Abstract: The present disclosure is directed to systems and methods that include and/or leverage one or more machine-learned language models that generate intermediate textual analysis (e.g., including usage of structural tools such as APIs) in service of contextual text generation. For example, a computing system can obtain a contextual text string that includes one or more contextual text tokens. The computing system can process the contextual text string with the machine-learned language model to generate one or more intermediate text strings that include one or more intermediate text tokens. The computing system can process the one or more intermediate text strings with the machine-learned language model to generate an output text string comprising one or more output text tokens. The one or more intermediate text strings can include textual analysis of the contextual text string that supports the output text string.

Abstract: A first document is stored in a first format having an ordered sequence of data elements including characters and embedded objects accessible by index location into the ordered sequence. The first document in the first format is converted into a second document in a second format having a second ordered sequence including a subset of the data elements of the first document including the characters but not the embedded objects, such that the index location into the ordered sequence of the characters differs between the first and second documents. The second document is analyzed in the second format as plain text to identify a string of the data elements of interest. A document location of the string of the data elements of interest is represented as a relative location into the second document. The relative location into the second document is mapped into the index location of the first document.

Abstract: According to an embodiment, a storage method includes: a step of generating scan data by scanning of a document with a reading device; a step of acquiring index information that is different from information that is based on the document; and a step of assigning the acquired index information to the scan data and storing the scan data to which the index information is assigned in a storage device.

Abstract: A frequency of occurrence for each term in a training data set is determined in relation to the entire training data set. A relational data structure is generated that associates each term in the training data with its respective frequency. Any term that has a frequency below a threshold frequency is then added to a list of relevant words. When a natural language search query is received, a plurality of terms in the natural language search query are identified and compared with the list of relevant words. If any term of the natural language search query is included in the relevant words list, that term is identified as a keyword. The natural language search query is then interpreted based on any identified keywords.

Abstract: Methods for generating content summaries in a web content management service, wherein in one embodiment a digital page editor and a component browser are launched to enable selection of a first content item. A summary of the first content item is automatically generated according to parameters that may have default values or values set by a user. The parameters may specify a size for the summary as a percentage of the first content item's size, as a particular number of lines, characters or words, as a size for a particular type of device, etc. The automatically generated summary is provided to the digital page editor, which can edit it and add it to the digital page. The summary is stored in a content repository as an independent summary content item with its own metadata.

Abstract: Systems, methods and products for accessing a set of electronic document templates, identifying instances of common document content such as content items which are semantically similar, and generating component templates containing the common content. Semantically similar content may be identified by analyzing content for factors such as expressed sentiment, included keyphrases, recognizable entities, expressed topics, assigning values to content based on these factors, and determining similarity based on comparisons of the assigned values. Component templates may also be generated based on types of content that include identical text or images, content that has a predefined level of similarity rather than being identical, content that has common rules, scripting logic or variables, metadata, etc. The component templates may be generated automatically, or in response to user instructions.

Abstract: A stateless, on-demand semantic indexing and search service for a health information exchange is provided that anticipatorily fetches data from one or more external sources, extracts semantic concepts from textual documents, and serves an in-memory semantic search index to users to ease finding of relevant patient data across a set of distributed health record data sources. Clinicians may search for not only text, but for semantic concepts, and the search index is able to return relevant results regardless of whether the source information is located inside or outside the local electronic health record. Predictive algorithms may be incorporated to pre-fetch, semantically process and index data for patients likely to be searched in an upcoming time window. Activity indicators for clinicians and patients may be utilized to pre-fetch, semantically process, and index data.

Abstract: An electronic apparatus and a control method thereof are provided. The electronic apparatus includes a communication interface configured to receive content comprising image data and speech data; a memory configured to store a language contextual model trained with relevance between words; a display; and a processor configured to: extract an object and a character included in the image data, identify an object name of the object and the character, generate a bias keyword list comprising an image-related word that is associated with the image data, based on the identified object name and the identified character, convert the speech data to a text based on the bias keyword list and the language contextual model, and control the display to display the text that is converted from the speech data, as a caption.

