Patent Grant
10671815
Embodiments of the present invention relate generally to natural language generation technologies and, more particularly, relate to a method, apparatus, and computer program product for operator text generation.
Natural language generation (NLG) is sometimes referred to as a subfield of artificial intelligence and computational linguistics that focuses on the production of understandable texts in English or other understandable language. In some examples, a natural language generation (NLG) system is configured to transform raw input data that is expressed in a non-linguistic format into a format that can be expressed linguistically, such as through the use of natural language (e.g., the conversion from data to text). In some cases the data is high frequency numerical data. For example, raw input data may take the form of a value of a stock market index over time and, as such, the raw input data may include data that is suggestive of a time, a duration, a value and/or the like. Other examples, may include the generation of textual weather forecasts base on numerical weather prediction data. Therefore, an NLG system may be configured to input the raw input data and output text that linguistically describes the value of the stock market index; for example, “securities markets rose steadily through most of the morning, before sliding downhill late in the day.” Importantly, for use in an NLG system, data must be analysed and interpreted in a way in which the analysis and interpretation can be linguistically communicated. For example, data that indicates the price of a stock market rising may be represent linguistically as rising, spiking or the like. A human may then make decisions based on how that human interprets rising versus spiking.
Data that is input into a NLG system may be provided in, for example, a recurrent formal structure. The recurrent formal structure may comprise a plurality of individual fields and defined relationships between the plurality of individual fields. For example, the input data may be contained in a spreadsheet or database, presented in a tabulated log message or other defined structure, encoded in a ‘knowledge representation’ such as the resource description framework (RDF) triples that make up the Semantic Web and/or the like. In some examples, the data may include numerical content, symbolic content or the like. Symbolic content may include, but is not limited to, alphanumeric and other non-numeric character sequences in any character encoding, used to represent arbitrary elements of information. In some examples, the output of the NLG system is text in a natural language (e.g. English, Japanese or Swahili), but may also be in the form of synthesized speech.
Methods, apparatuses, and computer program products are described herein that are configured to generate an operator text in response to an alarm that is either received from an alarm or alert system or that is self-generated based on an analysis of one or more data feeds. The method of an example embodiment may include determining whether an operator text is to be generated in response to a received alert condition by performing data analysis operations comprising: analyzing, using a processor, a primary data feed and at least one confirmatory data feed to identify one or more features; and determining based on the detection of a feature in the primary data feed or the at least one confirmatory data feed satisfies at least one predetermined constraint. The method may further include generating an output text that is displayable in a user interface that describes at least a diagnosis for the feature that satisfied that at least one predetermined constraint.
In some example embodiments, a computer implemented method is disclosed herein that includes determining whether an operator text is to be generated in response to a received alert condition by performing data analysis operations. In some examples, the data analysis operations may include: analyzing, using a processor, a primary data feed and at least one confirmatory data feed to identify one or more features; determining whether an alert condition in the primary data feed is confirmed by at least one confirmatory data feed, wherein the alert condition is validated in an instance in which the primary data feed and the confirmatory data feed satisfy a correlation threshold; determining whether the feature in the primary data feed is explainable by at least one diagnostic data feed; and traversing, using one or more features in the primary data feed or the at least one confirmatory data feed, a decision tree, wherein the decision tree is operable to determine that at least a portion of an operator text is to be generated in an instance in which a feature evaluates as true for at least one node of the decision tree. In further examples, the computer implemented method may include generating the operator text for at least one feature that evaluates as true by performing language generation operations. The language generation operations may include, but are not limited to: arranging one or more messages in a document plan data structure in an order in which they are to be linguistically described in the operator text, wherein each of the one or more messages are data structures that are linguistically describable using at least one of a word or phrase and are instantiated based on the one or more features in the primary data feed or the at least one confirmatory data feed; converting, using a processor, at least one of the one or more messages into a text specification data structure that represents one or more data structures that are representative of at least one syntactic constituent and syntactic feature of a sentence; and applying a grammar to the text specification data structure to generate the operator text that is displayable in a user interface, wherein the operator text describes at least a diagnosis based on the at least one diagnostic data feed for the feature that evaluated as true.
In some example embodiments, an apparatus is disclosed herein that includes at least one processor; and at least one memory including computer program code, the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus to at least determine whether an operator text is to be generated in response to a received alert condition by performing data analysis operations. In some examples, the data analysis operations may be performed by: analyzing a primary data feed and at least one confirmatory data feed to identify one or more features; determining whether an alert condition in the primary data feed is confirmed by at least one confirmatory data feed, wherein the alert condition is validated in an instance in which the primary data feed and the confirmatory data feed satisfy a correlation threshold; determining whether the feature in the primary data feed is explainable by at least one diagnostic data feed; and traversing, using one or more features in the primary data feed or the at least one confirmatory data feed, a decision tree, wherein the decision tree is operable to determine that at least a portion of an operator text is to be generated in an instance in which a feature evaluates as true for at least one node of the decision tree. In further examples, the apparatus may further be configured to generate the operator text for at least one feature that evaluates as true by performing language generation operations. The language generation operations are performed by arranging one or more messages in a document plan data structure in an order in which they are to be linguistically described in the operator text, wherein each of the one or more messages are data structures that are linguistically describable using at least one of a word or phrase and are instantiated based on the one or more features in the primary data feed or the at least one confirmatory data feed; converting at least one of the one or more messages into a text specification data structure that represents one or more data structures that are representative of at least one syntactic constituent and syntactic feature of a sentence; and applying a grammar to the text specification data structure to generate the operator text that is displayable in a user interface, wherein the operator text describes at least a diagnosis based on the at least one diagnostic data feed for the feature that evaluated as true.
In some example embodiments, a computer program product is disclosed herein that includes at least one computer readable non-transitory memory medium having program code instructions stored thereon, the program code instructions which when executed by an apparatus cause the apparatus at least to determine whether an operator text is to be generated in response to a received alert condition by performing data analysis operations. In some examples, the data analysis operations may be performed by analyzing a primary data feed and at least one confirmatory data feed to identify one or more features; determining whether an alert condition in the primary data feed is confirmed by at least one confirmatory data feed, wherein the alert condition is validated in an instance in which the primary data feed and the confirmatory data feed satisfy a correlation threshold; determining whether the feature in the primary data feed is explainable by at least one diagnostic data feed; and traversing, using one or more features in the primary data feed or the at least one confirmatory data feed, a decision tree, wherein the decision tree is operable to determine that at least a portion of an operator text is to be generated in an instance in which a feature evaluates as true for at least one node of the decision tree. In further examples, the computer program prodcut may further be configured to generate the operator text for at least one feature that evaluates as true by performing language generation operations. The language generation operations are performed by arranging one or more messages in a document plan data structure in an order in which they are to be linguistically described in the operator text, wherein each of the one or more messages are data structures that are linguistically describable using at least one of a word or phrase and are instantiated based on the one or more features in the primary data feed or the at least one confirmatory data feed; converting at least one of the one or more messages into a text specification data structure that represents one or more data structures that are representative of at least one syntactic constituent and syntactic feature of a sentence; and applying a grammar to the text specification data structure to generate the operator text that is displayable in a user interface, wherein the operator text describes at least a diagnosis based on the at least one diagnostic data feed for the feature that evaluated as true.
In some example embodiments, an apparatus is disclosed herein that includes means for determining whether an operator text is to be generated in response to a received alert condition by performing data analysis operations. In some examples, the means for data analysis operations may include means for analyzing a primary data feed and at least one confirmatory data feed to identify one or more features; means for determining whether an alert condition in the primary data feed is confirmed by at least one confirmatory data feed, wherein the alert condition is validated in an instance in which the primary data feed and the confirmatory data feed satisfy a correlation threshold; means for determining whether the feature in the primary data feed is explainable by at least one diagnostic data feed; and means for traversing, using one or more features in the primary data feed or the at least one confirmatory data feed, a decision tree, wherein the decision tree is operable to determine that at least a portion of an operator text is to be generated in an instance in which a feature evaluates as true for at least one node of the decision tree. In further examples, the apparatus may include means for generating the operator text for at least one feature that evaluates as true by performing language generation operations. The means for language generation operations may include, but are not limited to: means for arranging one or more messages in a document plan data structure in an order in which they are to be linguistically described in the operator text, wherein each of the one or more messages are data structures that are linguistically describable using at least one of a word or phrase and are instantiated based on the one or more features in the primary data feed or the at least one confirmatory data feed; means for converting at least one of the one or more messages into a text specification data structure that represents one or more data structures that are representative of at least one syntactic constituent and syntactic feature of a sentence; and means for applying a grammar to the text specification data structure to generate the operator text that is displayable in a user interface, wherein the operator text describes at least a diagnosis based on the at least one diagnostic data feed for the feature that evaluated as true.
Having thus described embodiments of the invention in general terms, reference will now be made to the accompanying drawings, which are not necessarily drawn to scale, and wherein:
Example embodiments will now be described more fully hereinafter with reference to the accompanying drawings, in which some, but not all embodiments are shown. Indeed, the embodiments may take many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements. Like reference numerals refer to like elements throughout. The terms “data,” “content,” “information,” and similar terms may be used interchangeably, according to some example embodiments, to refer to data capable of being transmitted, received, operated on, and/or stored. Moreover, the term “exemplary”, as may be used herein, is not provided to convey any qualitative assessment, but instead merely to convey an illustration of an example. Thus, use of any such terms should not be taken to limit the spirit and scope of embodiments of the present invention.
In some examples, one or more operators, technicians or the like may be tasked with responding to or otherwise addressing a series of alerts over a given time period. In particular examples, the operator may be responding to alerts that relate to machinery or equipment that could have a catastrophic result on people, infrastructure, other machines, platforms or the like if it were to fail. Given the importance of alert detection, the operator is generally inundated with alarming systems, data output system (dials, warning lights, graphs, indicators or the like) and various other outputs. As such, the operator may become overwhelmed or may indeed not interpret or understand the data (or warning signs) that are being presented before him/her. In some example cases, constant alarms or intermittent alarms may result in an operator simply ignoring or not responding to alarms.
Indeed, the problem facing the operator is not only the ability to detect patterns, trends or signals in the data (e.g., high frequency sensor data) that signify a fault, but also then determining if he/she needs to take a particular action to stave off a catastrophe. Most operators do not have the training, experience, time or the like to identify these patterns, trends or signals (that may act as an early warning signal) and make the necessary decisions. In other examples, the operator may see an alarm signal, but may be unable to diagnose the problem within a critical time period. Another more troubling problem is that the operator may not deduce a fault even when the warning signs are showing in the data. Operators are faced with each of these problems in real time as they are generally responsible for multiple systems, personnel and/or the like over long and demanding shifts.
In many instances, an operator may be required to interpret data, symbols or information in order to make a decision or otherwise gain an understanding of a current context, a situation or an event. In order to make such a decision or to otherwise react to an event, an operator may need to obtain an awareness (e.g. situational awareness) of the current context, the situation or the event before taking an action. Situational awareness may be defined as the perception of environmental elements with respect to time and/or space, the comprehension of their meaning, and the projection of their status after some variable has changed, such as time, or based on the happening of an event such as an alarm or alert. In other words, situational awareness is a state achieved when information that is qualitatively and quantitatively determined as suitable for a particular purpose is made available to a user by engaging them in an appropriate information exchange pattern or mental model. Situational awareness involves being aware of what is happening in the vicinity of a person or event to understand how information, events, and/or one's own actions may impact goals and objectives, both immediately and in the near future. Situational awareness may also be related to the perception of the environment critical to decision-makers in complex, dynamic areas from aviation, air traffic control, power plant operations, military command and control, engineering, machine monitoring, oil and gas, power plant monitoring, nuclear energy and emergency services such as firefighting and policing. The lack of or inadequate situational awareness has been identified as one of the primary factors in accidents attributed to human error.
Operators, technicians, nurses, and many other users in the modern world spend a great deal of time reacting to computer-generated alarms, alerts, and other indications, which result in a high false-positive rate. In some cases the false positive rate may be over 99%. One extreme example may be found with respect to airport baggage-screeners, where most screeners never see a valid alert in their entire working career. Identifying rare events is a difficult thing for users to do well, because the human brain is not well-suited to dealing with this kind of situation. For example, if a user has only ever seen or generally only sees certain alarms as false positives, the user may have a tendency to assume that all similar future alarms will also be false positives. Such cases occur, especially in instances in which operators deal with a large number of alarms, and, as such, may only have a few minutes for each individual alarm.
As such, the methods, apparatus, and computer program products described herein are configured to process incoming data into one or more data feeds and generate a diagnosis of a particular fault and actionable instructions for that operator to resolve the particular fault. In some examples, the operator may be operating on an oil extraction platform, operating an airplane, monitoring a patient, watching stock data or the like. In the case of the oil platform operator, the methods, apparatus, and computer program products described herein are configured to communicate with that operator so that operator is able to respond and address a problem or alert condition that is to occur (e.g., a piece of machinery may fail) within a given time period, such as within 30 minutes. Additionally, not only must that operator be alerted, but any such alert, such as the text generated by the methods, apparatus, and computer program products described herein, is configured to provide analytics, such as interpretation of an alarm system, an equipment failure diagnosis, a recommendation or the like. Additionally, the methods, apparatus, and computer program products described herein may also be designed to generate a text only for those critical alerts to avoid spamming, overwhelming or otherwise distracting the operator.
As described herein, the methods, apparatus, and computer program products are operable to solve or otherwise address the aforementioned problems by generating an output text in response to an alarm that is either received from an alarm or alert system or that is self-generated based on an analysis of one or more data feeds. The methods, apparatus, and computer program products described herein may then analyze one or more data feeds (e.g. one or more data feeds generated by sensors attached to machinery) to diagnose the fault and ultimately linguistically describe that diagnosis and provide, in some cases, recommendations, situational analysis and/or the like in an output text.
Advantageously, and in some examples, the output text that is generated by the methods, apparatus, and computer program products described herein is concise, reliable and tailored to the operator (e.g., based on domain rules, operator education level, context and/or the like). In some examples, the output text provides the operator with a decision or course of action so that the operator is no longer required to make a decision, but instead may act (e.g., shutdown compressor A, activate backup cooling system G, etc.). Advantageously, the operator text is designed to be constrained by a predetermined number of characters to ensure that the operator will read and address the fault without extraneous distractors in the text.
In order to generate the output text, an NLG architecture may be used that is tuned or otherwise designed for the generation of an operator text. As is described herein, the NLG system is operable to process data in the form of high frequency numerical data, such as a data feed received from a temperature sensor that is monitoring a compressor, and transform one or more observations or abstractions (e.g., features) taken from the high frequency numerical data into a linguistic representation of that data. Alternatively or additionally, the NLG system described herein may be configured to generate texts, such as operator texts, engineer texts, situational awareness texts, alert validation texts and/or the like for the following non-exclusive list of oil platform equipment: reservoirs (e.g., modeling, well production forecasting, drainage performance, drive performance, gas cap maintenance, injection profiling and/or the like), wells (e.g., performance management, integrity assurance, completion integrity, inflow/outflow, impairment prediction and/or the like), oil plant (rotating equipment surveillance, strainers and filters, water quality monitoring, injection pump surveillance, injector profiling and/or the like), gas plant (compressor monitoring, dehydration, vapor recovery, gas quality assurance and/or the like), utilities (e.g., power generation, control system performance, control and safety, valve performance and/or the like), drilling (pore pressure prediction, real time drilling information, trouble avoidance, borehole safety, well control and/or the like), and/or the like.
