. The query input processor is used for accepting an initial user query. The question processor processes the initial user query to identify a set of possible well-formed questions selected from the question database, where a well-formed question is a question in the database that is coupled to at least one answer reference. The answer reference is typically either an answer or a pointer to a possible location of an answer. This system uses general mathematical linguistics approach that is common for Internet search systems that implies collecting huge amount of natural language samples and possible rules of language object transformation.
It is apparent from the foregoing that the prior art deals with isolated cases of general problem of effective automatic speech recognition when it is taken in a complex with a problem of computer human language understanding and generation.
-4-The prior art suggests partial solutions that cannot be effective or sometimes applicable in general case.
SUMMARY OF THE INVENTION
The device is a virtual unit comprising the data processing methods of the invention and the means for composing the device functional time diagram and the device external interface.
The device processes symbol sequences in series symbol by symbol using a voting-like procedure to generate every new predicted symbol (P). The currently processing symbol of an input sequence is the entrance symbol (E). The next-to-it symbol of an input sequence is the desired symbol (D). The device does not control entrance and desired symbols of an input sequence, but it only reads them. At any point in time only two symbols of an input sequence (namely the desired and entrance symbols) are directly considered by the device. The desired symbol might not be set to a particular value. On the contrary, the entrance symbol has to have a concrete value at any time when it is processed by the device.
The device begins working after receiving the outside control signal (C) which indicates that the device has to start working.
The data are processed and remembered in the device by the device independent information agents. The said agents receive an entrance symbol that becomes the current symbol they have to process. Each agent has its unique name that is its serial number in the device boundaries. The said agents take part in the weighted voting procedure when they vote for every symbol that has to be predicted.
Information about weights for a weighted poll is kept by each agent separately in its own independent rating list. So weights for a weighted poll can be different for different device agents. Agent's rating lists are collections of combinations of a poll symbol and a numeric rank as a numeric value, where the latter represents the rank for a weighted poll, and poll symbol is a symbol. Each agent can vote only once after
SUMMARY OF THE INVENTION
The device is a virtual unit comprising the data processing methods of the invention and the means for composing the device functional time diagram and the device external interface.
The device processes symbol sequences in series symbol by symbol using a voting-like procedure to generate every new predicted symbol (P). The currently processing symbol of an input sequence is the entrance symbol (E). The next-to-it symbol of an input sequence is the desired symbol (D). The device does not control entrance and desired symbols of an input sequence, but it only reads them. At any point in time only two symbols of an input sequence (namely the desired and entrance symbols) are directly considered by the device. The desired symbol might not be set to a particular value. On the contrary, the entrance symbol has to have a concrete value at any time when it is processed by the device.
The device begins working after receiving the outside control signal (C) which indicates that the device has to start working.
The data are processed and remembered in the device by the device independent information agents. The said agents receive an entrance symbol that becomes the current symbol they have to process. Each agent has its unique name that is its serial number in the device boundaries. The said agents take part in the weighted voting procedure when they vote for every symbol that has to be predicted.
Information about weights for a weighted poll is kept by each agent separately in its own independent rating list. So weights for a weighted poll can be different for different device agents. Agent's rating lists are collections of combinations of a poll symbol and a numeric rank as a numeric value, where the latter represents the rank for a weighted poll, and poll symbol is a symbol. Each agent can vote only once after
-5-the device starts working until the device stops. Each agent has its own weight (W) that has to be set from outside of the device for each agent. Each agent has its rating and calling lists. The device has a post-active agent list that is common for all device agents.
Each agent becomes active and can vote after it receives its personal invitation from any other agent of the device or from any outside device source. The invitation for an agent is a binary signal, whose state means whether at least one invitation has been sent to this particular agent. Active agents can produce invitations after they finish voting. Every particular active agent produces invitations if the current symbol suits the agent's calling list that is a collection of combinations of a table symbol and a new agent name, where the latter is a name of agent that should receive an invitation and a table symbol is a symbol. It is said that the current symbol suits the agent's calling list if the current symbol coincides with one of table symbols in the agent's calling list combinations. After generating invitations for other agents, the agent that generated invitation becomes passive. All invitations are valid for one vote only.
Each agent that has invitation votes for all symbols those are in its rating list.
Each agent calculates the rank of each symbol through multiplying its own weight and the numeric value of rating list slot, which contains the symbol from poll symbol field that coincides with the current symbol. After all agents have completed voting, the symbol that has got the maximum sum of weighted votes becomes a predicted symbol.
If the desired symbol is not set at this moment, the device deletes everything from the post-active agent list; and then copies the serial number of a randomly selected agent, which was active during processing of the current symbol, to the post-active agent list; and then stops and waits for outside signal to process new entrance and desired symbols. The post-active agent list contains no serial number when the device begins functioning for the first time.
If the desired symbol is set and the desired symbol doesn't coincide with the predicted symbol, in order to achieve the coincidence of the desired and predicted symbols, the device fulfils one of three randomly selected alternative acts:
first, allows _(_ the device agent with a serial number in the post-active agent list to add a new combination that consists of a serial number of one randomly selected currently passive device agent and the current symbol to the said agent's calling list;
or, second, to add a new combination that consists of the entrance symbol and the numeric value ' "one" to the said agent's rating list if there is no such combination in the said list; or, ..
third, to increment the voting rank in the slot of the said agent's rating list, which contains symbol in poll symbol field that coincides with the current symbol.