Abstract: A method and server for ranking digital documents for a query are disclosed. T server hosts a search engine associated with an inverted index. The method includes accessing inverted index for retrieving query-independent data for a first document-term pair and a second document-term pair, where the query-independent data is indicative of (i) a term-specific occurrence of the first term in content of the document and (ii) a term-specific occurrence of the second term in the content of the document. The method includes generating a query-dependent feature using the query-independent data and which is indicative of a group occurrence of the first term with the second term in the content of the document. The method includes generating a ranking feature for the given document based on at least the first term, the second term, and the query-dependent feature, and ranking the given document based on at least the ranking feature.

Abstract: A system and method for content management, may include receiving content created by a creator; creating a container to contain the created content and storing the created content in the container; determining a content type for the content; based on the determined content type, identifying items of metadata specified for the content type (name metadata; support data; corresponding; determining whether an identified item of specified metadata corresponding to the content is missing from the container; and prompting a user to supply identified missing metadata corresponding to the container.

Abstract: Some embodiments provide a non-transitory machine-readable medium that stores a program. The program receives a plurality of documents. The program further uses a visual model to generate a vector representation for each document in the plurality of documents. The program also clusters the plurality of documents into a set of clusters based on the vector representations of the plurality of documents. The program further determines a sample set of documents from the plurality of documents based on the set of clusters.

Abstract: A database system may receive a natural language query that is associated with a tenant of a multi-tenant system. The natural language query may be parsed into a set of tokens, and the set of tokens may be tagged, using a tenant specific tagging model associated with the tenant, the set of tokens with at least one pre-configured data type identifier that is configured for the plurality of tenants. A global tagging model that supports the plurality of tenants of the multi-tenant system may tag the set of tokens with at least one category identifier. The global tagging model may use the natural language query and the pre-configured data type identifier to identify the at least one category identifier. The system may execute a database query on a database associated with the tenant using the at least one pre-configured data type identifier and the at least one category identifier.

Abstract: A method and system for engramic indexing of information technology (IT) infrastructure. Specifically, the method and system disclosed herein enable the efficient access, search, and/or management of enterprise-scale IT infrastructure and topologies using semantic context and natural language processing. That is, query context may be optimized using mapped semantics based on organizational constructs and machine learning, thereby reducing query overhead, increasing response performance, and improving contextual display capabilities in mass-scale environments.

Abstract: An entity combination and a text representation are obtained as a first fact, and the entity combination and a related predicate are obtained as a second fact. Word distributed representations are input to a neural network, real vectors at appearance positions of entities are specified and used as distributed representations. A first score indicating a degree of establishment of the first fact is calculated based on the distributed representations and on entity distributed representations. A second score indicating a degree of establishment is calculated with respect to an entity combination and a text representation that are not the first fact. A third score indicating a degree of establishment of the second fact is calculated based on predicate distributed representations and on entity distributed representations. A fourth score indicating a degree of establishment is calculated also with respect to an entity combination and a predicate that are not the second fact.

Abstract: Techniques described herein allow for accurate translation of natural language (NL) queries to declarative language. A syntactic dependency parsing tree is generated for an NL query, which is used to map tokens in the query to logical data model concepts. Relationship-type mappings are completed based on relationship constraints. Final mappings are identified for any relationship tokens that are associated with multiple candidate mappings by identifying which candidate mappings have the lowest cost metrics. An NL query-specific query graph is generated based on the mapping data for the NL query and the logical data model. The query graph represents an NL query-specific version of the logical data model where grammatical dependencies between NL query words are translated to the query graph. A query graph is annotated with information, from the mapping data, that is not represented paths in the query graph. The query graph is used generate a computer-executable translation of the NL query.

Abstract: The present application discloses a method and an apparatus for recognizing text content, and an electronic device, and relates to a text recognition technique in the field of computer technology. The specific implementation is as follows: acquiring a dial picture; detecting at least one text centerline and a bounding box corresponding to each text centerline in the dial picture; and recognizing text content in each line of text in the dial picture based on the at least one text centerline and the bounding box corresponding to each text centerline.

Abstract: A computer-based method of reducing or limiting data transmissions from a computer to a remote network destination includes receiving an indication, at an agent on a computer, that a recent user activity has occurred at the computer. The indication typically includes data relevant to user context when the user activity occurred. The method further includes determining, with the agent, whether the data relevant to the user's context when the user activity occurred indicates that a change in user context relative to a user activity at the computer immediately prior to the recent user activity and conditioning a transmission of data relevant to the recent user activity from the computer to a remote network destination based on an outcome of the determination.