In other words, the methods, apparatus, and computer program products described herein are configured to manipulate data that is stored in a memory of a computing system to transform or otherwise represent that data in natural language. For example, the data may be interpreted as illustrating a temperature rising, a temperature spiking, or a temperature sustaining a dangerous level for a period of time. In addition to simply providing a linguistic description of data trends or features, the output texts may also include a linguistic description of related machinery, a historical analysis of this type of alert in this machine or related machine, context (such as weather) that may be effecting the current alert or the like.
In some examples, and in order to generate an output text (e.g., operator text), the methods, apparatus, and computer program products as described herein may be configured to monitor one or more data feeds. These data feeds may be generated from sensors that are monitoring one or more pieces of equipment, may be data that is captured in stock trades, weather prediction data or the like. In some cases, the methods, apparatus, and computer program products described herein are operable to detect a trend in the one or more data feeds that is indicative of an alert condition. In other cases, the alert condition may be identified based on the activation of an alarm, the receipt of alarming indication or the like.
In response to the identification of an alarm condition related to a particular piece of equipment, the process described herein is configured to evaluate whether an output or operator text should be generated. In order to make such a decision, the methods, apparatus, and computer program products as described herein are operable to access a hierarchy (or other type of model that describes relationships between one or more data feeds) to determine the piece of equipment (e.g., the equipment or machine to which the sensor is attached) that is experiencing the alarm condition. In some examples, other related sensors or data feeds may also be loaded and/or analyzed, they include, but are not limited to: data feeds from confirmatory data feeds (e.g., one or more data feeds derived from sensors that are monitoring the same item, equipment or the like), explanatory data feeds (e.g., one or more data feeds that may be used to explain a particular condition), diagnostic data feeds (e.g., data feeds that are used to diagnose a fault or issue on a piece of equipment, such as, for example, providing an indication whether the equipment on or off, indicative of a fault based on a pressure level satisfying a threshold, and/or the like) or the like.
In some example embodiments, the alert condition or fault may be verified using the related data feeds (e.g., confirmatory data feeds, explanatory data feeds, diagnostic data feeds or the like). For example, the methods, apparatus, and computer program products as described herein may determine whether the alert condition is a result of a harmless cause as explained by, in some examples, the explanatory data feeds. For example, a determination is made as to whether an increase in temperature is the result of an intentional increase in speed by the machine (e.g., the operator intentionally increased speed to increase production). In such an example, an operator likely would expect a temperature reading to rise in an instance in which the equipment is working harder (e.g., speed is increasing) and thus would not want an operator text because the machine is likely behaving normally. Additionally, the alert condition may be validated based on an analysis of confirmatory data feeds. The analysis of the confirmatory data feeds should be the same or similar as the primary data feed for the alert to be valid. For example, if the primary data feed (e.g., the data feed that caused or otherwise is the basis for the alert condition) indicates an increase in temperature, but the confirmatory data feed s indicate temperature is normal, the alert or primary data feed may be experiencing a sensor glitch. In such a case, the operator text would likely indicate the need to replace the sensor instead of a more serious problem.
In some example embodiments, a diagnostic data feed may also be analyzed. In such examples, the diagnostic data feed may indicate a problem or fault with a piece of equipment, a component of the equipment (e.g. a cooling system for a compressor), an entity or the like. In such cases an alert condition directed to high temperature may be a result of a failed cooling system or a cooling system that was not operating (e.g., the cooling system is switched off). As such, the diagnostic data feed may assist in developing a diagnosis and/or a recommendation to such a problem in an output text (e.g., directing an operator to activate the cooling system).
In some example embodiments, and in conjunction with a determination of a shutdown threshold, a determination may be made as to a time until failure occurs (e.g., system will overheat and fail in 25 minutes). In particular, based on the trend analysis, the methods, apparatus, and computer program products described herein may be operable to provide an output text that describes the time until critical or catastrophic event occurs.
In the event that the alert condition is verified, the methods, apparatus, and computer program products described herein are configured to diagnose the underlying cause of the alert condition (e.g., fault) and then generate an output text or operator text that provides a recommendation or solution that addresses or otherwise solves the fault. The generation of an output text is further described with respect to
In some example embodiments an alert reception system 102 is configured to receive an indication of an alert condition (e.g. an alert received from a source such as, but not limited to, another system, an alarm system, a monitoring system or the like), a violation of a constraint (e.g. a data value over a threshold, within a threshold for a period of time and/or the like), a user input or the like. The alert reception system 102 may in some example embodiments be in data communication with an alert monitoring system, machine monitoring systems, equipment monitoring server, alarm centers, sensors and/or the like. In examples in which the alert reception system 102 is in communication with a monitoring system, the alert reception system 102 may receive a report (e.g., metadata), that includes but is not limited to a report identifier, an identification of a machine, sensor or the like that is experiencing the alert, the type of alert, a description, a title, and dates for maintenance requests related to the alert and/or the date time group (e.g. a set of characters, usually in a prescribed format, which may express the year, the month, the day of the month, the hour of the day, the minute of the hour, and/or the time zone) for the alert. Other alert condition information may be received and/or otherwise accessed in an alert database or data store via an alert monitoring system. However, in some example embodiments, the methods, apparatus and computer products described herein may operate without an alert reception system and/or the alert reception system 102 may be embodied by the data analysis system 104.
Alternatively or additionally, the alert reception system 102 is further configured to gather information, such as via raw input data 110, historical data 112, event log 116 or the like, on the history of an alert condition over a particular time period; for example, how often has the alert condition been triggered, how long was the alert condition active, or how often was the alert validated or not validated, responses to the alert, maintenance activity or the like. In some examples, the alert reception system 102 may further identify if a maintenance request is currently pending for an alert condition and, if so, a current status of the maintenance request. Closed or completed maintenance requests may also be determined by the alert reception system 102. A time period used to search for historical information about the alert condition may be predetermined, set by a user, set according to a domain model and/or the like. In some examples, the time period may be configured to exclude intermittent alert condition activity as described herein.
Alternatively or additionally, the alert reception system 102 may further gather or otherwise access information and/or data that refers to the status of a machine or unit having the alert condition. The information and/or data includes but is not limited to the functionality of the machine or unit, current operation of components of the machine, dependency diagrams that indicate whether power, cooling or the like is being supplied, the most recent start or stop times, maintenance times and/or the like. For example, the alert reception system 102 may determine whether a machine or unit has been started or stopped in the last 24 hours.
Referring now to
As is described herein, the machines 202 and 204, the components 210-216 and/or the sensors 220-226 may be defined in terms of relationships (e.g., based on an equipment hierarchy). For example sensor 220 may be related to sensor 222 with a relationship defined as the data generated by sensor 220 moves in the opposite direction of sensor 222 (e.g. 220 is falling and 222 is rising). Other relationships, such as a relation between sensor 220 and 224, may also be defined in some example embodiments. Each of the relationships may be given an importance and/or otherwise may be weighted based on the importance level of the relationship between the sensors, such as 220 and 222, components 210 and 212 and/or the like. Relationship and importance may be determined for each machine, component, sub component, sensor and/or the like. Metadata and/or an ontology may also be specified that includes a linguistic description or title for each component, sub component or the like.
Referring again to
In some example embodiments, the data analysis system 104 and/or the referenced alert module 232 may be configured to receive relevant data about an alert and gather any relevant metadata about the alert. In some examples, the alert condition is at least one of received from an alert monitoring system over a data communication link or is determined based on the violation of a predetermined constraint by the primary data feed. Based, on the detection of an alert condition, the referenced alert module 232 may input raw data segmented into one or more data feeds. The data feed is correlated, in some examples, to a particular sensor, group of sensors or the like. The one or more data feeds may be contained in or otherwise received from the raw input data 110.
In some example embodiments, the receipt or input of the one or more data feeds may occur in response to an alert condition, such as is indicated by or otherwise received in the alert from the alert reception system 102. Alternatively or additionally the data analysis system 104 or the referenced alert module 232 may be configured to receive or input the one or more data feeds continuously or semi-continuously and may determine an importance of the one or more data feeds (e.g., whether the data violates a constraint, satisfies a threshold and/or the like) in order to detect or otherwise determine the presence of an alert condition. In other words, in some example embodiments, the data analysis system 104, the referenced alert module 232 or the like may function as or otherwise embody the alert reception system 102 and/or an alert detection system.
The one or more data feeds (e.g., the raw data may include data such as, but not limited to, time series data that captures variations across time (e.g. profits, rainfall amounts, temperature or the like), spatial data that indicates variation across location (e.g. rainfall in different regions), or spatial-temporal data that combines both time series data and spatial data (e.g. rainfall across time in different geographical output areas)). The one or more data feeds may be provided in the form of numeric values for specific parameters across time and space, but the raw input data may also contain alphanumeric symbols, such as the RDF notation used in the semantic web, or as the content of database fields.
In some example embodiments, the data analysis system 104 or the referenced alert module 232 is further configured to determine a main or primary data feed. In some examples, a main or primary data feed may be selected based on a selection by a user, via a user interface, may be selected based on the happening of a condition such as, but not limited to, an alert, an alarm, an anomaly, a violation of a constraint, a warning, may be predetermined and/or the like (e.g., features). In some cases, the primary data feed is generally related to, for example, the raw input data and/or data feed that caused the alert condition.
In some example embodiments, the data analysis system 104, the hierarchy module 134 or the like is configured to access a hierarchy (e.g., a part-of-hierarchy) to identify or otherwise read information about the equipment that is related to the primary data feed. The hierarchy in some examples describes the relationship of a particular piece of equipment within a larger grouping of equipment. For example, a turbine may be a part of a helicopter which is part of a squadron, etc. As such, in an instance in which a primary data feed is identified (e.g., the data feed related to the sensor that detected a condition that caused the current alert condition), the piece of equipment is then identified in the hierarchy.
In some example embodiments, the equipment status detector 236 is configured is determine current and/or historical data for a particular piece of equipment. In some examples, the equipment status detector 236 may provide additional context or additional details for a particular piece of equipment or machine. For example, the maintenance history for the machine, the usage history (e.g. time on line), the on/off times and/or the like.
In some example embodiments, a related data feed reader 238 is configured to load or otherwise access one or more related data feeds, one or more historical data feeds and/or the like based on the data received from the hierarchy module 134. In some example embodiments, the one or more related data feeds are loaded based on an identification of one or more confirmatory data feeds as defined by the accessed hierarchy. In some examples, the historical data feeds may be related to the primary data feed (e.g., the data feed providing the indication of the alert condition) or the one or more confirmatory data feeds. In some examples, the related data feeds reader may load or otherwise access one or more explanatory data feeds and/or one or more diagnostic data feeds.
Relationships between data feeds may be defined as important or unexpected or may represent anomalous behavior as defined by a qualitative model such as the domain model 114. In some examples, the domain model 114 is a representation of information about a particular domain. For example, a domain model may contain an ontology that specifies the kinds of objects, concepts and/or the like that may exist in the domain in concrete or abstract form, properties that may be predicated of the objects, concepts and the like, relationships that may hold between the objects concepts and the like, and representations of any specific knowledge that is required to function in the particular domain. In some examples, multiple domain models may be provided for a single domain. For example, domains may include, but are not limited to, medical, rotating machinery, oil and gas, logistics, news and reporting, industrial, power, weather, legal, financial, nuclear and/or the like.
In some examples, relationships may be defined with respect to one or more sensors, sensor types, monitoring equipment or the like as defined by the hierarchy. For example, an alert on a first sensor may have a relationship with a second sensor on the same machine or on other related machines. The relationship may further include an indication that the related sensor generally detects movement in the same direction as the primary sensor or that the related sensor generally moves in an opposite direction from the primary sensor. An importance level may be also be assigned to a particular sensor, such as by a user, the domain model or the like. In some examples, this importance level may be used to determine an importance of a particular sensor event. In some cases, one or more sensors may be indicated as always to be included in a text.
Alternatively or additionally, the method, apparatus and computer program product described herein may further provide and/or otherwise access a rule language that is configured to qualitatively specify expected relationships between data feeds (e.g. a primary data feed and one or more related data feeds). The rule language may further be configured to specify or otherwise indicate instances in which related data feeds, related events and/or the like are mentioned in an operator text. In some examples, the rules may be built via a user interface, using the rule language that may extract or otherwise receive information related to the data feeds and the relationships between them. In some cases the data feeds and/or relationships may be discovered by analyzing one or more data sources. The rule language may further provide information related to the context of one or more data feeds and alerts that may be generated or accessed based on those data feeds, for example equipment status, alert history, and status of similar equipment elsewhere. Alternatively or additionally, the rule language may be derived and/or tuned using supervised or unsupervised learning models.
In some example embodiments, the signal analysis controller 240 is configured to analyze the one or more data feeds provided by the primary data feed, the confirmatory data feeds, the explanatory data feeds, the diagnostic data feeds and/or the like. Initially, the signal analysis controller may be configured to determine one or more problems or inconsistencies with the one or more data feeds that may be indicative of an error or invalid data. In some cases, the data problems may be in the data feed itself, in other cases the data problems may be a result of a communications error. In some examples, the signal analysis controller 240 may detect missing data in an instance in which data was not received over a predetermined duration (e.g. exceeds the time in which values are reported). In other examples, a sensor's values may not have changed over a predetermined period before an alert condition was indicated; such a case may indicate a frozen data condition. Frozen data may be identified in an instance in which a machine or unit is determined to be running during a period (e.g. based on a determined equipment status), the sensors are float valued and/or the sensor is not excluded from the frozen check. In other examples, raw input data may be checked to determine if the values fall outside those physically possible (e.g. temperatures that would cause melting). Other verification may include error tokens, inconsistent values and/or the like.
In some example embodiments, the signal analysis controller 240 is configured to detect patterns and trends in the one or more data feeds that are derived from the raw input data and/or the historical data to provide a set of abstractions. For example, a time-series dataset may contain tens of thousands of individual records describing the temperature at various time points on a component piece of machinery over the course of a day with a sample once every two or three seconds. Trend analysis may then be used to identify that the temperature changes in a characteristic way throughout certain parts of the day. As such, the trend analysis is configured to abstract those changes over time into an abstraction that is representative of the change over time.