Then, after the said agent completes its modifications of its lists or completes incrementing the rank of current symbol in its rating list, the device deletes everything from the post-active agent list, copies the serial number of a randomly selected agent, which was active during processing of the current symbol, to the post-active agent list, stops and waits for the outside signal to process new entrance and desired symbols.
The agents' invitations can also be set from outside the device at any time when the device is waiting for the outside signal to begin working.
The device Selection, Comparison and Control blocks regulate the time sequence of data exchange between the device functional parts independently of each other. These blocks use data in the device memories and the device signal values.
Generally the device can function according to two time diagrams. The first diagram occurs when the desired symbol is set and doesn't coincide with the predicted symbol. Otherwise the device works according to the second diagram. The beginning stages of both diagrams are identical until the predicted symbol is generated based on the calculation of the weighted votes of all active agents. Then, acxording to the first diagram, the active agents send their invitations to other device agents; or, according to the second diagram, when the desired and predicted symbols don't coincide, the device requires one randomly selected agent to modify its lists in order to achieve coincidence of the desired and predicted symbols and after that allows the active agents to send their invitations. At the end in both diagrams the device confirms the generated predicted symbol and stops.
_7_ The internal structure of the device blocks guarantees that there are no contrary data streams in the device at the same time, and there are no collisions amidst the device signals.
The data streams at the device inputs, combined input/outputs and outputs should be regulated from the outside of the device.
The method and the according device may be implemented in software, hardware, or a combination of software and hardware. The device is useful in real-time speech recognition and artificial intelligence systems such as natural language processing (NLP) or other systems with natural language input and requests. In NLP
case the outside feedback might be created to transfer the predicted symbol to the entrance symbol for the next device cycle. Agents' weights and agents' external invitation signals can be useful for supplementary adjustments if necessary.
Otherwise all agents' weights can be of the same value, and the external invitation signals can be absent.
BRIEF DESCRIPTION OF THE DRAWINGS
In the drawings, which form a part of this specification, Fig 1 is the device representative schema;
Fig 2 is the device data stream diagram;
Fig 3 is the device agent internal structure.
DETAIL DESRIPTION OF INVENTION
In the device description the specific signals and the specific signal processing blocks that form the device functional diagram are set forth in order to provide detail understanding of the invention only. It is apparent that the invention may be practiced _g_ without these specific signals and blocks if a proper device functional diagram is formed by any other means.
The device consists of the central unit and agents that exchange data and signals between each other and the outside of the device. The central unit, agents and signals can be realized by software programs and software program variables or any other appropriate means such as triggers and logical devices. In the actual invention the term memory is considered as an abstract named data storage that can be a part of computer RAM or any other dedicated or shared equivalent storage means; and the agents and blocks of the invention are considered as methods of data transformation.
Also the device inputs and outputs can be considered as software program entries.
The number of agents is not limited in the device. Every device agent has its unique name in the device boundaries. The agents' names are presented in the device by numeric values.
The initial state of the device is waiting. In the said state the device does nothing save accepting external data and waiting for the outside signal, which indicates that the device has to start working. The only other state of the device is working. The external data should not be altered while the device is working.
The device Central unit consists of:
~ Memory 1 for storing an entrance symbol. The content of memory 1 has to be set from outside the device through the first device input while the device is in the waiting state;
~ Memory 2 for storing the desired symbol and signal 5. The desired symbol is the next- to-entrance symbol in the current symbol sequence. The initial value of signal 5 is a low-level value. The only other value of the said signal is a high-level value. The desired symbol in memory 2 and the value of signal 5 have to be set from the outside of the device through the second device input while the device is in the waiting state. If desired symbol is not stored in memory 2, the signal 5 should keep a low-level value.
Otherwise the said signal should be assigned a high-level value;
~ Memory 3 for storing the predicted symbol and signal 6. Initial value of signal 6 is a low-level value. The only other permitted value of the said signal is a high-level value.
Memory 3 can be read from outside the device through the first device output;
~ Memory 4 for storing the signal 1; the initial value of this signal is a low-level value;
only other value is a high-level value;
~ Memory 5 for storing a numeric value in the range from 0 to 1 inclusively and signal 7. The initial value of signal 7 is a low-level value. The only other permitted value of the said signal is a high-level value;
~ Memory 6 for storing a name of device agent. The default value of memory 6 is zero;
~ Memory 7 for storing the signal 3; the initial value of this signal is a low-level value;
only other permitted value is a high-level value. Signal 3 is accessible for reading and writing from ou~ide the device through the first combined input/output;
~ Memory 8 for storing the temporary list of the Selection block. Memory 8 is subdivided into slots. Each slot of the said memory has its left and right fields;
~ Memory 9 for storing a name of a device agent. The default value of memory 9 is zero;
~ Memory 10 for storing signal 4 that serves to form the device functional diagram.
The initial value of this signal is a low-level value. The only other permitted value is a high-level value;
~ Memory 11 for storing signal 2 that serves to form the device functional diagram.