Abstract: Techniques described herein are directed to analyzing intellectual-property data according to provide various intellectual property related services to organizations. In particular implementations, information related to products and/or services may be obtained from a number of data sources. Additionally, information related to intellectual-property assets, such as patents, trademarks, copyrights, trade secrets, and know-how, may be obtained. In various situations, the intellectual-property assets may be mapped to respective products and/or services. The mappings between the products and/or services and intellectual-property assets may be used to provide intellectual property related services that correspond to the intellectual-property assets, such as valuation services, strategy-related services, or risk-related services.

Abstract: Search assistance personalized to a user, may be afforded by combining the application of scoring functions, with the training a Machine Learning (ML) model based upon feedback from that user. A plurality of scoring functions are applied to query results to calculate a plurality of score vectors. The ML model is then applied to generate a ranked list of vectors. Feedback from the user, is incorporated to develop an evolving memory. The feedback may be explicit, or may be implicit—e.g., based upon user selection of particular score vector(s) from the list or based upon user selection of particular query results corresponding to the score vectors. Embodiments may enhance relevance of search assistance by removing past feedback from the evolving memory that is used to retrain the ML model. Embodiments can provide assistance to search text corpuses utilizing scoring functions considering frequency of occurrence of particular words or terms.

Abstract: In general, embodiments of the present invention provide systems, methods and computer readable media configured to use a per-set level optimization of the rank order of promotions to be recommended to a consumer. In some embodiments, machine learning is used offline to generate a predictive diversity model that receives one or more similarity rank features associated with a promotion (e.g., category, price band) as input, and produces an output multiplier to be applied to the promotion's respective associated relevance score (e.g., a relevance score representing a prediction of the promotion's conversion rate without diversity features). At run time, per-set optimization of the ordering of a set of promotions is implemented by adjusting the respective associated relevance scores of the promotions using the diversity model and then re-ordering the set of promotions based on their respective adjusted relevance scores.

Abstract: Disclosed herein are systems and methods capable of performing text exploration on large volume of corpus without prior knowledge in an accurate and efficient manner and may also provide any number of additional or alternative benefits and advantages. In particular, embodiments described herein provide a text exploration executable environment that uses unsupervised machine-learning to assist a human analyst with distilling key emerging themes from a corpus of hundreds or thousands of text files presented in a time series graphical user interface (GUI). A document may be a unit of text under analysis received from a particular data source, such as word-processing documents, paragraphs, sentences, chat sessions, speech-to-text call segments, online texts, social media postings (e.g., Tweets®), and other machine-readable text.

Abstract: Observable data points are collected and organized into a link-oriented data set comprising nodes and links. Information is abstracted for use in link analysis by generating links between the collected data points, including deriving links and inducing links. A link can be induced by linking together a pair of nodes that satisfy a distance function. Exemplary distance functions that can be used to induce links include geospatial proximity, attribute nearness, and name similarity. Paths can be identified between selected nodes of interest through a dataset operation, and nodes and/or links can be selectively included or excluded from the data set operation. The dataset can be augmented with pedigree information or one or more association nodes. Link information, including a trajectory and a connected path that selectively produces or excludes one or more intermediate nodes, can be displayed and/or produced in a specified format.

Abstract: Provided are a method, system, and computer program product in which operations are performed to receive a question that includes a descriptor and an indication that indicates that a unique answer to the question is expected. A determination is made of instances of matching descriptors and descriptor targets from a set of documents. The determined descriptor targets are compared for consistency. In response to determining that the determined descriptor targets are inconsistent, more restrictive descriptors are iteratively generated via a selection model based on metadata associated with the question until the descriptor targets are consistent. An answer to the question is returned from consistent descriptor targets.

Abstract: Embodiments of this application disclose a chunking method and an apparatus for implementing the method. According to the method provided in the embodiments of this application, a first data segment of a first length may be determined starting from a header of a to-be-chunked data flow, a data distribution characteristic of the first data segment is determined based on character values of all characters in the first data segment, and then a chunking position is determined for different data distribution characteristics by using different methods. In this way, a data flow can be better chunked, so as to enhance a deduplication effect.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for collaboratively training an interaction prediction machine learning model using a plurality of user devices in a manner that respects user privacy. In one aspect, the machine learning model is configured to process an input comprising: (i) a search query, and (ii) a data element, to generate an output which characterizes a likelihood that a given user would interact with the data element if the data element were presented to the given user on a webpage identified by a search result responsive to the search query.