In some example embodiments, the signal analysis controller 240 may be configured to fit a piecewise linear model to the data received in the primary data feed, related data feed or the like. The fitting of the piecewise linear model may include filtering in some examples. For each trend in the raw input data, the signal analysis controller 240 may determine a start and/or end time values, qualitative direction (e.g. up, down, steady), qualitative stability (e.g. stable, unstable), threshold status (e.g. normal, high, low, unknown at start, end of trend and/or the like). The signal analysis controller 240 may be configured to perform the fitting of the piecewise linear model for one or more time scales; for example, over a short term (e.g. 6 hours) using the time of the alert condition to determine how the machine or unit was acting at the time the alert became active. A longer time period (e.g. 2 months) may also be analyzed. In some examples, the longer time period may ignore equipment off periods and/or the like.
The signal analysis controller 240 may alternatively or additionally then identify trends, spikes, steps, oscillations or other patterns in the data feeds to generate abstractions that summarize the patterns determined in the primary data feed and/or the other related data feeds. Alternatively or additionally, the data analysis system 104 may also be configured to perform pattern detection on the raw input data irrespective of data feeds or the receipt of an alarm condition.
In other example embodiments, the signal analysis controller 240 may determine whether an alert condition in the primary data feed is confirmed by at least one confirmatory data feed. For example, the alert condition may be confirmed if the behavior in the primary data feed is replicated, at least in part, by the behavior in the confirmatory data feed). Additionally or alternatively, the signal analysis controller may determine whether an alert condition in the primary data feed is to be discarded based on at least one explanatory data feed. For example, the alert condition may not be confirmed if the increase in temperature is a result of an intentional increase in speed by the machine. Additionally, the signal analysis controller 240 may determine whether the feature in the primary data feed is explainable by at least one diagnostic data feed. For example, a diagnostic data feed may report that the cooling system is disabled for a machine. As such, the temperature increase may be explained by the disabled cooling system and could be rectified by activating the cooling system.
By way of example, and in an instance in which a temperature alert has been received, the signal analysis controller 240 may determine if the pattern or feature (increase, decrease or stable) of the primary data feed is explained by explanatory data feeds. If the main data feed is not explained by the explanatory data feeds, the signal analysis controller 240 may determine if the primary data feed and the confirmatory data feeds are correlated. If the primary data feed and its confirmatory data feeds are correlated: the signal analysis controller 240 may look-up the cooling system that is currently configured to be related the system and return any fault associated with that cooling system.
In some examples, the correlation processor 242 may be operable to analyze and, in some examples, score the signal correlation between the primary data feed and the one or more of its related data feeds (e.g., confirmatory data feeds, explanatory data feeds, diagnostic data feeds and/or the like). Determining the signal correlation between the primary data feed and its related data feeds allows for the assessment of the operational condition preceding the alarms (e.g., an increase/decrease in production, the start of an equipment failure, etc.). Signal correlation quantifies the strength of a linear relationship between two signals. If no signal correlation exists between two signals, then there is no tendency for the values of the variables to increase or decrease in tandem. The implemented signal correlation function produces a matrix of correlation coefficients R, for a signal matrix (where each column represents a separate signal). The correlation coefficients may range from −1 to 1, in some example embodiments. In other examples, correlation coefficients above a predetermined threshold (e.g., above 0.02) indicate that there is a positive linear relationship or correlation between the data columns. The signal correlation function also returns P, a matrix of p-values for testing the hypothesis of no correlation. Each p-value is the probability of getting a correlation as large as the observed value by random chance, when the true correlation is zero. If P(i,j) is small, for example less than 0.05, then the correlation R(i,j) is significant.
In some examples, the recommendation and diagnosis processor 244 may be operable to determine whether to generate a text, and if so, what type of a recommendation or diagnosis to include in that text. In some examples, in order to determine whether an operator text is to be generated a decision tree, a machine learning model or the like may be used. In the decision tree embodiment, a series of if statements may be defined (and may ultimately be traversed) that identify one or more problems, features, patterns or the like (e.g., indication in the data that violate a constraint or the like) that if evaluated as true would trigger a text. In other examples, a machine learning model may be trained such that the presence of one or more features (e.g., indications of events, patterns detected or like) would generate a score that signifies the likelihood that a text should be generated. In an instance in which that score satisfied an operator text threshold score, then the system may generate the text. In some cases, these embodiments may be informed or constrained by one or more business rules.
In some examples, recommendations are built through a series of Decision Points. A Decision Point is a check that can be true or false. For example, the “BeingShutdown” decision point recommends that nothing should be done if the equipment is being currently shut down. In example embodiments herein, Decision Points enable the recommendation and diagnosis processor to determine whether a text to the operators. These are typically based on specific faults. For example, have all of the temperatures suddenly started rising together for no apparent reason? If yes, an operator text should be generated, for example, that indicates the problem as a clogged strainer issue. As such, should any decision point be true, a text may be generated for the operator in some example embodiments. In other examples, if the primary data feed is a temperature related data feed and it has an increasing trend, the system may decide there is a cooling issue and thus would generate an operator text with that diagnosis. If the primary data feed is related to a pressure tag and it has a decreasing trend, the system may determine there is a leakage issue and would generate an operator text with that diagnosis.
In some examples, a Decision Point is made up of four things: a certainty, an explanation, a piece of logic, and a set of features. The certainty of a Decision Point is a relative value that is used by the overall recommendation. Certainties may have values between 1 and −1. Problem Decision Points make the most use of their certainty. Their certainty is a relative measure of whether this Decision Point indicates a problem or not. A 1 means any time this Decision Point is true, there is definitely a problem. Similarly, −1 means there is no chance that there is a problem. The total probability that there is a problem is created by combining all of the different Problem Decision Points. One method for combining decision points includes, but is not limited to:
t=running total of certainty
for (each decision point) {
}
In some examples, action Decision Points use their certainty as a ranking system. The highest certainty Decision Point is used in the total recommendation, and the others are just included in the action reasons. Error Decision Points don't use their certainty. If there has been any problem in the recommendation processor, such is reported in the text. Error Decision Points should be set as 0. A Decision Point's explanation is the explanation supplied to the reader for why it is/isn't a problem, why that action was recommended, why there may be an error, etc. This a string which is used directly by the output text.
The logic of a Decision Point is what calculates whether the Decision Point is true or not. For example, a decision point related to temperature may include a test as to whether temperature is greater than a given threshold on a primary data feed and at least one secondary data feed. If the logic evaluates to true a Reason is created. A Reason is a message (e.g., described below) which wraps around the explanation. The features of a Decision Point are parameters that may be saved and used with machine learning that is described herein.
In some examples, the recommendation and diagnosis processor 244 may include a diagnostic model may be configured to determine or otherwise identify one or more possible causes of an alert condition. Based on the one or more possible causes of the alert condition, the diagnostic model may be further configured to determine which of the one or more data feeds should be examined to verify and/or disprove the one or more possible causes. For example, the diagnosis model may identify allergic reaction to medication as one reason for a spike in heart rate. It may also determine that if allergic reaction is suspected, a respiration rate data feed should be examined to verify and/or disprove this diagnosis, because respiration rate would become erratic if the underlying cause was allergic reaction. As such, in this example, respiration rate may be identified as a related data feed.
In some examples, machine learning may be used to build and/or modify the diagnostic model. In some examples, features for the machine learning model may be defined and may include: information about a data feed, results of historical decision point decisions, anything the decision points compare against a threshold and/or the like. Such a model can be trained by mapping the features to one or more historical alerts that have resulted in a service or maintenance request. As such, features can be selected based on their likelihood to result in an operator text. In some examples classifiers on subsets of the features (e.g., a J48 tree can be used). Because of a potential of an imbalance of services-non services, a costSensitive classifier can be used. This meta-classifier makes it so different false positives have different costs.
In some example embodiments, a natural language generation system, such as natural language generation system 106, is configured to generate phrases, sentences, text or the like which may take the form of natural language text. The natural language generation system 106 comprises a document planner 130, a microplanner 132 and/or a realizer 134. Other natural language generation systems may be used in some example embodiments, such as a natural language generation pipeline as described in Building Natural Language Generation Systems by Ehud Reiter and Robert Dale, Cambridge University Press (2000), which is incorporated by reference in its entirety herein.
In some examples, natural language generation system 106 may be configured to populate or otherwise instantiate one or more messages based on data or information in the primary data feed, the one or more related data feeds, the historical data, the contextual data feed, one or more events and/or the like. In some examples, messages are language independent data structures that correspond to informational elements in a text and/or collect together underling data in such a way that the underlying data can be linguistically expressed. In some examples, messages are created based on a requirements analysis as to what is to be communicated for a particular scenario (e.g. for a particular domain). A message typically corresponds to a fact about the underlying data (for example, the existence of some observed event) that could be expressed via a simple sentence (although it may ultimately be realized by some other linguistic means). For example, to linguistically describe wind, a user may want to know a speed, a direction, a time period or the like, but also the user wants to know changes in speed over time, warm or cold fronts, geographic areas and or the like. In some cases, users do not even want to know wind speed; they simply want an indication of a dangerous wind condition. Thus, a message related to wind speed may include fields to be populated by data related to the speed, direction, time period or the like, and may have other fields related to different time points, front information or the like. The mere fact that wind exists may be found in the data, but to linguistically describe “light wind” or “gusts” different data interpretation must be undertaken as is described herein.
In some examples, a message is created by the natural language generation system 106 in an instance in which the data in the one or more data feeds warrants the construction of such a message. For example, a wind message would only be constructed in an instance in which wind data was present in the raw input data. Alternatively or additionally, while messages may correspond directly to observations taken from the raw data input, others, however, may be derived from the observations by means of a process of inference. For example, the presence of rain may be indicative of other conditions, such as the potential for snow at some temperatures. Alternatively or additionally, in some example embodiments, the natural language generation system 106 may embody all or portions of the data analysis system 104.
The concepts and relationships that make up messages may be drawn from an ontology (e.g. a domain model) that formally represents knowledge about the application scenario. For example, message structures may be defined by the domain model 114 based on a particular alert condition and/or the raw input data, such as but not limited to the primary and/or related data feeds. Messages may also be derived from another data structure, may be user defined and/or the like. Each type of message may also be represented by a message template, which expresses a relationship between instances of a number of concepts; the message template contains slots which may be filled in, or instantiated, using particular values that are derived from the raw input data.
As such, the natural language generation system 106 is configured to instantiate a plurality of messages based on the one or more data feeds. In order to determine the one or more messages, the importance level of each of the messages and relationships between the messages, the natural language generation system 106 may be configured to access the domain model 114 directly or indirectly via the data analysis system 104 or the like. The domain model 114 may contain information related to a particular domain or industry. In some examples, the domain model 114 may provide importance levels, single data feed limits related to normal behaviors in a domain (e.g. normal ranges), information related to anomalous behaviors and/or the like. In other examples, the domain model 114 may describe relationships between various events and/or phenomena in multiple data feeds. For example in a weather domain, a domain model may indicate or otherwise instantiate an extreme weather message in an instance in which wind speeds that are related to hurricane type events or temperatures that may cause harm to humans or other animals or may cause damage or interference to shipping are present in the data. The extreme weather message may then be labeled as important, whereas typical temperatures or a typical wind message may not be marked as important in some examples. Alternatively or additionally, the domain model 114 may be configured to contain or otherwise have access to the diagnostic model.
In some example embodiments, the natural language generation system 106 may be configured to annotate messages with an indication of their relative importance; this information can be used in subsequent processing steps or by the natural language generation system 106 to make decisions about which information should be conveyed and which information may be suppressed, such as by using the domain model 114. The natural language generation system 106 may assign an importance level to the one or more messages based on the pattern itself (e.g. magnitude, duration, rate of change or the like), defined constraints (e.g. defined thresholds, constraints or tolerances), temporal relationships between the pattern in the primary data feed and patterns in other related data feeds and/or the like. For example, a heart rate over 170 beats per minute, or 100 mile per hour winds, may be assigned a high level of importance. In some examples, messages that describe other patterns and/or constraints may be defined by the domain model 114. Alternatively or additionally, the natural language generation system 106 may also be configured to annotate messages with information about how they are related to each other; for example, the natural language generation system 106 might indicate that an event described in one message is assumed to have been caused by the event described in another message.
Using the importance level, the natural language generation system 106 may assign certain ones of the messages that describe or are otherwise are instantiated with patterns or other data in the primary data feed as including key events. A key event may be selected or otherwise identified based on a pre-determined importance level threshold, such as a threshold defined by a user, a constraint defined by the domain model 114, or the like. Alternatively or additionally, key events may be selected or otherwise identified based on those patterns in the primary data feed with the highest level of importance, those patterns that exceed or otherwise satisfy the pre-determined importance level threshold and/or the like. For example, a domain model or user preference may indicate that any messages having wind readings over 50 miles per hour may be designated as key events, whereas in other examples only a message with highest wind reading over a defined time period may be a determined to include a key event. In further examples, the importance level determination may be performed over a plurality of time scales that may be user defined, defined by the domain model or the like (e.g., one hour, one day, one week, one month and/or the like).
In some example embodiments, the natural language generation system 106 may also be configured to determine the importance of messages that describe patterns or events detected in one or more secondary or related data feeds. In some examples, the natural language generation system 106 may determine one or more messages that describe patterns or events in the related data feeds that overlap time-wise or occur within the same time period as the patterns in the primary data feed. For example, during the same time period as rain is detected, another data feed may detect temperature falling below the freezing point. The natural language generation system 106 may then mark the one or more messages that describe patterns or events in the related channels as important, expected, unexpected or as having or not having some other property based on the domain model 114. For example, the domain model may suggest that the one or more patterns in the related data feed were expected to rise as they did in the primary channel. By way of example, as winds are rising, a wave height may then be expected to rise. In other cases, the behavior of the one or more related channels may be unexpected or may be anomalous when compared to the behavior of the primary data feed.
The one or more messages may be marked as including significant events based on the importance level, domain model 114, constraints, user settings or the like. For example, messages that include patterns or events in the related data feed that have an importance level above a predetermined threshold defined by the domain model 114, a user or the like, and may be marked as including significant events. In some example embodiments, messages including unexpected patterns or messages may also be categorized as significant events as they are suggestive of a particular condition or fault. Other messages including patterns or events may be determined to be significant events based on one or more constraints on channel value (e.g. expected range of values or the like), data anomalies, patterns marked as neither expected or unexpected that satisfy an importance level, and/or the like.
In some example embodiments, the natural language generation system 106 may also be configured to determine the importance of messages built or otherwise instantiated using historical data, such as historical data 112, background information, event data, and/or the like. For example, historical data may contain information related to a previous alert condition and the actions taken or a result. Historical data may also provide indicators of the validity of an alert and/or provide additional information that may provide additional situational awareness or may assist diagnosing the fault.
In further example embodiments, the natural language generation system 106 may be configured to generate one or more messages based on determined or otherwise inferred events from the one or more data feeds, historical data, event data and/or the like. Events may include specific activities that may influence the one or more key events and/or may have caused the one or more significant events. In some examples, the one or more events may be inferred based in context with the one or more patterns in the primary and/or related data feeds. Alternatively or additionally events may be provided as a separate channel, such as a contextual data feed, in the raw input data 110, the event log 116 or may be provided directly to the natural language generation system 106. Alternatively or additionally, one or more messages may be generated based on the contextual data feed.