The initial value of this signal is a low-level value. The only other value is a high-level value;
~ The Comparison block for comparing symbols in memories 2 and 3 and producing signal 1;
- 1~-~ The Selection block for selecting the predicted symbol on the base of the entrance symbol and the device agents' rating lists and for storing the predicted symbol in memory 3;
~ The Control block that serves to form the device functional diagram, ~ The Random block for random choosing of a device agent's serial number among all serial numbers of agents that are currently in an active state, ~ The Choosing block for random choosing of a device agent's serial number among all serial numbers of agents that are currently in a passive state.
The Comparison block waits while signal 6 has a low-level value. If the value of signal 6 becomes a high-level value, the Comparison block compares the symbol in memory 2 with the symbol in memory 3 and then assigns a high-level value to signal 1 if the said symbols coincide or signal 5 has a low-level value. Otherwise the Comparison block assigns a low-level value to signal 1.
Signal 3 regulates the device functioning from outside the device. Low-level value of signal 3 also means that the device completed functioning and switched itself to the waiting state and waits for new symbols to process.
Before the device begins functioning after receiving the high-level of signal 3, the entrance symbol should be stored in memory 1 from outside the device. Then if necessary, desired symbol should be stored in memory 2 from outside the device and signal 5 should be set to a high-level value from outside the device.
Otherwise signal 5 should be set to a low-level value.
The device begins functioning after signal 3 has been set to high-level value from outside the device. At this moment, when signal 3 changes its value from low-level value to high level value, the Control block simultaneously changes signals 1 and 2 to low-level value and signal 4 to high-level value, and changes the value of signal 6 to low-level value, and changes the value of signal 7 to low-level value.
-11_ Every device agent has two lists: namely calling and rating lists. Calling and rating lists are kept in the said agent's memory. Every device agent has its own memory for keeping its lists.
Every agent can stay in two alternative states: active and passive.
Every agent has its weight that has to be set from outside the device through the second combined input/output of the device. Each agent keeps its weight in its own memory.
Agent can be in active state if the said agent has received an invitation from any other device agent or outside the device. The invitation for any particular agent is a signal that can have only low or high value. High-level value of the said signal means that at least one invitation has been sent to this particular agent. The invitation for any particular agent can be sent by any other device agent or from outside the device through the second combined input/output of the device. Agent is active if it is in an active state.
Passive state is the state of the device agents, which have not received an invitation from any other device agent or outside the device. Agent is passive if it is in passive state.
Every agent has its own agent's memory to store the calling and rating lists.
The said memory consists of plurality of data slots. Each slot is divided into two fields - left and right. At the very beginning the memory of any device agent is empty, e.g. it does not contain slots.
Every calling list consists of a set of combinations: a table symbol in the left field and a new agent name in the right field of the said slots of agent's calling list. New agent name means a serial number of some agent that is presented in the device.
Table symbol is a symbol. Each table symbol is unique for the calling list of any concrete agent, e.g. there are no coinciding symbols in table symbol slot fields of the calling list of any concrete agent of the device.
The rating list consists of a set of combinations: a poll symbol in the left field and a numeric rank in the right field of the said slots of agent's rating list. Numeric rank slot field keeps numbers only. Poll symbol is a symbol. Each poll symbol is unique for the rating list of any concrete agent, e.g. there are no coinciding symbols in poll symbol slot fields of the rating list of any concrete device agent.
If signal 4 has high-level value and signal 1 has low-level value, the Selection block does in series:
~ Waits for the combination of low-level value of signal 2 and low-level value of signal
Each agent becomes active and can vote after it receives its personal invitation from any other agent of the device or from any outside device source. The invitation for an agent is a binary signal, whose state means whether at least one invitation has been sent to this particular agent. Active agents can produce invitations after they finish voting. Every particular active agent produces invitations if the current symbol suits the agent's calling list that is a collection of combinations of a table symbol and a new agent name, where the latter is a name of agent that should receive an invitation and a table symbol is a symbol. It is said that the current symbol suits the agent's calling list if the current symbol coincides with one of table symbols in the agent's calling list combinations. After generating invitations for other agents, the agent that generated invitation becomes passive. All invitations are valid for one vote only.
Each agent that has invitation votes for all symbols those are in its rating list.
Each agent calculates the rank of each symbol through multiplying its own weight and the numeric value of rating list slot, which contains the symbol from poll symbol field that coincides with the current symbol. After all agents have completed voting, the symbol that has got the maximum sum of weighted votes becomes a predicted symbol.
If the desired symbol is not set at this moment, the device deletes everything from the post-active agent list; and then copies the serial number of a randomly selected agent, which was active during processing of the current symbol, to the post-active agent list; and then stops and waits for outside signal to process new entrance and desired symbols. The post-active agent list contains no serial number when the device begins functioning for the first time.
If the desired symbol is set and the desired symbol doesn't coincide with the predicted symbol, in order to achieve the coincidence of the desired and predicted symbols, the device fulfils one of three randomly selected alternative acts:
first, allows _(_ the device agent with a serial number in the post-active agent list to add a new combination that consists of a serial number of one randomly selected currently passive device agent and the current symbol to the said agent's calling list;
or, second, to add a new combination that consists of the entrance symbol and the numeric value ' "one" to the said agent's rating list if there is no such combination in the said list; or, ..
third, to increment the voting rank in the slot of the said agent's rating list, which contains symbol in poll symbol field that coincides with the current symbol.