Abstract: An information handling system operating a private individual data integration protection system comprising a processor receiving a query, submitted electronically, involving a request about private individual data and parsing elements the query, the processor correlating the query with private individual data manipulated within a business integration process modeled and deployed with a business integration application management system identifying data set field names via metadata and labels from the modeled business integration process matching private individual data identified in the query as being manipulated from a source input dataset to a destination dataset by operation of the business integration process, and transmitting a responsive report on the private individual data matching the received query indicating manipulation of the matching private individual data during the executed business integration process.

Abstract: According to an embodiment of the present invention, there is provided an artificial intelligence (AI) apparatus for mutually converting a text and a speech, including: a memory configured to store a plurality of Text-To-Speech (TTS) engines; and a processor configured to: obtain image data containing a text, determine a speech style corresponding to the text, generate a speech corresponding to the text by using a TTS engine corresponding to the determined speech style among the plurality of TTS engines, and output the generated speech.

Abstract: A system and method for automated theming of thought objects is disclosed. In a preferred embodiment, an object theming computer creates aggregated text by aggregating text associated with a first thought object, the aggregated text comprising descriptive and important text associated with the first thought object. The object theming computer then tokenizes the aggregated text into tokens. Further, a theme is associated to thought objects using one or more previously themed thought objects. If one or more thought objects have no associated theme, then a theme is generated and associated to the un-themed thought objects using a current plurality of the one or more thought objects.

Abstract: Embodiments of the application disclose a summary generation method and apparatus. A specific embodiment of the method comprises: acquiring a target article including a headline and a body of the article; determining whether a question is included in the headline; determining, in the body of the article, an information-satisfied-paragraph including an answer to the question, in response to determining that the question is included in the headline; and generating a summary of the target article based on the determined information-satisfied-paragraph. The above embodiment may generate a summary that directly satisfies the users' demand for information.

Abstract: Disclosed herein are computer-implement systems and methods for determining whether an electronic document transiting a computer network is safe for viewing and does not contain malicious code, and automatically routing or deleting such electronic document based on such determination, thus providing quick risk decisions relating to the electronic document using an objective, quantitative method. A search algorithm is employed to detect the presence of words or phrases in the subject electronic document that are possible threats to the computer network. Threat updates may be provided through publicly available threat database. A weight is added to the generated score that accounts for the time that the particular vulnerability has been in the subject threat database. Finally, a PNN is used to quickly and objectively classify the subject electronic document for acceptance, quarantine, or denial.

Abstract: A system includes application of respective compression types to first data associated with each of a plurality of columns to generate compressed column data, determination of a first compression ratio for each of the plurality of columns based on the compressed column data, storage of the determined first compression ratios, application, for each of the plurality of columns, of the determined compression type to second data associated with the column to generate second compressed column data, determination of a second compression ratio for each of the plurality of columns based on the second compressed column data, determination of a value for each column based on the stored first compression ratio and the second compression ratio determined for the column, determination of a representative value of the determined values, and determination, based on the representative value, whether to re-determine a compression type for each of the plurality of columns.

Abstract: Systems, methods, and apparatuses are presented for a novel natural language tokenizer and tagger. In some embodiments, a method for tokenizing text for natural language processing comprises: generating from a pool of documents, a set of statistical models comprising one or more entries each indicating a likelihood of appearance of a character/letter sequence in the pool of documents; receiving a set of rules comprising rules that identify character/letter sequences as valid tokens; transforming one or more entries in the statistical models into new rules that are added to the set of rules when the entries indicate a high likelihood; receiving a document to be processed; dividing the document to be processed into tokens based on the set of statistical models and the set of rules, wherein the statistical models are applied where the rules fail to unambiguously tokenize the document; and outputting the divided tokens for natural language processing.

Abstract: A display control device includes a processor. The processor displays plural areas. The plural areas include a first area, a second area, a third area and a fourth area. The first area is related to a search for content data by a user. The second area is related to a selection of content data by the user. The third area is related to a display of content of the content data selected in the second area. The fourth area is related to a user's edit of information associated with the content data displayed in the third area. The processor hides the first area and the second area and does not hide the third area and the fourth area when the content data is selected by the user in the second area.