In some examples, the data analysis system 104, the data analysis system 104 or the like may receive raw input data, such as the data in the following table, that illustrates a primary data feed (e.g. heart rate) and a related data feed (e.g. respiration rate):
As is demonstrated by the raw input data in the table above, heart rate went above 115 beats per minute (bpm) at time point 13, thus causing an alert condition. As such, the alert reception system 102, the data analysis system 104 and/or the like may receive an indication of an alarm condition, such as by a patient monitoring system, patient monitoring equipment and/or based on the determination by the data analysis system 104 that the data indicates an alert situation. In response to the alert condition, the data analysis system 104 may cause the heart rate data feed to be the primary data feed. In other embodiments, a user, the domain model or the like may indicate that the primary data feed is the heart rate data feed. In some example embodiments, the data analysis system 104 may abstract or otherwise identify the rapid change of heart rate between time point 10 and time point 11 lasting to time point 15 for use by the natural language generation system 106.
The data analysis system 104 may also determine whether a secondary or related data feed (e.g. respiration rate) has a pattern (e.g. no change when a change is generally expected) in a corresponding time period. In some examples, the corresponding time period may be the same time period or may be a later time period when compared to the time period of the key events. Further, the corresponding time period may, in some examples, be defined by a domain model, such as domain model 114. In some example embodiments, the data analysis system 104 may abstract or otherwise identify the relatively flat and/or steady respiration rate between time point 10 and time point 15 for use by the natural language generation system 106.
In some example embodiments, the natural language generation system 106 is configured to generate one or more messages based on the raw input data in the one or more data feeds. As described herein, messages are language independent data structures that correspond to informational elements in a text and/or collect together underlying data in such a way that the underlying data can be linguistically expressed. Using the heart rate example, a message may include portions of the raw input data, to include abstractions of the data, but may also include additional distinctions necessary for the generation of text as the raw input data is likely to be insufficient for such a purpose. For example, a HeartRateSpike message may be instantiated using the raw input data and such a message may include: a time and relative variation in terms of heart rate change or peak heart rate, a time period and a direction. In some examples, another message may be generated on related channels, historic data, events and/or the like. In some examples, the HeartRateSpike message may be related to an Alert Message that contains information relating to the alert itself. For example, in an instance in which caffeine was applied prior to the heart rate spike, a message may be generated to identify such an event. Such a message may be an Event message that is instantiated with an event time and an event description, such as from the event log 116; for example, a message that indicates that caffeine had been orally administered prior to the spike in heart rate. Other messages such as RespirationRate (e.g. respiration rate stable=yes), HeartRateAlertHistorical (e.g. previous alert condition quantity=2, time=yesterday), HeartRateHistorical (e.g. heart rate trend=no change, time period=10 days) may be instantiated to include information about the related data feeds and/or historical data. Alternatively or additionally, the natural language generation system 106, the document planner 130 and/or the like may be configured to generate the one or more messages.
The document planner 130 is configured to input the one or more messages that are generated and/or instantiated by the natural language generation system 106. The document planner 130 is further configured to determine how to arrange those messages to describe the patterns in the one or more data feeds derived from the raw input data. The document planner 130 may comprise a content determination process that is configured to select the messages, such as based on the decisions of the recommendation and diagnosis processor.
The document planner 130 may also comprise a structuring process that determines the order of messages to be included in a natural language text. In some example embodiments, the document planner 130 may access one or more text schemas for the purposes of content determination and document structuring. A text schema is a rule set that defines the order in which a number of messages are to be presented in a document. For example, an event message (e.g. medication injection) may be described prior to a key event message (e.g. rise in heart rate). In other examples, a significant event message (e.g. falling respiration rate) may be described after, but in relation to, a key event message (e.g. rise in heart rate). By way of further example a document plan may include, but is not limited to, an AlertMessage, a HeartRateSpike message and then a RespirationRate message. An Event message, HeartRateAlertHistorical message and HeartRateHistorical message may then follow in the example document plan.
In some examples, a document plan may take the form of:
Paragraph 1
alert description, including setpoint information and temporal relation to recent restart (if any)
description of previous stable period
related data feeds with similar behavior (aggregated)
related data feeds with different behavior (aggregated)
Recommendation/analysis
Actions
Paragraph 2
Summary of other related operator texts
Most recent related operator text
The output of the document planner 130 may be a tree-structured object or other data structure that is referred to as a document plan. In an instance in which a tree-structured object is chosen for the document plan, the leaf nodes of the tree may contain the messages, and the intermediate nodes of the tree structure object may be configured to indicate how the subordinate nodes are related (e.g. elaboration, consequence, contrast and/or the like) to each other. A sample document plan may include, but is not limited to, document plan 250 of
In some example embodiments, the microplanner 132 is configured to modify a document plan, to create a text specification for input into a realizer. As is shown in some examples, a document plan may contain one or more leaf nodes that contain messages. An example message may comprise a plurality of slots that contain a named attribute and a value (e.g. channel and “HeartRate”). A message may also comprise slots that contain a named attribute and a set of named attributes and their values. Other messages may include additional named attributes and values.
Initially and in some example embodiments, the text specification may include a tree structure that matches or is otherwise structured in the same or similar manner as a document plan tree. In some examples, one or more messages may be combined (e.g. one or more document plan nodes) to form a single phrase specification (e.g. to form a single text specification node). Each leaf node of a text specification may include a phrase specification with one or more empty elements. The microplanner 132 may be configured to populate those element values by applying genre parameters, lexicalization rules, reference rules, aggregation rules and the like.
In some example embodiments, the microplanner 132 may be configured to input a series of genre parameters that are representative of genre conventions. Genre conventions are rules about the use of language which apply throughout texts in that particular genre. In some examples, however, the rules may be overridden by a user, by lexicalization rules and/or the like. The genre conventions specify default behavior for the realizer so that these aspects of language use do not have to continually re-specified by a user. Examples of genre parameters include, but are not limited to, the particular tense (e.g. past, present or future) that should be used consistently throughout the text to be generated; a convention on the use of pronouns in the text to be generated; and/or a convention as to whether or not abbreviated names are to be used in the text to be generated. Alternatively or additionally, other elements of the phrase specification may be set by the one or more genre conventions.
Genre conventions may be applied by the microplanner 132 as a first step in the initialization of the phrase specification that corresponds to an individual message. In such a case, subsequently applied lexicalization rules may override the results of application of the genre parameters. Alternatively or additionally, genre parameters may be applied by the microplanner 132 once all the lexicalization rules have been applied to a given message. In such a case, the genre parameters are configured to populate the elements of the phrase specification that have not been specified or otherwise populated by the lexicalization rules. For example, a tense equal to past, may be set by genre parameter and/or a lexicalization rule.
In additional example embodiments, one or more lexicalization rules may be input. Lexicalization rules are rules that determine how the content of individual messages may be mapped into phrase specifications. In some examples, lexicalization rules may include, but are not limited to, message-level rules that are configured to apply to messages as a whole. Lexicalization rules may also be configured to apply to one or more slots within each message. For example, message-level rules may specify how the overall form of a phrase is to be constructed from the contents of a message (e.g. heart rate is rising, falling or staying steady). Slot-level rules may specify how specific kinds of entities that are present in a message should be described (e.g. heart rate is expressed via a prepositional phrase such as “to 118 bpm”) or otherwise referred to (e.g. refer to a machine by its machine ID or full machine title). For example a message-level rule may map a name value and high rate value from a message to a phrase specification.
For a given domain, there may be at least one message-level lexicalization rule for each type of message in the ontology for that domain that may be applied b. The one or more lexicalization rules for a message type define one or more constraints that are configured to test the message itself, the discourse model (e.g. a model that is configured to store the relevant aspects of the discourse context, such as a list of entities mentioned in the text so far, and the lexicalization of the previous sentence in a text), parameters set by the document planner 130 and/or the genre parameters. In an instance in which the one or more lexicalization rules matches the constraints, a default lexicalization rule may be defined for each message type and/or slot type.
In one example, a message-level rule may be configured to specify a canned text string to be used whenever a message of the specified type is received as input. For example, a GREETING message might result in the simple text string “Hello friend”. Message-level lexicalization rules may also be configured to assign the contents of the slots of a message to particular syntactic constituents (e.g. a word or group of words that function as a single unit, such as a noun phrase, a verb phrase, a prepositional phrase or the like, within a hierarchical structure) in a sentence as represented by a phrase specification. For example, a lexicalization rule, or the one or more lexicalization rules, may be configured to specify the verb to be used to express a particular type of message, and slots in the message might be assigned to the subject and object positions in the sentence. In some examples, a user may allocate information in the one or more slots of a message to the elements of a phrase specification by using the following non-exhaustive list of syntactic constituents, subject: typically the first position in the sentence; verb: the main action described in the sentence; object: typically the position following the verb; indirectobject: used in those cases where a verb has three arguments, as in “John gave the cat a bath”; frontmodifier: used to provide information that will be placed at the beginning of the sentence, as in “yesterday, John gave the cat a bath”; premodifier: used to provide information that will be placed immediately in front of the verb, as in “John reluctantly gave the cat a bath”; postmodifier: used to provide information that will be placed immediately after the object, as in “John took a bus to the city” and/or the like.
Alternatively or additionally, a slot-level rule may be configured to specify a canned text string when a slot of a specified type is received and/or specify a slot to be mapped to a particular syntactic constituent in a sentence as represented by a phrase specification.
Alternatively or additionally, a message-level rule may also specify particular syntactic features of the sentence to be generated, such as by overriding default values for those features either as provided by the realizer itself or by the genre parameters. Typical features include but are not limited to tense, which may be set to PAST, PRESENT or FUTURE; aspect, which may be set to PERFECTIVE or PROGRESSIVE; passive, which may be set to either TRUE or FALSE; negation and/or the like. In some example embodiments, a slot-level rule may specify a particular feature of a sentence to be generated, such as by overriding a default value. Alternatively or additionally, tense and aspect may be computed, such as by using a Reichenbachian model which is based on the time of the message (e.g. when the event described by the message happened), the time the text is generated, and/or a reference time. In some examples, reference time can be computed using one or more of the following non-exhaustive list: setting a reference time to the time of the previous message in the text specification, setting the reference time as the time of the first message expressed in a current paragraph and/or the like.
In some example embodiments, the microplanner may also apply slot-level rules. Slot-level rules may be applied to each slot in each message to enable the slot to be mapped to an element of a phrase specification. In some example embodiments, the message-level rules described herein may also be expressed as slot-level rules, allowing recursive embedding. However, in some examples the value of the slot itself may be used to fill corresponding element in a phrase specification.
In some examples, the microplanner is configured to determine whether two or more phrase specifications can be combined together linguistically to produce a more complex sentence. For example, one or more other phrase specifications may be combined with phrase specification to form a more complex sentence. In some examples, a reference system is configured to determine how to refer to an entity so that it can be unambiguously identified by the reader. For example, in a first sentence “John Smith” may be used where “he” or “his” may be used in subsequent sentences.
Alternatively or additionally, a slot-level rule may be executed. In such cases, the slot-level rule may specify how the value of the slot should be described based on the reference rules. Possible reference rules include, but are not limited to, StringValue: indicating that a string value associated with the object should be used to refer to the object; Named Entity: indicating that a predefined reference strategy for named entities should be used to refer to the object and may include the choice between a full name or description, a reduced form of description, or a pronoun, on the basis of information about the other entities that have been referred to in the text; NumericValue: indicating that a predefined strategy for referring to numeric values should be used; TimeValue: indicates that a predefined reference strategy for referring to time values should be used to refer to the object; DurationValue: indicating that a predefined reference strategy for referring to durations should be used to refer to the object; EnumValue: indicating how specific values of an enumerated type should be expressed and/or the like.
In some example embodiments, the microplanner may also use a slot-level rule to specify content for each of a number of syntactic constituents within a linguistic element that is to be realized as a noun phrase. For example, the following non-exhaustive example list of positions may be available: determiner, specifier, noun, modifier, premodifier, postmodifier and/or the like. In some examples, a slot-level rule may also contain conditions that determine its applicability; amongst other things, these may be used to determine when the rule should have a null output, resulting in the constituent being elided in the sentence being planned.
In some example embodiments, the microplanner may also use one or more slot-level rules to specify syntactic features. For example, a slot level rule may specify the following non-exhaustive example list of syntactic features: a pronominal (e.g. force a use of a pronoun), number (e.g. singular or plural), an indication of definite or indefinite and/or the like.
The output of the microplanner 132, in some example embodiments, is a tree-structured text specification whose leaf-nodes are phrase specifications, and whose internal nodes express rhetorical relations between the leaf nodes. A tree-structured text specification may include, but is not limited to text specification 260 of
A realizer 134 is configured to traverse the tree-structured text specification to express the tree-structured text specification in natural language. The realization process that is applied to each phrase specification in a text specification makes use of a grammar which specifies the valid syntactic structures in the language and further provides a way of mapping from text specifications into the corresponding natural language sentences. The output of the process is, in some example embodiments, a well-formed natural language text. In some examples, the natural language text may include embedded mark-up. The output of the realizer 134, in some example embodiments, is the operator text. The realizer may also output situational analysis text or a narrative that is configured to describe or otherwise summarize the one or more key events, the one or more significant events, the one or more contextual data feed s, and/or the one or more events.
By way of example, the realizer may output the following text in response to the text specification shown above:
Alternatively or additionally and by way of further example, an operator text may take the form of:
Impact:
Reason for Escalation:
Diagnosis:
Situational Analysis:
Alternatively or additionally, the natural language generation system 106 may be configured to generate a graph to display one or more key events that are detected in a data feed. In some example embodiments, the graph may also include one or more significant events in one or more related feeds and/or events. In further examples, a time period or duration of the data shown in the graph may be selected such that the displayed graph illustrates the portion of the data feed that contains the one or more key events. The output graph is further configured to include textual annotations that provide a textual comment, phrase or otherwise is configured to explain, using text, the one or more key events, the one or more significant events and/or the events in a contextual data feed in natural language. In further examples, the textual annotations are generated from the raw input data and further are designed, in some examples, to textually describe identified patterns, anomalies and/or the context of the graph. In some examples, a narrative (e.g. situational analysis text) may be included with the graph that provides situational awareness or an overview of the data/patterns displayed on and/or off of the graph.
Alternatively or additionally, the environment 100, may be configured to generate different types of texts based on the same one or more data feeds in some examples, operator texts or situational awareness texts for engineers may be generated, business level texts may be generated for executives or the like. In some examples, a situational analysis text may include, but is not limited to: an example paragraph containing one or more of an alert information message (e.g. alert name, unit, time), an intermittent alert message in an instance in which an alert has been on intermittently (e.g. start time), and/or an intermittent validation summary message in an instance in which the alert has been on intermittently and alert validation has been completed, such as by the data analysis system 104 (e.g. number of times validated, number of times not validated). An example paragraph relating to the behavior of the sensor via the raw input data. This paragraph may include but is not limited to messages related to data problems if there are data problems, messages related to primary trends in the form of key events and/or a machine status message (e.g. on/off, on duration, multiple starts or the like). In an instance in which the document planner 130 is configured to provide a recommendation, one or more recommendation messages and/or explanations may be included by the document planner 130.