Then, after the said agent completes its modifications of its lists or completes incrementing the rank of current symbol in its rating list, the device deletes everything from the post-active agent list, copies the serial number of a randomly selected agent, which was active during processing of the current symbol, to the post-active agent list, stops and waits for the outside signal to process new entrance and desired symbols.
The agents' invitations can also be set from outside the device at any time when the device is waiting for the outside signal to begin working.
The device Selection, Comparison and Control blocks regulate the time sequence of data exchange between the device functional parts independently of each other. These blocks use data in the device memories and the device signal values.
Generally the device can function according to two time diagrams. The first diagram occurs when the desired symbol is set and doesn't coincide with the predicted symbol. Otherwise the device works according to the second diagram. The beginning stages of both diagrams are identical until the predicted symbol is generated based on the calculation of the weighted votes of all active agents. Then, acxording to the first diagram, the active agents send their invitations to other device agents; or, according to the second diagram, when the desired and predicted symbols don't coincide, the device requires one randomly selected agent to modify its lists in order to achieve coincidence of the desired and predicted symbols and after that allows the active agents to send their invitations. At the end in both diagrams the device confirms the generated predicted symbol and stops.
_7_ The internal structure of the device blocks guarantees that there are no contrary data streams in the device at the same time, and there are no collisions amidst the device signals.
The data streams at the device inputs, combined input/outputs and outputs should be regulated from the outside of the device.
The method and the according device may be implemented in software, hardware, or a combination of software and hardware. The device is useful in real-time speech recognition and artificial intelligence systems such as natural language processing (NLP) or other systems with natural language input and requests. In NLP
case the outside feedback might be created to transfer the predicted symbol to the entrance symbol for the next device cycle. Agents' weights and agents' external invitation signals can be useful for supplementary adjustments if necessary.
Otherwise all agents' weights can be of the same value, and the external invitation signals can be absent.
BRIEF DESCRIPTION OF THE DRAWINGS
In the drawings, which form a part of this specification, Fig 1 is the device representative schema;
Fig 2 is the device data stream diagram;
Fig 3 is the device agent internal structure.
DETAIL DESRIPTION OF INVENTION
In the device description the specific signals and the specific signal processing blocks that form the device functional diagram are set forth in order to provide detail understanding of the invention only. It is apparent that the invention may be practiced _g_ without these specific signals and blocks if a proper device functional diagram is formed by any other means.
The device consists of the central unit and agents that exchange data and signals between each other and the outside of the device. The central unit, agents and signals can be realized by software programs and software program variables or any other appropriate means such as triggers and logical devices. In the actual invention the term memory is considered as an abstract named data storage that can be a part of computer RAM or any other dedicated or shared equivalent storage means; and the agents and blocks of the invention are considered as methods of data transformation.
Also the device inputs and outputs can be considered as software program entries.
The number of agents is not limited in the device. Every device agent has its unique name in the device boundaries. The agents' names are presented in the device by numeric values.
The initial state of the device is waiting. In the said state the device does nothing save accepting external data and waiting for the outside signal, which indicates that the device has to start working. The only other state of the device is working. The external data should not be altered while the device is working.
The device Central unit consists of:
~ Memory 1 for storing an entrance symbol. The content of memory 1 has to be set from outside the device through the first device input while the device is in the waiting state;
~ Memory 2 for storing the desired symbol and signal 5. The desired symbol is the next- to-entrance symbol in the current symbol sequence. The initial value of signal 5 is a low-level value. The only other value of the said signal is a high-level value. The desired symbol in memory 2 and the value of signal 5 have to be set from the outside of the device through the second device input while the device is in the waiting state. If desired symbol is not stored in memory 2, the signal 5 should keep a low-level value.
Otherwise the said signal should be assigned a high-level value;
~ Memory 3 for storing the predicted symbol and signal 6. Initial value of signal 6 is a low-level value. The only other permitted value of the said signal is a high-level value.
Memory 3 can be read from outside the device through the first device output;
~ Memory 4 for storing the signal 1; the initial value of this signal is a low-level value;
only other value is a high-level value;
~ Memory 5 for storing a numeric value in the range from 0 to 1 inclusively and signal 7. The initial value of signal 7 is a low-level value. The only other permitted value of the said signal is a high-level value;
~ Memory 6 for storing a name of device agent. The default value of memory 6 is zero;
~ Memory 7 for storing the signal 3; the initial value of this signal is a low-level value;
only other permitted value is a high-level value. Signal 3 is accessible for reading and writing from ou~ide the device through the first combined input/output;
~ Memory 8 for storing the temporary list of the Selection block. Memory 8 is subdivided into slots. Each slot of the said memory has its left and right fields;
~ Memory 9 for storing a name of a device agent. The default value of memory 9 is zero;
~ Memory 10 for storing signal 4 that serves to form the device functional diagram.