Abstract: The dialogue system includes a ranking determination unit configured to determine a priority ranking of the resulting sentence candidates on the basis of a degree of coincidence between the input key group and the storage key group, and a presentation unit configured to present a response for prompting a user to make an additional input on the basis of a resulting sentence candidate of a highest priority ranking. In a case where there is the stored keyword which is in a predetermined correspondence relation with the character string, the keyword acquisition unit acquires the input keyword obtained by adding additional information to the stored keyword. In a case where the additional information is added to the input keyword, the ranking determination unit reduces an influence of the input keyword to which the additional information is added on the degree of coincidence.

Abstract: Artificial intelligence based document analysis systems and methods are disclosed. Embodiments of document analysis systems may allow the manipulation of datasets and associated codes by determining representations for these codes or datasets based on a machine learning model. The codes or datasets can then be manipulated using the associated representations.

Abstract: At least some embodiments of the present disclosure relate to a method of training an artificial neural network (ANN) for an artificial intelligence recognition. The method includes producing, by an ANN, outputs by feeding inputs of a training data set to the ANN; determining errors of the generated outputs from target outputs of the training data set; generating a first-order derivative matrix including first-order derivatives of the errors and a second-order derivative matrix including second-order derivatives of the errors; obtaining an approximation of the first-order derivative matrix or an approximation of the second-order derivative matrix by compressing the first-order derivative matrix or the second-order derivative matrix; and updating weights of the ANN based on the approximation of the first-order derivative matrix or the approximation of the second-order derivative matrix.

Abstract: Methods, systems, and apparatus, including computer program products, for processing search query suggestions. In one aspect, a search service provides query suggestions responsive to a query suggestion request from a client device, and determines if a prediction criterion is met. The prediction criterion is independent of a user selection of a query suggestion provided in response to one or more query suggestion requests. In response to determining that the prediction criterion is met, the search system provides search results to the client device. The search results are responsive to one of the query suggestions provided in response to the query suggestion request or one or more previous query suggestion requests.

Abstract: Systems, methods and products for accessing a set of electronic document templates, identifying instances of common document content such as content items which are semantically similar, and generating component templates containing the common content. Semantically similar content may be identified by analyzing content for factors such as expressed sentiment, included keyphrases, recognizable entities, expressed topics, assigning values to content based on these factors, and determining similarity based on comparisons of the assigned values. Component templates may also be generated based on types of content that include identical text or images, content that has a predefined level of similarity rather than being identical, content that has common rules, scripting logic or variables, metadata, etc. The component templates may be generated automatically, or in response to user instructions.

Abstract: A search is performed based on a voice input combined with user selection of entities displayed on a display screen as well as real-world entities. A voice input is received from the user by a media device, as well as a selection of a first entity being displayed on the media device. A gesture made by the user is also identified, and a second, real-world entity corresponding to the gesture is determined. A search query is constructed based on a search operator in the voice input, the first entity, and the second entity. The search query is transmitted to a database and, in response, the media device receives at least one identifier of a least one content item. The at least one identifier is then generated for display to the user.

Abstract: Systems, devices, and methods of the present invention relate to text classification. A text classification system accesses an utterance of text. The utterance includes at least one word. The text classification system generates a parse tree for the utterance. The parse tree includes at least one terminal node with a word type. The terminal node represents a word of the utterance. The text classification system applies one or more rules to the text. The text classification system then classifies the utterance as a question or a request for an autonomous agent to perform an action.

Abstract: Systems and methods are provided for presenting aggregate data in response to a natural language user input. In one example, a system includes a display and a computing device coupled to the display and storing instructions executable to receive a natural language user input, process the natural language user input, in response to determining that the user input includes a request to display two different plots of record data specific to the subject, generate, with the virtual assistant, a single graph including the two different plots of record data based on the processed natural language user input, the two different plots of record data plotted from two different record data sets, one or more aspects of the single graph selected based on an overlapping parameter for each of the two different record data sets, and output, to the display, the single graph as part of a communication thread.

Abstract: A computer-implemented method of transcribing an audio stream can include transcribing the audio stream using a first transcribing instance having a first predetermined transcription size that is smaller than the total length of the audio stream. The first transcribing instance can provide a plurality of consecutive first transcribed text data snippets of the audio stream and the size of the first transcribed text data snippets can respectively corresponding to the first predetermined transcription size. The audio stream can also be transcribed using at least a second transcribing instance having a second predetermined transcription size that is smaller than the length of the audio stream. The second transcribing instance can provide a plurality of consecutive second transcribed text data snippets each corresponding to the second predetermined transcription size.