Alternatively or additionally, the situational analysis text may include one or more of a concurrent alerts summary, in an instance in which there are similar alerts on related units; a similar alert summary; active maintenance request messages in an instance in which there are active maintenance requests for this alert condition on the machine (e.g. number, status) and related maintenance request messages in an instance in which there are active maintenance requests for this alert condition on related units; a data problems message and/or previous history messages.
In some example embodiments, the situational analysis text may include, but is not limited to, messages describing a key event in the primary data feed, key event of a paired sensor (e.g. sensors monitoring the same component), messages related to a machine startup in an instance in which the key event is related to the start of a machine; messages related to related data feeds with an importance level exceeding a threshold (e.g. significant messages) sorted based on direction, paired sensors, sensor type or the like; messages related to unexpected behavior (e.g. significant events) also based on direction, paired sensors, sensor type or the like; and/or other messages related to events (e.g. an event summary), related feeds, historical information and/or the like. A situational analysis text may further include messages as described herein for each time duration included.
In further example embodiments, a situational analysis text and/or an operator text may include a recommendation with a supporting explanation. For example, a recommendation may include a recommendation to create a maintenance service on a particular piece of machinery, and the supporting explanation may include, but is not limited to, describing via text that there are no current maintenance requests for a particular machine as well as describing that previous alert conditions similar to a current alert condition resulted in a need for a maintenance request. Other recommendations, such as future actions, recommended investigations and/or the like may be provided via text with an explanation or data to support or otherwise provide context for the recommendation. Alternatively or additionally, the recommendation may take the form of a diagnosis of the alert condition. In such cases, a situational analysis text may include the diagnosis, based on a diagnostic model, one or more possible causes and/or an explanation of the diagnosis.
In some example embodiments, the NLG system that is described herein is an advanced computational linguistics application into which, in some examples, knowledge and expertise of experienced experts, machine learning models, analysis models and the like is permanently and continuously captured. In further examples, the NLG system described herein has the ability to continuously ingest and analyze current and historical data inputs from multiple, complex systems and technology platforms from any point in the service delivery stack (e.g., sensor data historian, alert monitoring system, equipment & alert metadata, engineer maintenance history, operator logs, diagnostic business rules and/or the like). In further still examples, and, from such inputs, write expert plain language (e.g., English narratives that provide instant support to operational decision making processes.
In some examples, the NLG system described herein is configured to input structured and unstructured data and information from multiple disparate sources and systems. The NLG system, in some examples, is then configured to replicate the thought processes of senior engineers to generate the highest priority information to be reported based on the historical context. The NLG system, in some examples, automatically writes rich and technically relevant, plain language summary reports that a reader would believe were written by an experienced engineer, operator, expert, analyst or the like based on what is important within the time period.
By way of example, the NLG system described herein, in some examples, is configured to generate and/or otherwise write reports that: speak in a defined “voice” in a consistent manner tailored to different audiences; is programmable for all languages, is exportable to any portal (e.g. user interface, communication medium or the like) and may be run based on alerts, periodic requests and or the like.
In some examples, an NLG system, such as the one described herein, may include input from various diverse data sources (e.g., sensor data historian, alert monitoring system, equipment and alert metadata, engineer maintenance history, operator logs, diagnostic business rules and/or the like). The example NLG engine may perform data analytics (e.g., such as based on domain knowledge, subject matter expertise, reasoning rules, interpretation rules and/or the like) to generate a message pool (e.g. atoms of informative content that can be used to tell a story). An example document planner may select and/or organize the various messages in the message pool (e.g., such as based on an importance assessment, a relevance assessment, genre conventions, story-telling rules and/or the like). The overall document plan may take the form of a contextually appropriate plan for the overall structure of the text. In some examples a microplanner (e.g., aggregation rules, semantic rules, reference conventions, lexical rules and/or the like) generates sentence plans (e.g., specifications for how information should be packaged into individual sentence). A linguistic realizer (e.g., syntax rules, morphological rules, formatting rules, orthographic rules and/or the like) may be used to generate the NLG output (e.g. decision support narratives, instant support for human decisions and actions, alert driven, periodic, operations, engineering and/or the like).
In some examples, the NLG system described herein is configured to analyze and prioritizes alerts generated by a monitoring system; combine pattern analysis, knowledge based reasoning and natural language processing; determine patterns in data, identify which of the patterns are most important and then analyzes what these patterns tell about the current operational situation. In some examples, the system may then generate expert, accurate, real time engineering and operational decision support recommendations and delivers them to onshore and offshore engineers and/or operators to action; may support day to day engineering and operational performance improvement and/or may, in some example, reduce the decision support process from hours to seconds.
In the example embodiment shown, computing system 300 comprises a computer memory (“memory”) 301, a display 302, one or more processors 303, input/output devices 304 (e.g., keyboard, mouse, CRT or LCD display, touch screen, gesture sensing device and/or the like), other computer-readable media 305, and communications interface 306. The processor 303 may, for example, be embodied as various means including one or more microprocessors with accompanying digital signal processor(s), one or more processor(s) without an accompanying digital signal processor, one or more coprocessors, one or more multi-core processors, one or more controllers, processing circuitry, one or more computers, various other processing elements including integrated circuits such as, for example, an application-specific integrated circuit (ASIC) or field-programmable gate array (FPGA), or some combination thereof. Accordingly, although illustrated in
The alert reception system 102, the data analysis system 104 and/or the natural language generation system 106 are shown residing in memory 301. The memory 301 may comprise, for example, transitory and/or non-transitory memory, such as volatile memory, non-volatile memory, or some combination thereof. Although illustrated in
In other embodiments, some portion of the contents, some or all of the components of the alert reception system 102, the data analysis system 104 and/or the natural language generation system 106 may be stored on and/or transmitted over the other computer-readable media 305. The components of the alert reception system 102, the data analysis system 104 and/or the natural language generation system 106 preferably execute on one or more processors 303 and are configured to generate operator texts, as described herein.
Alternatively or additionally, other code or programs 330 (e.g., an administrative interface, a Web server, and the like) and potentially other data repositories, such as data repository 340, also reside in the memory 301, and preferably execute on one or more processors 303. Of note, one or more of the components in
The alert reception system 102, the data analysis system 104 and/or the natural language generation system 106 are further configured to provide functions such as those described with reference to
In an example embodiment, components/modules of the alert reception system 102, the data analysis system 104 and/or the natural language generation system 106 are implemented using standard programming techniques. For example, the alert reception system 102, the data analysis system 104 and/or the natural language generation system 106 may be implemented as a “native” executable running on the processor 303, along with one or more static or dynamic libraries. In other embodiments, the alert reception system 102, the data analysis system 104 and/or the natural language generation system 106 may be implemented as instructions processed by a virtual machine that executes as one of the other programs 330. In general, a range of programming languages known in the art may be employed for implementing such example embodiments, including representative implementations of various programming language paradigms, including but not limited to, object-oriented (e.g., Java, C++, C#, Visual Basic.NET, Smalltalk, and the like), functional (e.g., ML, Lisp, Scheme, and the like), procedural (e.g., C, Pascal, Ada, Modula, and the like), scripting (e.g., Perl, Ruby, Python, JavaScript, VBScript, and the like), and declarative (e.g., SQL, Prolog, and the like).
The embodiments described above may also use synchronous or asynchronous client-server computing techniques. Also, the various components may be implemented using more monolithic programming techniques, for example, as an executable running on a single processor computer system, or alternatively decomposed using a variety of structuring techniques, including but not limited to, multiprogramming, multithreading, client-server, or peer-to-peer, running on one or more computer systems each having one or more processors. Some embodiments may execute concurrently and asynchronously, and communicate using message passing techniques. Equivalent synchronous embodiments are also supported. Also, other functions could be implemented and/or performed by each component/module, and in different orders, and by different components/modules, yet still achieve the described functions.
In addition, programming interfaces to the data stored as part of the alert reception system 102, the data analysis system 104 and/or the natural language generation system 106, such as by using one or more application programming interfaces can be made available by mechanisms such as through application programming interfaces (API) (e.g. C, C++, C#, and Java); libraries for accessing files, databases, or other data repositories; through scripting languages such as XML; or through Web servers, FTP servers, or other types of servers providing access to stored data. The raw input data 110, historical data 112, the domain model 114 and/or the event log 116 may be implemented as one or more database systems, file systems, or any other technique for storing such information, or any combination of the above, including implementations using distributed computing techniques. Alternatively or additionally, the raw input data 110, historical data 112, the domain model 114 and/or the event log 116 may be local data stores but may also be configured to access data from the remote data sources/alert systems 356.
Different configurations and locations of programs and data are contemplated for use with techniques described herein. A variety of distributed computing techniques are appropriate for implementing the components of the illustrated embodiments in a distributed manner including but not limited to TCP/IP sockets, RPC, RMI, HTTP, Web Services (XML-RPC, JAX-RPC, SOAP, and the like). Other variations are possible. Also, other functionality could be provided by each component/module, or existing functionality could be distributed amongst the components/modules in different ways, yet still achieve the functions described herein.
Furthermore, in some embodiments, some or all of the components of the alert reception system 102, the data analysis system 104 and/or the natural language generation system 106 may be implemented or provided in other manners, such as at least partially in firmware and/or hardware, including, but not limited to one or more ASICs, standard integrated circuits, controllers executing appropriate instructions, and including microcontrollers and/or embedded controllers, FPGAs, complex programmable logic devices (“CPLDs”), and the like. Some or all of the system components and/or data structures may also be stored as contents (e.g., as executable or other machine-readable software instructions or structured data) on a computer-readable medium so as to enable or configure the computer-readable medium and/or one or more associated computing systems or devices to execute or otherwise use or provide the contents to perform at least some of the described techniques. Some or all of the system components and data structures may also be stored as data signals (e.g., by being encoded as part of a carrier wave or included as part of an analog or digital propagated signal) on a variety of computer-readable transmission mediums, which are then transmitted, including across wireless-based and wired/cable-based mediums, and may take a variety of forms (e.g., as part of a single or multiplexed analog signal, or as multiple discrete digital packets or frames). Such computer program products may also take other forms in other embodiments. Accordingly, embodiments of this disclosure may be practiced with other computer system configurations.
Accordingly, blocks of the flowchart support combinations of means for performing the specified functions and combinations of operations for performing the specified functions. It will also be understood that one or more blocks of the flowcharts', and combinations of blocks in the flowchart, can be implemented by special purpose hardware-based computer systems which perform the specified functions, or combinations of special purpose hardware and computer instructions.
In some example embodiments, certain ones of the operations herein may be modified or further amplified as described below. Moreover, in some embodiments additional optional operations may also be included (some examples of which are shown in dashed lines in
Many modifications and other embodiments of the inventions set forth herein will come to mind to one skilled in the art to which these inventions pertain having the benefit of the teachings presented in the foregoing descriptions and the associated drawings. Therefore, it is to be understood that the inventions are not to be limited to the specific embodiments disclosed and that modifications and other embodiments are intended to be included within the scope of the appended claims. Moreover, although the foregoing descriptions and the associated drawings describe example embodiments in the context of certain example combinations of elements and/or functions, it should be appreciated that different combinations of elements and/or functions may be provided by alternative embodiments without departing from the scope of the appended claims. In this regard, for example, different combinations of elements and/or functions than those explicitly described above are also contemplated as may be set forth in some of the appended claims. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.