The initial value of this signal is a low-level value. The only other permitted value is a high-level value;
~ Memory 11 for storing signal 2 that serves to form the device functional diagram.
The initial value of this signal is a low-level value. The only other value is a high-level value;
~ The Comparison block for comparing symbols in memories 2 and 3 and producing signal 1;
- 1~-~ The Selection block for selecting the predicted symbol on the base of the entrance symbol and the device agents' rating lists and for storing the predicted symbol in memory 3;
~ The Control block that serves to form the device functional diagram, ~ The Random block for random choosing of a device agent's serial number among all serial numbers of agents that are currently in an active state, ~ The Choosing block for random choosing of a device agent's serial number among all serial numbers of agents that are currently in a passive state.
The Comparison block waits while signal 6 has a low-level value. If the value of signal 6 becomes a high-level value, the Comparison block compares the symbol in memory 2 with the symbol in memory 3 and then assigns a high-level value to signal 1 if the said symbols coincide or signal 5 has a low-level value. Otherwise the Comparison block assigns a low-level value to signal 1.
Signal 3 regulates the device functioning from outside the device. Low-level value of signal 3 also means that the device completed functioning and switched itself to the waiting state and waits for new symbols to process.
Before the device begins functioning after receiving the high-level of signal 3, the entrance symbol should be stored in memory 1 from outside the device. Then if necessary, desired symbol should be stored in memory 2 from outside the device and signal 5 should be set to a high-level value from outside the device.
Otherwise signal 5 should be set to a low-level value.
The device begins functioning after signal 3 has been set to high-level value from outside the device. At this moment, when signal 3 changes its value from low-level value to high level value, the Control block simultaneously changes signals 1 and 2 to low-level value and signal 4 to high-level value, and changes the value of signal 6 to low-level value, and changes the value of signal 7 to low-level value.
-11_ Every device agent has two lists: namely calling and rating lists. Calling and rating lists are kept in the said agent's memory. Every device agent has its own memory for keeping its lists.
Every agent can stay in two alternative states: active and passive.
Every agent has its weight that has to be set from outside the device through the second combined input/output of the device. Each agent keeps its weight in its own memory.
Agent can be in active state if the said agent has received an invitation from any other device agent or outside the device. The invitation for any particular agent is a signal that can have only low or high value. High-level value of the said signal means that at least one invitation has been sent to this particular agent. The invitation for any particular agent can be sent by any other device agent or from outside the device through the second combined input/output of the device. Agent is active if it is in an active state.
Passive state is the state of the device agents, which have not received an invitation from any other device agent or outside the device. Agent is passive if it is in passive state.
Every agent has its own agent's memory to store the calling and rating lists.
The said memory consists of plurality of data slots. Each slot is divided into two fields - left and right. At the very beginning the memory of any device agent is empty, e.g. it does not contain slots.
Every calling list consists of a set of combinations: a table symbol in the left field and a new agent name in the right field of the said slots of agent's calling list. New agent name means a serial number of some agent that is presented in the device.
Table symbol is a symbol. Each table symbol is unique for the calling list of any concrete agent, e.g. there are no coinciding symbols in table symbol slot fields of the calling list of any concrete agent of the device.
The rating list consists of a set of combinations: a poll symbol in the left field and a numeric rank in the right field of the said slots of agent's rating list. Numeric rank slot field keeps numbers only. Poll symbol is a symbol. Each poll symbol is unique for the rating list of any concrete agent, e.g. there are no coinciding symbols in poll symbol slot fields of the rating list of any concrete device agent.
If signal 4 has high-level value and signal 1 has low-level value, the Selection block does in series:
~ Waits for the combination of low-level value of signal 2 and low-level value of signal
6, and then ~ Creates the temporary list of combinations in memory 8. Each combination consists of data in memory slots: a poll symbol at the left field and a numeric rank at the right field of the said slots. Poll symbol is a symbol, and numeric rank is a number. Just after the temporary list is created the temporary list does not contain slots;
the temporary list will be filled in at the next step, ~ Sequentially looks through all rating lists of currently active agents by sequentially selecting the agents and then the agents' slots in their rating lists one by one (each slot is viewed only once); if the symbol in the left field of the selected slot in the rating lists of currently selected agent is already presented in one of the left fields of the slots in temporary list, then the Selection block adds the product of the selected agent's weight and the numeric value in the right field of the selected slot of currently selected agent to the numeric rank value in the right field of the slot of the temporary list that contains the said symbol. Otherwise the Selection block adds a new slot combination to the temporary list. The said combination consists of the said symbol in the left field of the new slot of the temporary list and the product of the weight of currently selected agent and the said numeric value in the according right field of the same temporary list slot. Formally, let R(c) be the sum of weighted votes for symbol c. Then R(c) _ ~ f Vij(c).Wj , (1 ) i j where Vij(c) is the numeric value in the i-th rating list slot of the j-th agent that contains symbol c as poll symbol in the left field of the said slot, and Wj is a weight of the j-th agent, ~ Selects the symbol in the poll symbol parts of slots of the temporary list that has the maximum value in the according numeric rank field of the slot among all slots of the temporary list, e.g. finds symbol c with maximum R(c), and then stores the said symbol in memory 3, ~ Switches value of signal 6 to high-level value to let the Comparison block to compare symbols and produce the value of signal 1, and then ~ Generates a random number in the range from 0 to 1 inclusively and stores it in the memory 5, ~ Randomly chooses (through the Choosing block) a name of agent among all names of agents that are currently in the passive state and stores the said name, represented by the agent serial number, in memory 9, ~ Sets value of signal 7 to high-level value to allow agents to change the calling or rating list if necessary, ~ Waits for one of two events: low-level value of signal 7 or high-level value of signal 1, and then ~ Assigns high-level value to signal 2 to let the agents to send their invitation signals, ~ Deletes the temporary list of combinations in memory 8, ~ Assigns low-level value to signal 4.
