Abstract: A first candidate answer to a time-sensitive question received by a question answering (QA) system is identified using a first document. A first temporal confidence subscore of the first candidate answer is then calculated using a first temporal confidence rule. A first general confidence score is then generated based on at least the first temporal confidence subscore. A second candidate answer to the question is identified using a second document. A second temporal confidence subscore of the second candidate answer is then calculated using a second temporal confidence rule. A second general confidence score is then generated based on at least the second temporal confidence subscore. The first general confidence score and the second general confidence score are then compared. Based on this comparison, a final answer is selected from the first candidate answer and the second candidate answer.

Abstract: Users of online services may avoid having to gain knowledge of the intricacies of specific Application Programming Interface (API) syntaxes and elements by using natural language requests which are processed to extract components of the request. The components are then used to find one or more matches of known APIs, and then at least one structured API call is constructed by mapping the request's components to the structured API call's elements. The system can then invoke the online service on behalf of the client, and return the results to the client, or it can return the constructed structured API call to the client. The known API details can be initialized and updated by ingesting the description documents for the structured APIs, and then using natural language processing to extract components from the descriptions, which can then be utilized in the matching process to further enhance the results of the tool.

Abstract: Users of online services may avoid having to gain knowledge of the intricacies of specific Application Programming Interface (API) syntaxes and elements by using natural language requests which are processed to extract components of the request. The components are then used to find one or more matches of known APIs, and then at least one structured API call is constructed by mapping the request's components to the structured API call's elements. The system can then invoke the online service on behalf of the client, and return the results to the client, or it can return the constructed structured API call to the client. The known API details can be initialized and updated by ingesting the description documents for the structured APIs, and then using natural language processing to extract components from the descriptions, which can then be utilized in the matching process to further enhance the results of the tool.

Abstract: Users of online services may avoid having to gain knowledge of the intricacies of specific Application Programming Interface (API) syntaxes and elements by using natural language requests which are processed to extract components of the request. The components are then used to find one or more matches of known APIs, and then at least one structured API call is constructed by mapping the request's components to the structured API call's elements. The system can then invoke the online service on behalf of the client, and return the results to the client, or it can return the constructed structured API call to the client. The known API details can be initialized and updated by ingesting the description documents for the structured APIs, and then using natural language processing to extract components from the descriptions, which can then be utilized in the matching process to further enhance the results of the tool.

Abstract: Mechanisms are provided for resolving annotation conflicts in a portion of content. A portion of content is received and annotation operations are performed on it using a plurality of annotators. Each annotator in the plurality of annotators generates a set of annotations for the portion of content, thereby generating a plurality of sets of annotations for the portion of content. The plurality of sets of annotations are analyzed to identify a conflict between a first annotation and a second annotation in the plurality of sets of annotations. The conflict between the first annotation and the second annotation is resolved to generate a resultant set of annotations for the portion of content. The resultant set of annotations is stored in an annotations data structure associated with the portion of content.

Abstract: A watermarketing system and method is proposed for still or moving pictures in which a watermark is embedded robustly and simply into DC-values, but without causing unacceptable visible picture degradation. The watermark is secure and can be readily detected, and the method can be used to convey additional data. Also disclosed is a method for watermarking a sequence of pictures in which the visibility of artifacts is reduced. The watermark may carry data and may be used to label a programme.