Abstract: This disclosure describes a technique to determine whether a client computing device accessing an API is masquerading its device type (i.e., pretending to be a device that it is not). To this end, and according to this disclosure, the client performs certain processing requested by the server to reveal its actual processing capabilities and thereby its true device type, whereupon—once the server learns the true nature of the client device—it can take appropriate actions to mitigate or prevent further damage. To this end, during the API transaction the server returns information to the client device that causes the client device to perform certain computations or actions. The resulting activity is captured on the client computing and then transmitted back to the server, which then analyzes the data to inform its decision about the true client device type.

Abstract: A blank template form generation method and system may employ synthetically generated blank template forms, differing from each other in one or more respects, to train a neural network to recognize relevant differences between otherwise similar forms, including types and locations of keywords and potential locations of values corresponding to the keywords. In an embodiment, filled or partly filled forms as well as blank template forms may be used later in training. Forms are input in pairs to identify differences between the two. Depending on the differences, weights of a neural network may be adjusted. After training, when a form is input into the system, whether the form is filled or blank, a blank template may be generated for future use.

Abstract: A computer-implemented method for performing cross-document coreference for a corpus of input documents includes determining mentions by parsing the input documents. Each mention includes a first vector for spelling data and a second vector for context data. A hierarchical tree data structure is created by generating several leaf nodes corresponding to respective mentions. Further, for each node, a similarity score is computed based on the first and second vectors of each node. The hierarchical tree is populated iteratively until a root node is created. Each iteration includes merging two nodes that have the highest similarity scores and creating an entity node instead at a hierarchical level that is above the two nodes being merged. Further, each iteration includes computing the similarity score for the entity node. The nodes with the similarity scores above a predetermined value are entities for which coreference has been performed in input documents.

Abstract: A mechanism is provided in a data processing system to implement a cognitive natural language processing (NLP) system with descriptor uniqueness identification to support named entity mention clustering. The mechanism annotates a set of documents from a corpus of documents for entity types and mentions, collects descriptor usages from all documents in the corpus of documents, analyzes the descriptor usages to classify the descriptors as base terms or modifier terms, generates compatibility scores for the descriptors, and performs entity merging of entity clusters based on the compatibility scores.

Abstract: Operating a low-latency database analysis system with phrase translation may include obtaining a locale-specific phrase localization rule and a canonical phrase localization rule for a phrase, generating a locale-specific index and a locale-specific finite state machine for the locale using the localization definition data and a canonical finite state machine, generating a resolved-request by obtaining a locale-specific token representing locale-specific input data by traversing the locale-specific index, obtaining a canonical token associated with locale-specific token, obtaining a locale-specific phrase by traversing the locale-specific finite state machine, obtaining a canonical phrase corresponding to the locale-specific phrase, the canonical phrase including the canonical token, generate a data-query based on the canonical phrase, obtaining results data responsive to the data expressing the usage intent by executing a query corresponding to the data-query by an in-memory database of the low-latency datab

Abstract: A virtual repository system with robust item management automatically derives item data from accessed current and past transactions. The system interfaces with merchant systems to receive current and archived transaction data, scans emails for current and past transaction data, monitors browser data for online transaction data, and accepts manual input. Data obtained from all sources is collated and stored in a cache for user validation, whereupon it is added to a virtual repository. Triggers prompt the delivery of responsive results including information from shared virtual repositories.

Abstract: Embodiments of the present disclosure provide a method for generating an information assessment model, a method for determining the usefulness of comment information, apparatus, electronic device, and computer-readable medium.

Abstract: To enhance the accuracy of a learner in semi-supervised learning, learning means of a learning system (S) causes the learner, which is configured to classify symbol information included in each of a plurality of documents, to learn based on training data indicating an attribute value of each of a plurality of attributes. Acquisition means inputs each of the plurality of documents to the learner to acquire the symbol information classified by the learner as an attribute value candidate. Determination means determines whether a symbol or a symbol string indicated by the attribute value candidate satisfies a predetermined condition. Additional learning control means controls, based on a determination result obtained by the determination means, additional learning by the learner using the attribute value candidate.