Each device agent accepts the active state if signal 1 has low-level value and the said agent has received at least one invitation signal from any other agent or outside the device before the device functioning has begun. After accepting the active state the said agent discards the invitation, which the said agent received from other device agents or outside the device, and keeps its own lists and weight.
In an active state the agent waits for high-level value of signal 2, and then sends its own invitation to other agents, which serial numbers are in the said agent calling list slots and the table symbol of the said slots coincides with the device current symbol, and then the said agent accepts a passive state. Agents do not generate invitation signals in a passive state.
If signal 1 has high-level value and there are no device agents in the active state, the Control block changes signals 1, 2 and 3 to low-level value. At the said moment the device completes functioning and the predicted symbol in memory 3 can be read outside the device. When signal 3 has high-level value the symbol stored in memory 3 is not reliable.
In a passive state the device agent keeps its lists, weight and invitation signal received from other device agents or outside the device in the said agent own memory.
If signal 1 has low-level value and the value of the signal 7 is high-level value and memory 5 contains value '1', the agent with the serial number stored in memory 6 does in series:
~ Increments numeric value of numeric rank field of the said agent rating list in the slot that contains the symbol in the poll symbol slot field that coincides with the symbol in memory 2; if such combination does not exists in the rating list of the said agent, the agent creates the said combination and adds the according slot with numeric value '1' in its right field and the symbol from memory 2 in the left field to the rating list in the memory of the said agent, ~ Randomly chooses (through the Random block) a name of agent among all names of agents that are currently in the active state and stores the said name, represented by the agent serial number, in memory 6, ~ Sets signal 7 to low-level value.
-1$-If signal 1 has low-level value, and signal 7 has high-level value, and memory contains value '0', the agent with a serial number in memory 6 does in series:
~ Replenishes its calling list by adding a new slot with the combination, in which the symbol in the table symbol slot field coincides with the symbol in memory 1, and the according new agent name slot field accepts the agent serial number from memory 9, ~ Randomly chooses (using the Random block) a name of agent among all names of agents that are currently in the active state and stores the said name, represented by the agent serial number, in the memory 6, ~ Sets signal 7 to low-level value.
In the foregoing specification the present invention has been described with reference to embodiments thereof shown in Figures 1, 2 and 3. The drawings are to be regarded in an illustrative rather than restrictive sense.
the temporary list will be filled in at the next step, ~ Sequentially looks through all rating lists of currently active agents by sequentially selecting the agents and then the agents' slots in their rating lists one by one (each slot is viewed only once); if the symbol in the left field of the selected slot in the rating lists of currently selected agent is already presented in one of the left fields of the slots in temporary list, then the Selection block adds the product of the selected agent's weight and the numeric value in the right field of the selected slot of currently selected agent to the numeric rank value in the right field of the slot of the temporary list that contains the said symbol. Otherwise the Selection block adds a new slot combination to the temporary list. The said combination consists of the said symbol in the left field of the new slot of the temporary list and the product of the weight of currently selected agent and the said numeric value in the according right field of the same temporary list slot. Formally, let R(c) be the sum of weighted votes for symbol c. Then R(c) _ ~ f Vij(c).Wj , (1 ) i j where Vij(c) is the numeric value in the i-th rating list slot of the j-th agent that contains symbol c as poll symbol in the left field of the said slot, and Wj is a weight of the j-th agent, ~ Selects the symbol in the poll symbol parts of slots of the temporary list that has the maximum value in the according numeric rank field of the slot among all slots of the temporary list, e.g. finds symbol c with maximum R(c), and then stores the said symbol in memory 3, ~ Switches value of signal 6 to high-level value to let the Comparison block to compare symbols and produce the value of signal 1, and then ~ Generates a random number in the range from 0 to 1 inclusively and stores it in the memory 5, ~ Randomly chooses (through the Choosing block) a name of agent among all names of agents that are currently in the passive state and stores the said name, represented by the agent serial number, in memory 9, ~ Sets value of signal 7 to high-level value to allow agents to change the calling or rating list if necessary, ~ Waits for one of two events: low-level value of signal 7 or high-level value of signal 1, and then ~ Assigns high-level value to signal 2 to let the agents to send their invitation signals, ~ Deletes the temporary list of combinations in memory 8, ~ Assigns low-level value to signal 4.
Each device agent accepts the active state if signal 1 has low-level value and the said agent has received at least one invitation signal from any other agent or outside the device before the device functioning has begun. After accepting the active state the said agent discards the invitation, which the said agent received from other device agents or outside the device, and keeps its own lists and weight.