The present application is a continuation of U.S. patent application Ser. No. 14/914,461, titled “TEXT GENERATION FROM CORRELATED ALERTS,” filed Feb. 25, 2016, which was a national stage entry of International Application No. PCT/IB2013/058131, titled “TEXT GENERATION FROM CORRELATED ALERTS,” filed Aug. 29, 2013, the contents of which are incorporated herein by reference in their entirety.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5181250 | Morgan et al. | Jan 1993 | A |
| 5237502 | White et al. | Aug 1993 | A |
| 5311429 | Tominaga | May 1994 | A |
| 5321608 | Namba et al. | Jun 1994 | A |
| 5629687 | Sutton et al. | May 1997 | A |
| 5794177 | Carus et al. | Aug 1998 | A |
| 5802488 | Edatsune | Sep 1998 | A |
| 6023669 | Suda et al. | Feb 2000 | A |
| 6078914 | Redfern | Jun 2000 | A |
| 6138087 | Budzinski | Oct 2000 | A |
| 6266617 | Evans | Jul 2001 | B1 |
| 6374224 | Horiguchi et al. | Apr 2002 | B1 |
| 6442485 | Evans | Aug 2002 | B2 |
| 6466899 | Yano et al. | Oct 2002 | B1 |
| 6629340 | Dale et al. | Oct 2003 | B1 |
| 6665640 | Bennett et al. | Dec 2003 | B1 |
| 6668281 | Ayyadurai | Dec 2003 | B1 |
| 6717513 | Sandelman et al. | Apr 2004 | B1 |
| 6947885 | Bangalore et al. | Sep 2005 | B2 |
| 6958746 | Anderson et al. | Oct 2005 | B1 |
| 7043420 | Ratnaparkhi | May 2006 | B2 |
| 7062483 | Ferrari et al. | Jun 2006 | B2 |
| 7111018 | Goodrich et al. | Sep 2006 | B1 |
| 7167824 | Kallulli | Jan 2007 | B2 |
| 7231341 | Bangalore et al. | Jun 2007 | B2 |
| 7238313 | Ferencz et al. | Jul 2007 | B2 |
| 7305336 | Polanyi et al. | Dec 2007 | B2 |
| 7310969 | Dale | Dec 2007 | B2 |
| 7346493 | Ringger et al. | Mar 2008 | B2 |
| 7418447 | Caldwell et al. | Aug 2008 | B2 |
| 7424363 | Cheng et al. | Sep 2008 | B2 |
| 7444287 | Claudatos et al. | Oct 2008 | B2 |
| 7490042 | Eide et al. | Feb 2009 | B2 |
| 7493253 | Ceusters | Feb 2009 | B1 |
| 7496621 | Pan et al. | Feb 2009 | B2 |
| 7526424 | Corston-Oliver et al. | Apr 2009 | B2 |
| 7533089 | Pan et al. | May 2009 | B2 |
| 7562005 | Bangalore et al. | Jul 2009 | B1 |
| 7653545 | Starkie | Jan 2010 | B1 |
| 7657424 | Bennett | Feb 2010 | B2 |
| 7684991 | Stohr et al. | Mar 2010 | B2 |
| 7711581 | Hood et al. | May 2010 | B2 |
| 7783486 | Rosser et al. | Aug 2010 | B2 |
| 7809552 | Pan et al. | Oct 2010 | B2 |
| 7849048 | Langseth et al. | Dec 2010 | B2 |
| 7849049 | Langseth et al. | Dec 2010 | B2 |
| 7856390 | Schiller | Dec 2010 | B2 |
| 7873509 | Budzinski | Jan 2011 | B1 |
| 7921091 | Cox et al. | Apr 2011 | B2 |
| 7930169 | Billerey-Mosier | Apr 2011 | B2 |
| 7933774 | Begeja et al. | Apr 2011 | B1 |
| 7966172 | Ruiz et al. | Jun 2011 | B2 |
| 7966369 | Briere et al. | Jun 2011 | B1 |
| 7970601 | Burmester et al. | Jun 2011 | B2 |
| 7979267 | Ruiz et al. | Jul 2011 | B2 |
| 8015006 | Kennewick et al. | Sep 2011 | B2 |
| 8019610 | Walker et al. | Sep 2011 | B2 |
| 8024331 | Calistri-Yeh et al. | Sep 2011 | B2 |
| 8037000 | Delmonico et al. | Oct 2011 | B2 |
| 8082144 | Brown et al. | Dec 2011 | B1 |
| 8090727 | Lachtarnik et al. | Jan 2012 | B2 |
| 8117261 | Briere et al. | Feb 2012 | B2 |
| 8150676 | Kaeser | Apr 2012 | B1 |
| 8175873 | Di Fabbrizio et al. | May 2012 | B2 |
| 8180647 | Walker et al. | May 2012 | B2 |
| 8180758 | Cornali | May 2012 | B1 |
| 8204751 | Di Fabbrizio et al. | Jun 2012 | B1 |
| 8229937 | Kiefer et al. | Jul 2012 | B2 |
| 8355903 | Birnbaum et al. | Jan 2013 | B1 |
| 8374848 | Birnbaum et al. | Feb 2013 | B1 |
| 8425325 | Hope | Apr 2013 | B2 |
| 8457950 | Gardner | Jun 2013 | B1 |
| 8473911 | Baxter | Jun 2013 | B1 |
| 8494944 | Schiller | Jul 2013 | B2 |
| 8495675 | Philopott et al. | Jul 2013 | B1 |
| 8515733 | Jansen | Aug 2013 | B2 |
| 8515737 | Allen | Aug 2013 | B2 |
| 8521512 | Gorman et al. | Aug 2013 | B2 |
| 8548814 | Manuel-Devadoss | Oct 2013 | B2 |
| 8548915 | Antebi et al. | Oct 2013 | B2 |
| 8561014 | Mengusoglu et al. | Oct 2013 | B2 |
| 8566090 | Di Fabbrizio et al. | Oct 2013 | B2 |
| 8572173 | Briere et al. | Oct 2013 | B2 |
| 8589148 | Atallah et al. | Nov 2013 | B2 |
| 8589172 | Alonso et al. | Nov 2013 | B2 |
| 8616896 | Lennox | Dec 2013 | B2 |
| 8620669 | Walker et al. | Dec 2013 | B2 |
| 8626613 | Dale et al. | Jan 2014 | B2 |
| 8630844 | Nichols et al. | Jan 2014 | B1 |
| 8655889 | Hua et al. | Feb 2014 | B2 |
| 8660545 | Redford et al. | Feb 2014 | B1 |
| 8676691 | Schiller | Mar 2014 | B2 |
| 8688434 | Birnbaum et al. | Apr 2014 | B1 |
| 8689176 | Bagheri et al. | Apr 2014 | B2 |
| 8700396 | Mengibar et al. | Apr 2014 | B1 |
| 8711732 | Johnson | Apr 2014 | B2 |
| 8719696 | Duncan et al. | May 2014 | B2 |
| 8738384 | Bansal et al. | May 2014 | B1 |
| 8738558 | Antebi et al. | May 2014 | B2 |
| 8762134 | Reiter | May 2014 | B2 |
| 8762133 | Reiter | Jun 2014 | B2 |
| 8775161 | Nichols | Jul 2014 | B1 |
| 8825533 | Basson et al. | Sep 2014 | B2 |
| 8843363 | Birnbaum et al. | Sep 2014 | B2 |
| 8849670 | Di Cristo et al. | Sep 2014 | B2 |
| 8886520 | Nichols et al. | Nov 2014 | B1 |
| 8892417 | Nichols et al. | Nov 2014 | B1 |
| 8892419 | Lundberg et al. | Nov 2014 | B2 |
| 8898063 | Sykes et al. | Nov 2014 | B1 |
| 8903711 | Lundberg et al. | Dec 2014 | B2 |
| 8903718 | Akuwudike | Dec 2014 | B2 |
| 8909595 | Gandy et al. | Dec 2014 | B2 |
| 8914452 | Boston et al. | Dec 2014 | B2 |
| 8924330 | Antebi et al. | Dec 2014 | B2 |
| 8930178 | Pestian et al. | Jan 2015 | B2 |
| 8930305 | Namburu et al. | Jan 2015 | B2 |
| 8935769 | Hessler | Jan 2015 | B2 |
| 8977953 | Pierre et al. | Mar 2015 | B1 |
| 8984051 | Olsen et al. | Mar 2015 | B2 |
| 9002695 | Watanabe et al. | Apr 2015 | B2 |
| 9002869 | Riezler et al. | Apr 2015 | B2 |
| 9015730 | Allen et al. | Apr 2015 | B1 |
| 9028260 | Nanjiani et al. | May 2015 | B2 |
| 9092276 | Lallen et al. | Jul 2015 | B2 |
| 9104720 | Rakshit et al. | Aug 2015 | B2 |
| 9110882 | Overell et al. | Aug 2015 | B2 |
| 9110977 | Pierre et al. | Aug 2015 | B1 |
| 9111534 | Sylvester et al. | Aug 2015 | B1 |
| 9135244 | Reiter | Sep 2015 | B2 |
| 9135662 | Evenhouse et al. | Sep 2015 | B2 |
| 9146904 | Allen | Sep 2015 | B2 |
| 9164982 | Kaeser | Oct 2015 | B1 |
| 9173005 | Redford et al. | Oct 2015 | B1 |
| 9190054 | Riley et al. | Nov 2015 | B1 |
| 9208147 | Nichols et al. | Dec 2015 | B1 |
| 9229927 | Wolfram et al. | Jan 2016 | B2 |
| 9240197 | Begeja et al. | Jan 2016 | B2 |
| 9244894 | Dale et al. | Jan 2016 | B1 |
| 9251134 | Birnbaum et al. | Feb 2016 | B2 |
| 9251143 | Bird et al. | Feb 2016 | B2 |
| 9263039 | Di Cristo et al. | Feb 2016 | B2 |
| 9268770 | Kursun | Feb 2016 | B1 |
| 9323743 | Reiter | Apr 2016 | B2 |
| 9336193 | Logan et al. | May 2016 | B2 |
| 9396181 | Sripada | Jul 2016 | B1 |
| 9405448 | Reiter | Aug 2016 | B2 |
| 9600471 | Bradshaw et al. | Mar 2017 | B2 |
| 9640045 | Reiter | May 2017 | B2 |
| 9754051 | Bastide et al. | Sep 2017 | B2 |
| 9904676 | Sripada et al. | Feb 2018 | B2 |
| 9946711 | Reiter | Apr 2018 | B2 |
| 9990360 | Sripada | Jun 2018 | B2 |
| 10026274 | Reiter | Jul 2018 | B2 |
| 20020026306 | Bangalore et al. | Feb 2002 | A1 |
| 20020143742 | Nonomura et al. | Oct 2002 | A1 |
| 20030131315 | Escher | Jul 2003 | A1 |
| 20030182102 | Corston-Oliver et al. | Sep 2003 | A1 |
| 20030195740 | Tokuda et al. | Oct 2003 | A1 |
| 20030212545 | Kallulli | Nov 2003 | A1 |
| 20030229605 | Herrera et al. | Dec 2003 | A1 |
| 20030233230 | Ammicht et al. | Dec 2003 | A1 |
| 20040044515 | Metcalf et al. | Mar 2004 | A1 |
| 20040246120 | Benner | Dec 2004 | A1 |
| 20050015681 | Strole | Jan 2005 | A1 |
| 20050039107 | Hander et al. | Feb 2005 | A1 |
| 20050108001 | Aarskog | May 2005 | A1 |
| 20050228635 | Araki et al. | Oct 2005 | A1 |
| 20050256703 | Markel | Nov 2005 | A1 |
| 20060020886 | Agrawal et al. | Jan 2006 | A1 |
| 20060020916 | Allison et al. | Jan 2006 | A1 |
| 20060074597 | Raphael | Apr 2006 | A1 |
| 20060085414 | Chai et al. | Apr 2006 | A1 |
| 20060085667 | Kubota et al. | Apr 2006 | A1 |
| 20060136196 | Brun et al. | Jun 2006 | A1 |
| 20060178868 | Billerey-Mosier | Aug 2006 | A1 |
| 20060184888 | Bala | Aug 2006 | A1 |
| 20060224638 | Wald et al. | Oct 2006 | A1 |
| 20060229872 | Eide et al. | Oct 2006 | A1 |
| 20060242563 | Liu et al. | Oct 2006 | A1 |
| 20060259293 | Orwant | Nov 2006 | A1 |
| 20070038643 | Epstein | Feb 2007 | A1 |
| 20070078655 | Semkow et al. | Apr 2007 | A1 |
| 20070106628 | Adjali et al. | May 2007 | A1 |
| 20070129942 | Ban et al. | Jun 2007 | A1 |
| 20070136301 | Jardin | Jun 2007 | A1 |
| 20070143099 | Balchandran et al. | Jun 2007 | A1 |
| 20070143278 | Srivastava et al. | Jun 2007 | A1 |
| 20070150806 | Hartmann | Jun 2007 | A1 |
| 20070156677 | Szabo | Jul 2007 | A1 |
| 20070169021 | Huynh et al. | Jul 2007 | A1 |
| 20070219773 | Roux et al. | Sep 2007 | A1 |
| 20070287929 | Goedje | Dec 2007 | A1 |
| 20080005005 | Billieux | Jan 2008 | A1 |
| 20080077653 | Morris | Mar 2008 | A1 |
| 20080221865 | Wellmann | Sep 2008 | A1 |
| 20080221870 | Attardi et al. | Sep 2008 | A1 |
| 20080281781 | Zhao et al. | Nov 2008 | A1 |
| 20080312954 | Ullrich et al. | Dec 2008 | A1 |
| 20090076799 | Crouch et al. | Mar 2009 | A1 |
| 20090089100 | Nenov et al. | Apr 2009 | A1 |
| 20090089126 | Odubiyi | Apr 2009 | A1 |
| 20090111486 | Burstrom | Apr 2009 | A1 |
| 20090138258 | Neale | May 2009 | A1 |
| 20090144609 | Liang et al. | Jun 2009 | A1 |
| 20090156229 | Hein et al. | Jun 2009 | A1 |
| 20090177929 | Sijelmassi | Jul 2009 | A1 |
| 20090182549 | Anisimovich et al. | Jul 2009 | A1 |
| 20090198496 | Denecke | Aug 2009 | A1 |
| 20090281839 | Lynn et al. | Nov 2009 | A1 |
| 20090286514 | Lichorowic et al. | Nov 2009 | A1 |
| 20090287567 | Penberthy et al. | Nov 2009 | A1 |
| 20090313240 | Gile et al. | Dec 2009 | A1 |
| 20100010802 | Ruano et al. | Jan 2010 | A1 |
| 20100146491 | Hirano et al. | Jun 2010 | A1 |
| 20100153095 | Yang et al. | Jun 2010 | A1 |
| 20100153105 | Di Fabbrizio et al. | Jun 2010 | A1 |
| 20100174545 | Otani | Jul 2010 | A1 |
| 20100191658 | Kannan et al. | Jul 2010 | A1 |
| 20100203970 | Hope | Aug 2010 | A1 |
| 20100210379 | Shelley | Aug 2010 | A1 |
| 20100217737 | Shama | Aug 2010 | A1 |
| 20100241421 | Funakoshi | Sep 2010 | A1 |
| 20100313149 | Zhang et al. | Dec 2010 | A1 |
| 20100325608 | Radigan | Dec 2010 | A1 |
| 20100332235 | David | Dec 2010 | A1 |
| 20110010164 | Williams | Jan 2011 | A1 |
| 20110035210 | Rosenfeld et al. | Feb 2011 | A1 |
| 20110055687 | Bhandar et al. | Mar 2011 | A1 |
| 20110068929 | Franz et al. | Mar 2011 | A1 |
| 20110087486 | Schiller | Apr 2011 | A1 |
| 20110113382 | Cannon et al. | May 2011 | A1 |
| 20110160986 | Wu et al. | Jun 2011 | A1 |
| 20110179006 | Cox et al. | Jul 2011 | A1 |
| 20110184959 | Maxwell, III et al. | Jul 2011 | A1 |
| 20110218822 | Buisman et al. | Sep 2011 | A1 |
| 20110225185 | Gupta | Sep 2011 | A1 |
| 20110257839 | Mukherjee | Oct 2011 | A1 |
| 20110307435 | Overell et al. | Dec 2011 | A1 |
| 20110313757 | Hoover et al. | Dec 2011 | A1 |
| 20110314060 | Sinha et al. | Dec 2011 | A1 |
| 20120004903 | Pillai et al. | Jan 2012 | A1 |
| 20120078888 | Brown et al. | Mar 2012 | A1 |
| 20120084027 | Caine | Apr 2012 | A1 |
| 20120131008 | Ahn et al. | May 2012 | A1 |
| 20120136649 | Freising et al. | May 2012 | A1 |
| 20120158089 | Bocek et al. | Jun 2012 | A1 |
| 20120173475 | Ash et al. | Jul 2012 | A1 |
| 20120174018 | Ash et al. | Jul 2012 | A1 |
| 20120232919 | Wilson et al. | Sep 2012 | A1 |
| 20120290289 | Manera et al. | Nov 2012 | A1 |
| 20120290310 | Watson | Nov 2012 | A1 |
| 20120310990 | Viegas et al. | Dec 2012 | A1 |
| 20130013290 | Funakoshi et al. | Jan 2013 | A1 |
| 20130030810 | Kopparapu et al. | Jan 2013 | A1 |
| 20130041921 | Cooper et al. | Feb 2013 | A1 |
| 20130066873 | Salvetti et al. | Mar 2013 | A1 |
| 20130095864 | Marovets | Apr 2013 | A1 |
| 20130138428 | Chandramouli et al. | May 2013 | A1 |
| 20130144606 | Birnbaum et al. | Jun 2013 | A1 |
| 20130145242 | Birnbaum et al. | Jun 2013 | A1 |
| 20130145255 | Birnbaum et al. | Jun 2013 | A1 |
| 20130151238 | Beaurpere et al. | Jun 2013 | A1 |
| 20130174026 | Locke | Jul 2013 | A1 |
| 20130185050 | Bird et al. | Jul 2013 | A1 |
| 20130185056 | Ingram et al. | Jul 2013 | A1 |
| 20130205195 | Dekhil et al. | Aug 2013 | A1 |
| 20130211855 | Eberle et al. | Aug 2013 | A1 |
| 20130238329 | Casella dos Santos | Sep 2013 | A1 |
| 20130238330 | Casella dos Santos | Sep 2013 | A1 |
| 20130238987 | Lutwyche | Sep 2013 | A1 |
| 20130251233 | Yang et al. | Sep 2013 | A1 |
| 20130268263 | Park et al. | Oct 2013 | A1 |
