International Business Machines Corporation

Hui Wan

Sasha P Caskey

Cheng Wu

Andrzej Sakrajda

Jian Ni

Luis Alfonso Lastras-Montano

Sachindra Joshi

Bishwaranjan Bhattacharjee

Haifeng Qian

Radu Florian

Xiaodong Cui

John Ronald Kender

Jatin Ganhotra

Chulaka Gunasekara

Song Feng

Parul Awasthy

A computer system, product, and method are provided. The computer system includes an artificial intelligence (AI) platform operatively coupled to a processor. The AI platform includes tools in the form of a machine learning model (MLM) manager, a metric manager, and a training manager. The MLM manager accesses a plurality of pre-trained source MLMs, and inputs a plurality of data objects of a test dataset into each of the source MLMs. The test dataset includes the plurality of data objects associated with respective labels. For each source MLM, associated labels are generated from the inputted data objects and a similarity metric is calculated. The MLM manager selects a base MLM to be used for transfer learning from the plurality of source MLMs based upon the calculated similarity metric. The training manager trains the selected base MLM with a target dataset for the target domain.

Predictive data and model selection for transfer learning in natural language processing

From metadata of a corpus of natural language text documents, a relativity matrix is constructed, a row-column intersection in the relativity matrix corresponding to a relationship between two instances of a type of metadata. An encoder model is trained, generating a trained encoder model, to compute an embedding corresponding to a token of a natural language text document within the corpus and the relativity matrix, the encoder model comprising a first encoder layer, the first encoder layer comprising a token embedding portion, a relativity embedding portion, a token self-attention portion, a metadata self-attention portion, and a fusion portion, the training comprising adjusting a set of parameters of the encoder model.

Transformer-based encoding incorporating metadata

Techniques facilitating interpretation of deep neural model based dialogue agents are provided. In one example, a computer-implemented method comprises extracting, by a device operatively coupled to a processor, features from a dialogue model independently from the dialogue model, the features comprising input features provided to the dialogue model and output features produced via the dialogue model in response to the input features, resulting in extracted features; and analyzing, by the device, a dialogue context associated with the extracted features by identifying pairwise interactions between respective ones of the extracted features.

Interpretation of deep neural model based dialogue agents

A method of providing real-time speech analysis to a user includes capturing a speech input, performing a real-time recognition of the speech input including converting the speech input to a text, analyzing the recognized speech input to identify an error in a voice of the user, the analyzing including comparing a voice of a correct text generated by an automated speech generation system with the captured speed input, and processing the text to extract a context dialog prompt.

Real-time speech analysis method and system using speech recognition and comparison with standard pronunciation

A method of translating information using crowdsourcing includes evaluating a metric related to a content to be translated, determining a priority for the content, queuing the content for the crowdsourcing based on the priority determined from the metric, and translating the information from a language to another language using the crowdsourcing.

Selective machine translation with crowdsourcing

A method of providing real-time speech analysis for a user includes capturing a speech input, performing a real-time recognition of the speech input including converting the speech input to a text using an automatic speech recognition component, analyzing the recognized speech input, by a processing unit of a computer in a speech recognition and analyzing system, to identify an error in the user's speech, and by comparing a voice of a correct text generated by a speech generation and analyzing system with the captured speech input, and providing a real-time correction to the user based on a result of the comparing the voice of the correct text with the captured speech input. The comparing the voice of the correct text with the captured speech input includes comparing a standard pronunciation of the correct text with a pronunciation of the user in the captured speech input to identify the error.

Real-time speech analysis and method using speech recognition and comparison with standard pronunciation

Mechanisms are provided for performing a cognitive operation. An input graph is received having a plurality of first nodes, where subsets of first nodes are coupled to one another via first edges and each first edge has an associated weight. A blinking graph model is generated based on the graph, where blink rates are associated with second edges and are calculated based on weights of corresponding first edges in the input graph. The blink rate specifies a fraction of time a corresponding second edge is determined to be present in the blinking graph model. A relatedness metric is calculated for a target node relative to a node of interest based on the blink rates of the second edges. The relatedness metric indicates a degree of relatedness of the target node to the node of interest. A cognitive operation is then performed based on the relatedness metric.

Ranking related objects using blink model based relation strength determinations

A structure and method for crowdsourcing includes evaluating a metric related to a content to be translated, determining a priority for the content based on the metric related to the content, and queuing the content for crowdsourcing based on the priority determined from the metric.

Growing community of inventors

White Plains, NY, United States of America

Hui Wan