In an active state the agent waits for high-level value of signal 2, and then sends its own invitation to other agents, which serial numbers are in the said agent calling list slots and the table symbol of the said slots coincides with the device current symbol, and then the said agent accepts a passive state. Agents do not generate invitation signals in a passive state.
If signal 1 has high-level value and there are no device agents in the active state, the Control block changes signals 1, 2 and 3 to low-level value. At the said moment the device completes functioning and the predicted symbol in memory 3 can be read outside the device. When signal 3 has high-level value the symbol stored in memory 3 is not reliable.
In a passive state the device agent keeps its lists, weight and invitation signal received from other device agents or outside the device in the said agent own memory.
If signal 1 has low-level value and the value of the signal 7 is high-level value and memory 5 contains value '1', the agent with the serial number stored in memory 6 does in series:
~ Increments numeric value of numeric rank field of the said agent rating list in the slot that contains the symbol in the poll symbol slot field that coincides with the symbol in memory 2; if such combination does not exists in the rating list of the said agent, the agent creates the said combination and adds the according slot with numeric value '1' in its right field and the symbol from memory 2 in the left field to the rating list in the memory of the said agent, ~ Randomly chooses (through the Random block) a name of agent among all names of agents that are currently in the active state and stores the said name, represented by the agent serial number, in memory 6, ~ Sets signal 7 to low-level value.
-1$-If signal 1 has low-level value, and signal 7 has high-level value, and memory contains value '0', the agent with a serial number in memory 6 does in series:
~ Replenishes its calling list by adding a new slot with the combination, in which the symbol in the table symbol slot field coincides with the symbol in memory 1, and the according new agent name slot field accepts the agent serial number from memory 9, ~ Randomly chooses (using the Random block) a name of agent among all names of agents that are currently in the active state and stores the said name, represented by the agent serial number, in the memory 6, ~ Sets signal 7 to low-level value.
In the foregoing specification the present invention has been described with reference to embodiments thereof shown in Figures 1, 2 and 3. The drawings are to be regarded in an illustrative rather than restrictive sense.
Claims (29)
PROPERTY OR PRIVILEGE IS CLAIMED AS FOLLOWS:
1. The method of predicting symbols in correlated symbol sequences and the according device that predicts symbols in a symbol sequence on the base of the previous sequence parts and the device experience gained by the device during processing all previous symbol sequences.
2. Memory of the device of claim 1 for storing the entrance symbol.
3. Memory of the device of claim 1 for storing the desired symbol and a signal that can have either low-level or high-level value.
4. Memory of the device of claim 1 for storing the predicted symbol and a signal that can have either low-level or high-level value.
5. Memory of the device of claim 1 for storing a signal that can have either low-level or high-level value each.
6. Memory of the device of claim 1 for storing a numeric value in the range from 0 to 1 inclusively.
7. Memory of the device of claim 1 for storing the serial number of device agent that is in active state.
8. Memory of the device of claim 1 for storing the serial number of device agent that is in passive state.
9. Memory of the device of claim 1 for storing and keeping the agent's weights, invitation signals that agents receive from other agents, and two agent's lists wherein the said first (polling) list consists of slots that are subdivided each into left and right fields, where the right field of said slots stores numerical values and the left field of said slot stores symbols, and wherein the said second (calling) list consists of slots that are subdivided each into left and right fields, where the right field of said slots stores symbols and the left part of said slots stores the agent serial numbers.
Each agent of the device has the said memory.
Each agent of the device has the said memory.
10. Memory of the device of claim 1 for storing a temporary list of combinations that is subdivided into slots that have left and right fields each where the right field of said slot stores numerical values and the left field of the said slot stores symbols.
11. Memory of claim 8 wherein the said memory comprises the Choosing block for randomly choosing the serial number of agent among all serial numbers of the device agents that are in passive state.
12. Memory of claim 7 wherein the said memory comprises the Random block for randomly choosing the serial number of agent among all serial numbers of the device agents that are in active state.
13. Memory of claim 9 wherein the said memory comprises a block for adding the first memory 9 list with the combination of agent serial number and the symbol where the said combination is composed of the desired symbol and the serial number of agent that is stored in memory of claim 6, and wherein the said memory comprises a block for adding the second memory list with the combination of poll symbol and numeric value where the said combination is composed of the desired symbol and the numeric value one in the respective slot fields.
14. The Selection block of the device of claim 1 for selection of the predicted symbol on the base of the entrance symbol in memory of claim 2 and the device agents' rating lists in memory of claim 9 using the method given by formula (1), which determines the calculation of sums of weighted votes;
and for the further selection of the predicted symbol that is a symbol that has got the maximum weighted vote sum; and for the further storing of the predicted symbol in memory of claim 4.
and for the further selection of the predicted symbol that is a symbol that has got the maximum weighted vote sum; and for the further storing of the predicted symbol in memory of claim 4.
15. The Comparison block of the device of claim 1 for comparison of the symbol in memory of claim 3 and the symbol in memory of claim 4 using the signals of memories of claims 3 and 4 wherein the said Comparison block comprises the means for producing a value of the signal in memory of claim 5.