| 20130293363 | Plymouth et al. | Nov 2013 | A1 |
| 20130311201 | Chatfield et al. | Nov 2013 | A1 |
| 20140019531 | Czajka et al. | Jan 2014 | A1 |
| 20140025371 | Min | Jan 2014 | A1 |
| 20140039878 | Wasson | Feb 2014 | A1 |
| 20140052696 | Soroushian | Feb 2014 | A1 |
| 20140062712 | Reiter | Mar 2014 | A1 |
| 20140067377 | Reiter | Mar 2014 | A1 |
| 20140072947 | Boguraev et al. | Mar 2014 | A1 |
| 20140072948 | Boguraev et al. | Mar 2014 | A1 |
| 20140089212 | Sbodio | Mar 2014 | A1 |
| 20140100846 | Haine | Apr 2014 | A1 |
| 20140100901 | Haine et al. | Apr 2014 | A1 |
| 20140100923 | Strezo et al. | Apr 2014 | A1 |
| 20140143720 | Dimarco et al. | May 2014 | A1 |
| 20140149107 | Schilder | May 2014 | A1 |
| 20140149596 | Emerson, III | May 2014 | A1 |
| 20140164303 | Bagchi et al. | Jun 2014 | A1 |
| 20140164304 | Bagchi et al. | Jun 2014 | A1 |
| 20140188477 | Zhang | Jul 2014 | A1 |
| 20140278358 | Byron et al. | Sep 2014 | A1 |
| 20140281935 | Byron et al. | Sep 2014 | A1 |
| 20140281951 | Megiddo et al. | Sep 2014 | A1 |
| 20140297268 | Govrin et al. | Oct 2014 | A1 |
| 20140316768 | Khandekar | Oct 2014 | A1 |
| 20140328570 | Cheng et al. | Nov 2014 | A1 |
| 20140358964 | Woods et al. | Dec 2014 | A1 |
| 20140375466 | Reiter | Dec 2014 | A1 |
| 20140379322 | Koutrika et al. | Dec 2014 | A1 |
| 20140379378 | Cohen-Solal et al. | Dec 2014 | A1 |
| 20150006437 | Byron et al. | Jan 2015 | A1 |
| 20150032443 | Karov et al. | Jan 2015 | A1 |
| 20150081299 | Jasinschi et al. | Mar 2015 | A1 |
| 20150081307 | Cederstrom et al. | Mar 2015 | A1 |
| 20150081321 | Jain | Mar 2015 | A1 |
| 20150095015 | Lani et al. | Apr 2015 | A1 |
| 20150106307 | Antebi et al. | Apr 2015 | A1 |
| 20150142418 | Byron et al. | May 2015 | A1 |
| 20150142421 | Buurman et al. | May 2015 | A1 |
| 20150154359 | Harris et al. | Jun 2015 | A1 |
| 20150163358 | Klemm et al. | Jun 2015 | A1 |
| 20150169522 | Logan et al. | Jun 2015 | A1 |
| 20150169548 | Reiter | Jun 2015 | A1 |
| 20150169659 | Lee et al. | Jun 2015 | A1 |
| 20150169720 | Byron et al. | Jun 2015 | A1 |
| 20150169737 | Byron et al. | Jun 2015 | A1 |
| 20150179082 | Byron et al. | Jun 2015 | A1 |
| 20150227508 | Howald et al. | Aug 2015 | A1 |
| 20150242384 | Reiter | Aug 2015 | A1 |
| 20150261744 | Suenbuel et al. | Sep 2015 | A1 |
| 20150261836 | Madhani et al. | Sep 2015 | A1 |
| 20150279348 | Cao et al. | Oct 2015 | A1 |
| 20150310013 | Allen et al. | Oct 2015 | A1 |
| 20150310112 | Allen et al. | Oct 2015 | A1 |
| 20150310861 | Waltermann et al. | Oct 2015 | A1 |
| 20150324343 | Carter et al. | Nov 2015 | A1 |
| 20150324347 | Bradshaw et al. | Nov 2015 | A1 |
| 20150324351 | Sripada et al. | Nov 2015 | A1 |
| 20150324374 | Sripada et al. | Nov 2015 | A1 |
| 20150324413 | Gubin et al. | Nov 2015 | A1 |
| 20150325000 | Sripada | Nov 2015 | A1 |
| 20150326622 | Carter et al. | Nov 2015 | A1 |
| 20150331845 | Guggilla et al. | Nov 2015 | A1 |
| 20150331846 | Guggilla et al. | Nov 2015 | A1 |
| 20150332670 | Akbacak et al. | Nov 2015 | A1 |
| 20150347400 | Sripada | Dec 2015 | A1 |
| 20150356127 | Pierre et al. | Dec 2015 | A1 |
| 20150363363 | Bohra et al. | Dec 2015 | A1 |
| 20150363364 | Sripada | Dec 2015 | A1 |
| 20150363382 | Bohra et al. | Dec 2015 | A1 |
| 20150363390 | Mungi et al. | Dec 2015 | A1 |
| 20150363391 | Mungi et al. | Dec 2015 | A1 |
| 20150371651 | Aharoni et al. | Dec 2015 | A1 |
| 20160019200 | Allen | Jan 2016 | A1 |
| 20160027125 | Bryce | Jan 2016 | A1 |
| 20160055150 | Bird et al. | Feb 2016 | A1 |
| 20160132489 | Reiter | May 2016 | A1 |
| 20160140090 | Dale et al. | May 2016 | A1 |
| 20160179755 | Desai et al. | Jun 2016 | A1 |
| 20160232152 | Mahamood | Aug 2016 | A1 |
| 20160328371 | Logan et al. | Nov 2016 | A1 |
| 20160328381 | Reiter | Nov 2016 | A1 |
| 20160328385 | Reiter | Nov 2016 | A1 |
| 20170018107 | Reiter | Jan 2017 | A1 |
| 20170075884 | Sripada et al. | Mar 2017 | A1 |
| 20190035232 | Reiter | Jan 2019 | A1 |
| Number | Date | Country |
|---|---|---|
| 2011247830 | Dec 2011 | AU |
| 2011253627 | Dec 2011 | AU |
| 2013201755 | May 2013 | AU |
| 2013338351 | May 2015 | AU |
| 2577721 | Feb 2006 | CA |
| 2826116 | Mar 2006 | CA |
| 103999081 | Aug 2014 | CN |
| 104182059 | Dec 2014 | CN |
| 104881320 | Feb 2015 | CN |
| 1336955 | May 2006 | EP |
| 2707809 | Mar 2014 | EP |
| 2750759 | Sep 2014 | EP |
| 2849103 | Mar 2015 | EP |
| 2518192 | Mar 2015 | GB |
| 61-221873 | Oct 1986 | JP |
| 2004-21791 | Jan 2004 | JP |
| 2014165766 | Sep 2014 | JP |
| WO 2000074394 | Dec 2000 | WO |
| WO 2002031628 | Apr 2002 | WO |
| WO 2002073449 | Sep 2002 | WO |
| WO 2002073531 | Sep 2002 | WO |
| WO 2002031628 | Oct 2002 | WO |
| WO 2006010044 | Jan 2006 | WO |
| WO 2007041221 | Apr 2007 | WO |
| WO 2009014465 | Jan 2009 | WO |
| WO 2010049925 | May 2010 | WO |
| WO 2010051404 | May 2010 | WO |
| WO 2012071571 | May 2012 | WO |
| WO 2012009613 | Jan 2013 | WO |
| WO 2013042115 | Mar 2013 | WO |
| WO 2013042116 | Mar 2013 | WO |
| WO 2014102568 | Jul 2013 | WO |
| WO 2013177280 | Nov 2013 | WO |
| WO 2014035402 | Mar 2014 | WO |
| WO 2014098560 | Jun 2014 | WO |
| WO 2014140977 | Sep 2014 | WO |
| WO 2014187076 | Nov 2014 | WO |
| WO 2014028844 | Mar 2015 | WO |
| WO 2015113301 | Aug 2015 | WO |
| WO 2015148278 | Oct 2015 | WO |
| WO 2015159133 | Oct 2015 | WO |
| WO 2015164253 | Oct 2015 | WO |
| WO 2015175338 | Nov 2015 | WO |
| WO 2016004266 | Jan 2016 | WO |
| Entry |
|---|
| Portet, Frangois, et al. “Automatic generation of textual summaries from neonatal intensive care data.” Artificial Intelligence 173.7-8 (2009): 789-816. (Year: 2009). |
| Gatt, Albert, et al. “From data to text in the neonatal intensive care unit: Using NLG technology for decision support and information management.” Ai Communications 22.3 (2009): 153-186. (Year: 2009). |
| Yu, Jin, et al. “SumTime-turbine: a knowledge-based system to communicate gas turbine time-series data.” International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems. Springer Berlin Heidelberg, 2003. (Year: 2003). |
| Radev, Dragomir R., and Kathleen R. McKeown. “Generating natural language summaries from multiple on-line sources.” Computational Linguistics 24.3 (1998): 470-500. (Year: 1998). |
| Alawneh et al., “Pattern Recognition Techniques Applied to the Abstraction of Traces of Inter-Process Communication,” Software Maintenance and Reengineering (CSMR), 2011 15th European Conference on Year: 2011, IEEE Conference Publications, pp. 211-220, (2011). |
| Andre et al., “From Visual Data to Multimedia Presentations,” Grounding Representations: Integreation of Sensory Information in Natural Language Processing, Artificial Intelligence and Neural networks, IEE Colloquium On, pp. 1-3, (1995). |
| Andre et al., “Natural Language Access to Visual Data: Dealing with Space and Movement,” Report 63, German Research Center for Artificial Intelligence (DFKI) SFB 314, Project VITRA, pp. 1-21, (1989). |
| Barzilay et al.; “Aggregation via Set Partitioning for Natural Language Generation”, Proceedings of the Human Language Technology Conference of the North American Chapter of the ACL; pp. 359-266; (2006). |
| Bhoedjang et al., “Optimizing Distributed Data Structures Using Application-Specific Network Inteface Software,” Parallel Preocessing, Proceedings; International Conference on Year: 1998, IEEE Conference Publications, pp. 485-492, (1998). |
| Cappozzo et al., “Surface-Maker Cluster Design Criteria for 3-D Bone Movement Reconstruction,” IEEE Transactions on Biomedical Engineering, 44(12):1165-1174, (1997). |
| Chang-Jie et al., “Interactive Location-based Services Combined with Natural Language,” International Conference on Wireless Communications, Networking and Mobile Computing, pp. 3015-3018, (2007). |
| Dalianis et al.; “Aggregation in Natural Language Genration;” Trends in Natural Language Generation, an Artificial Intelligence Perspective, pp. 88-105; (1993). |
| Dragon et al., “Multi-Scale Clustering of Frame-to-Frame Correspondences for Motion Segmentation,” Compute Vision ECCV, Springer Berling Heidelberg, pp. 445-458, (2012). |
| Gatt et al., “From Date To Text In The Neonatal Intensive Care Unit: Using NLG Technology for Decision Support and Information Management,” AI Communications, IOS Press, ISSN 0921-7126, 33 pages, (2009). |
| Gatt et al., “From Data to Tect in the Neonatal intensive Care Unit: Usiing NLG Technology for Decision Support and Information Management, ” AI Communication, pp. 153-186, (2009). |
| Gorelov et al., “Search Optimization in Semiconstructed Databases Using Hierarchy of Document Schemas,” Programmin and Computer Software, 31(6):321-331, (2005). |
| Guoqiang et al., “The Research on Interactive short Message Response,” Workshop on Intelligent Information Technology Application, IEEE Conference Publications pp. 206-209, (2007). |
| Hezog et al., “Combining Alternatives in the Multimedia Presentation of Decision Support Information for Real-Time Control,” IFIP, 15 pages,(1998). |
| Kottke et al., “Motion Estimation Via Cluster Matching,” 8180 IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(11):1128-1132, (1994). |
| Krahmer et al., “Computational Generation of Referring Expressions: A Survey,” In Computational Linguistics, 38:173-218, (2012). |
| Kukich, “Knowledge-Based Report Generation: A Knowledge-Engineering Approach to Natural Language Report Generation,” Dissertation to The Interdisciplinary Department of Information Science, University of Pittsburg, 260 pages, (1983). |
| Leonov et al., “Construction of an Optimal Relational Schema for Storing XML Documents in an RDBMS Without Using DTD/XML Schma,” Programming and Computer Software, 30(6):323-336, (2204). |
| Paraboni, “Generating Referring Expressions: Making Referents Easy to Identify,” In Computational Linguistics, 33(2):229-254, (2207). |
| Paraboni, “Generating refernces in hierarchical domains: the case of Document Deixis,” University of Brighton PhD thesis, pp. 1-207, (2003). |
| Perry et al., “Automatic Realignment of Data Structures to Improve MPI Performance,” Networks (ICN), Ninth International Conference on Year: 2010, IEEE Conference Publications, pp. 42-47, (2010). |
| Portet et al., “Automatic generation of textual summaries from neonatal intensive care data,” Artificial Intelligence, 173:789-816, (2009). [Available online Dec. 25, 2008]. |
| Premchaiswadi et al., “Enhancing Learning Systems by using Virtual Interactive Classrooms and Web-based Collaborative Work,” Education Engineering (EDUCON) IEEE Conference Publications, pp. 1531-1537, (2010). |
| Quinlan, “Induction of Decision Trees, ” Machine Learning, Kluwer Academic Publishers, 1(1):81-106, (1986). |
| Radev et al., “Generating Natural Language Summaries from Multiple On-Line Sources,” Association of Computational Linguistics, 24(3):469-500, (1998). |
| Reiter et al., “Building Applied Natural Languange Generation Systems,” Natural Language Engineering 1 (1), 31 pages, (1995). |
| Reitler et al., “Choosing words in computer-genrated weather forecasts,” Artificial Intelligence, 167:137169, (2005). |
| Reiter et al.; “Studies in Natural Language Processing—Building Natural Language Generation Systems,” Cambridge University Press, (2000). |
| Reiter, “An Architecture for Data-to-Text Systems,” Proceedings of ENLG-2007, pp. 97-104, (2007). |
| Reiter, “Chapter 4: Document Planning (early draft), Building Natural Language Generation Systems,” pp. 73-113, (2005). [Retrieved from the Internet Nov. 2, 2017: <http://ling.helsinki.fi/˜gwilcock/Tartu-2003/ReiterDale/4-DocumentPlanning.pdf>]. |
| Seki, “XML Tranfsormation-based three-stage pipelined Natural Language Generation System,” Proc. of 6th NLP Pacific Rim Symposium (NLPRS 2001), pp. 767-768, (2001). [Retrieved from the Internet Nov. 2, 2017: <http://www.afnlp.org/archives/nlprs2001/pdf/exh-04-01.pdf>]. |
| Shaw, “Clause Aggregation Using Linguistic Knowledge;” Proceedings of the IWNLG, pp. 138-147, (1998). Retrieved from <http://acl.ldc.upenn.edu/W/W98/W98-1415.pdf>. |
| Siddharthan et al., “Generating referring expressions in open domains,” In Proceedings of ACL 2004, pp. 1-8, (2004). |
| Spillner et al., “Algorthims for Dispersed Processing,” Utility and Cloud Computing (UC) 204 IEEE/ACM 7th International Conference on year: 2014, IEEE Conferenced Publications, pp. 914-921, (2014). |
| Takeuchi et al., “Human Prosocial Response to Emotive Facial Expression of Interactive Agent,” The 15th IEEE International Symposium on Robot and Human Interactive Communication, pp. 680-685, (2006). |
| Voelz et al., “Rocco: A RoboCup Soccer Commentator System,” German Research Center for Articficial Intelligence DFKI GmbH, 11 pages, (1999). |
| Wilcock, “An Overview of Shallow XML-Based Natural Language Generation,” Baltic HLT, pp. 67-78, (2005). [Retrieved from the Internet Nov. 2, 2017: <https://www.ling.helsinki.fi/˜gwilcock/Pubs/2005/BalticHLT-05.pdf>]. |
| Yu et al. “SumTime-turbine: a knowledge-based system to communicate gas turbine time-series data.” International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems, 6 pages, (2003). |
| Yu et al., “Choosing the Context of Textual Summaries of Large-Time Data Sets,” Natural Language Engineering, 13:1-28, (2007). |
| Applicant Initiated Interview Summary for U.S Appl. No. 14/822,349 dated Feb. 13, 2018. |
| International Preliminary Report on Patentability for Application No. PCT/IB2012/056513 dated May 19, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/IB2012/056514 dated May 19, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/IB2012/057773 dated Jun. 30, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/IB2012/057774 dated Jun. 30, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/IB2013/050375 dated Jul. 21, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/IB2013/058131 dated May 5, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/IB2014/060846 dated Oct. 18, 2014. |
| International Preliminary Report on Patentability for Application No. PCT/US2012/053115 dated Mar. 3, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/US2012/053127 dated Mar. 3, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/US2012/053128 dated Mar. 3, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/US2012/053156 dated Mar. 3, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/US2012/053183 dated Mar. 3, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/US2012/061051 dated Mar. 3, 2015. |
| International Preliminary Report on Patentability for Application No. PCT/US2012/063343 dated May 5, 2015. |
| International Search Report and Written Opinion for Application No. PCT/IB2012/056513 dated Jun. 26, 2013. |
| International Search Report and Written Opinion for Application No. PCT/IB2012/056514 dated Jun. 26, 2013. |
| International Search Report and Written Opinion for Application No. PCT/IB2012/057773 dated Jul. 1, 2013. |
| International Search Report and Written Opinion for Application No. PCT/IB2012/057774 dated Sep. 20, 2013. |
| International Search Report and Written Opinion for Application No. PCT/IB2013/050375 dated May 7, 2013. |
| International Search Report and Written Opinion for Application No. PCT/IB2013/058131 dated Jul. 3, 2014. |
| International Search Report and Written Opinion for Application No. PCT/IB2014/060846 dated Feb. 4, 2015. |
| International Search Report and Written Opinion for Application No. PCT/US2012/053115 dated Jul. 24, 2013. |
| International Search Report and Written Opinion for Application No. PCT/US2012/053127 dated Jul. 24, 2013. |
| International Search Report and Written Opinion for Application No. PCT/US2012/051128 dated Jun. 27, 2013. |
| International Search Report and Written Opinion for Application No. PCT/US2012/053156 dated Sep. 26, 2013. |
| International Search Report and Written Opinion for Application No. PCT/US2012/053183 dated Jun. 4, 2013. |
| International Search Report and Written Opinion for Application No. PCT/US2012/061051 dated Jul. 24, 2013. |
| International Search Report and Written Opinion for Application No. PCT/US2012/063343 dated Jan. 15, 2014. |
| Notice of Allowance for U.S. Appl. No. 14/023,023 dated Apr. 11, 2014. |
| Notice of Allowance for U.S. Appl. No. 14/023,056 dated Apr. 29, 2014. |
| Notice of Allowance for U.S. Appl. No. 14/027,684 dated Mar. 21, 2016. |
| Notice of Allowance for U.S. Appl. No. 14/027,775 dated Aug. 12, 2015. |
| Notice of Allowance for U.S. Appl. No. 14/027,775 dated Sep. 10, 2015. |
| Notice of Allowance for U.S. Appl. No. 14/311,806 dated Dec. 28, 2016. |
| Notice of Allowance for U.S. Appl. No. 14/311,998 dated Dec. 22, 2015. |
| Notice of Allowance for U.S. Appl. No. 14/311,998 dated Jan. 21, 2016. |
| Notice of Allowance for U.S. Appl. No. 14/634,035 dated Mar. 30, 2016. |
| Notice of Allowance for U.S. Appl. No. 14/634,074 dated Jun. 30, 2015. |
| Notice of Allowance for U.S. Appl. No. 14/634,119 dated Feb. 2, 2016. |
| Notice of Allowance for U.S. Appl. No. 14/634,158 dated Jan. 11, 2016. |
| Notice of Allowance for U.S. Appl. No. 14/914,461 dated Dec. 12, 2017. |
| Notice of Allowance for U.S. Appl. No. 15/421,921 dated Mar. 14, 2018. |
| Office Action for U.S. Appl. No. 14/023,023 dated Mar. 4, 2014. |
| Office Action for U.S. Appl. No. 14/023,056 dated Nov. 21, 2013. |
| Office Action for U.S. Appl. No. 14/027,684 dated Oct. 6, 2015. |
| Office Action for U.S. Appl. No. 14/027,775 dated Jul. 13, 2015. |
| Office Action for U.S. Appl. No. 14/311,806 dated Jun. 10, 2016. |
| Office Action for U.S. Appl. No. 14/311,998 dated Feb. 20, 2015. |
| Office Action for U.S. Appl. No. 14/311,998 dated Oct. 7, 2015. |
| Office Action for U.S. Appl. No. 14/634.305 dated Aug. 28, 2015. |
| Office Action for U.S. Appl. No. 14/634,035 dated Dec. 10, 2015. |
| Office Action for U.S. Appl. No. 14/634,035 dated Mar. 30, 2016. |
| Office Action for U.S. Appl. No. 14/634,074 dated Apr. 17, 2015. |
| Office Action for U.S. Appl. No. 14/634,119 dated Apr. 21, 2015. |
| Office Action for U.S. Appl. No. 14/634,119 dated Oct. 23, 2015. |
| Office Action for U.S. Appl. No. 14/760,848 dated May 11, 2017. |
| Office Action for U.S. Appl. No. 14/822,349 dated Jan. 20, 2017. |
| Office Action for U.S. Appl. No. 14/822,349 dated Jun. 27, 2018. |
| Office Action for U.S. Appl. No. 14/822,349 dated Nov. 13, 2017. |
| Office Action for U.S. Appl. No. 14/822,349 dated Sep. 2, 2016. |
| Office Action for U.S. Appl. No. 14/914,461 dated May 18, 2017. |
| Office Action for U.S. Appl. No. 14/961,222 dated Mar. 3, 2018. |
| Office Action for U.S. Appl. No. 15/022,420 dated Feb. 13, 2018. |
| Office Action for U.S. Appl. No. 15/022,420 dated May 18, 2017. |
| Office Action for U.S. Appl. No. 15/022,420 dated Sep. 28, 2018. |
| Office Action for U.S. Appl. No. 15/074,425 dated Feb. 26, 2018. |
| Office Action for U.S. Appl. No. 15/074,425 dated May 10, 2017. |
| Office Action for U.S. Appl. No. 15/093,337 dated Apr. 4, 2018. |
| Office Action for U.S. Appl. No. 15/093,337 dated Jun. 29, 2017. |
| Office Action for U.S. Appl. No. 15/093,365 dated Aug. 31, 2018. |
| Office Action for U.S. Appl. No. 15/093,365 dated Nov. 16, 2017. |
| Office Action for U.S. Appl. No. 15/186,927 dated Jul. 3, 2018. |
| Office Action for U.S. Appl. No. 15/186,927 dated May 1, 2017. |
| Office Action for U.S. Appl. No. 15/186,927 dated Nov. 17, 2017. |
| Office Action for U.S. Appl. No. 15/188,423 dated Jul. 20, 2018. |
| Office Action for U.S. Appl. No. 15/188,423 dated Oct. 23, 2017. |
| Office Action for U.S. Appl. No. 15/421,921 dated Sep. 27, 2017. |
| Statement in accordance with the Notice from the European patent Office dated Oct. 1, 2007 concerning business methods (OJ EPO Nov. 2007, 592-593, (XP002456414) 1 page. |
| U.S. Appl. No. 12/779,636; entitled “System and Method for Using Data to Automatically Generate a Narrative Story” filed May 13, 2010. |
| U.S. Appl. No. 13/186,308; entitled “Method and Apparatus for Triggering The Automatic Generation of Narratives” filed Jul. 19, 2011. |
| U.S. Appl. No. 13/186,329; entitled “Method and Apparatus for Triggering the Automatic Generation of Narratives” filed Jul. 19, 2011. |
| U.S. Appl. No. 13/186,337; entitled “Method and Apparatus for Triggering the Automatic Genereation of Narratives” filed Jul. 19, 2011. |
| U.S. Appl. No. 13/186,346; entitled “Method and Apparatus for Triggering the Automatic Generation of Narratives” filed Jul. 19, 2011. |
| U.S. Appl. No. 13/464,635; entitled “Use of Tools and Abstraction in a Configurable and Portable System for Generating Narratives” filed May 4, 2012. |
| U.S. Appl. No. 13/464,675; entitiled “Configurable and Portable System for Generating Narratives” filed May 4, 2012. |
| U.S. Appl. No. 13/464,716; entitled “Configurable and Portable System for Generating Narratives” filed May 4, 2012. |
| U.S. Appl. No. 14/023,023; entitled “Method and Apparatus for Alert Valtidation;” filed Sep. 10, 2013. |
| U.S. Appl. No. 14/023,056; entitled “Method and Apparatus for Situational Analysis Text Generation;” filed Sep. 10, 2013. |
| U.S. Appl. No. 14/027,684; entitled “Method, Apparatus, and Computer Program Product for User-Directed Reporting;” filed Sep 16, 2013. |
| U.S. Appl. No. 14/027,775; entitled “Method and Apparatus for Interactive Reports;” filed Sep. 16, 2013. |
| U.S. Appl. No. 14/311,806; entitled Method and Apparatus for Alert Validation; In re: Reiter, filed Jun. 23, 2014. |
| U.S. Appl. No. 14/311,998, entitled Method and Apparatus for Situational Analysis Text Gerneation; In re: Reiter; filed Jun. 23, 2014. |
| U.S. Appl. No. 14/634,035, entitled Method and Apparatus for Annotating a Graphical Output; In re: Reiter; filed Feb. 27, 2015. |
| U.S. Appl. No. 14/634,074, entitled Method and Apparatus for Configurable Microplanning; In re: Reiter; filed Feb. 27, 2015. |
| U.S. Appl. No. 14/760,848, enititled Method and Apparatus for Document Planning; In re: Sripada; filed Jul. 14, 2015. |
| U.S. Appl. No. 14/822,349; entitled Method and Apparatus for Configurable Microplanning; In re: Reiter, filed Aug. 10, 2015. |
| U.S. Appl. No. 14/914,461, filed Feb. 25, 2016; In re: Reither et al., entitled Text Generation From Correlated Alerts. |
| U.S. Appl. No. 14/961,222, enititled Method and Apparatus for Interactive Reports; in re: Date et al., filed Dec. 7, 2015. |
| U.S Appl. No. 15/022,420, filed Mar. 16, 2016; In re: Mahamood, entitled Method and Apparatus for Document Planning. |
| U.S. Appl. No/ 15/074,425, filed Mar. 18, 2016; In re: Reiter, entitled Method and Apparatus for Situational Analysis Text Generation. |
| U.S. Appl. No. 15/093,227, filed Apr. 7, 2016; In re: Reiter, entitled Method and Apparatus for Referring Expression Generation. |
| U.S. Appl. No. 15/093,365, filed Apr. 7, 2016; In re: Logan et al., entitled Method and Apparatus for Updating a Previously Generated Text. |
| U.S. Appl. No. 15/188,423, filed Jun. 21, 2016; In re: Reither, entitled Method and Apparatus for Annotating a Graphical Output. |
| U.S. Appl. No. 15/421,921, filed Feb. 1, 2017; In re: Reither, entitled Method and Apparatus for Alert Validation. |
| U.S. Appl. No. 15/186,927; filed Jun. 20, 2016; in re: Sripada, entitled Mathod, Apparatus, and Computer Program Product for User-Directed Reporting. |
| Buschmeier et al, “An alignment-capable microplanner for natural language generation,” Proceedings of the 12th European Workshop on Natural Language Generation. Association for Computational Linguistics, pp. 82-89, (2009). |
| Theune, “Natural Language Generation for dialogue: system survey,” Thesis , University of Twene, pp. 1-47, (2003). |
| Notice of Allowance for U.S. Appl. No. 14/961,222 dated Nov. 16, 2018. |
| Notice of Allowance for U.S. Appl. No. 15/093,365 dated Jun. 28, 2019. |
| Notice of Allowance for U.S. Appl. No. 15/186,927 dated Dec. 20, 2018. |
| Notice of Allowance for U.S. Appl. No. 15/188,423 dated Dec. 28, 2018. |
| Notice of Allowance for U.S. Appl. No. 16/009,006 dated Jul. 31, 2019. |
| Office Action for U.S. Appl. No. 14/760,848 dated Oct. 3, 2018. |
| Office Action for U.S. Appl. No. 14/822,349 dated Dec. 26, 2018. |
| Office Action for U.S. Appl. No. 14/822,349 dated Mar. 22, 2019. |
| Office Action for U.S. Appl. No. 15/022,420 dated Apr. 22, 2019. |
| Office Action for U.S. Appl. No. 15/074,425 dated Nov. 27, 2018. |
| Office Action for U.S. Appl. No. 15/093,337 dated Dec. 14, 2018. |
| Office Action for U.S. Appl. No. 15/093,365 dated Dec. 17, 2018. |
| Office Action for U.S. Appl. No. 15/188,423 dated Oct. 30, 2018. |
| Office Action for U.S. Appl. No. 16/009,006 dated Dec. 3, 2018. |
| Notice of Allowance and Applicant Initiated Interview Summary for U.S. Appl. No. 14/822,349 dated Oct. 3, 2019 |
| Office Action for U.S. Appl. No. 15/074,425 dated Oct. 4, 2019. |
| Number | Date | Country | |
|---|---|---|---|
| 20180276204 A1 | Sep 2018 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 14914461 | US | |
| Child | 15918336 | US |