16. Memory of the device of claim 1 for storing a signal that can have either low-level or high-level value that serves to form the device functional diagram.
17. Device agents of the device of claim 1 for processing the entrance symbol and the desired symbol, calling and rating lists and invitation signals.
18. Memory of the device of claim 1 for storing a signal that can have either low-level or high-level value that serves to form the device functional diagram.
19. A block of the agent of the claim 17 for receiving the invitation signals from other agents or outside the device before the device functioning has begun.
Each agent of the device has the said block.
Each agent of the device has the said block.
20. A block of the agent of claim 17 for generating the invitation signals to other device agents using the invention method, which requires that each agent that is in active state generates the invitation signals for other device agents if these agents' serial numbers are in the said agent memory (claim 9) second list slots, which also contains the symbol that coincides with the symbol in memory of claim 2. Each agent of the device has the said block.
21. A block of the agent of claim 17 for incrementing a numeric value of numeric rank field in the slot of the first list in the memory of claim 9 that contains the symbol that coincides with the symbol in memory of claim 2 in the poll symbol slot field if there is such slot in the said agent first list and the said agent is in active state; and for adding the slots to the said agent first list in memory of claim 9 if there is no such slot in the said agent first list and the said agent is in active state; and for adding the slots to the said agent second list in memory of claim 9 if the said agent second list doesn't have a slot with symbol in the memory of claim 2 and with the agent number in memory of claim 7 and the said agent is in active state. Each agent of the device has the said block.
22. The Selection block of claim 14 wherein the said block comprises the means for producing values of signals in memories of claims 4 and 16 using the signal of memory of claim 16 and the signal in memory of claim 4.
23. The Control block of the device of claim 1 for checking the device agents' states and for the setting up values to signals in memories of claims 5 and 18, and for the changing the signals in memories of claims 16, 3 and 4.
24. The device of claim 1 wherein the said device comprises the input for entrance symbol.
25. The device of claim 1 wherein the said device comprises the input for the desired symbol.
26. The device of claim 1 wherein the said device comprises the output for the predicted symbol.
27. The device of claim 1 wherein the said device comprises the combined input/output for the control signal.
28. The device of claim 1 wherein the said device comprises the combined input/output for the agent weights and for the external invitation signals.
29. Memory of the device of claim 1 for storing a signal that can have either low-level or high-level value that regulates the device functioning from outside the device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CA002507260A CA2507260A1 (en) | 2005-05-18 | 2005-05-18 | Device and method of predicting symbols in correlated symbol sequences |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CA002507260A CA2507260A1 (en) | 2005-05-18 | 2005-05-18 | Device and method of predicting symbols in correlated symbol sequences |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CA2507260A1 true CA2507260A1 (en) | 2006-11-18 |
Family
ID=37450446
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002507260A Abandoned CA2507260A1 (en) | 2005-05-18 | 2005-05-18 | Device and method of predicting symbols in correlated symbol sequences |
Country Status (1)
| Country | Link |
|---|---|
| CA (1) | CA2507260A1 (en) |
-
2005
- 2005-05-18 CA CA002507260A patent/CA2507260A1/en not_active Abandoned
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20210397610A1 (en) | Machine learning system for digital assistants | |
| CN108255934B (en) | Voice control method and device | |
| CN110457432B (en) | Interview scoring method, interview scoring device, interview scoring equipment and interview scoring storage medium | |
| CN110096567B (en) | QA knowledge base reasoning-based multi-round dialogue reply selection method and system | |
| US7289950B2 (en) | Extended finite state grammar for speech recognition systems | |
| CN1953056B (en) | Conversation controller | |
| CN111916111B (en) | Intelligent voice outbound method and device with emotion, server and storage medium | |
| CN112100354B (en) | Man-machine conversation method, device, equipment and storage medium | |
| CN108711420A (en) | Multilingual hybrid model foundation, data capture method and device, electronic equipment | |
| CN1953057A (en) | Conversation controller | |
| CN111694940A (en) | User report generation method and terminal equipment | |
| CN111191002A (en) | Neural code searching method and device based on hierarchical embedding | |
| DE19635754A1 (en) | Speech processing system and method for speech processing | |
| CN110853626A (en) | Bidirectional attention neural network-based dialogue understanding method, device and equipment | |
| CN111078847A (en) | Power consumer intention identification method and device, computer equipment and storage medium | |
| CN109461437A (en) | The verifying content generating method and relevant apparatus of lip reading identification | |
| CN109522397A (en) | Information processing method and device based on semanteme parsing | |
| CN112527955A (en) | Data processing method and device | |
| KR20170122755A (en) | Hammer pair expansion device, a computer program for it, and a question answering system | |
| CN114818729A (en) | Method, device and medium for training semantic recognition model and searching sentence | |
| CN113590078A (en) | Virtual image synthesis method and device, computing equipment and storage medium | |
| CN112836030A (en) | Intelligent dialogue system and method | |
| CN110717027B (en) | Multi-round intelligent question-answering method, system, controller and medium | |
| JP6810580B2 (en) | Language model learning device and its program | |
| CN114153942A (en) | Event time sequence relation extraction method based on dynamic attention mechanism |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request | ||
| FZDE | Discontinued |